Poverty and Discrimination This Page Intentionally Left Blank Poverty and Discrimination

A PRINCETON UNIVERSITY PRESS E-BOOK Poverty andDiscrimination Kevin Lang Poverty and Discrimination This page intentionally left blank Poverty and Discrimination

KEVIN LANG

Published by Princeton University Press, 41 William Street, Princeton, New Jersey 08540 In the United Kingdom: Princeton University Press, 3 Market Place, Woodstock, Oxfordshire OX20 1SY

Library of Congress Cataloging-in-Publication Data Lang, Kevin. Poverty and discrimination / Kevin Lang. p. cm. Includes bibliographical references and index. ISBN-13: 978-0-691-11954-0 (cloth : alk. paper) ISBN-10: 0-691-11954-6 (cloth : alk. paper) 1. Poverty—United States. 2. Economic policy—United States. 3. Social policy— United States. 4. Public welfare—United States. 5. Discrimination—United States. I. Title. HC110.P6L36 2007 362.5′5610973—dc22 2006020888

British Library Cataloging-in-Publication Data is available

This book has been composed in Adobe Garamond, Scala Sans, ITC Stone Sans, and ITC Stone Serif by Princeton Editorial Associates, Inc., Scottsdale, Arizona

Printed on acid-free paper. ∞ pup.princeton.edu

Printed in the United States of America

10987654321 contents

Acknowledgments xi

Chapter 1 Introduction 1 1. The Content of This Book 2 2. Recent Developments in the Study of Poverty and Discrimination 4 3. The Object of This Book 8 4. Why Do Policy Analysts Disagree? The Limits of Statistical Arguments 10 5. Why Do Policy Analysts Disagree? The Role of Values 12 6. A Case Study: Retention in Grade 13 7. Concluding Remarks 17 8. Further Reading 18 9. Questions for Discussion 18 10. Appendix: A Quick Guide to Statistics 19

Part 1 POVERTY Chapter 2 Who Is Poor? 31 1. Evidence on the Importance of Relative Income 36 2. How the Government Measures Poverty 37 3. Valuing Nonmonetary Income 39 4. Over What Time Period Should We Measure Poverty? 40 5. Other Problems with the Official Measure 41 6. The National Academy of Sciences Report 42 7. Gathering the Data 42 8. Who Is Poor (By the Official Measure)? 43 9. Extreme Poverty 45 10. Homelessness 45 11. Hunger and Food Insecurity 48 12. Alternative Measures of Poverty 51 13. The Dynamics of Poverty 53 14. Why Does Poverty Last So Long for Some People? 56 15. Concluding Remarks 58 16. Further Reading 58 17. Questions for Discussion 59 18. Appendix: A Brief Note on Data 61

Chapter 3 The Evolution of Poverty Policy 63 1. Federal Poverty Programs, 1970–2000 63 2. Incentives under AFDC 66 3. The Earned Income Tax Credit 69 4. Cash or In-Kind Transfer: Which Is Better? 78 5. Concluding Remarks 81 6. Further Reading 81 7. Questions for Discussion 82

Chapter 4 Trends in Poverty 83 1. Trends Using the Official Measure 83 2. Trends in Poverty under Alternate Measures 86 3. Accounting for Trends 87 4. Concluding Remarks 102 5. Further Reading 103 6. Questions for Discussion 104 7. Appendix: Multivariate Analysis 104

Chapter 5 Labor Market Policies 108 1. Understanding Wage Inequality 108 2. Minimum Wage Laws 115 3. Living Wage Laws 120 4. Job Training Programs 121 5. Can Job Training Programs Reduce Poverty? 123 6. Evaluating the JTPA 125 7. Evaluating the Job Corps and Other Youth Programs 129 8. Training Programs and Tagging 133 9. Welfare to Work: Work First 134 10. Employer-Based Subsidies 136 11. Concluding Remarks 140 12. Further Reading 140 13. Questions for Discussion 140 14. Appendix: Adjusting for Program Nonparticipation 141

Chapter 6 Family Composition 143 1. Births to Single Mothers 144 2. Declining Marriage 146 3. Changing Social Attitudes 150

vi Contents 4. The Role of Welfare 156 5. Features of Welfare 158 6. Teenage Childbearing 161 7. Effects of Growing Up with a Single Parent 168 8. Intergenerational Transmission of Poverty 172 9. Policies Aimed at Infants and Toddlers 174 10. Preschool Programs 177 11. Programs for School-Age Children 182 12. Medicaid and SCHIP 190 13. Concluding Remarks 192 14. Further Reading 194 15. Questions for Discussion 196 Chapter 7 Concentrated Poverty 197 1. Life in High-Poverty Neighborhoods 198 2. Do Neighborhoods Matter? 198 3. The Gautreaux Program 201 4. Moving to Opportunity 202 5. Public Housing 203 6. Gangs 205 7. Community Development 206 8. Concluding Remarks 208 9. Further Reading 209 10. Questions for Discussion 210 Chapter 8 Education and Education Reform 211 1. Education and Earnings 212 2. Testing 213 3. Decentralization and School Quality 221 4. Using Tests to Increase School and District Accountability 236 5. Concluding Remarks 239 6. Further Reading 240 7. Questions for Discussion 241 Chapter 9 Welfare Reform 243 1. The Case for Reform 243 2. The Welfare Reform Act of 1996 245 3. Assessing the Effects of Welfare Reform 251 4. Effect on Welfare Receipt 252 5. Welfare Reform and Well-Being 254 6. Living Arrangements 258 7. Effects on Children and Adolescents 259 8. Concluding Thoughts 259 9. Further Reading 260 10. Questions for Discussion 261

Contents vii Part 2 DISCRIMINATION Chapter 10 Discrimination: Theory 265 1. What Is Discrimination? 265 2. Theories of Discrimination: Prejudice 269 3. Prejudice in Imperfect Labor Markets 272 4. Transaction Costs Models 273 5. Statistical Discrimination 274 6. Self-Confirming Expectations 277 7. Concluding Remarks 280 8. Further Reading 281 9. Questions for Discussion 282 Chapter 11 Race Discrimination in the Labor Market 283 1. Trends in Black-White Earnings Differentials 283 2. Explaining the Decline in the Wage Gap 287 3. Evidence on Current Discrimination 293 4. Testing for Discrimination: Legal Perspectives 307 5. Affirmative Action in Employment 311 6. Affirmative Action in Public Employment 313 7. Concluding Remarks 314 8. Further Reading 315 9. Questions for Discussion 316 Chapter 12 Race Discrimination and Education 317 1. The Black-White Test Score Gap 317 2. Discrimination in Education 325 3. Affirmative Action in Education 330 4. Concluding Remarks 332 5. Further Reading 333 6. Questions for Discussion 333 Chapter 13 Race Discrimination in Customer Markets and the Judicial System 334 1. Housing 335 2. Discrimination in Other Markets 345 3. Discrimination in the Justice System 349 4. Concluding Remarks 351 5. Further Reading 352 6. Questions for Discussion 352 Chapter 14 Sex Discrimination 354 1. Theory 354 2. Is There Discrimination against Women in the Labor Market? 360 3. Discrimination, Marriage, and Children 364 4. Sexual Orientation 366

viii Contents 5. Trends in the Female/Male Wage Ratio 368 6. Comparable Worth 373 7. Concluding Remarks 375 8. Further Reading 377 9. Questions for Discussion 378

Chapter 15 Conclusion: An Agenda to Decrease Poverty and Discrimination? 379 1. The Value and Limits of Research 379 2. The Value and Limits of a Strong Labor Market 381 3. Family and Early Childhood Programs 383 4. Education 385 5. Addressing the Needs of High-Poverty Neighborhoods 385 6. Race Discrimination and Inequality 386 7. Addressing Inequality 387 8. Health Care 388 9. Concluding Remarks 388

Author Index 391 Subject Index 395

Contents ix This page intentionally left blank acknowledgments

L  , I have devoted much of my career to writing academic articles aimed at an academic audience. Such articles typically include only cursory thanks to those who have helped with those particular projects. Therefore, my debt of acknowledgments for broader intellectual influences has grown. Those who know both me and my parents, Kurt Lang and Gladys Engel Lang, will recognize their influence. After all, I learned sociology at the dinner table and, as a young child, observed them both studying and participating in school desegregation efforts. One of my first opportunities to earn pocket money was by clipping newspaper articles for a study of race riots that they were working on in the 1960s. My parents also introduced me to the late Yvan Corbeil, who, while I was taking time off before college, hired me to work at his survey research firm, CROP Inc., and then kept a job open for me during college vacations and after graduation. At CROP, I worked closely with Michael Saykaly, who introduced me to survey design and analysis. With this background, I entered Oxford certain that I wanted to focus on sociology, and I was encouraged in this direction by my sociology tutor, Anthony Heath, who exposed me to rational choice sociology. But Nick (now Sir Nicholas) Stern and Sudhir Anand exposed me to the rigors of economics and convinced me that I could pursue my interest in social issues more effectively as an economist. There were so many fellow students and faculty who helped me and influenced me during my graduate studies at the University of Montreal and the Massachusetts Institute of Tech- nology that to mention a few would be to slight others, so I will just recognize them collectively. It is difficult to limit my thanks to more recent intellectual influences. I have had the good luck to work with excellent and supportive colleagues, first at the University of California–Irvine, and later at Boston University as well as at MIT during two one- year visits. I have coauthored articles with about thirty different people, all of whom are important in some way. Bill Dickens has most frequently been my coauthor over a period of more than two decades. His work is frequently cited in this book. The late

xi Bob Rosenthal taught me game theory while coauthoring two articles with me. I was the beneficiary of both his unflinching honesty and his patience. Bob and Michael Manove have changed the way I write theory papers. I have probably argued more about methods and methodology with Josh Angrist than with anyone else, but his enormous influence on my thinking about research is reflected in this book. Paula England and George Farkas invited me to a conference on sociology and economics, an event that sparked a friendship, a book series we edited collectively, and many stimulating exchanges. That this book combines theory and empirical work with an emphasis on natural experiments and yet contains a significant element of sociology reflects their influence. Although I have never published an article with any of them, there are three people without whom it is difficult to imagine I would have written this book. Andy Weiss convinced Boston University to hire me at the same time it hired him, then convinced me that we should create a course on poverty and discrimination that we would teach together. Our lectures frequently turned into impromptu debates or brainstorming sessions that found their way into this book. Later, when I restarted the course on my own, I cribbed much of the syllabus from Becky Blank, used her own wonderful book as my initial text, and interrupted her duties as dean when I needed help or advice. My former colleague Glenn Loury thinks deeply and widely about issues of race. Our con- versations exposed me to ideas and writings that I would otherwise have missed. I have been fortunate to have many excellent graduate students. While writing the book, I have benefited from the research assistance of Carlos Sepulveda-Rico. Carlos checked and corrected many of my calculations and then field tested the book for his summer course. Yu Ping Tsai took over many of the thankless tasks that accompany the final stages of production and did so efficiently and with good humor. Peter Dougherty of Princeton University Press had worked with me on an earlier book that Jonathan Leonard and I edited and remained in touch with me over the years. He strongly encouraged me on this project. I was hesitant to publish with the Press because I feared that Peter would be promoted and lose interest in the project. Despite becoming director of the Press, he remained involved and supportive. The staff at the Press and at Princeton Editorial Associates have made the production stage less painful than I feared. The constant in my personal and professional lives since graduate school has been my fellow student, colleague, coauthor, and wife, Shulamit Kahn. Shu is the person I rely on for both honest criticism and support. For over a quarter of a century, we have read, criticized, and edited each other’s papers even when we did not coauthor them. My thinking is clearer and my writing more sequential as a result of all her tutoring. Moreover, she knows better than most spouses the demands of academia and has tol- erated the vacations I skipped and the many weekends spent at the office so that I could write this book. She has shared the times when I was certain I would never finish it and assured me that I would. This book is intended to have two audiences: advanced undergraduates and masters- level students in the social sciences and public policy, and individuals involved in policy. In thinking about these two audiences, it has been helpful to think not just about

xii Acknowledgments the students in my classes, but about specific individuals. Current and past members of the Brookline School Committee (the elected school board) with whom I have been proud to serve will undoubtedly hear me addressing them. My daughters, Ariella and Jenya, currently twenty and sixteen years old, were often in my thoughts as I wrote this book. Would they understand a particular explanation or argument? What would they think of the book if they used it in one of their courses? There are countless other friends and past and present colleagues and students who deserve my thanks and recognition, and I apologize for not including them here even if many of their thoughts and ideas are found in the body of the book. I hope and trust that in this book they will find not only an honest and accurate portrayal of the state of research on poverty and discrimination but also some new ideas and ways of thinking about the issues, and that many will recognize their own influence in the development of the perspectives presented.

Acknowledgments xiii This page intentionally left blank Poverty and Discrimination This page intentionally left blank chapter 1

Introduction

T   T M  D R earned economics the nickname the dismal science. In An Essay on the Principle of Population (1798), Malthus argued that population growth inevitably outstrips the growth of food production, so eventually population becomes too great for food supply. In The Principles of Political Economy and Taxation (1817), Ricardo expanded this argument to the “iron law of wages.” At wages above the subsistence level, population grows, driving down wages. At wages below the subsistence level, the poor face starvation, population declines, and wages rise. Poverty policies that transfer money to the poor must be ineffective in the long run because wages must always end up at the subsistence level. Writing less than a century later, Alfred Marshall was much more optimistic. He believed it was possible to eliminate poverty, if not within a generation, at least within two generations.1 Marshall thought poverty policy consisted of increasing the demand for labor and reducing the supply of unskilled labor. He argued that economic growth increased the demand for labor. Increased education would lower the supply of unskilled labor, both directly, by moving workers from the unskilled to the skilled labor force, and indirectly, by reducing population growth. To further reduce the size of the unskilled labor force, society should encourage later marriages and childbearing among the “lower strains.” Finally, government should address the environment in which the poor lived. Marshall commented on the benefits of the suburbs, “where excellent systems of drainage, water supply and lighting, together with good schools and opportunities for open air play, give conditions at least as conducive to vigour as are to be found in the country,” and continued, “There is no better use for public and private money than in providing public parks and playgrounds in large cities, in contracting with railways to increase the number of workmen’s trains run by them, and in helping

1. Alfred Marshall, “Three Lectures on Progress and Prosperity,” 1883; reprinted in Journal of Law and Economics 12 (April 1969): 184–212.

1 those of the working classes who are willing to leave the large towns to do so, and to take their industries with them.”2 Marshall saw little role for redistribution of income to the poor. Giving the poor money would reduce their industriousness. Only the victims of misfortune were good candidates for such charity. In sharp contrast, Henry George, in Progress and Poverty (1879), maintained that because land was fixed in supply and necessary for production, only landowners would benefit from progress. According to George, the solution was to raise taxes on land (not on structures or other improvements) in proportion to their value. In this way, the benefits from the increased value of land could be shared with workers.

1. The Content of This Book From the vantage point of another century of experience, it is clear that Marshall’s op- timism was in part justified and in part exaggerated. Standards of living are much higher than they were in the late nineteenth century. On the other hand, poverty did not disappear in two generations. In large part, this is because our understanding of what it means to be poor has changed over time. Much of chapter 2 is devoted to exploring this issue. But the issue is not only that we have redefined who is poor. Using a constant definition of poverty, over the past three decades there has been little change in the proportion of Americans who are poor, despite dramatic increases in average incomes. Thus the Marshall-George debate remains relevant. Has the relation between economic growth and poverty broken down? Did some other trend hide the positive effects of economic growth? Or is the problem the way we measure poverty? Chapters 3 and 4 address these issues. Many of Marshall’s concerns remain relevant today. As chapter 4 shows, there is a strong relation between poverty and the state of the labor market for low-wage workers. Chapter 5 addresses the effectiveness of different policies designed to raise after-tax wages for low-wage workers. Some of these policies follow in the Marshallian tradition of increasing skills and demand for low-wage workers. Others have a somewhat more Georgist flavor (although certainly not based on taxing land), using the tax system to support low-wage workers or intervening directly in the wage-setting process. We also see renewed focus on the family and on increasing the age of mothers at first childbirth. Chapter 6 addresses issues such as out-of-wedlock births and teenage childbearing. It looks at policies designed to help children. As discussed earlier, Mar- shall also believed that poverty was closely related to place. He advocated getting the poor out of crowded cities. Today, many analysts believe that concentrated poverty is a particular problem. Chapter 7 addresses the issues associated with such poverty and programs designed to alleviate it. Since Marshall’s time, the availability of public schooling has increased dramatically. Not only public primary schooling but public secondary schooling is universally

2. Marshall, 199–200.

2 Chapter 1 available in the United States. Yet the effectiveness of public education, particularly in high-poverty areas, is hotly debated in this country. Chapter 8 discusses the debate over education reform. Finally, the concern that programs that support the poor will “sap their industriousness” has been a recurring theme over the centuries of poverty policy. The 1996 Personal Responsibility and Work Opportunity Reconciliation Act, more commonly known as welfare reform, was a response to concerns that welfare was hurting the very people it was designed to help, very much as Marshall believed that the Poor Laws passed in the late eighteenth century in England had hurt laborers by encouraging them to rely on support for the poor. It is still somewhat early to assess the impact of welfare reform, but chapter 9 discusses the background of reform and what we know about its effects. Although most of this book is about poverty, the last third is about discrimination, and of that last third, three of the four chapters focus on race discrimination while the last is concerned with sex discrimination. There are two reasons for combining discussions of poverty and discrimination in a single book. The first is that the methods researchers use to study the two topics are closely related. Discrimination research relies on a combination of theory, observational studies, and experimental methods similar to those used in poverty research (discussed later). Although discrimination research relies more heavily on theory and less heavily on actual or quasi-experiments than does poverty research, it is easy to make the transition from the latter to the former. For the most part, those of you reading this book do not need a new set of analytical tools. In- deed, when we discuss sex discrimination, we will return to some of the same theories that we will have discussed when analyzing the relation between the decline of marriage and poverty. The second reason for discussing poverty and discrimination in the same book is that, although the topics are distinct, they are also related. Most of the poor are not black, but the poverty rate is much higher among black Americans than among white Americans. To the extent that discrimination contributes to lower incomes among blacks, it contributes to poverty and helps to account for their higher poverty rate. But it is also likely that higher poverty rates among blacks contribute to discrimination. Chapter 10 discusses a variety of theories of discrimination. Most rest on perceived or actual differences (or both) between blacks and whites. If blacks tend to come from more disadvantaged backgrounds than do whites, this can affect the flow of information from potential employees to potential employers and can, through multiple mechanisms, reduce the employment prospects of blacks relative to whites, even whites from the same background. On the other hand, these differences could promote prejudice, but in many settings people are unable to act on their prejudice. Therefore, the existence of prejudice need not lead to worse outcomes for blacks than for otherwise equivalent whites. Chapter 11 examines the evidence for and against the existence of discrimination in the labor market as well as the role of policy in addressing labor market discrimination. Many researchers believe that differences between blacks and whites in labor market outcomes primarily reflect differences in the skills that people bring to the labor

Introduction 3 market. Chapter 12 explores the black-white test score gap and issues regarding differences in access to schooling and desegregation. Chapter 13 reviews the evidence on discrimination in other domains, including the justice system and customer markets. Just as poverty is more common among blacks than among whites, it is more common among women than among men and is particularly common among female- headed households. Thus differences in earnings capacity between men and women may have an important impact on the prevalence of poverty. Chapter 14 addresses the debate over the source of this differential and policies designed to reduce it. Some readers of earlier drafts of this book have commented that it would benefit from a lengthier discussion of inequality. After all, one important explanation for the lack of a decline in poverty over the past three decades is the dramatic increase in inequality. In the concluding chapter, I argue that reducing inequality must be an important component of any policy that reduces poverty. Nevertheless, I have made only a modest attempt to accommodate readers who have requested more on this subject. There are two reasons that I have not added a complete chapter or more on the study of inequality. The first is simply a matter of space and time. This book is already longer than either the publisher or I anticipated, and it may well contain more material than can be covered in a typical semester course (although I do cover most of the material in a semester). The second reason is that the study of inequality in many ways relies on a different set of tools and methods than does the study of poverty and discrimination. By its nature, the theory of inequality requires a more global approach. The empirical analysis of inequality, for the most part, uses a different statistical approach and, in particular, makes less use of actual and quasi-experiments, which are the focus of much of this book.

2. Recent Developments in the Study of Poverty and Discrimination The earlier discussion may have given the impression that little has changed since Mar- shall and George debated progress and poverty. In fact, the study of poverty in general and poverty policy in particular has changed dramatically over the past thirty years. A large part of the impetus for the change can be traced to the debate over the negative income tax. In 1962, the future Nobel laureate Milton Friedman proposed that the welfare system in the United States be replaced with a negative income tax.3 Under a negative income tax, all individuals or households with incomes below a certain level would receive a basic guaranteed annual income from the government. As household income increased, the government grant (or negative income tax) would be scaled back. When household income was sufficiently high, the household would not receive any grant from the government but instead would pay income tax as it would in a standard income tax system. Although Friedman was a well-known conservative, he was also a highly

3. Milton Friedman, Capitalism and Freedom (Chicago: Chicago University Press, 1962).

4 Chapter 1 respected economist, and his proposal was subject to considerable analysis in professional journals. Support for the negative income tax crossed political boundaries. The first proposal for a negative income tax came from the Johnson administration in 1965, and the 1967 reforms to the welfare system reflect some of the spirit of the negative income tax proposal. Nixon’s Family Assistance Plan proposed a form of negative income tax that was opposed by welfare rights activists on the left.4 James Tobin, also a Nobel laureate but, in contrast with Friedman, a recognized liberal, designed a negative income tax proposal for the McGovern campaign in 1972.5 The key aspects of the theoretical analysis can be summarized briefly. If the negative income tax were to be affordable, the rate at which the grant would be reduced as income increased would have to be substantial. At a minimum, the grant would decline by one dollar for every three dollars of income and more probably by one dollar for every two dollars of income. In today’s terms, even using the lower rate, a worker earning nine dollars an hour would see his family’s grant fall by three dollars for each hour that he worked. Thus, in effect, he would be earning only six dollars an hour. Be- cause the after-tax hourly wage of working families would be much lower but their overall income would be higher, these families might work less under a negative income tax than they would otherwise.6 On the other hand, under traditional welfare, benefits were typically reduced by one dollar for every one dollar a recipient earned. Thus people with very low potential earnings who were therefore unlikely to earn much more than they would receive from welfare had little or no incentive to work. Under a negative income tax, they would keep some of their earnings and thus have some incentive to work. The advocates of the negative income tax hoped that it would encourage very low-income families to work and not reduce labor supply very much among somewhat higher-income households. But of course it was possible that just the opposite would happen. Perhaps current welfare recipients would be little affected if their incentive to work were increased and there would be a big reduction in work effort among near-poor families who did not receive assistance through traditional welfare. This issue could not be resolved on the basis of theory alone. Economists began with observational studies, that is, they looked at the relation between after-tax wages and labor supply in the population. They relied on surveys of individuals who reported, among other variables, their hours of work and earnings or wage rate. We can think of an observational study in the following way. Suppose we find a sample of people who seem to be otherwise similar but some of whom earn six dollars

4. Walter Williams, “The Continuing Struggle for a Negative Income Tax: A Review Article,” Journal of Human Resources 10 (Fall 1975): 427–44. 5. Holcomb B. Noble, “James Tobin, Nobel Laureate in Economics and an Adviser to Kennedy, Dies at 84,” New York Times, March 13, 2002. 6. For a formal discussion of the effects of income and wage rates on labor supply, see the discussion of the earned income tax credit in chapter 3.

Introduction 5 an hour and some of whom earn nine dollars an hour. If the only difference between the two groups is that some people were lucky and got jobs paying nine dollars an hour and others were unlucky and got jobs paying six dollars hour, it may be reasonable to assume that if we cut the pay of the lucky people to six dollars an hour, they would act like the unlucky people. Suppose that similar people earning six dollars an hour worked two hours per week less than those earning nine dollars an hour. Then we might conclude that if we were to tax workers earning nine dollars an hour so that they ended up earning six dollars an hour, they would reduce their labor by two hours per week. The implicit assumption that the only difference between the groups is how lucky they were is very strong. It is likely that even though they look similar on paper, the people earning nine dollars an hour are somehow different from those earning six dollars an hour. For example, they may be more skilled or work harder even though they have similar educations. Or their jobs may be different. The higher-paying job may be more dangerous or more demanding. So the real challenge for the statistician, and one to which we will devote a great deal of time in this book, is figuring out ways to obtain samples of people who differ only along the dimension we are trying to study. A classic book by Glen Cain and Harold Watts brought together seven papers focused on predicting the effect of a negative income tax on labor supply.7 In their conclusion, Cain and Watts point out the large range in the estimated effects of very similar programs. One study found that a $3,000 guarantee coupled with a 50 percent tax rate would have a negligible effect on the labor supply of husbands. In contrast, a second study predicted that a less generous program with a $2,400 guarantee and a 50 percent tax rate would reduce the labor supply of male family members by 37 percent. The former study implied that the negative income tax would be a cost-effective approach to reducing poverty. The latter implied that it would be very costly. As reflected in this example, it is often very difficult to use observational data to obtain convincing evidence of the causal effect of one variable (such as the after-tax wage or parental absence) on a second variable (such as labor supply or adult outcomes). If we could conduct an experiment in which we randomly assigned some people to have high wages and some people to have low wages, we would have much more convincing evidence regarding the relation between labor supply and after-tax wages. Because the stakes involved in instituting a negative income tax were so high, policy analysts convinced the federal government to conduct experiments in which some people were randomly assigned to be eligible for the negative income tax while others were randomly assigned to remain subject to traditional welfare. In the experimental group, there was also variation in the generosity of the program. Four experiments were conducted in the United States and a fifth in Canada. The largest of these is known as SIME/DIME (the Seattle Income Maintenance Experiment / Denver Income Mainte- nance Experiment). Many policy analysts found the results of the experiments disappointing. Although the labor supply response was modest, it added substantially to the cost of the program.

7. Glen G. Cain and Harold W. Watts, Income Maintenance and Labor Supply (New York: Acad- emic Press, 1973).

6 Chapter 1 Depending on the generosity of the program evaluated in SIME/DIME, the labor supply response could account for over half of the costs.8 The least generous program would save $4 billion but would make 95 percent of recipients worse off. A program that would guarantee support at the poverty level and tax-back benefits at a rate of only 50 percent would still make one-fourth of recipients worse off and would exceed the cost of the welfare program then current by $30 billion, an enormous increase.9 There were also some “unintended consequences,” the social science equivalent of medical side effects. In particular, the divorce rate rose among recipients randomly assigned to the negative income tax.10 The combination of the costs of the labor supply effects and the effect on marriage led Senator Daniel Patrick Moynihan, an early sup- porter of the negative income tax, to withdraw his support. Despite the failure of the policy proposal, the negative income tax experiments established the value of large-scale experiments and helped to create the infrastructure to carry them out. Randomized experiments have become the “gold standard” by which empirical work on poverty and discrimination is judged. The welfare reform passed by Congress in 1996 was heavily influenced by the outcomes of a large number of experiments that evaluated potential reforms. Of course, it is not possible to use experiments to evaluate all policy proposals. Some experiments would be immoral if not impossible to perform. And a badly conducted experiment is still bad research even though that research uses an experiment. Other experiments are simply too expensive compared with the benefit they are expected to bring. In such cases, researchers can sometimes rely on what are called “natural experiments” or “quasi-experiments.” These experiments attempt to find situations that mimic randomized trials. In these situations, it is as if the researcher had randomly assigned participants to a treatment or control group. Participants are said to be as good as randomly assigned. We will discuss later many of the issues associated with particular quasi-experiments. In general, however, we will be concerned with whether participants are truly “as good as randomly assigned” and with whether the control group may be affected by the experiment. The answers to these questions are frequently unclear even in the case of true experiments but especially when we examine quasi-experiments. Despite the difficulties associated with particular experimental or quasi-experimental studies, such studies, when done well, are generally more convincing than observational studies. This does not mean that there is no longer a role for observational studies in

8. Michael C. Keeley, Philip K. Robins, Robert G. Spiegelman, and Richard W. West, “The Labour Supply Effects and Costs of Alternative Negative Income Tax Programmes,” Journal of Hu- man Resources 13 (Winter 1978): 3–36. 9. Robert G. Spiegelman and K. E. Yaeger, “The Seattle and Denver Income Maintenance Ex- periments: Overview,” Journal of Human Resources 15 (Fall 1980): 463–79. 10. Lyle P. Groeneveld, Nancy B. Tuma, and Michael T. Hannan, “The Effects of Negative In- come Tax Programs on Marital Dissolution,” Journal of Human Resources 15 (Fall 1980): 654–74; but see Glen G. Cain and Douglas A. Wissoker, “A Reanalysis of Marital Stability in the Seattle-Denver Income-Maintenance Experiment,” American Journal of Sociology 95 (March 1990): 1235–69, for a reanalysis challenging this conclusion.

Introduction 7 research on poverty and discrimination. Observational studies are often the basis for justifying more careful experimental study. Or, if it is impossible to design a convincing experiment or quasi-experiment, we may be forced to fall back on an observational study. Moreover, for several reasons, experiments are only imperfect substitutes for theory. First, theory is an important guide to which experiments to conduct and how to interpret them. Second, suppose we learn (as we do in chapter 11) that if they apply randomly for jobs, individuals with names that reveal that they are black are less likely to be called for an interview. Theory will tell us that the effect of this discriminatory behavior depends a great deal on how the labor market works. If applicants have a good idea of which firms discriminate and which do not, they will not apply randomly, and the experiment will not give a clear impression of the effect of this discriminatory behavior on applicants. We might follow up the initial experiment with a study of how black applicants decide where to apply for jobs and, in particular, see whether they avoid the firms for which we have evidence of discriminatory behavior. By now it may occur to you that the study of poverty and discrimination increasingly resembles the study of medicine in its reliance on experiments. It would be false to view medicine as atheoretical. Theory guides the choice of medicines and medical procedures to study. Nor is empirical analysis in medicine purely experimental. We learn a great deal from observational studies. Often observational studies justify experimental studies. But ultimately, experimental studies are the strongest guide to good practice. New medicines are subject to clinical trials in which outcomes from those receiving the medicine are compared with outcomes from those receiving either no treatment or the standard treatment. And researchers watch for unintended consequences in the form of side effects. Research on poverty and discrimination certainly does not rival experimental medical research, let alone research in microbiology. This reflects in part the difficulty of working with human beings who can see through experiments. A participant in a negative income tax experiment who knows that the experiment will last only three years need not respond to the experiment in the same way that she would if a negative income tax were established permanently. In contrast, we expect that the body’s response to medication is not affected by the participant’s understanding that the experiment is of limited duration (although willingness to take the medication consistently may be affected). Research in both medicine and the social sciences suffers from the fact that the environment changes in response to our policies. Getting a flu shot reduces not only the chance of getting the flu but also the probability of spreading it to others. Thus the effect of a policy making flu shots freely available would be different from the effect that would be predicted by a small-scale experiment. Similarly, with a universal negative income tax, the decline in labor force participation would affect the availability of jobs and the wages they paid. Again, this means that we cannot dispense with theory.

3. The Object of This Book What do we know about poverty and discrimination? And how do we know it? The body of literature on this topic is enormous. Unfortunately, much of what has been

8 Chapter 1 written is designed to further a political agenda, and much of the rest is just not very good. The goal of this book is to help you distinguish the good research from the rest. I discuss in detail a small number of the best studies on each topic. This approach will allow you to understand not only the principal findings of the study but also the weaknesses that limit our confidence in its results. As the introduction to this chapter should make clear, even the best studies are imperfect. I try to avoid summarizing a large number of studies, many of which suffer from significant shortcomings. When choosing among good studies, I have tried to select the one that is most accessible. Inevitably, on occasion I am forced to say “most researchers believe that” or “most studies show that” or “the evidence is mixed” because it would be too time-consuming to discuss the individual studies. In general, however, I try to resist the temptation to be the ultimate judge of a body of literature rather than giving you the tools and information you need to evaluate it. You may come away from the book feeling less sure about what you know than you did before you read it. That is good, not bad. Although many people offer simple solutions to the problems of poverty and discrimination, these problems are genuinely complex. If there were simple solutions, we probably would have done away with both of these problems. After all, most people think that poverty and discrimination are bad and would like to get rid of them if they could. We will discuss various policies that have been tried in the attempt to end poverty and/or discrimination. We will discuss whether they worked. In some cases, we will consider arguments that the very policies intended to reduce poverty and discrimination have increased them. This should give you an idea of just how hard it will be to solve these problems. There are probably no “right” answers to the questions that we will raise. But there are better and worse answers and even wrong answers. Some people will find this conclusion depressing. A number of students have made comments to me like “How can you stand working in this area? The answer is always ‘We don’t know.’” But it is precisely because there is so much that we do not know and understand about such an important topic that it is fun to study. And of course, except in some deep metaphysical sense, it is not true that we do not know anything. We know a great deal, and we have learned much of it in the past ten years. Still, it is also important to recognize what we do not know and that the evidence for what we think we know is often weak. This book concentrates on statistical analyses and formal theories. Certainly our understanding of poverty and discrimination is informed by careful and thoughtful descriptions of the poverty experience. Reading ethnographic studies11 gives enormous insight into the lives of poor people, and this book will draw on these insights. But we will not talk a lot about what it is like to be poor and how it feels to be the victim of discrimination. Instead, we will spend a little time doing economic theory and a lot of time looking at arguments based on statistical analysis.

11. Ethnographic studies provide rich descriptions of the functioning of human societies. Several excellent ethnographies are included in the section on additional readings at the end of this chapter.

Introduction 9 In a sense, this book is about statistical analysis. By this I do not mean that it covers statistical techniques, although the appendix to this chapter covers the basics you need to read this book. I mean that it is about how to assess the quality of statistical arguments. Statistical arguments are generally imperfect. There are usually other explanations for a result than the one presented by a researcher. Sometimes those other explanations are equally or even more plausible than the one presented. Understanding how to assess statistical arguments is a skill that is valuable well beyond the study of poverty and discrimination. Businessmen, doctors, policy analysts, and many others rely on statistical analysis. Understanding the quality of this analysis may be important for making the right business decision or choosing the right medical approach. Understanding the limits of statistical analysis can also help you understand why policy analysts disagree about policies. I refer to “policy analysts.” I could say “social scientists” or “economists,” but the poverty field is highly policy oriented, and this book follows in that tradition.

4. Why Do Policy Analysts Disagree? The Limits of Statistical Arguments Ultimately statistical analysis is about correlation, the degree to which characteristics tend to vary together: more educated people tend to have higher incomes than do less educated people; teenage mothers tend to have lower incomes than women who did not have children as teenagers; users of the leading asthma medicine are more likely to have bad asthma than people who do not use that medicine. It is the policy analyst who interprets these correlations as showing a causal relation: getting more education increases income; having a child as a teenager lowers a woman’s income; using the leading asthma medicine worsens rather than improves asthma. But the opposite interpretations are also possible: people who are going to earn more money do not feel as much pressure to start working soon and thus get more education; women who expect to have low earnings are more likely to become teenage mothers; people who have bad asthma are more likely to use the leading asthma medicine. Or some other factor may cause both characteristics: smart people tend to choose to get more education, and smart people tend to have higher earnings; women from disadvantaged backgrounds are more likely to become single mothers and are more likely to have low incomes; doctors who treat people who live in the Bronx like to prescribe the leading asthma medicine, and people who live in the Bronx are more likely than other Americans to have asthma. In each case, it may seem obvious to you which of these explanations is correct. I certainly believe that the most likely reason that heavy users of the leading asthma medicine are more likely to have asthma is that asthmatics are the people most likely to use the medicine. But I cannot prove that my belief is correct simply by pointing to the relation between medicine use and asthma. If another policy analyst is convinced that use of the medicine and asthma are related because the medicine causes (or worsens) asthma, we will have to look for new evidence to help us distinguish between the two explanations.

10 Chapter 1 Perhaps we can find an experiment in which a researcher randomly gave some people the asthma medicine and others a placebo. If we find that asthma was no more common among those receiving the medicine than among those who received the placebo, we may feel justified in concluding that the medicine does not cause asthma and that my original belief was correct: people who use the leading asthma medicine are more likely to have asthma than are people who do not because people with asthma are more likely to use the medicine. But the experiment is not definitive. Perhaps the reason that asthma is no more common among those who get the medicine than among those who get the placebo is because the placebo also causes asthma. Policy analysts disagree in part because they have different beliefs about how likely these alternative explanations are. All policy analysts work with implicit models of how the world works that reflect both their life experience and their academic training. So- cial workers and economists do not necessarily view the world in the same way. Faced with the finding that teenage mothers have lower incomes than women who did not give birth as teens, one researcher may see evidence that teenage motherhood causes poverty while another sees evidence that poverty causes teenage motherhood. An economist is likely to respond that if having a child as a teenager is extremely costly, only those with strong reasons to have children as teenagers will do so. They may therefore be inclined to believe an explanation indicating that the poor economic prospects of some women lead them to have children at an early age. Child psychologists and social workers are less likely to believe that teenagers make rational decisions, which may make them more likely to believe that having a child at an early age has major negative consequences for the mother. Of course, both or neither may be true. It would be easy for you to translate the last paragraph as “It’s all a matter of opinion.” I hope that you will not. There are weaker and stronger statistical arguments. The best way to find out which, if either, of these arguments is correct would be to conduct two experiments. In one experiment, we would randomly assign girls to poor and wealthy families and see if there was a difference in the proportions becoming teenage mothers. In the other experiment, we would force some teenage girls to become mothers and ensure that others did not. If the teenage mothers ended up with lower incomes than the other women, we would be reasonably confident that teenage motherhood lowered their incomes. If there were no difference in the earnings of the two groups, we would conclude that teenage motherhood did not cause lower incomes. But even here there would be a problem. Perhaps some girls receive lower incomes as a result of becoming teenage mothers and others receive higher incomes. Those who will receive higher incomes become teenage mothers while those who would be hurt financially do not become teenage mothers. So our experiment would give the wrong answer, because it asks what is the effect of teenage motherhood on the average teenage girl rather than on the type of teenager who becomes a teenage mother. Of course, both of these experiments would be totally immoral and would not be conducted by any ethical person. However, we will see that clever researchers often try to imitate experiments by comparing groups that are similar except for the factor they are examining. Perhaps we could compare twin sisters, one of whom gave birth while a teenager and the other of whom did not. A study discussed later in this book

Introduction 11 compares women who were teenage mothers with women who had miscarriages as teenagers.

5. Why Do Policy Analysts Disagree? The Role of Values To some extent, policy analysts disagree about policies because they disagree about their effects. We have seen that it is essentially impossible to prove the case for or against a policy based on data alone. However, it is frequently the case that policy analysts do agree about the effects. Later in the book, we will examine the effects of minimum wage laws. There is considerable consensus regarding these effects. Most (although certainly not all) economists agree that minimum wage laws (at the levels found in the United States) reduce employment but that the effect is small. They also agree that minimum wage laws reduce wage inequality but do not have a large effect on income inequality or on the poverty rate. Given this consensus, why do economists disagree about whether the minimum wage should be raised? One explanation is values.12 Minimum wage laws are inefficient— they reduce employment—and most of us agree that is bad. But minimum wage laws reduce wage inequality, may reduce family income inequality, and make those people working in low-wage jobs more capable of supporting themselves. Many people, perhaps most, think that those are good things. We must now decide what weight to put on the good and bad effects. Reasonable people can arrive at different judgments. Most policy analysts have concluded that the negative income tax is too expensive because it reduces work effort too much. However, others disagree. The philosopher and economist Philippe van Parijs argues that the reduction in work effort is a benefit, not a cost, of the negative income tax. In his view, society’s objective should be to have people work only for the innate pleasure they derive from working and not for the income they receive. The fact that people work less when they can receive a subsidy from the government makes the financial cost of the negative income tax higher but also lessens the rat race.13 Similarly, when welfare reform was passed during the Clinton administration, most people believed that it would encourage families to leave welfare and find jobs (which most people felt was good). Most people also believed that it would leave those people who did not find jobs worse off. Part of the disagreement over the reform had to do

12. Victor R. Fuchs, Alan B. Krueger, and James M. Poterba, in “Economists’ Views about Para- meters, Values, and Policies: Survey Results in Labor and Public Economics,” Journal of Economic Lit- erature 36 (September 1998): 1387–425, examine the relation among values, beliefs about the effects of policies, and support for policies. They find that among economists specializing in labor economics and public economics, the two areas they study, policy disagreements are influenced much more by differences in values than by differences in beliefs about the effects of policies. 13. See, for example, Robert Van der Veen and Philippe van Parijs, “A Capitalist Road to Com- munism,” Theory and Society 15 (September 1986): 635–55.

12 Chapter 1 with how large each of those effects would be. Part reflected different values—how much some valued increasing the number of children growing up in families that were independent of welfare and how much others valued making it less likely that children would go hungry. Put differently, almost all policy analysts would agree that if there is a way to make some people better off without making anyone worse off, we should do it. This is known as the Pareto principle. However, most policies do not fit nicely into this category. Policies that do not help anyone will generally not receive much support, while those that help at least some people and do not hurt anyone will be quickly enacted. What is left is a set of policies that help some people and hurt others. The policy analyst can help determine who is helped and who is hurt, and by how much and in what ways. Ultimately, however, the policy advocate (who may be the same person) takes over and makes the case that the benefits to some outweigh the costs to others or vice versa.

6. A Case Study: Retention in Grade Writing in 1989, one analyst concluded, “There is probably no widespread educational practice as thoroughly discredited as retention in grade. If the research undercutting this practice is sound, the task is to uproot outdated misconceptions appealing to educators’ ‘common sense’ wisdom.”14 How did researchers “know” retention was bad, and why would policy makers not listen?

6.1. The Early Research. What we knew was that students who were retained in grade did not catch up with their peers. If anything, they fell further behind. They had lower self-esteem than other students and were more likely to drop out. If retaining students in grade caused these differences, for most of us, this would make a compelling case that retention is a bad policy. However, there are good reasons for questioning whether the relation is causal. Students who struggle in school are more likely to be retained in grade. Students who have difficulty with school one year are more likely than other students to have difficulty in other years. We would not be surprised to find that these students also are more likely to drop out and to have low self-esteem even if they are not retained in grade. To use an analogy, compared with those who receive high grades, students who receive low grades in high school do worse, on average, on the SAT (Scholastic Aptitude Test) or the ACT (American College Test). Few people would suggest that giving everyone As would improve SAT scores. Serious researchers understood this problem. They tried to find students who were not retained in grade but who looked on paper a lot like the students who were retained in grade. They could match students on factors such as their race, sex, month

14. Roy P. Doyle, “The Resistance of Conventional Wisdom to Research Evidence: The Case of Retention in Grade,” Phi Delta Kappan 71 (November 1989): 215–20.

Introduction 13 of birth, region, parents’ marital status, and education. They still found that being retained in grade was associated with bad outcomes. How convincing is this? Suppose we found a sample of same-sex twins in which one twin was retained in grade and the other was not. We would still be worried that the twin who was retained in grade was more academically challenged than the one who was not and would therefore have done worse anyway. Back to our analogy. If the twin who got lower grades in school did worse on the SAT, would we blame the low grades for the lower performance on the SAT?

6.2. Recent Research. We can do better if we find a setting in which there is a relatively sharp cutoff establishing who is promoted and who is retained. Suppose a large school system sets a rule that students who score 65 or higher on a citywide test will be promoted, while all those who score 64 or less will be retained in grade. We will assume that the test is administered in a standard manner across schools and is fairly graded. All tests have a random component. Was the student lucky or unlucky in his choice of topics to study? Did she accidentally mark the wrong box on the answer sheet? Did he make a lot of lucky guesses? For this reason, students who score 65 on the test should be a lot like students who score 64 on the test. We can compare the future performance of students who scored 64 on the test and were therefore retained with the future performance of those who scored 65 and were therefore promoted. If retention helps students, we would expect the future scores of those with a test score of 64 to be higher. If it hurts, we would expect the opposite. If there is no difference, retention neither helps not hurts, but given its cost, it is probably not a good idea. Of course, in the real world, the line between being promoted and being retained in grade is likely to be a little fuzzy. The district may have a waiver policy that allows some students who fail the test to be promoted anyway, and it probably has other requirements that can cause a student who passes the test to be retained. But the basic idea remains the same. If, for example, 95 percent of students who pass the test are promoted and only 20 percent of those who fail it are promoted, if retention is good, we should see better future outcomes for students with 64s than for students with 65s.15 Jenny Nagaoka and Melissa Roderick used essentially this approach. In Chicago, third graders who, after a summer remedial program, were more than one year below grade level on the Iowa Test of Basic Skills reading test were retained in grade. For sixth graders the cutoff was one and a half years below grade level. Nagaoka and Roderick compared students up to three-tenths of a year below this cutoff with students at this cutoff or an equivalent amount above. For both grades, as the rules imply, those just above the cutoff were much more likely to be promoted than those just below it.16

15. There is a way to adjust these differences to obtain an estimate of the effect of retention. We discuss this later in the book. 16. Jenny Nagaoka and Melissa Roderick, “Ending Social Promotion: The Effects of Retention” (Consortium of Chicago School Research, Chicago, 2004).

14 Chapter 1 Nagaoka and Roderick found that third graders who scored just below the cutoff showed somewhat more improvement in performance the following year but that most of this difference had disappeared by the end of two years. For sixth graders, those scoring just above the cutoff had larger gains one year later, and this difference was similar two years later. Is this the end of the discussion? Should we now be convinced that retention has only transitory benefits for third graders and longer-lasting negative effects for sixth graders? We will talk about more global reasons for not drawing this conclusion later in this section, but for now there are two issues that must be addressed. The first is specific to the study and helps to underline the importance of examining studies in detail rather than merely summarizing results. The second addresses a more general issue with the approach. The first point is that Chicago had high-stakes testing in the third, sixth, and eighth grades. Students who were retained in grade therefore faced a high-stakes test one year after being retained. We might expect them to take the exam more seriously than someone taking a low-stakes exam and therefore to perform better. It is plausible that the one-year difference overstates the gain to retention. Conversely, sixth graders who were promoted took another high-stakes test two years later, while most of the sixth graders who were not promoted were taking a low-stakes test. This is likely to mean that the harm from retention two years later was exaggerated for sixth graders. The second problem is that even though students above and below the cutoff were similar, they were not identical. There are good reasons both for believing students who did particularly badly on a test would show more improvement than other students and for believing that they would fall further behind. How can we get around this problem? One way is to look at what happens when the cutoff is not important. Although they did not focus on it, Nagaoka and Roderick showed that in 2000, when the cutoff used in the study was not in effect, those below the study cutoff showed very slightly less improvement over one year than those slightly above the study cutoff. Unfortu- nately, data for two years later were not available. The second approach is to look at differences with those slightly more above and slightly more below the cutoffs. To return to our earlier example, we could look at the difference between those scoring 63 and those scoring 64 on the test and between those scoring 65 and those scoring 66 on the test. If the difference between those scoring 64 and those scoring 65 on the test was similar to these other two differences, we would conclude that retention had no effect. If the improvement in score was greater going from 64 to 65 than from 63 to 64 or from 65 to 66, we would conclude that retention hurt the students, and if it was smaller, we would conclude that it helped them. Brian Jacob and Lars Lefgren used essentially this approach to look at the Chicago data.17 They examined both reading and mathematics scores because students had to

17. Brian A. Jacob and Lars Lefgren, “Remedial Education and Student Achievement: A Regression-Discontinuity Analysis,” Review of Economics and Statistics 86 (February 2004): 226–44.

Introduction 15 pass a threshold in each. Their conclusions are similar but not identical to those reached by Nagaoka and Roderick. In the third grade, they found that retention had a large positive one-year effect on reading that was mostly gone by the end of two years. In math, the positive one-year effect was diminished but not completely gone after two years. In the sixth grade they found a small negative one-year effect in reading that increased in the second year. For math, there was no clear negative effect in either year.

6.3. What Should We Conclude? The results of the two Chicago studies are reasonably consistent. For those close to the cutoff, retention in grade has some positive but probably transitory effects on third graders and some negative and possibly transitory effects on sixth graders. Although we might make a case that we should continue to study the effects on third graders, surely at least for older students, these new studies support the quotation at the beginning of this section that describes retention as “thoroughly discredited.” Sorry, but we should not be so fast to agree with that conclusion. There are a large number of questions we can ask that might stop us from drawing that conclusion even about retention of sixth graders. Both studies looked at the effect of retention on performance at the same age. What would happen if we looked at performance in the same grade? Did sixth graders who were retained do better or worse on the eighth-grade test than similar students who were promoted? We do not know. Because the mastery of math for the two groups was similar when the retained students were in seventh grade and the promoted students were in eighth grade, it is a good bet that the retained students did better on the eighth-grade test, but we cannot be sure until we check. And it is very uncertain whether they did better or worse on the eighth-grade reading test. And, to make matters more complex, suppose that we do the study and we conclude that, in the long run, students retained in sixth grade do better than their promoted peers at each grade level but worse than their promoted peers at each age level. What policy should we favor? Our evaluation becomes very complex. There is good evidence that being old for their grade makes students more likely to drop out.18 So some students will get less education because they were retained in grade, and, therefore, based on the assumption at the beginning of the paragraph, leave school with fewer skills. Other students will not reduce their education, and under these assumptions, will leave school with more skills. But to acquire these additional skills, they will have spent an extra year in school at great cost to both themselves and to the public. Even if retention does increase the skills of these students, it may be a very cost-ineffective approach. There are at least three additional reasons that an advocate of retention might ad- here to that position in the light of this research. The first is that, by their nature, the

18. Joshua D. Angrist and Alan B. Krueger, “The Effect of Age at School Entry on Educational Attainment: An Application of Instrumental Variables with Moments from Two Samples,” Journal of the American Statistical Association 87 (June 1992): 328–36, and Susan E. Mayer and David Knut- son, “Does the Timing of School Affect How Much Children Learn?” in Susan E. Mayer and Paul E. Peterson, eds., Earning and Learning: How Schools Matter (Washington, DC: Brookings Institution Press, 1999), 79–102.

16 Chapter 1 Nagaoka-Roderick and Jacob-Lefgren studies tell us only about the effect of retention on students with scores near the cutoff chosen by the Chicago Public Schools (CPS). Per- haps retention is helpful to some students but the CPS set the promotion bar too high. The second is that we have looked only at the effect on students who were retained, but there are important incentive effects from high-stakes testing.19 If students work harder in order to pass the exam, perhaps the benefit from the extra effort outweighs any harm from a retention policy. We do not know that the incentive effects are important (or even positive), but if they are, they may make the effects positive for some students and negative for others. Even if policy analysts agreed on all the effects, they might reach different conclusions. Finally, someone could conclude that the Chicago experience suggests that retention is not generally bad but rather that it is a bad policy as practiced by the CPS. They could maintain that a more sophisticated retention policy with special programs for retained students or one that relied on multiple and better indicators of academic mastery of the material would produce positive results.

7. Concluding Remarks At this point you may be feeling frustration. You want to know whether retention is good or bad. Are the critics of “social promotion” correct, or are they just promoting a policy that sounds good, putting politics ahead of the sound judgment of most education professionals? Throughout this book, I will try to play the role of objective arbiter. I regret to tell you that we will often end up where we are with the retention versus social promotion debate. There is some research that points in a particular direction, but there is a good deal that we do not know. For the most part, in this book I have resisted the temptation to discuss research that I consider bad and to elaborate on its obvious weaknesses. I certainly point out the weaknesses of many of the studies that I do discuss, but I view that differently. Many of the weaknesses are simply unavoidable, and it is often better to rely on a weak lamp than on none at all. Still, there is ample ground for reasonable people to disagree, both because the available research leaves many unanswered questions and because some policy positions depend on value judgments. I am concerned that, having learned that it will not teach you how to solve poverty and discrimination, you will stop reading this book and drop the course in which it is assigned. But if you resist that temptation, I truly believe that by the end of this book you will be better at evaluating social policy options, and because you care enough about poverty and discrimination to be reading this book, that is important. Therefore, in the last chapter, I will drop my cloak of academic distance and outline my conclusions, based on the material in this book, about what policies we should pursue. But that chapter is the least important chapter in the book. For now, because the subject is only marginally related to poverty and discrimination policy, let me appease you by addressing what I think we should conclude about

19. We discuss this issue in more detail in the chapter on education reform.

Introduction 17 retention in grade and social promotion. My strongest conclusion is that it is time to design a randomized study of retention and social promotion. Under the impetus of the No Child Left Behind Act and the standards movement, the United States is currently moving rapidly in the direction of widespread use of high-stakes testing and ending social promotion. Enormous resources are being directed at such testing. To me, it is unthinkable that we would not be devoting substantial resources to determining the effectiveness of the policy. This may seem like a cheap conclusion for a researcher to reach, but it is not. There are strong moral restrictions on the types of experiments to which we should subject children. If I believed that retention was “thoroughly discredited,” advocating experimental research would be immoral, as it would be if I believed the opposite. But I do believe that it is possible that we have been too lax historically (although the public probably underestimates the extent to which retention has become more common over the past twenty years).20 I also believe that even if retention in grade proves less harmful than its critics maintain, it is unlikely to be a cost-effective policy for providing remediation. If we are moving in the direction of using retention more frequently, we should have better evidence of its effectiveness.

8. Further Reading Anderson, Elijah. Code of the Street. New York: Norton, 1999. Eaton, Susan E. The Other Boston Busing Story. New Haven, CT: Yale University Press, 2001. Fuchs, Victor R., Alan B. Krueger, and James M. Poterba. “Economists’ Views about Parameters, Values, and Policies: Survey Results in Labor and Public Economics.” Journal of Economic Literature 36 (September 1998): 1387–425. Kotlowitz, Alex. There Are No Children Here. New York: Doubleday, 1991. Liebow, Elliot. Tally’s Corner. Boston: Little, Brown and Company, 1967. Lukas, J. Anthony. Common Ground. New York: Knopf, 1985.

9. Questions for Discussion 1. What is the difference between correlation and causality? If two events tend to occur together or sequentially, must one of them cause the other? 2. Explain what is meant by the Pareto principle. What are the limitations of this principle as a guide to policy? 3. A friend tells you that you should never be a patient in a teaching hospital because the death rate among patients in teaching hospitals is higher than in other hospitals. How do you respond? 4. Suppose that, relative to policy analysts who oppose more funding for job training programs, policy analysts who support more funding for job

20. See Robert M. Hauser, “Should We End Social Promotion? Truth and Consequences” (Work- ing Paper 99-06, Center for Demography and Ecology, University of Wisconsin–Madison, 1999).

18 Chapter 1 training programs, on average, believe that job training programs generate a larger increase in employment. Does this mean that policy analysts’ beliefs are biased by their values? 5. Suppose that policy analysts who support and those who oppose more funding for Head Start, on average, have similar beliefs about the effectiveness of the program. Does this mean that policy analysts’ views about policy are not influenced by their beliefs about the scientific evidence?

10. Appendix: A Quick Guide to Statistics This appendix covers what you need to know to understand the statistics used in this book. It focuses on how to interpret the statistics rather than how they are calculated. It is not intended as a substitute for a standard statistics course that teaches the theory underlying the statistics.

10.1. Randomness. Before we discuss statistics, we need to think about what we mean when we say something is random. Suppose somebody shuffles a standard fifty- two-card deck. What is the probability that the top card is the ace of spades? In some sense, the probability is either one or zero. Either the top card is the ace of spades, in which case the probability is one, or the top card is not the ace of spades, in which case the probability is zero. But assuming that the person shuffled the deck fairly and that we have not looked at the top card or otherwise “cheated,” from our perspective, the probability is 1/52. With more information, the outcome might not be random, but given our information it is. The same will be true of many of the phenomena we study throughout this book. Test scores may or may not be random in some deep sense, but there are certainly many factors that affect test scores and that we do not measure. Therefore, from our perspective, test scores are random. Suppose we find one hundred pairs of students. We choose the pairs so that they look as similar as is feasible. To be part of the same pair, the students must be of the same sex, age, and race; have the same family structure; go to the same school; and have the same sixth-grade Iowa Test of Basic Skills math and reading scores. We will assume that under the rules of the school system, all are supposed to be retained in grade. As part of an experiment, one member of each pair is given a waiver from the school system rules and is promoted, while the other is retained in grade. Unless promotion or retention is much better for all students (which seems unlikely), some of the “experimental” students who are promoted will do better and some will do worse than the “control” students who are retained in grade. Maybe one student in the pair was generally a good student but had been sick the night before the test that determined promotion or retention and did poorly. The other member of the pair was even weaker than the test scores suggest but had made a lot of lucky guesses. The first member of the pair will probably do better next time whether he is the one who is promoted or the one who is retained. From our perspective, the number of promoted students who do better than their matched retained counterparts is a random variable. We could also look at the difference

Introduction 19 between the test scores of the student who was promoted and the student who was retained. This, too, would be a random variable. And because each difference is a random variable, the average difference across the one hundred pairs is also a random variable.

10.2. The Mean and Standard Deviation. Suppose we give a test to a large number of students. We could list the entire distribution. That is, we could say that twenty got a 0, twenty-eight got a 1, thirty-five got a 2, and so on, up to the top score. If there are a lot of possible scores, this would be tedious and hard to interpret. One thing every student knows to ask is “What was the average on the test?” Statisticians call the average the mean. But knowing the mean on the test is not enough. If the mean was 60, is a 50 an okay grade or a terrible grade? If the grades are all spread out from 0 to 100, a 50 is not too bad. If everyone got very close to 60, 50 is near the bottom. We would like to have a way of summarizing how dispersed the grades are without listing all the grades. There are many ways to do this. One way statisticians measure dispersion is the standard deviation. The formula for the standard deviation is given in the next paragraph, but it is more important to understand how the standard deviation relates to the dispersion of the variable. We will discuss this relation after giving a formal definition of the standard deviation and describing the normal distribution. To obtain the standard deviation, we first take all the observations of the random variable and calculate their mean. We then take the value of each observation of the random variable and subtract the mean. We then take this difference and multiply it by itself. This gives us the squared deviation of the measurement from the mean measurement. Next we add up all the squared deviations and divide by the number of measurements to get the mean squared deviation from the mean.21 This is called the variance. The square root of the variance is the standard deviation.

10.3. The Normal Distribution. Many random variables have what is called a normal distribution. The normal distribution is sometimes referred to as a bell curve. The normal distribution has a very useful feature: it can be fully described by just two values, its mean and its standard deviation. We can use the mean and standard deviation to describe how likely it is that, if we pick randomly from a normal distribution, we will obtain a particular value of set of values. In particular, 95 percent of the time, a random variable drawn from a normal distribution will lie within 1.96 (or approximately two) standard deviations of the mean. Furthermore, 2.5 percent of the time it will be more than 1.96 standard deviations above the mean, and 2.5 percent of the time it will be more than 1.96 standard deviations below the mean. Similarly, 90 percent of the time it will be within 1.64 standard deviations of the mean, with the re-

21. We can also divide by the number of observations minus one. The measures have slightly different properties but for most practical purposes are indistinguishable.

20 Chapter 1 maining 10 percent split equally between values more than 1.64 standard deviations above and below the mean.

10.4. Two Key Theorems. One important theorem in statistics says that (in most cases) if we take a lot of observations of a random variable and then take their mean, the estimated mean will be close to the true mean. So if we flip a coin a lot of times, the fraction of times that it comes up heads will be very close to the true probability of its coming up heads. A second important theorem in statistics says that if we take the mean of a large number of independent random variables, the mean will be approximately normally distributed. Many random variables are the result of the offsetting effects of a large number of very small factors. For example, how tall someone is depends on both genetics and a large number of environmental factors. It is not surprising that we frequently observe the normal distribution in nature. These two theorems are very helpful. They tell us, for example, that if we have enough pairs, the average difference in the test scores in our sample will be close to the true average difference and that the average will be approximately normally distributed. Acting as if a distribution is normal often yields quite accurate results. It turns out that if we flip a fair coin one hundred times, the number of heads will have a standard deviation of five. Equivalently, the coin will come up heads an average of half (.5) of the time, with a standard deviation of .05. Using the normal distribution, we would expect the number of heads to be more than forty and less than sixty 95 percent of the time. We can show that the true probability is 94.3 percent, so the approximation is pretty accurate. Of course, not all distributions are normal. Wages are not normally distributed, but if we take the logarithm of the wage, it is approximately normally distributed. Other distributions cannot be made normal even by redefining the variable. Despite this caveat, in most cases we will be quite accurate if we treat an estimate based on a large number of observations as normally distributed.

10.5. The Standard Error of an Estimate. Because an estimate is likely to be a normally distributed random variable, if we know its mean and standard deviation, we will know a great deal about it. In some cases, we can figure out the mean and standard deviation by relying on statistical theory. If promotion and retention are equally good, half the time the person who is promoted will do better and half the time the person retained will do better. If this is true, the mean number of our one hundred pairs in which the person promoted does better should be fifty. And, as mentioned earlier for the coin toss, the standard deviation will be five. In other cases, we will have to estimate the standard deviation based on information about the observations in our data. Statisticians have developed methods for estimating what the standard deviation of the recorded mean will be without actually estimating lots of different means. We will not discuss exactly how they do this, but we can look at the general principle. If the test scores in our pairs differ by almost the same amount (e.g., in every pair the retained student does two points better than the

Introduction 21 It is important to remember that whether or not they are reported along with the estimate, all estimates based on samples have associated standard errors. In this book and in newspapers, you will often find statements of this form: “According to the Bureau of the Census, the poverty rate in the United States in 2001 was 11.7 percent.” In fact, the Census Bureau provides information to help readers calculate standard errors of its estimates.22 Based on this information, we can calculate that the standard error of this estimate is about .14. We believe that 95 percent of the time if the Census Bureau had used a different (but similarly drawn) sample of the population, the estimated poverty rate would have fallen between 11.4 percent and 12.0 percent. For many purposes, it is unimportant whether the poverty rate is 11.7 percent or 11.4 percent, so being casual about standard errors causes no harm, but if we wish to make a point based on small differences in poverty rates, we will have to be aware that poverty rates are estimates and not exact. And we should remember that estimated poverty rates for smaller groups such as blacks or Hispanics are more imprecise.

promoted student), the standard deviation of the data will be low and our estimate of the standard deviation of the mean will also be low. If there is a lot of variation across pairs, so that in some pairs the retained student does a lot better and in others the promoted student does a lot better, the standard deviation of the data will be high, and so will the estimated standard deviation of the mean. We refer to our estimate of the standard deviation as the standard error of our estimate of the mean. Thus, for example, we might report that our estimate of the difference in achievement between retained and promoted students was two points with a standard error of four. This is often written with the estimate on top and the standard error in parentheses underneath: 2 . (4)

10.6. Confidence Intervals and Statistical Significance. Our best estimate of the average of all the means we would record is the one mean we have actually calculated. Using the fact that the distribution of the mean is approximately normal, we estimate that if we were to estimate the mean many times, 95 percent of the time the estimated mean would lie within two standard errors of our estimate of the mean. That is a lot of “estimates,” and it is important to keep this in mind when we look at real data. In our earlier example, the students who were retained did better on average than those who were promoted. We would like to know how likely it is that we would obtain the same result if we did the experiment over again. Recall that our estimate of the average difference is that retained students do two points better than promoted stu-

22. U.S. Census Bureau, “Source and Accuracy of Estimates for Poverty in the United States: 2001,” appendix to Poverty in the United States: 2001, Current Population Report P60-219 (Wash- ington, DC: Government Printing Office, 2002).

22 Chapter 1 dents. Our estimate of the standard error is four, and the distribution of our estimate is approximately normal. Therefore, we estimate that if we repeated the experiment, 95 percent of the time our estimate would lie between 2 – 1.96 standard errors or 2 – 1.96 × 4, or about –6, and 2 + 1.96 standard errors, or about 10. This range is called the 95 percent confidence interval because we believe (or are confident) that if we repeated the exercise, 95 percent of the time we would obtain an estimate in this range. To determine the 90 percent confidence interval, we would multiply the standard error by 1.64 instead of 1.96. We could turn the question on its head by asking what our estimated standard error would be if the true average difference between the retained and promoted students were zero. Suppose that the estimated standard error in this case were also four. We would know that if the true difference were zero, 95 percent of the time we would obtain an estimate between –8 and 8. The probability of obtaining an estimate outside this range is 5 percent, or .05. In practice, it usually makes little difference whether we ask whether two lies outside the confidence interval we would have if the true value were zero or whether zero lies outside the 95 percent confidence interval based on our estimate of a difference of two.23 Because it is usually simpler, we more frequently ask whether zero is outside the 95 percent confidence interval based on our estimate of two, but we conclude that if the true value were zero, it is (un)likely that we would obtain an estimate of two.

10.7. Statistical Significance. In statistical jargon, we say that the difference between our estimate and some value is statistically significant if the value lies outside the 95 percent confidence interval. Because the probability of something outside the 95 percent confidence interval is 5 percent, we will say that the difference is statistically significant at the 5 percent or .05 level. We may also decide to use a different confidence interval, 90 percent or 99 percent, in which case we will say that the difference between our estimate and the value is significant at the .1 or the .01 level. If some value falls within the confidence interval of our estimate, we will say that the difference between our estimate and that value is statistically insignificant. Whether our estimate is statistically significantly or insignificantly different from a value depends on the level of significance that we choose. The difference may be statistically significant at the .1 level but insignificant at the .05 level (but, of course, not the reverse). In the earlier example, our estimate of a two-point difference is statistically insignificant whether we choose the .05 or the .1 level. It is important to recognize that statistical significance does not mean statistical importance. It is unfortunate that statisticians adopted the word “significant.” Suppose we had a very large sample of wages of men and women and we estimated that, relative to men, women on average earned five dollars less per year. If the standard error of this

23. The rationale for this is as follows. If the true difference is zero, our estimated difference will be close to zero and the estimated standard error will be very similar. If the estimate is very different from zero, the estimated standard error may be quite different, but because the estimate is a long way from zero, in either case, the estimate and zero will fall outside each other’s confidence interval.

Introduction 23 estimate were only two dollars, a difference of zero dollars would lie outside the 95 percent confidence interval of one to nine, and the difference would be statistically significant but probably of no social significance whatsoever. On the other hand, with a small sample, a large estimated difference may not be statistically significant. In this case, if correct, the difference might be socially important, but because of the small sample, we have little confidence in the precision of our estimate. The true difference might be much larger, or there might be no difference whatsoever. Statistics texts used to say that before performing a test, a researcher should choose the significance level and then report whether the effect being studied was statistically significant at that level. This approach is problematic. First, if one researcher chose a significance level of .1 and another a significance level of .05, they could conduct identical experiments and draw different conclusions. In addition, the reader had no way of verifying that the researcher chose the significance level before learning the results and often suspected that the significance level was chosen ex post facto on the basis that best suited the researcher. Moreover, recall that we do not usually know the standard error and must estimate it. Therefore, we only have an estimate of how improbable an estimate is. Perhaps most important, our goal is to ask a question like “Do poverty rates differ between blacks and whites?” Our answer takes the form of a probability based on the data: “If there were no difference in the poverty rates of blacks and whites, the probability of finding a difference this large would be less than [a given number].” Our conclusion about whether blacks and whites have different poverty rates should not differ substantially if that number is 5.0001 percent or 4.9999 percent. Therefore, it is best to think about the significance level of a difference and not focus too much on whether it is above or below some critical value.

10.8. The t-Statistic. If we ask whether some value lies within the x percent confidence interval for our estimate, we are asking whether estimate + t* × standard error > value and value > estimate – t* × standard error, where t* is the value that determines the size of the confidence interval. If we were interested in the 95 percent confidence interval, t* would be 1.96. For the 90 percent confidence interval, it would be 1.64. A little algebra shows that asking whether the value is in the confidence interval is the same as asking whether

absolute value estimate—————— – value < t*. ( standard error ) Thus, if we want to know whether our two-point difference lies within the 95 percent confidence interval, we divide it by the estimated standard error (four):

24 Chapter 1 absolute value estimate—————— – value = absolute value —–—2 – 0 = .5. ( standard error ) ( 4 ) Because .5 is less than 1.96, two lies inside the 95 percent confidence interval, and our estimate of two is not significantly different from zero at the .05 level. Indeed, it is not significant at the .1 level. Note that in order to determine whether the difference is significant, all we have to do is divide the difference by the standard error. This ratio is called the t-statistic. We then compare the absolute value of the t-statistic to our chosen critical value, or we can report the significance level of the t-statistic based on statistical tables. For most purposes, it is sufficient to remember that the probability of a t-statistic greater in absolute value than 1.64 is about .1 and the probability of a t-statistic greater in absolute value than 1.96 is about .05 and that the probability declines rapidly as the absolute value of the t-statistic exceeds two. Often, as in our example, we are interested in whether some estimate equals zero. Is the difference in earnings between men and women statistically significantly different from zero? In this case, our t-statistic becomes

t-statistic = ——————estimate . ( standard error) We have noted that our tables will frequently present an estimate with its standard error in parentheses underneath. This allows us to calculate the t-statistic quickly. Some authors use the same format but report the t-statistic instead of the standard error in parentheses. Be careful to check which convention is used when reading different sources.

10.9. Relations among Variables. Often we are interested in questions that relate one variable to another. We implicitly ask such questions when we ask about differences among groups, such as “Is the poverty rate higher for blacks than for whites?” But we may be interested in questions like “How does the poverty rate vary with the state of the economy?” We might observe the poverty rate and the unemployment rate (a measure of the state of the economy) over a period of years. We can plot these combinations of poverty and unemployment rates on a two-dimensional diagram as in figure 1.1. This gives us a good visual sense of whether there is a relation between unemployment and the poverty rate, but we still require some way of summarizing the relation. We would like to say something like “For each percentage point increase in the unemployment rate, the poverty rate increases by x percentage points.” To summarize the data, we can fit a line to the points. Obviously a straight line will not fit all of the points perfectly, but we can choose the line on the basis of how well it fits. Figure 1.1 fits one possible line. Statistics courses focus on different techniques for choosing the best line. For this book, you will not need to know how to fit a line or the various advantages and disadvantages of different techniques. You will need to know how to read a table showing

Introduction 25 Poverty Rate

11 46810 Unemployment Rate Figure 1.1 Relation between Poverty Rate and Unemployment Rate

the results of the line-fitting exercise. You will also need to remember that it is the analyst, not the data, who argues the direction of causality. The results of the line-fitting in figure 1.1 can be summarized in a table that describes the line we have fitted. Table 1.1 is presented in a way similar to the estimates we presented above. The first column in the table (the one headed “Without Median Male Earnings”) says that the poverty rate increases by about .39 percentage points for every one percentage point increase in the unemployment rate. The coefficient, .39, is only an estimate of the relation between the unemployment rate and the poverty rate. It is based on a sample of years. If we had chosen different years, we might have obtained different answers. Recall that we will be much more confident of the precision of our estimate of the difference in test scores if they are clustered around the same value than if they vary substantially. Similarly, we will be more confident of the precision of our coefficient estimate if the points are clustered around our fitted line. As with the mean, we summarize the precision of our coefficient estimate by the standard error. If the coefficient estimate is normally distributed (and theory shows that it generally will be), if we looked at a different 1967 through 2000, 95 percent of the time we would obtain an estimate within 1.96 standard errors of .39. In our example, the standard error is about .12. The 95 percent confidence interval therefore runs from about .39 – 1.96 × .12, or about .15, to .39 + 1.96 × .12, or about .63. We can divide the coefficient by the standard error to obtain a t-statistic of about three. It is very unlikely that if the true coefficient were zero we would have obtained a t-statistic this large in absolute value by chance. We are therefore reasonably confident that the relation is not due to random sampling error.

26 Chapter 1 Table 1.1 Relation between Poverty Rate and Unemployment Rate (Sample Table)

Without Median Male Earnings With Median Male Earnings

Unemployment Rate 0.39 0.15 (0.12) (0.12) Median Male Earnings — –0.47 (thousands of dollars) (0.12) Constant 10.65 26.17 (0.73) (4.10)

Note: Standard errors are in parentheses.

It is important to remember that finding that a relation is unlikely to be the result of random sampling does not establish that it is causal. As discussed in the introduction to this book, we have merely established that the poverty rate and the unemployment rate have tended to change in the same direction. The poverty rate could be changing the unemployment rate, or both rates could be influenced by some other factor.

10.10. Controlling for Other Factors. In our fictional example earlier in this appendix, we assumed that we were able to match students on the basis of their sex, age, race, family structure, school, and sixth-grade ITBS math and reading scores. In practice, we are unable to match people exactly. We need techniques that allow us to ask an “if” question such as “What would the effect of promotion be if two individuals were identical in all these dimensions?” If we think that some other factor might account for the relation between the unemployment rate and the poverty rate, we want to ask, “What would be the effect on the poverty rate of an increase in the unemployment rate if this other factor did not change?” For example, there is some evidence that wages fluctuate over the business cycle, so periods of high unemployment might also be periods of low wages.24 Are the changes in the poverty rate driven by changes in the unemployment rate, prevailing wages, or both? Conceptually, to answer this question we would like to compare periods with differing unemployment rates in which prevailing wages were constant and also to compare the poverty rates in periods with differing prevailing wages but similar unemployment rates. Of course, such perfect correspondence may not occur in the data. However, statistical techniques allow us to do something comparable. The right-hand column of table 1.1 shows the results of one such technique. Each coefficient should be interpreted as the effect on the poverty rate of varying that factor

24. Gary Solon, Robert Barsky, and Jonathan A. Parker, “Measuring the Cyclicality of Real Wages: How Important Is Composition Bias?” Quarterly Journal of Economics 109 (February 1994): 1–25.

Introduction 27 while holding the other factors constant. Thus, the results show that, holding median male earnings constant, a one-point increase in the unemployment rate raises the poverty rate by about .15 percentage points. The standard error of this estimate is .12, which gives a t-statistic of

—–.15 = 1.25, .12 well below 1.96 or even 1.64. Obtaining a t-statistic of this magnitude is quite likely even if the true effect of the unemployment rate on the poverty rate is zero. We therefore do not have any good evidence that the unemployment rate affects the poverty rate. On the other hand, even holding the unemployment rate constant, increases in median male earnings are associated with quite noticeable reductions in the poverty rate. If we were to compare two periods with the same unemployment rate in one of which median male earnings exceeded those in the other period by $1,000, we would expect that, on average, the period with the higher earnings would have a poverty rate about .5 percentage points lower than the period with the lower earnings. The standard error of this estimate is also .12, so the t-statistic is close to four. It is very unlikely that we would observe a t-statistic of this magnitude if there were no real relation between median male earnings and the poverty rate. Of course, the relation between median male earnings and the poverty rate, conditional on the unemployment rate, may still be due to some other factor. We address this issue in greater depth elsewhere in this book.

28 Chapter 1 part 1 POVERTY This page intentionally left blank chapter 2

Who Is Poor?

A  ,       of this chapter may seem trivial. Although there is no magic income level below which people are poor and above which they are not, there are people who are clearly poor and people who are not. Bill Gates is not poor. The homeless person who survives by collecting bottles and returning them for their deposits almost certainly is. The World Bank says: “Poverty is hunger. Poverty is lack of shelter. Poverty is being sick and not being able to see a doctor. Poverty is not being able to go to school and not knowing how to read. Poverty is not having a job, is fear for the future, living one day at a time. Poverty is losing a child to illness brought about by unclean water. Poverty is powerlessness, lack of representation and freedom.”1 It is not obvious whether this is meant as a definition or as a description of poverty. Regardless, it is both a useful starting point and inadequate. As a starting point, we can summarize the statement by saying, “Someone is poor if he or she lacks adequate food, clothing, shelter, and medical care.” The definition is incomplete because it does not define adequate or address why and for how long the individual lacks these necessities. The World Bank uses two poverty cutoffs to define adequate—$365 per year and $730 per year adjusted for purchasing power parity.2 Few Americans would doubt that someone living on less than $1,000 per year was poor, but most would want to set the bar much higher. It is interesting to consider why someone living on $3,000 per year

1. http://www1.worldbank.org/prem/poverty/mission/up1.htm. 2. If we convert a dollar to pesos or yen, we will not necessarily find that we can buy the same amount of goods in Mexico or Japan as in the United States. The World Bank converts currencies using purchasing power parity so that, in principle, a dollar should have the same purchasing power everywhere. This comparison is difficult. It would not be unusual to find that electronics cost the same in a poor country as in the United States, while restaurant meals are less expensive in the poor country and that certain staple foods common in the poor country are simply unavailable in the United States.

31 in a developing country might be considered affluent while someone with the same income in the United States would almost certainly be viewed as poor. We will return to this question shortly after considering some examples solely from the United States. Most of us agree fairly quickly that people are poor if they do not have adequate food, clothing, shelter, and medical care. But, as applying the World Bank standard to the United States suggests, the problem is more complicated. In my experience with classes, it is easy to think of cases that elicit considerable disagreement:

• A homeless schizophrenic man who has $500,000 in a bank account but lives off the money he receives for bottles that he finds and returns for the deposit.

This example is difficult, because the man clearly has the lifestyle of a poor person. In some sense, the fact that he has $500,000 in the bank is irrelevant. He lives in a manner that is identical to the way that he would live if he had no bank account at all, so his ownership of significant financial resources seems irrelevant. Yet if we decide that this man is poor, we will have difficulties. Is anyone with bad nutrition poor? If so, some people who are generally regarded as very rich would have to be counted as poor. Suppose that, instead of being mentally ill, the man had decided to take an extended wilderness trip and live off the land for a year. We certainly want to capture elements of lifestyle with our definition of poverty, but we probably only want to call people with these lifestyles poor if they are forced into them by the lack of financial resources. Based on this example, we can revise our initial definition to say, “Someone is poor if he or she lacks sufficient financial resources to obtain adequate food, clothing, shelter, and medical care.” Consider now

• A young woman who lives with her wealthy aunt in a beautiful home. She receives free room and board. She and her aunt enjoy each other’s company. She accompanies her aunt to various social events, such as the theater, and she happily helps her aunt by doing various chores around the house (shopping, some light cleaning). When she needs money, her aunt gives it to her. In a typical year, her aunt gives her $3,000.

This young woman seems to have a rather good life. She has a nice house and presumably has good food. Although $3,000 is not a great deal of money, from the description, it appears to be adequate for her needs. Despite the fact that she has relatively little money, many students are happy to conclude that she is not poor. But now consider

• A similar young woman who lives with her wealthy aunt in a beautiful home and has similar meals and comforts to those enjoyed by the previous young woman. Like the young women in the last example, she receives her room and board and $3,000 in return for doing the set of chores also done by the previous young woman, including escorting her aunt to various social

32 Chapter 2 events, such as the theater. However, this young woman does not enjoy her aunt’s company and receives room and board and the $3,000 as explicit payment for performing the chores and serving as a companion for her aunt.

Although classes conclude that the first young woman is not poor, they find the case of this young woman troublesome. There is a sense in which the two young women have identical lifestyles. But although the first description evokes a lifestyle that many people would find agreeable, the second suggests a state that might even be considered degrading. Although the two young women consume the same goods (e.g., living with a wealthy aunt), they experience their consumption very differently. We may be tempted to conclude that the second woman, but not the first, is poor. Yet we do not wish to reduce poverty to an absence of happiness. Perhaps the solution is to conclude that the first woman is poor but happy while the second is poor and unhappy. However, there is another way of looking at this pair of examples that concerns the place of the young women in society. Before making this argument, let us consider another pair of examples:

• A graduate student couple that lives on a total of $11,000. They share a room in a group apartment. The couple pays $200 per month for the room. • A similar couple living in similar quarters but neither of whom is a student. The husband works at McDonald’s for $11,000 per year. The wife has been unemployed for over a year and has given up looking for work.

In my experience, American students are divided over whether the graduate student couple is poor, but usually agree that the second couple is poorer. How can we explain this perception when the two couples have the same income? One possible argument is that the graduate students expect to be earning more in the future and can borrow against their future income. Although this is possible, nothing in the example suggests that the student couple is, in fact, living on more than the stated $11,000. Even if we rule out borrowing, the graduate couple seems better off. A second argument might be that the graduate students are consuming education, whereas the husband, at least, of the other couple is working. Yet it is not obvious that the McDonald’s worker would prefer to be in graduate school. Indeed, it is plausible that he dropped out of school because he disliked it so much. It seems even less likely that the unemployed wife would rather be in graduate school. Some people will argue that the difference between the two couples is that the graduate students are making a choice: “They are not poor because they have chosen to be poor.” But in some sense, both couples are living in poverty because of choices they have made. Perhaps the husband who works at McDonald’s and his wife are both high school dropouts. They could have chosen to remain in school, not get married, live with their parents, and so on. We can certainly fill in details of the scenario to make it seem that they are “choosing to be poor.” Suppose that the husband’s parents are willing to have him live at home and support him while he goes back to school if he divorces his wife, and her parents are willing to make a similar offer. Does this change our belief about whether they are poor?

Who Is Poor? 33 Other people will argue that the graduate students are not poor because they expect not to be poor in the future even if they cannot borrow against their future income. Cer- tainly being poor may be less painful if one knows that it is temporary. Many people (immigrants, students) are willing to accept temporary poverty as an investment. How- ever, the mere fact that their poverty is likely to be temporary does not eliminate it. If our couple with the husband working at McDonald’s knew that one of them was likely to inherit a large amount of money in five years, it would not make them less poor unless they were able to use that knowledge to increase their current consumption. Instead, the graduate students experience their consumption differently because society responds to them differently. Many of their friends are also graduate students with similar incomes. Their standards of consumption are based on that shared income level. Their high school and college friends, even if earning considerably more money, recognize the graduate students’ period of low income as a stage through which they are passing. The graduate students may envy their friends for their cars, houses, and other goods, but they expect to have cars, nice houses, and the other goods after they finish graduate school and begin working. In contrast, the McDonald’s worker and his wife seem trapped in their poverty. Al- though they may have friends who are similarly situated, they are likely to be excluded from social interaction with more successful friends. They are unable to reciprocate with equivalent hospitality when invited to more successful friends’ homes or to go out to even moderately expensive restaurants with them. They may be unable to exchange appropriate presents on birthdays, holidays, and special occasions. Let us turn now to one last example: • An eighty-year-old widow who has lived for sixty years in a house that is now worth $500,000. Her total income is barely adequate to pay the taxes on the house. She survives by keeping the house heated to no more than 60 degrees in the winter, by skipping meals or eating at the local soup kitchen. There are no inexpensive houses in her neighborhood, which is the only place where she has a network of friends and feels comfortable. Experience shows that classes have the hardest time agreeing on this last example. Clearly the widow has the resources to live comfortably. If she sold her home and invested it in bonds, she would be able to live on at least $30,000 per year, possibly as much as $50,000 per year, even if she lived to be 100 years old. On the other hand, her consumption level is that of a very poor person. Ultimately, whether or not we conclude that the widow is poor is likely to depend on whether we believe that remaining in her home is essential to her participation in society. We understand that she does not want to leave her friends and her familiar neighborhood. Many of us will feel that with an income of $30,000 or more, she could find an apartment in a community where she could make new friends. Others of us will feel that if she values her present community so much that she is willing to suffer the hardships she bears, owning her present home must be essential to her participation in society.

34 Chapter 2 We can now further refine our definition of poverty. We must add participation in society as one of the basic human needs that the financial resources must be adequate to address. Thus we can say, “An individual is poor if he or she lacks sufficient financial resources to obtain adequate food, clothing, shelter, and medical care and to participate in society.” There are three important points to make about this definition. First, it solves the mystery we associated with the World Bank definition—why someone with an income of $3,000 is well off in some countries and poor in others. In a society in which per capita income is $100, an income of $3,000 ensures the capacity to purchase the goods necessary to participate in society. In a society in which per capita income is $30,000, though, $3,000 is woefully inadequate for that purpose. Note, however, that our revised definition does not imply that being poor is purely relative. Victor Fuchs has suggested that anyone with an income below half of the median for that society is poor.3 Thus, in a society in which the median family income is $50,000, anyone with a family income below $25,000 is poor. In a society in which the median family income is $5,000, anyone with a family income above $2,500 is not poor. Obviously, these cutoffs must be adjusted for family size, but let us ignore this complexity. If for each family in the first country there were an identical family with one-tenth the income in the second country, according to the Fuchs definition, the poverty rate would be identical in the two countries. Yet many of the nonpoor (by Fuchs’s definition) in the second country would be poor by the World Bank definition. They would indeed suffer from absolute poverty. Thus the Fuchs definition is certainly inadequate if we want to compare poverty rates across countries with very different per capita incomes. The second important point to make about our working definition is that it is not yet operational. We have not defined what constitutes adequate financial resources. In this respect, Fuchs’s definition is attractive. In a wealthy country such as the United States, it is plausible that relative poverty is the important consideration. Anyone who suffers from absolute poverty also suffers from relative poverty. Yet there are many more people who lack the financial resources to fully participate in society, even if they do not suffer from a lack of food, clothing, shelter, or medical care. Of course, there is nothing magical about Fuchs’s choice of 50 percent of the median income as the level of income necessary to participate in society. Any precise definition is arbitrary. It would be very hard to argue that 50 percent was the right fraction and that 49 percent or 51 percent was clearly wrong. The Fuchs definition has the further advantage that it allows poverty rates to vary across countries. For well-to-do countries in which the richest of the poor do not suffer from absolute poverty, the poverty rate can vary from zero to almost 50 percent. It is perfectly possible for no family to be below 50 percent of the median family income,

3. Victor R. Fuchs, “Redefining Poverty and Redistributing Income,” The Public Interest 8 (Sum- mer 1967): 88–95.

Who Is Poor? 35 Minimum Needed ($1982–84) 18,000 1980 1976 1979 1972 1981 1978 1973 1985 1975 1977 16,000 1970 1982 1969 1974 1971 1983 1984 1965 1966 1967

14,000 1968 1960

1959 1964 1957 12,000 25,000 30,000 35,000 40,000 Real Median Income: Married Couples Figure 2.1 Average Perceived Need and Median Income Sources: Gallup Polls, 1957–1985.

while in another country with very unequal income almost half could be below 50 percent of the median.4

1. Evidence on the Importance of Relative Income Survey data suggest that both relative and absolute poverty are important. Between 1957 and 1985, almost every year, the Gallup Poll asked what was the minimum weekly money income that a family of four (mother, father, two children) needed to get by in their community.5 Figure 2.1 multiplies the average response to this question by fifty- two to determine annual need and by the Consumer Price Index to determine real as opposed to nominal need. The figure plots this need against real median family income for married-couple families. It looks like a straight line. If we fit a straight line to these data, we obtain a constant term of about $6,000 and a slope coefficient of about .27. If we believe these results, they suggest that even if people did not have to participate in society, they would require (for a family of four) an income of about $6,000 in 1982– 1984 dollars, or about $11,400 in current dollars. The cost of participating in society rises by a little more than a quarter for every dollar increase in median family income.

4. Because by definition exactly half are below the median family income, the fraction below half of the median income cannot quite attain 50 percent. 5. George H. Gallup, Gallup Poll (New York: Random House, 1972), and Gallup Opinion Index and Gallup Report, various issues.

36 Chapter 2 The third point about our working definition is that it remains imperfect. By focusing on financial resources, we miss important sources of consumption. We miss gifts from friends and family, such as those received by the woman with the wealthy aunt. Nor do we cover all basic human needs. We might easily have added love, affec- tion, sex, and so on. But we prefer to think that someone who lacks these basic human needs but has adequate resources to participate in society is suffering not from poverty but from some other source of deprivation. In the next section we will see that the official measure of poverty differs quite a bit from the definition developed here. Most of what we know about poverty is based on the government definition. It is important to remember this difference when we talk about government poverty statistics.

2. How the Government Measures Poverty Most people do not realize that the poverty thresholds were essentially developed by one person, a woman named Mollie Orshansky who worked for the Social Security Administration. Orshansky began with a minimal food plan that had been developed by the Department of Agriculture. She took the cost of this plan as the minimal cost of a nutritionally adequate diet, allowing for reasonable variety. Given the cost of food, the question was how to account for minimal expenditure on other necessities (clothing, shelter, medical care). Orshansky used the fact that families spend about one-third of their incomes on food and multiplied the cost of food by three to obtain an estimate of the minimum income needed, which became known as the poverty threshold. She adjusted this requirement for household size. Sub- sequently, Orshansky’s calculations, with only relatively minor changes, were adjusted by the Consumer Price Index, the index most commonly used to measure inflation. Whether a person is poor is calculated at the family level, because families are assumed to pool resources. Thus, if you live with your parents, even if you have no income, you will not be considered poor if your parents have sufficient income to lift the whole family above the poverty line. However, when we count the number of poor, we count the number of individuals. If a family’s income is below the poverty line for a family of its size, all members of the family are poor. Thus, an individual is poor if she lives in a family whose income is less than three times the cost of providing members of the family with a nutritionally adequate diet (as calculated by Mollie Orshansky based on the Department of Agri- culture’s minimal food plan and adjusted for inflation since then).6 Let us be clear. Whether or not a person is poor does not depend on his actual expenditure on food or on the proportion of his income that is spent on food. The poverty cutoff is simply a number. In 2004, for a single-person family in which the individual

6. See Gordon M. Fisher, “The Development and History of the Poverty Thresholds,” Social Security Bulletin 55, no. 4 (Winter 1992): 3–14, for more details.

Who Is Poor? 37 was less than 65 years old, the poverty threshold was $9,827.7 That means that, updating Orshansky’s figures, the government calculated that it cost a little less than $3,300 per year for a minimally adequate diet. If a person lived by herself and had an income below $9,827, she was poor regardless of whether she spent $2,000 per year on food or $4,000 per year on food. Similarly, if she earned more than $9,827 in 2003, she was not poor regardless of how much she actually spent on food. The poverty threshold for a couple without children (both adults under 65) was $12,649. For a four-person family with two adults and two children, it was $19,157. It should be evident from our previous discussion that there are significant problems with the government definition. 1. What constitutes an “adequate” diet is arbitrary. There is no magic level at which one ceases to be poor. 2. It is strictly a measure of absolute poverty. If we believe that part of the cost of living is the cost of participating in society, we would expect the real value of the poverty threshold to rise as society becomes richer. This second problem is primarily an issue when we try to follow poverty rates over time. At any given point in time, the poverty threshold may be too high or too low, but that is probably not a problem that can be resolved easily. Each of us has a different standard of what we consider minimal consumption. This standard tends to be higher for higher-income people. We would not necessarily expect everyone to agree. Seventy years ago, we would not necessarily have thought that someone without a telephone was poor. Today participating in society may require having a phone. In some groups, having a cell phone is increasingly becoming necessary in order to arrange social occasions.8 One check on how we are doing is to project the relation in the Gallup data to the present. In 2003, the median family income for a family with a husband and wife without children was $62,281. Using our estimate from Gallup, the poverty threshold would have been about $28,216. This suggests that the current poverty thresholds are set at levels below the ones that the average American would set, perhaps because the poverty thresholds have risen too slowly or perhaps because they were set too low initially. Gordon Fisher has reviewed the relation between measures of poverty in the United States and elsewhere.9 He reports that prior to the development of the current poverty line, estimates of the poverty cutoff consistently increased over time as society became wealthier.

7. Poverty thresholds can be found at http://www.census.gov/hhes/poverty/threshld/thresh04. html. You may substitute a more recent year for “04” in the URL. 8. The Boston Globe (July 27, 2003, Third Edition, City Weekly, p. 1) reported on a young homeless woman who has a cell phone as well as a Web page that she accesses through the local public library. 9. Gordon M. Fisher, “Is There Such a Thing as an Absolute Poverty Line over Time? Evidence from the United States, Britain, Canada, and Australia on the Income Elasticity of the Poverty Line,” http://www.census.gov/hhes/poverty/povmeas/papers/elastap4.html#C2.

38 Chapter 2 Although the failure to take into account the increasing cost of participating in society will tend to cause us to underestimate the poverty rate (if we accept that Mollie Orshansky was right back in the mid-1960s), there are other problems that would cause us to overestimate it.

1. It is widely (although not universally) believed that the Consumer Price Index overestimates inflation by about 1 percent per year. If this is correct, since 1963 we have overestimated the increase in the poverty threshold by almost one-third. 2. The official poverty rate is based solely on money income. There are other sources of income that are nearly equivalent to money income and might fall within our definition of financial resources—food stamps, section 8 housing vouchers. There is also no adjustment for other forms of nonmonetary income (primarily owner-occupied housing).

3. Valuing Nonmonetary Income How much is nonmonetary income worth? The answer is that it is worth whatever you would be willing to sell it for. However, most of the time people cannot sell their nonmonetary income, so we have to try to figure out how much they would sell it for. Some of the time we are pretty sure that the person would want the nominal cash value of the nonmonetary income. Consider a college student who is paying $200 per month to rent a room in an apartment. Her father says, “I don’t want to pay for your partying, but I will contribute $100 per month toward your rent.” How much would the student be willing to sell the restricted subsidy for? Put differently, how much unrestricted cash would the student need to be just as well off? Because the student would have spent the $100 on housing anyway, it makes no difference to her whether she “has to” spend the money on an apartment. Her behavior is no different from what it would have been if her father had given her $100 per month. He is paying for her partying. Now suppose the doting father says, “I still won’t pay for your partying, but I do want you to have a nicer room. I will pay $250 toward your rent.” Now our analysis is more complicated. We need to know what the student would do if her father gave her $250 per month cash rather than paying $250 toward her rent. It is possible that with an extra $250 per month income, she would choose to rent a much nicer room and pay $275 per month. With the “housing subsidy” she would take the room at $275 per month, apply her father’s $250 subsidy to the room, and have the same money to spend on other goods as she would have had if her father had given her $250 without telling her to spend it on housing. So the $250 housing subsidy is just as good as $250 cash. On the other hand, with an extra $250 in cash, she might not choose to spend $250 or more on rent. So she would prefer a cash payment of somewhat less than $250 to the $250 housing subsidy. In some cases, in-kind transfer presents a situation a lot like that of the father giving his daughter $100 toward her rent. Most food stamp recipients spend more on food than they receive in food stamps, which suggests that for most recipients a dollar

Who Is Poor? 39 of food stamps is worth one dollar. On the other hand, some recipients would not choose to spend all their food stamps on food and therefore value them at less than their nominal price. At least prior to recent changes designed to make it harder to sell food stamps, there was an active market for food stamps in most major cities with a going rate of between 50 and 75 cents per dollar.10 Because most food stamp recipients want all of their food stamps and because those who do not can sell the food stamps they do not want for a significant fraction of their nominal price, food stamps are probably worth close to their nominal value. One way to tell whether the cash value of food stamps equals their nominal value is to look at what happens when food stamps are replaced with cash. If people do not change their food consumption, food stamps are worth their full nominal value. Unfortunately, evidence on this issue is mixed. Thomas Fraker and his coauthors report on a series of experiments that found that replacing food stamps with cash payments significantly reduced food consumption. On the other hand, Robert Moffitt found no evidence of a reduction in food consumption when Puerto Rico converted from food stamps to cash payments.11 At the other extreme, the evidence suggests that people would not purchase insurance with the value of their Medicaid health insurance if it were not provided in its in-kind form. Given the very high cost of equivalent insurance purchased on the private market, it is very unlikely that most recipients of free insurance would purchase such insurance if they did not receive it for free. We would certainly not want to add the market cost of the equivalent of Medicaid insurance to people’s incomes when determining whether they are poor.12

4. Over What Time Period Should We Measure Poverty? The poverty rate is based on income per year. The period over which we choose to measure income has a significant effect on the poverty rate. Many people are paid in ways that cause their income to fluctuate significantly from day to day. A taxi driver may have a good day or a bad day, as may a plumber or a salesperson working on commission. A real estate broker who sold twenty houses per year with an average commission of $2,000 per house would be viewed as earning well above the poverty level on an annual basis. However, most days the broker would sell no houses and earn no money. It

10. Deborah Lutterbeck, “Food Stamp Fallout: Why Turning at Least One Federal Program Over to the States May Not Be Such a Great Idea,” Common Cause (Winter 1995). 11. Thomas M. Fraker et al., “The Effect of Food Stamp Cashout on Food Expenditures: An Assessment of the Findings from Four Demonstrations,” Journal of Human Resources 30 (Fall 1995): 633–49, and Robert Moffitt, “Estimating the Value of an In-Kind Transfer: The Case of Food Stamps,” Econometrica 57 (March 1989): 385–409. 12. Marilyn Manser, in “Estimates of Cash-Equivalent Values from a Flexible Demand System for the U.S.” (Working Paper 235, Bureau of Labor Statistics, Washington, DC, September 1992), estimates that for most recipients the value of food stamps is close to their nominal value, but the value of Medicaid is substantially below its nominal value.

40 Chapter 2 is likely that each year there would be one or two months in which he sold no houses. If we measured poverty on a monthly basis, the real estate broker would be viewed as poor in those months. In the same way, many of those who are poor for one year would not be poor if we measured income over a longer period. A businesswoman or farmer who loses money one year will be recorded as poor, but, knowing that her earnings fluctuate from year to year, she may have substantial savings to draw on. The graduate students in our earlier example probably would not be seen as poor if we measured their income over their lifetimes. What is the correct period over which to measure income? To answer this question, we need to return to our definition of poverty. Ultimately, we are interested in how someone’s financial resources affect her ability to consume. If fluctuations in income over some periods have no effect on consumption, the period considered is certainly not too long and may be too short. My consumption does not fluctuate systematically over the month even though I am paid on a monthly basis. With the spreading use of credit cards, that is probably increasingly the case for most people. However, there are some people, particularly among the poor, whose consumption does fluctuate over the month because they run short of money toward the end of the pay period. For many people with uncertain incomes, consumption may vary significantly over the course of the year, tracking good and bad income periods. Thus, for some people one year is too long a period, while for others it may be too short. Some people experience brief spells of poverty that are too short to be captured by the official measure, while others consume well above what is “adequate” despite a temporarily low income for a year or more. Even if one year were the “correct” time period, there is nothing magic about calendar years. Having a low income from July 1 to June 30 is not fundamentally different from having a low income from January 1 to December 31.

5. Other Problems with the Official Measure Having a single national poverty threshold is also problematic. There are significant variations in the cost of living (and possibly of participating in society) across states and between urban and rural areas. The cost of housing is much higher in Boston than in, say, Newcastle, Pennsylvania. Housing costs are the major source of regional differences in the cost of living, but they are significant. Differences in the cost of food and clothing across regions are very small. Another reason the poverty rate may be underestimated is that the poverty threshold does not take into account taxes and unavoidable costs of working (transportation costs, child care costs). At the same time, it may be overestimated because the threshold does not take account of subsidies received through the tax system. Rather than paying taxes, low-income working parents often receive cash payments from the government as part of the federal income tax system. Unlike welfare, which is paid “before tax,” such payments are paid “after tax” and therefore do not affect whether members of a family are classified as poor.

Who Is Poor? 41 6. The National Academy of Sciences Report In 1995 the National Academy of Sciences produced a set of suggested revisions to the official poverty measure. There were a number of technical points, but the report also addressed many of the criticisms described earlier. The main changes suggested were as follows: • The poverty line should be a fixed percentage of median consumption of food, clothing, and shelter (adjusted for family composition and size). • The income considered should be disposable income. Therefore, tax payments and certain necessary expenditures associated with employment, such as child care, should be subtracted from it, but the value of tax credits should be added to it. • Housing subsidies, food stamps, and similar near–cash transfers should be treated as cash and added to income, as should the imputed rents13 received by homeowners. • The poverty thresholds should be adjusted for regional differences in the cost of living. • Medical out-of-pocket expenses should be deducted from income. This last suggestion has been the most controversial. It avoids the problem of valuing Medicaid and Medicare. If individuals had no control over their medical expenditures, it would be noncontroversial. However, individuals with Medicare presumably receive higher-quality health care than low-income individuals with no health insurance. Two individuals with the same out-of-pocket medical expenses and the same incomes may have very different levels of real consumption if one has government-provided health insurance and the other has none. The person with government health insurance is likely to be receiving more health care at the same out-of-pocket cost.

7. Gathering the Data Most of the official statistics come from the Annual Demographic Supplement to the Current Population Survey (CPS). The CPS is a large survey conducted each month, primarily to determine the monthly employment and unemployment figures. Each March, the survey also gathers information about respondents’ earnings and income during the previous calendar year. The government uses this information about income to determine whether each family in the sample is poor. From the sample, it then projects the number of individuals who are poor. It also uses information, such as that on race, to determine the number of people in particular groups, such as non-Hispanic whites, who are poor.

13. Homeowners are their own landlords. If they make a profit on the “rental of their homes to themselves,” that should be treated as income. This is particularly important when we compare poverty rates between the elderly and nonelderly. The elderly are more likely to own their homes and have paid off their mortgages.

42 Chapter 2 The poverty rate for the United States is given by the estimated number of people who are poor (derived from the CPS) divided by the estimated population of the United States. The poverty rate for a group is defined in a similar way: estimated number of poor in group estimated poverty rate for a group = ——————————————. estimated population of group

8. Who Is Poor (By the Official Measure)? Although by now you are aware that the official measure of poverty is imperfect, it remains the most widely used measure in the United States and is our chief basis for describing poverty in this country. When they think of someone poor, older Americans may have an image of someone living in Appalachia or the rural south. Others may think of an elderly person. Younger Americans are likely to think of a black or Hispanic person living in a single- parent family located in an inner-city neighborhood. Although there are poor people who fit each of these descriptions, each of these images describes a poor person who is in many ways atypical. Part of the problem is that we confuse how common a characteristic is within a group (the poverty rate among blacks) with how common that group is among those who have the characteristic (the fraction of the poor who are black). In the year 2004, the U.S. Census Bureau estimated that 12.7 percent of Americans lived in poverty.14 Of those, roughly thirty-seven million people, seventeen million were non-Hispanic whites, nine million were black, and nine million were Hispanic, including some who also described themselves as black.15 Thus, somewhat fewer than half of the poor are black or Hispanic. There are about as many poor non-Hispanic whites as there are poor blacks and Hispanics. Of course, there are many more non-Hispanic whites in the United States than there are blacks or Hispanics. Thus non-Hispanic whites are less likely than members of the other two groups to be poor. As shown in table 2.1, only 9 percent of non-Hispanic whites, compared with 24 percent of blacks and 22 percent of Hispanics of all races, are poor. About one-third of the poor are children, while only about 10 percent of the poor have reached age 65. The poverty rate is highest among children and young adults (about 18 percent). As can be seen in table 2.1, the poverty rate falls through the 45–54 age group and then begins to rise again. However, the poverty rate among the elderly is 10 percent, well below the poverty rate for children. Somewhat more than one-fourth of the poor live by themselves or in households with unrelated individuals. Roughly one-third live in households headed by a single

14. Carmen DeNavas-Walt, Bernadette D. Proctor, and Robert J. Mills, Income, Poverty, and Health Insurance Coverage in the United States: 2004, U.S. Census Bureau, Current Population Report P60-229 (Washington, DC: U.S. Government Printing Office, 2005). 15. The Census Bureau now allows people to classify themselves as multiracial (for example, both black and white). These numbers exclude individuals who reported themselves as multiracial.

Who Is Poor? 43 Table 2.1 Poverty of People with Assorted Characteristics, 2004

Percentage of Poor Poverty Extreme Characteristic with Characteristic Rate Poverty Rate

All Characteristics 100.0 12.7 5.4

Race Non-Hispanic White 45.6 8.6 3.7 Black 24.3 23.7 11.7 Hispanic 24.7 21.9 7.9

Age Younger than 18 35.2 17.8 7.6 18–24 13.7 18.1 7.4 25–34 13.3 12.5 3.9 35–44 11.5 9.9 3.4 45–54 9.5 8.4 3.4 55–59 3.8 8.4 3.2 60–64 3.6 10.4 4.3 65 or older 9.3 9.8 2.1

Family Structure Living with Family 71.8 11.0 4.4 Living Alone or with Unrelated Individuals 26.7 20.5 9.8

Type of Community* Central City 40.6 17.5 7.8 Other Metropolitan 38.5 9.1 3.9 Outside Metropolitan Area 20.9 14.2 5.4

Source: http://ferrit.bls.census.gov/macro/032004/pov/toc.htm.tables pov01, pov40, and pov41. *Data are for 2003.

parent, while the remainder live in households in which either both a husband and a wife are present or there are two related adults but no children.16 Thus most of the poor are not in single-parent families. But the poverty rate is substantially higher among single-parent families than among other groups. Over one-third of female-headed families with related children under 18 are poor. In contrast, only 6 percent of married-couple families and 7 percent of married-couple families with related children under 18 are poor.17

16. http://pubdb3.census.gov/macro/032005/pov/new03_100_01.htm. 17. http://pubdb3.census.gov/macro/032005/pov/new04_100_01.htm.

44 Chapter 2 Finally only, about 40 percent of the poor live in central cities, and many of these central city areas are not areas of high poverty. Mary Jo Bane and Paul Jargowsky argue that our common perception of an area of concentrated poverty corresponds closely to an area in which the poverty rate exceeds 40 percent.18 Jargowsky calculates that in 2000, just 10 percent of the poor, 19 percent of the black poor, and 14 percent of the Hispanic poor lived in census tracts (geographical areas with an average population of four thousand) in which at least 40 percent of the population was poor, and about one-tenth of these lived outside of metropolitan areas.19

9. Extreme Poverty Just as it is arbitrary to choose an income cutoff at which a family’s status changes from poor to nonpoor, it is arbitrary to choose a cutoff for extreme poverty. Nevertheless, anyone living in a family with an income level below half the poverty line is likely to be extremely poor. Noncash income is unlikely to make up for the shortfall in income. Therefore, to get a sense of the prevalence of extreme poverty, we can examine the fraction of individuals who live in families whose income falls below half of the poverty level. Recall that for an individual living alone this means an income of less than about $4,900 and for a family of four with two adults and two children it means no more than a little over $9,500. This extreme poverty rate was 5.4 percent in 2004 (see table 2.1). In contrast with what we saw previously for the poverty rate, the extreme poverty rate does not rise as we pass from middle age to retirement. Fully 7.6 percent of both children and 7.4 percent of young adults live in families (or by themselves) with incomes less than half of the poverty threshold. Thereafter, this rate tends to decline with age, so that among those 65 and older, only 2.1 percent of the population lives in families with incomes below this level. Social security, including supplemental security income (SSI), has been largely successful in eliminating extreme poverty among the elderly, at least that determined using the official measure. Comparing the second and third columns of table 2.1, the other difference to note is that the relative prevalence of extreme poverty among blacks is even greater than the relative prevalence of poverty overall. Although the poverty rate among blacks is less than three times the poverty rate among whites (already a large difference), the extreme poverty rate among blacks is more than three times that among whites.

10. Homelessness Recall that we defined poverty as lacking the financial resources to obtain adequate food, clothing, shelter, and medical care and to participate in society. One way we can

18. Paul Jargowsky and Mary Jo Bane, “Ghetto Poverty in the United States, 1970–1980,” in Christopher Jencks and Paul E. Peterson, eds., The Urban Underclass (Washington, DC: Brookings Institution, 1991), 235–73. 19. Paul A. Jargowsky, Stunning Progress, Hidden Problems: The Dramatic Decline of Concentrated Poverty in the 1990s (Washington, DC: Brookings Institution, 2003).

Who Is Poor? 45 address the extent of absolute poverty is by looking at the prevalence of people in our society who lack adequate food, clothing, shelter, or medical care. This section looks at shelter. The next will discuss the extent to which Americans lack adequate food. Of course, some of the homeless may, like the schizophrenic man in our example earlier in this chapter, have adequate financial resources to obtain housing. But our other measures of poverty are also imperfect. At first blush, despite the practical difficulties of finding and counting the homeless, the concept is straightforward. Someone who is homeless is someone who does not have a home. It seems that nothing could be simpler. But on further reflection, deciding who is homeless is more complicated. When I first moved from California to Massachusetts, I arrived two weeks before the start of the lease on my apartment. I did not have a home, but surely I was not homeless. I stayed at my sister’s house until I could move into my new apartment. The solution may seem easy: exclude from the homeless people who can stay with friends or family. But do we really want to exclude someone who moves from friend to friend, trying not to wear out his welcome, and who is always aware that only the kindness of friends keeps him from sleeping on the streets? Perhaps what distinguished me from the homeless person was that I could have rented a hotel room. I would not have slept outdoors. If so, is the person who rents a cubicle in what is essentially a dormitory also not homeless? The government routinely pays hotels to house people we call homeless. Are they no longer homeless once the government houses them in a hotel? Suppose the hotel stops serving the public and houses only the previously homeless, in which case it becomes a homeless shelter. Are the homeless living in the homeless shelter no longer homeless? Moreover, if we include people living in “welfare hotels” as homeless but exclude those living with friends and families, increasing the availability of welfare hotels will appear to increase homelessness because people who otherwise would have lived with friends and family (not homeless under the strict definition) now live in welfare hotels and are homeless. The point here is not to reach some agreed-upon definition of homelessness. In- stead, it is to clarify why estimates of the number of homeless can vary so widely. If we restrict our definition of the homeless to those regularly sleeping outside of places intended for sleeping (outdoors or bus stations, etc.) or in shelters to which they do not have access during the day, we will find far fewer homeless than if we also count people living in hotel rooms at government expense. But if we add those who are living off the “charity” of friends or family, we will find even more homeless. Although 9 percent of Americans report ever having lived on the streets or in a shelter, 15 percent report having considered themselves homeless at some time.20 In addition to the usual concerns about reaching an appropriate definition, obtaining an accurate count of the homeless is extraordinarily difficult. The census provides decennial counts of individuals in emergency and transitional shelters, but on any given

20. Bruce Link et al., “Reconsidering the Debate about the Size of the Homeless Population,” cited in Christopher Jencks, The Homeless (Cambridge, MA: Harvard University Press, 1994), 6.

46 Chapter 2 night, a significant fraction of those who sometimes use shelters are elsewhere. In 1990, the Census Bureau attempted to count the homeless on the street, but this effort was widely criticized and was not repeated in 2000. If we use a more expansive definition of homelessness that includes people living with family or friends, counting becomes even more complicated because we have to distinguish between people voluntarily living together and those who would be unable to obtain shelter otherwise. The census counted approximately 180,000 individuals in emergency and transitional shelters in 1990 and about 170,000 in 2000.21 This tells us little about what happened to homelessness over that period because changes in the availability of shelter beds can change the number of people sleeping in shelters without any change in the extent of homelessness. However, it does at least provide a lower bound on the number of homeless. The McKinney Homeless Act of 1987 defines as homeless “an individual who lacks a fixed, regular, and adequate nighttime residence” or whose primary nighttime residence is “(A) a supervised publicly or privately operated shelter designed to provide temporary living accommodations (including welfare hotels, congregate shelters, and transitional housing for the mentally ill); (B) an institution that provides a temporary residence for individuals intended to be institutionalized; or (C) a public or private place not designed for, or ordinarily used as, a regular sleeping accommodation for human beings.” The McKinney Act definition is relatively narrow. Such definitions produce correspondingly low estimates of the number of homeless. Christopher Jencks, using a similar definition, estimates that the number of homeless in March 1987 was almost surely between three hundred thousand and four hundred thousand.22 For the same period, Martha Burt estimates the number to have been between five hundred thousand and six hundred thousand. The discrepancy appears to reflect differences in their counts of the number of people served in homeless shelters.23 There are a number of reasons to be cautious about these numbers. First, estimates of the number of homeless can be sensitive to the period during which the estimates are made. Using similar information and techniques in both cases, Martha Burt and her colleagues estimate annual rates of homelessness that are roughly 50 percent to 100 percent higher based on October/November 1996 data than are those based on February 1996 data.24 One reason may be that it is easier to find the homeless in shelters than outdoors, and there are more shelter spaces in the winter than in the fall. The second reason to be cautious is that for many people, homelessness is a very short-term phenomenon. Burt’s team also estimates that the number of people experiencing homelessness in a year is about four to five times the number experiencing it in a week. Jencks’s estimates imply a ratio of about three or four.

21. Annetta C. Smith and Denise I. Smith, Emergency and Transitional Shelter Population: 2000, U.S. Census Bureau, Census Special Reports, Series CENSR/01-2 (Washington, DC: U.S. Govern- ment Printing Office, 2001). 22. Jencks, 13. 23. Jencks, 146. 24. Martha Burt et al., Helping America’s Homeless (Washington, DC: Urban Institute, 2001), 47.

Who Is Poor? 47 Finally, anyone who thinks these estimates are too low can always claim that many homeless live on the streets and do not use shelters, soup kitchens, and so on. As Jencks puts it, “It is hard enough finding the proverbial needle in the haystack. It is far harder to prove that a haystack contains no more needles.”25 What does all this say about our poverty counts? The first point is that the number of people experiencing homelessness in the course of the year is almost certainly a small fraction of the roughly 5 percent who experience extreme poverty as measured by having an annual income less than half of the poverty cutoff. The second point is that although the homeless are not covered by the Current Population Survey and thus not included in the poverty count, counting the homeless would have only a small effect on the poverty rate. Even if we take the number of homeless as six hundred thousand (a relatively high estimate) and assume that all the homeless are poor by the official definition, this will add only .2 percent to the poverty rate. And this estimate is too high, because not all of the homeless have annual incomes below the poverty level.

11. Hunger and Food Insecurity Since 1995, the U.S. Department of Agriculture has included a supplement on food insecurity in the Current Population Survey. Currently the information in the supplement is collected each December.26 For households without children, the food insecurity supplement consists of ten questions: 1. We worried whether our food would run out before we got money to buy more. Was that often, sometimes, or never true for you in the last 12 months? 2. The food that we bought just didn’t last and we didn’t have money to get more. Was that often, sometimes, or never true for you in the last 12 months? 3. We couldn’t afford to eat balanced meals. Was that often, sometimes, or never true for you in the last 12 months? 4. In the last 12 months, did you or other adults in the household ever cut the size of your meals because there wasn’t enough money for food? 5. (If yes to Question 4) How often did this happen—almost every month, some months but not every month, or in only one or two months? 6. In the last 12 months, did you ever eat less than you felt you should because there wasn’t enough money for food?

25. Jencks, 12. 26. The material in this section is based on Mark Nord, Margaret Andrews, and Steven Carlson, Household Food Security in the United States, 2003 (Food Assistance and Nutrition Research Report 42, Food and Rural Economics Division, Economic Research Service, U.S. Department of Agriculture, Washington, DC, 2004).

48 Chapter 2 7. In the last 12 months, were you ever hungry but didn’t eat, because you couldn’t afford enough food? 8. In the last 12 months, did you lose weight because you didn’t have enough money for food? 9. In the last 12 months, did you or other adults in your household ever not eat for a whole day because there wasn’t enough money for food? 10. (If yes to Question 9) How often did this happen—almost every month, some months but not every month, or in only one or two months?

The Department of Agriculture uses the answers to these questions to classify households without children as food secure, food insecure, or food insecure with hunger. They are classified as having a food-insecure condition if they answer “Often” or “Sometimes” to any of questions 1–3, “Almost every month” or “Some months but not every month” to question 5 or 10, and “Yes” to any of the remaining questions. The department then counts the number of food-insecure conditions the household experiences. If a household has two or fewer food-insecure conditions, it is classified as food secure. If it has three to five food-insecure conditions, it is classified as “food insecure,” and if it has more than five conditions, as “food insecure with hunger.” For households with children, the survey poses additional questions: 11. We relied on only a few kinds of low-cost food to feed our children because we were running out of money to buy food. Was that often, sometimes, or never true for you in the last 12 months? 12. We couldn’t feed our children a balanced meal, because we couldn’t afford that. Was that often, sometimes, or never true for you in the last 12 months? 13. The children were not eating enough because we just couldn’t afford enough food. Was that often, sometimes, or never true for you in the last 12 months? 14. In the last 12 months, did you ever cut the size of any of the children’s meals because there wasn’t enough money for food? 15. In the last 12 months, were the children ever hungry but you just couldn’t afford more food? 16. In the last 12 months, did any of the children ever skip a meal because there wasn’t enough money for food? 17. (If yes to Question 16) How often did this happen—almost every month, some months but not every month, or in only 1 or 2 months? 18. In the last 12 months, did any of the children ever not eat for a whole day because there wasn’t enough money for food?

Answers of “Sometimes” or “Often” to questions 11–13, of “Almost every month” or “Some months but not every month” to question 17, and “Yes” to the other questions indicate the presence of a food-insecure condition. Like households without children, households with children are considered “food secure” if they report two or fewer food-

Who Is Poor? 49 insecure conditions, “food insecure” if they report three to five food-insecure conditions, and “food insecure with hunger” if they report six or more conditions based on the first ten questions. They are also food insecure with hunger if they report eight or more food-insecure conditions from the full set of eighteen possible conditions. More- over, households that report five or more food-insecure conditions from questions 11–18 are considered “food insecure with hunger among children.” There are a number of things to note before we examine the results of the survey. First, the Current Population Survey is based on a sample of dwelling units and the people living in them. The homeless are not covered by the survey. Although we have suggested that, using a strict definition, the number of homeless is probably modest, it is likely that the incidence of hunger is much higher among this group than among the population as a whole. Therefore, the survey underestimates the extent of hunger and food insecurity in the United States. Second, it could be argued that anyone who gives a positive answer to any of the questions has suffered from food insecurity, and, to some questions, from hunger. If so, requiring multiple food-insecure conditions for a household to be classified food insecure or food insecure with hunger would underestimate the extent of food insecurity and hunger in the United States. However, the fewer the number of food- insecure conditions required to classify the household as food insecure, the more likely it is that a household that is not really food insecure will be misclassified. A single food- insecure condition might reflect misunderstanding of the question or interviewer error (clicking “Yes” instead of “No” ). Finally, the questions refer to the previous twelve months. The number of people experiencing food insecurity or hunger on any given day is likely to be much lower than the number experiencing it at some time over the course of the year. For 2003, the Department of Agriculture reports that 11.2 percent of households were food insecure, including 3.5 percent that were food insecure with hunger. This figure includes 0.5 percent that were food insecure with hunger among children. Not surprisingly, food insecurity is much more common among poor households than among the nonpoor. More than one-third (35 percent) of poor households experienced food insecurity in the course of the year, and roughly one-quarter (24 percent) experienced hunger, compared with 8 percent and 2 percent for nonpoor households. This comparison is somewhat misleading because many near-poor households suffer from food insecurity. Among households with incomes above the poverty line but no more than 85 percent above the line, 23 percent are classified as experiencing food insecurity and 5 percent as experiencing hunger. Note that actual hunger declines with income much more rapidly than does food insecurity. The supplement used to measure food insecurity over the year also asked questions about food insecurity with hunger over the thirty days prior to the survey. Of course, fewer households experienced hunger over the previous thirty days than during some time in the previous twelve months. Just under 3 percent experienced hunger at some time during the previous thirty days, and each day roughly .5 percent of households were hungry.

50 Chapter 2 Jayanta Bhattacharya and his coauthors provide further evidence on hunger among poor families.27 They find that during cold weather, poor families reduce expenditures on food while other families increase expenditure. Among poor families, the reduction in food expenditure is roughly equal to the increased expenditure on fuel for home heating.

12. Alternative Measures of Poverty We have seen that there are significant problems with the official measure of poverty. The Census Bureau has estimated that adopting the National Academy of Sciences recommendations in 1997 would have increased the poverty rate from 13.3 percent to 15.4 percent.28 The Census Bureau now regularly produces poverty rates based on the National Academy of Sciences recommendations. Unfortunately, these rates were adjusted so that in 1993 they produced the same poverty rate as the official measure. Thus the estimates can be used to compare how the National Academy of Sciences and official measures change over time or to compare what they show about poverty rates among groups. They do not show how adopting the National Academy of Sciences recommendations would affect the level of the estimated poverty rate. Table 2.2 gives poverty rates in 2001 for different groups using three measures of poverty. The first is the official poverty rate that we have been studying. The second adjusts earnings for medical expenditures. Families who have medical expenditures that exceed the expected (or average) medical expenditures for a family with that composition have the additional expenditures subtracted from their income. Those with less than the expected expenditures have the difference added to their income. The last measure adjusts for differences in housing costs across regions in addition to adjusting for medical costs. We can see that accounting for medical costs has three substantial effects: 1. It increases the overall poverty rate. Many of the near-poor have significant medical costs. 2. It increases the poverty rate among the elderly relative to the young. After adjusting for medical costs, the poverty rate among the elderly exceeds the poverty rate among children. 3. It increases the poverty rate among Hispanics while decreasing it among blacks so that the poverty rate among Hispanics exceeds that among blacks.

27. Jayanta Bhattacharya et al., “Heat or Eat? Cold Weather Shocks and Nutrition in Poor American Families” (Working Paper 9004, National Bureau of Economic Research, Cambridge, MA, 2002). 28. Kathleen Short, Thesia Garner, David Johnson, and Patricia Doyle, Experimental Poverty Mea- sures: 1990 to 1997, U.S. Census Bureau, Current Population Report P60-205, Consumer Income (Washington, DC: U.S. Government Printing Office, 1999).

Who Is Poor? 51 Table 2.2 Poverty Rates for Alternative Measures of Poverty, 2001

Adjusted for Adjusted for Medical Medical Expenses and Official Expenditures Geographic Location

Total Poor 11.7 13.0 12.9

People Living in Families 9.9 11.1 11.1 In Married-Couple Families 5.7 7.1 7.1 In Female-Householder Families 28.6 28.1 28.3 In Families with No Wife Present 13.6 16.0 16.0

Age Younger than 18 16.3 15.3 15.4 18–64 10.1 11.3 11.3 65 or older 10.1 17.1 16.2

Race Non-Hispanic White 7.8 9.4 8.8 Black 22.7 22.2 21.8 Hispanic 21.4 23.1 25.9

Region Northeast 10.7 11.1 13.8 Midwest 9.4 10.4 9.0 South 13.5 15.4 13.0 West 12.1 13.5 16.1

Source: Bernadette D. Proctor and Joseph Dalaker, U.S. Census Bureau, Current Population Reports, P60-219, Poverty in the United States: 2001 (Washington, DC: U.S. Government Printing Office, 2002), table 8.

Adjusting for regional differences in housing costs changes the numbers somewhat: 1. It has little effect on the overall poverty rate. 2. It somewhat decreases the estimated poverty rate of the elderly. 3. It further increases the estimated poverty rate among Hispanics. 4. It increases the prevalence of poverty in the northeast and the west relative to the midwest and the south. The alternative measures shown in table 2.2 address only some of the criticisms of the official poverty rate. Gary Burtless and Timothy Smeeding have examined the effects of making some of the other changes recommended by the National Academy of

52 Chapter 2 Sciences.29 They treat near–income transfers as income but reduce income for employment-related costs. The addition of near-income is more important than the sub- traction of employment-related income, so their revised estimate of the poverty rate in 1998 falls to 12.0 from the official rate of 12.7. However, further adjusting the poverty threshold for medical costs significantly increases the poverty rate to 16.1. Burtless and Smeeding find less extreme effects on the relative poverty of the elderly and children, although their results point in the same direction as the Census Bureau results.

13. The Dynamics of Poverty Historically, when Americans thought about the poor, their images were of people who lived in Appalachia, dust bowl farmers, or ghetto dwellers mired in continual poverty. Increasingly, we have come to understand that the situation is complex. Many people experience quite short spells of poverty, but some experience very long spells of poverty. When we discussed the problems with the official measure of poverty, we noted that one year might be too long or too short a period to consider. Periods of low income far shorter than one year would result in considerable hardship for some people. Others might be able to experience a longer period of low income without substantially reducing their consumption. Does changing the time period we study alter our view of poverty? In a classic study that changed perceptions of poverty, Greg J. Duncan and his coauthors examined the poverty experience of roughly five thousand American families over a period of ten years.30 They found that the proportion of families experiencing poverty over this ten-year period was far greater than the annual poverty rate but that relatively few families experienced poverty for five or more years over the period. As a result, they concluded that poverty was a much more transitory experience than had previously been believed. In an updated version of the study, Rebecca Blank examined how much of the time people were poor over a period of thirteen years.31 She asked how many of the calendar years 1979–1991 each person was in poverty. She found that two-thirds of the people surveyed were never poor. One-sixth of all people surveyed, or about half of those who experienced poverty over the period, were poor for one to three years. Only about 5 percent of the people experiencing poverty (1.5 percent of the total population) were poor for all thirteen years. If we are prepared to call periods lasting no more than three years “short-term poverty,” it might appear that much poverty is short term. After all, half of Blank’s

29. Gary Burtless and Timothy Smeeding, “The Level, Trend, and Composition of Poverty,” in Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty (Cambridge, MA: Har- vard University Press, 2001), 27–68. 30. Greg J. Duncan et al., Years of Poverty, Years of Plenty: The Changing Economic Fortunes of American Workers and Families (Ann Arbor, MI: University of Michigan Press), 1984. 31. Rebecca M. Blank, It Takes a Nation (Princeton, NJ: Princeton University Press, 1997).

Who Is Poor? 53 sample of people experiencing poverty were poor for no more than three years over a period of thirteen years. However, Blank is careful not to draw that conclusion. The Duncan approach underestimates the importance of long-term poverty. To see this, suppose that all spells of poverty lasted exactly ten years. Most people would consider a poverty spell lasting a decade long term. To make the arithmetic easy, suppose that exactly one spell began each year from 1970 to 1991. The spell beginning in 1970 would have ended in 1979 and would have been recorded by Blank as lasting one year because she observed people starting in 1979. The spell beginning in 1971 would have ended in 1980 and been recorded as lasting two years. Continuing this process, we would observe twenty-two spells of poverty over the 1979–1991 period. Only four of these twenty-two spells would have been recorded as lasting ten years. Six would have been recorded as lasting three or fewer. This example is extreme. Obviously some poverty spells last for more than ten years and others for less time. The important point is that some of the people who were poor in 1979 had been poor for a long time. Even if they exited poverty over the next two years, their poverty spell would have been long. Similarly, some of those who were poor in 1991 would continue being poor for a long time. Even though they had not been poor for very long prior to 1991, their eventual poverty spell would be long. Thus, some of the people who were poor for three years or less over Blank’s sample period were at the beginning or end of much longer poverty spells. Thus, if we view poverty spells or repeated poverty spells that add up to more than three years as “long-term” poverty, one-half overestimates the fraction of the poor who experience short-term poverty. An- other reason that this type of analysis is misleading is that it focuses on poverty spells. Although it is true that most poverty spells are short, it is not true that most poverty is due to short spells. This might appear to be a contradictory statement, but it is not. Consider the following example. Suppose that every year, one person becomes poor and remains poor for exactly one year, and every decade, one person becomes poor and remains poor for exactly one decade. In such a world, between January 1, 1990, and December 31, 1999, ten people would experience poverty spells lasting one year and one person would experience a poverty spell lasting ten years. In this example, most spells are short—ten out of eleven spells last one year. The median duration of a spell is one year, and the average is about 1.8 years. However, if we ask how much poverty in our example is accounted for by long spells and by short spells, the answer is that every decade ten person-years of poverty are accounted for by the ten one-year spells and ten person-years are accounted for by the one ten-year spell. In other words, half of the poverty is due to the long spell. If instead of asking how long the average spell lasts, we ask how long the poverty spell will last for the average person who is currently poor, we get a very different answer. At any point in time, one person is in the middle of a one- year poverty spell, and one person is in the middle of a ten-year poverty spell. The average eventual duration of poverty for the currently poor in the example is 5.5 years. Which is the right perspective? To some extent, the answer is that it depends on why we are asking the question. People who are just entering poverty are not likely to remain poor for very long, but people who are currently poor, on average, will remain poor for a long time.

54 Chapter 2 Table 2.3 Distribution of Eventual Duration of Spells of Poverty

Eventual Duration People Beginning a Spell of Poverty People Presently Poor

1 Year 44.5 10.6 2 Years 15.8 7.6 3 Years 9.8 7.0 4 Years 6.2 5.9 5 Years 4.7 5.6 6 Years 2.8 4.0 7 Years 2.1 3.5 8 Years 1.0 2.0 9 Years 1.1 2.3 Over 9 Years 12.0 51.5 Average 4.2 12.3

Source: Mary Jo Bane and David Ellwood, “Slipping into and Out of Poverty: The Dynamics of Spells,” Journal of Human Resources 21 (Winter 1986), table 2.

Mary Jo Bane and David Ellwood have looked at the question both ways.32 Their results are summarized in table 2.3. They find that among people who have been poor for one year, the probability of not being poor the next year is 44.5 percent. Among those who have been poor for two years, 28.5 percent of the remaining 55.5 percent, or 15.8 percent overall, will not be poor the next year. So within two years, 60 percent of poverty spells will have ended. According to their calculations, the average poverty spell lasts a little over four years. However, if we ask what the eventual duration of poverty will be for the currently poor (or, equivalently, how much poverty is accounted for by spells of various lengths), Bane and Ellwood present a very different picture. Over half the poor are in the midst of spells that will last ten years or more. The average poor person is in the middle of a spell that will last 12.3 years. Yet even Bane and Ellwood in some sense underestimate the persistence of poverty. The problem is that many spells of nonpoverty do not last very long either. People who are on the edge of the poverty threshold are likely to slip back and forth between being above the poverty line and being below it. Ann Huff Stevens calculated that more than one-fourth of people who have been poor and escaped poverty for only one year will return to poverty the next year.33 Of those who have been out of poverty for exactly two years, one-sixth will return to poverty the following year, so almost 40 percent will be poor again within two years of leaving poverty. Almost two-thirds will experience another spell of poverty within ten years.

32. Mary Jo Bane and David Ellwood, “Slipping into and Out of Poverty: The Dynamics of Spells,” Journal of Human Resources 21 (Winter 1986): 1–24. 33. Ann Huff Stevens, “The Dynamics of Poverty Spells: Updating Bane and Ellwood,” Ameri- can Economic Review 84 (May 1994): 34–37.

Who Is Poor? 55 14. Why Does Poverty Last So Long for Some People? Many spells of poverty are short, but some people seem to be mired in poverty for very long periods, and the longer people have been in poverty, the less likely it seems that they will get out of it. People just entering poverty on average remain poor for less than five years. But people who have been poor for ten years, on average, remain poor for another ten years. Similarly, the probability of returning to poverty declines the longer one has been out of poverty. After one year, one-fourth of those who have escaped poverty return, but those who have escaped poverty for ten years have less than a 5 percent chance of returning to poverty the following year. There are at least three explanations for this:

1. Especially in the case of people who are poor only in one year, there is a good chance that some of those recorded as poor are not poor. The information has simply been misreported or misrecorded. Such people “leave poverty” at a very high rate. 2. The longer people have been poor, the harder it is for them to leave poverty. 3. People who have been poor for a long time have been in that state precisely because they are the sorts of people who have a hard time exiting poverty.

The second explanation is known as state dependence; the state (poverty or nonpoverty) a person is in directly affects the state that person will be in during the next period. The third explanation is simply due to differences across individuals and is known as individual heterogeneity. If we were able to account for all individual differences, we would, according to this explanation, find that among identical individuals, the probability of exiting poverty does not decline over time. To see how heterogeneity can make it seem as if the probability of exiting poverty declines over time, suppose that there are two types of people who enter poverty in equal numbers, say one thousand of each type. The quick-exit types have a 50 percent chance of getting out of poverty each year. The slow-exit types have a 25 percent chance of getting out of poverty each year. As we can see from table 2.4, even though no individual’s probability of exiting poverty declines over time, the pool of poor people becomes increasingly skewed toward those with a low probability of getting out of poverty. After eight years, in the example, almost all the remaining poor will be in the group that has greater difficulty exiting poverty. Distinguishing between individual heterogeneity and state dependence is both very difficult and very important for policy. It is difficult because in both cases we will observe a decline in exit rates from poverty (or nonpoverty) as an individual remains poor (or nonpoor) for a longer period of time. It is important because it affects our choice of policies. If the decline in exit rates is due to state dependence, we can derive a great deal of benefit from early intervention. By preventing people from staying poor for a long time and by supporting them in the early years of nonpoverty, we can get them out of the “poverty trap.” On the other hand, if the decline is due to individual heterogeneity, we have to decide on what type of people our efforts will most usefully be

56 Chapter 2 Table 2.4 Number of People in Poverty after N Years

N Exit Rate in Year Quick Leavers Remaining Slow Leavers Remaining

1 37.5 500 750 2 35.0 250 562 3 32.7 125 422 4 30.7 62 316 5 29.1 32 237 6 27.9 16 178 7 27.0 8 133 8 26.4 4 100 Infinity 25.0 0 0

focused. We do not wish to focus on those who will quickly exit poverty without our help. Whether we focus on the hard-core poor or those with more intermediate probabilities of exiting will depend both on the effectiveness of programs for different groups and on our values. Undoubtedly, both individual heterogeneity and state dependence play some role in generating the relation between time in poverty and the probability of leaving poverty. Some people pass through poverty as a very temporary, possibly even voluntarily en- dured state. Faced with real hardship, they are able to obtain reasonable employment rapidly. But this is not the case for all poor people. So there is some heterogeneity. At the same time, it would be surprising if there were no state dependence. The longer a person is in poverty, the less likely she is to have maintained the personal contacts and networks that will help her escape from poverty. Employers will be less recep- tive to an applicant who lacks a steady work history. Unfortunately, we have no compelling studies of the relative importance of these two explanations of why the probability of exiting poverty declines as the length of a poverty spell grows. Now for an aside. The failure or inability to distinguish between state dependence and individual heterogeneity is important for a wide range of public policies, not just poverty policy. Police departments have to decide how quickly to take up missing persons cases. Many police forces require that an individual have been missing for 48 hours before investigating. Police departments that use a 24-hour rule instead have a much higher clearance rate, that is, they find a much higher proportion of missing persons. But they also waste more resources on people who show up on their own or would have shown up. The critical issue is how much of the increased clearance rate is due to people who would have been found anyway but simply were not recorded as missing and how much is due to the benefit of investigating earlier. The question of whether persistence in a given state is caused by state dependence or individual heterogeneity arises in many statistical debates. Perhaps you went to a high school with tracking. If so, you undoubtedly observed that students who were in

Who Is Poor? 57 the most advanced tracks in the first year of high school were more likely than other students to be in the most advanced tracks in their senior year. Some people moved from less advanced tracks to more advanced tracks, and some (probably more frequently) moved from more to less advanced tracks, but this was not sufficient to offset the statistical advantage of starting in an advanced track. Opponents of tracking argue that it permanently traps students initially placed in low tracks (state dependence). Proponents of tracking dismiss the persistence of tracks as due to individual heterogeneity. They argue that if the tracking is done right, most of the students initially placed in high tracks will belong there and thus remain, and the same will be true for those placed in low tracks.

15. Concluding Remarks There is probably no definition of poverty that simultaneously perfectly captures our notions of who is poor and is workable for studying poverty at a national level. Gov- ernment cannot distinguish between the young woman who is happily living with her aunt and the one who is working for her aunt. Nor is it likely to be able to distinguish between the graduate student couple and the couple where the husband works at a fast food restaurant. Even if government could obtain this information, it is unlikely that we would want it to pry into our lives or those of the possibly poor to the degree necessary. Yet even recognizing these limitations, the official definition of poverty falls short. It is likely that at some point in the not too distant future, the official poverty measure will be revised to take account of many of the issues raised in the National Academy of Sciences report. When it is revised, some of our perceptions of who is poor and how the poverty rate has changed over time will also be revised. The lesson to be learned from all this is not that government statistics are misleading. Instead, it is that when using statistics, it is important to understand how they have been derived and how our interpretation must take this derivation into account. Is the poverty rate higher among the elderly than among nonelderly adults? We have seen that this critically depends on how we define poverty. It is equally important to understand how policy analysts derive their statistics. Is poverty long term or short term? The answer depends not only on what we mean by poverty and by short term and long term but also on whether we are talking about the experience of the typical person becoming poor or the experience of the typical person who is currently poor. We will continue with this theme in the next chapter, where we will examine how shifts in poverty policy may have affected measured poverty differently from actual poverty.

16. Further Reading Bane, Mary Jo, and David Ellwood. “Slipping into and Out of Poverty: The Dynam- ics of Spells.” Journal of Human Resources 21 (Winter 1986): 1–24.

58 Chapter 2 Burtless, Gary, and Timothy Smeeding. “The Level, Trend, and Composition of Poverty.” In Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty. Cambridge, MA: Harvard University Press, 2001, 27–68. Citro, Constance F., and Robert T. Michael, eds. Measuring Poverty: A New Approach. Washington, DC: National Academy Press, 1995, 1–89. DeNavas-Walt, Carmen, Bernadette D. Proctor, and Robert J. Mills. Income, Poverty, and Health Insurance Coverage in the United States: 2004. U.S. Census Bureau, Current Population Report P60-229. Washington, DC: U.S. Government Printing Office, 2005. Fisher, Gordon M. “The Development and History of the U.S. Poverty Thresholds.” http://www.ssa.gov/history/fisheconpoverty.html. Fuchs, Victor R. “Redefining Poverty and Redistributing Income.” Public Interest 8 (Summer 1967): 88–95. Jencks, Christopher. The Homeless. Cambridge, MA: Harvard University Press, 1994. Nord, Mark, Margaret Andrews, and Steven Carlson. Household Food Security in the United States, 2002. Food Assistance and Nutrition Research Report 35. Washing- ton, DC: Food and Rural Economics Division, Economic Research Service, U.S. Department of Agriculture. “Revising the Poverty Measure.” Focus 19 (Spring 1998): 1–20 and 47–52. http:// www.ssc.wisc.edu/irp/focus/focus.htm. Sen, Amartya. “Poor Relatively Speaking.” Oxford Economic Papers 2, no. 35 (July 1983): 153–69. Stevens, Ann Huff. “The Dynamics of Poverty Spells: Updating Bane and Ellwood.” American Economic Review 84 (May 1994): 34–37.

17. Questions for Discussion 1. The World Bank defines people as poor if their incomes are less than two dollars per day. Does this mean that the United States has almost entirely eliminated poverty? 2. What evidence is there that people adjust their perceptions of what constitutes poverty in response to changes in overall standards of living? 3. For each of the following people, explain whether they are poor under the official U.S. definition.

• A homeless schizophrenic man who has $500,000 in a bank account but lives off the money he receives from returning bottles that he finds for the deposit. • A young woman who lives with her wealthy aunt in a beautiful home. She receives free room and board. She and her aunt enjoy each other’s company. She accompanies her aunt to various social events, such as the theater, and she happily helps her aunt by doing various chores around the house (shopping, some light cleaning). When she needs money, her aunt gives it to her. In a typical year, her aunt gives her $3,000.

Who Is Poor? 59 • A similar young women who is not related to the owner of the house. • A graduate student couple that lives on a total of $11,000. They share a room in a group apartment. The couple pays $200 per month for the room. • That same couple if they spend more than one-third of their income on food. • That same couple if they spend less than one-third of their income on food. • An 80-year-old widow who has lived for sixty years in a house that is now worth $500,000. Her total income is barely adequate to pay the taxes on the house. She survives by keeping the house heated to no more than 60 degrees in the winter, by skipping meals or eating at the local soup kitchen. There are no inexpensive houses in her neighborhood, which is the only place where she has a network of friends and feels comfortable.

4. A group living in the United States in 2004 consisted of fifteen households with no elderly members. There were five single individuals living separately from each other. Their incomes were $7,000, $7,500, $8,000, $9,000, and $10,000, respectively. There were five married-couple households with incomes of $10,000, $11,000, $12,000, $13,000, and $14,000, respectively. And there were five two-parent, two-children households with incomes of $10,000, $15,000, $20,000, $25,000, and $30,000, respectively. What percentage of these households were poor? What was the poverty rate among this group? 5. Describe the principal criticisms of the official definition of poverty. For each, explain whether correcting it would lower or raise the official poverty estimate. 6. A family receives $100 per month in food stamps. It spends $150 per month on food that can be purchased with food stamps. How much are the food stamps worth relative to cash for the family? 7. A different family also receives $100 per month in food stamps but only spends $100 per month on food that can be purchased with food stamps. There is an active market in which food stamps can be sold (without fear of legal or social sanctions) at 80 cents per dollar of food stamps. What can we say about the value of the food stamps to the family? 8. Which sociodemographic groups (e.g., age groups, racial or ethnic groups) have the highest poverty rates? Which are a majority of the poor? Do these answers differ for people in extreme poverty? 9. Most poverty spells are short, but most of the poor are in long poverty spells. Explain. 10. One well-known research study looks at the distribution of years spent in poverty over a period of fourteen years. Explain why this underestimates the frequency of long-term poverty.

60 Chapter 2 11. We observe that people who have been poor for two years are more likely to be poor the following year than are people who have been poor for only one year. Why?

18. Appendix: A Brief Note on Data Many of the data referred to in this book are drawn from published sources. However, on occasion we will require information that has not been published by the Census Bureau, the Bureau of Labor Statistics, other government departments, or researchers. In those cases, I am sometimes able to obtain the information by calculating it myself from the original data. I rely on two principal data sources: the Current Population Survey (CPS) and the National Longitudinal Survey of Youth, 1979 (NLSY79).

CPS. We discussed the CPS briefly in section 7 of this chapter. As noted there, the primary purpose of the CPS is to measure employment and unemployment. It is based on a large sample, currently covering over 60,000 households. The CPS is based on a sample of dwelling units. Each dwelling unit that enters the sample is surveyed for four consecutive months, then dropped from the sample for the next eight months, and finally surveyed again for four consecutive months before being dropped permanently from the sample. In addition to basic information on the employment status of all persons aged sixteen and older living in the dwelling unit, the CPS gathers other valuable information. As discussed earlier, the Annual Demo- graphic Supplement is administered in March and collects information on earnings and employment for the previous calendar year, among other items. The December supplement collects information on food insecurity and hunger. In addition, in the fourth and eighth months that a dwelling unit is in the sample, information is collected about residents’ labor market earnings in the previous week and about their educational attainment. CPS data are available from a number of sources. The raw data for the surveys beginning in January 1994 can be downloaded from the Bureau of Labor Statistics (http://www.bls.census.gov/cps/datamain.htm), as can the Annual Demographic Sup- plement going back to March 1992 and the food security supplement going back to 1995. The National Bureau of Economic Research (http://www.nber.org/data/cps_ index.html) provides data from 1976 on and the Annual Demographic Supplement from 1962 (with the exception of 1970). It also sells a CD for $100 with data from the outgoing rotation groups starting in 1979. The so-called MORG files on CD are much easier to use than the basic monthly data. Unicon Research (http://www.unicon.com) has also developed CDs that simplify working with the CPS. They provide data and tools for extracting the data from, among others, the Annual Demographic Supplement, the food security supplement, and the outgoing rotation groups. Prices vary according to the affiliation and status of the purchaser.

Who Is Poor? 61 NLSY79. The sample for the NLSY79 was selected in 1979, when respondents were 14 to 21 years old, although some had turned 22 by the time of the actual interview. The original sample of 12,686 individuals consisted of a national cross-section sample plus supplemental samples of blacks, Hispanics, and poor whites as well as military personnel. Individuals were initially interviewed every year through 1994, and since then have been interviewed every other year. The military sample was dropped after 1984, and the poor white sample was dropped after 1990. Beginning in 1986, children of women in the NLSY79 sample were included in the study. There is extensive information on both the original NLSY79 sample and these children. The researchers have collected information on a broad variety of topics, including complete fertility histories, complete family composition histories, and information on cognitive performance, labor market outcomes, depression, alcohol use, sources of income, and so on. Data may be ordered or downloaded through http://www.nlsinfo.org/.

62 Chapter 2 chapter 3

The Evolution of Poverty Policy

I    teach students that it is better to give someone money than to give him a good of some kind, such as housing or food. Yet in this chapter we will see that over the past few decades, U.S. policy has shifted away from reliance on traditional welfare in the form of cash grants. It initially moved toward the use of in-kind transfers of goods and later toward subsidizing workers. What happened? Did foolish policy makers ignore the insights of economics? Or did economists have it wrong? The answer is probably a little bit of both. In some cases, as we have seen, in-kind transfers are nearly indistinguishable from cash. In other cases, policy makers were concerned that traditional welfare discouraged work. We will see that, perhaps surprisingly, subsidizing work reduces work effort by some workers. But it is likely that policy makers cared more about getting more people into the work force than they did about reducing work effort among those already working. In addition, in-kind transfers have other advantages, especially when identifying who is poor is difficult.

1. Federal Poverty Programs, 1970–2000 There are a large variety of programs primarily oriented toward the poor. Table 3.1 shows the evolution of spending in seven major categories. The first column shows spending on what most people think of as “welfare.” This program was originally known as Aid to Dependent Children and later as Aid to Families with Dependent Children (AFDC). After welfare reform, it was substantially revised and became Transitional Assistance to Needy Families (TANF). We can see that welfare spending increased rapidly between 1970 and 1975. After that, it stagnated for over twenty years and then declined dramatically after the passage of the welfare reform act in 1996. AFDC/TANF is one of the antipoverty programs that most effectively targets the poor. John Karl Scholz and Kara Levine estimate that 87 percent of all AFDC/

63 Table 3.1 Expenditure on Assorted Federal Antipoverty Programs (Billions of Real $1999)

1970 1975 1980 1985 1990 1995 1999

AFDC/TANF 17.5 26.0 24.2 22.6 23.6 24.1 13.4 Disability Insurance 13.2 26.1 31.2 29.2 31.6 44.7 51.3 Supplemental Security Income 12.6 18.2 16.1 17.1 20.5 30.7 29.7 Food and Nutrition 5.3 19.9 26.4 24.8 23.8 34.8 26.1 Housing 2.2 6.6 11.1 17.7 19.7 30.0 27.6 Medicaid 20.8 39.1 52.1 63.4 92.4 171.0 188.8* EITC 0.0 3.9 4.0 3.2 9.6 28.4 31.9

Source: John Karl Scholz and Kara Levine, “The Evolution of Income Support Policy in Recent Decades,” in Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty (Cam- bridge, MA: Harvard University Press), table 6A-1. *$1998.

TANF funds go to people who would be poor in the absence of the transfer. Moreover, 84 percent of the funds fill the poverty gap, that is, they reduce the additional funds families would need to be above the poverty cutoff.1 The 3 percent difference reflects the fact that 3 percent of transfers go to people who would be poor in the absence of transfers but raise these families to an income level that exceeds the poverty cutoff. The largest increase in cash transfer payments oriented toward the poor has been in the disability insurance program, which has grown by almost $40 billion, quadrupling in less than thirty years. Disability insurance is designed to provide social assistance to individuals who are unable to work for over a year because of a physical or mental disability. Because it is part of the social security program, in order to be eligible, individuals must have worked for a certain amount of time. Like social security, disability insurance is not primarily focused on the poor. However, the benefit rules favor the poor and, as in the case of social security, reduce the poverty rate. David Autor and Mark Duggan argue that the rise in disability insurance payments reflects two factors.2 First, disability eligibility rules were liberalized in 1984 so that more people qualified. Second, disability benefits are indexed to the mean wage in the economy. As the earnings of low-skilled workers fell relative to the average wage, disability payments became attractive relative to employment, especially for older workers experiencing a period of unemployment. Supplemental security income (SSI) serves low-income individuals who are either elderly or disabled. According to Scholz and Levine, it effectively targets the pre-

1. John Karl Scholz and Kara Levine, “The Evolution of Income Support Policy in Recent Decades,” in Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty (Cam- bridge, MA: Harvard University Press, 2001), 193–228. 2. David H. Autor and Mark G. Duggan, “The Rise in the Disability Rolls and the Decline in Unemployment,” Quarterly Journal of Economics 118 (February 2003): 157–205.

64 Chapter 3 transfer poor, albeit less effectively than AFDC/TANF. They calculate that 77 percent of SSI transfers are to the pretransfer poor and 69 percent serve to reduce or eliminate the poverty gap. AFDC/TANF, disability insurance, and supplemental security income are all counted as income for purposes of calculating the official poverty rate. All three were essentially flat from 1975 to 1985. Adding the three together, spending rose from $70.3 billion in 1975 to $75.7 in 1990. This rate of growth was less than the U.S. population growth rate. Thus, between 1975 and 1990, the principal cash transfer programs that would have reduced the official poverty rate grew less rapidly than the U.S. population and shifted toward programs that targeted the poor less. Had nothing else in the economy changed, the official U.S. poverty rate would have risen on the basis of this policy change alone. This point is reinforced by looking at the evolution of average AFDC/TANF monthly payments over this period. The average monthly benefit per recipient, in 2002 dollars, rose from $171 in 1962 to $224 in 1978 and then fell steadily. By 1990, it was down to $187, and by 2002 to $143.3 This, of course, does not mean that there was no increase in assistance to the poor. As we can see from the table, from 1975 to 1990, food and nutrition programs, housing, and Medicaid all grew significantly. Food aid increased by about 20 percent. Housing aid tripled, and Medicaid more than doubled. Although a significant part of this increase was due to the relative increase in medical costs, much of it represented an expansion of services. In the 1990s food stamp spending was flat, but there were significant continued increases in spending on housing and Medicaid. These programs are all in-kind transfer programs. As a result, they do not affect the official poverty rate. While the programs that affect the official poverty determination were stagnating, in- kind programs were growing significantly. The earned income tax credit (EITC) gives money to workers with low incomes. As such, it does not help the nonworking poor but increases after-tax income for the working poor. The EITC is viewed as a reduction in tax liability. Because the official definition of poverty is based on pretax income, the EITC is not counted in the official determination of poverty. However, because most poor families pay no income tax, the earned income tax credit is a kind of transfer. Moreover, because families care about their after-tax income, not their before-tax income, even if the EITC served only to reduce taxes, it would make families better off. As we can see, the EITC increased dramatically. However, because it targets low- income working families, the EITC is less effective in eliminating the poverty gap than are more traditional welfare programs. Scholz and Levine estimate that 61 percent of the EITC goes to pretransfer poor and that 54 percent serves to reduce or eliminate the poverty gap.

3. U.S. Department of Health and Human Services, Indicators of Welfare Dependence Annual Report to Congress, 2004 (Washington, DC: U.S. Department of Health and Human Services), table TANF-6.

The Evolution of Poverty Policy 65 Why did poverty programs shift from an emphasis on cash transfers to in-kind transfers and earned income credits? To understand this, we need to look carefully at how each type of program affects behavior.

2. Incentives under AFDC Until it was abolished as part of the Personal Responsibility and Work Opportunity Reconciliation Act (more commonly known as the Welfare Reform Act) of 1996, Aid to Families with Dependent Children was what most people meant when they referred to welfare. As mentioned earlier, AFDC was originally Aid to Dependent Children.4 Its original goal was to aid needy children who lived with their single mothers, although implicitly it aided their mothers as well. In 1950, support for the single mothers was made explicit. In 1962, the rules were changed so that states were allowed to provide aid to two-parent families in which the father was unemployed, and the new name was adopted. Later states were required to extend aid to two-parent families. Eligibility rules and levels of support were determined by individual states within broad guidelines established by the federal government. A key element of AFDC was how it treated other sources of income. Prior to 1968, except for a small allowance for work-related expenses, AFDC payments were reduced by $1 for each dollar of income from other sources. In 1968, the law was changed so that AFDC recipients could keep the first $30 they earned. Beyond $30, AFDC payments were reduced by $2 for every $3 earned. This meant that a family with the average AFDC payment of $162 could earn $273 per month before completely losing AFDC eligibility. Given the median earnings of $4,568 for full-time, year-round women workers,5 many single mothers would not lose their entire eligibility if they worked. In 1981, Congress again changed the rules so that after four months of work, AFDC payments were reduced by $1 for each dollar earned except for a $75 allowance and actual child care expenses. Families who anticipate having no more income as a result of working have little incentive to work and have a strong incentive to hide their earnings from the government. AFDC could cause women to drop out of the labor market even if, had they remained in the market, they would have earned noticeably more than their AFDC payment. Let us consider the case of a woman who, if she works, earns $7.50 per hour and pays $3.00 per hour for child care. Her transportation costs $2.00 per work day, which for the sake of simplicity we will treat as 25 cents per hour. We will ignore other taxes and subsidies.

4. The description of the history of AFDC in this section is based on Assistant Secretary for Plan- ning and Evaluation, U.S. Department of Health and Human Services, Aid to Families with Depen- dent Children: The Baseline, June 1998, http://aspe.hhs.gov/hsp/AFDC/baseline/1history.pdf, chapter titled “A Brief History of the AFDC Program.” 5. http://www.census.gov/hhes/income/histinc/p36.html.

66 Chapter 3 Net Monthly Income ($)

1,000

800

600 With AFDC

400

200

Without AFDC 0 0 50 100 150 200 250 Hours Worked per Month Figure 3.1 Income and Hours Worked by a Fictional AFDC Recipient

Suppose, first, that she is not eligible for AFDC. In the absence of AFDC, after paying for transportation and child care, she has a net income of $4.25 per hour. If she does not work at all, she consumes 720 hours of leisure (assuming a thirty-day month) but cannot buy any goods. If she chooses to work one hour, she will consume 719 hours of leisure and $4.25 of goods. At the other extreme, if she chooses to work 720 hours, she will consume no leisure and $3,060.00 of goods. Of course, neither extreme is realistic. If the woman in our example does not work at all, she will soon starve. If she works all the time, she will soon die of exhaustion. We are merely enu- merating the trade-offs. We represent these potential trade-offs between hours of work and income by a straight line called the budget line. This line is the straight (dashed) line in figure 3.1. We have not shown all the possibilities, but only those involving 250 hours of work per month or less. Now suppose that the woman in our example is eligible for AFDC payments of up to $500 per month. Her new budget line is shown as the solid line in figure 3.1. If she does not work at all, she receives $500 in AFDC payments. For the first few hours she works each month, she receives a net wage of $4.25 per hour ($7.50 minus $3.00 for child care and 25 cents for transportation). She does not lose any of her benefit until she has worked seventeen hours, because the government increases her allowable earnings by $3.00 per hour to cover her child care costs. Beyond this point, her benefit is reduced by $4.50 per hour (her earnings minus child care costs), but her net income is only $4.25 per hour because of the additional transportation expense she bears. If she works a little more than 125 hours per month, she will fully exhaust her AFDC benefit. Beyond this point, her net monthly income begins to rise again and the budget line is identical to the budget line without AFDC. If she works 175 hours per month (more

The Evolution of Poverty Policy 67 than the standard forty hours per week given holidays and vacations), her net gain relative to not working is less than $1 per hour. We would not be surprised if some women, like the one described in our example chose to forgo such meager additional income in order to spend more time with their children and/or more time on other leisure pursuits. This would be even more true if the woman could earn a little extra money without reporting it to the welfare authorities. In interviews with welfare mothers in the late 1980s and early 1990s, Kathryn Edin and Laura Lein found that few mothers survived on welfare alone. AFDC, food stamps, and supplemental security income (SSI) provided an average of 64 percent of their income. Reported work provided only 2 percent of their income, while “underground” work and other work that they did not report to the welfare authorities provided 14 percent. They obtained most of the rest of their income through family, friends, boyfriends, and absent fathers.6 Surprisingly, Edin and Lein did not find that low-wage women workers were better off than their sample of women receiving AFDC. Working mothers had substantially higher transportation and child care expenses than did the AFDC recipients. Their main job, on average, covered only 63 percent of their expenses. They supplemented this with a further 7 percent from food stamps and the EITC (discussed later). The remaining 30 percent came primarily from sources similar to those used by AFDC recipients: unreported work, friends, family, boyfriends and absent fathers, and social agencies.7 As the history of AFDC reveals, any cash transfer program faces a difficult trade- off. We cannot simultaneously provide a large basic transfer and strong incentives for work while restricting transfers to low-income families and individuals. Suppose we want to guarantee everyone at least half of the median income. If we want to encourage people to work at low-wage jobs, we should probably not take back more than one- third of their earnings. If we consider someone earning six dollars an hour, that means letting them keep four dollars an hour. But if we reduce the transfer payment by only one-third, we will be giving at least some transfer to all families with less than one and a half times the median income, but that will be roughly 70 percent of families.8 If we wish to limit transfers to families in the bottom third of the income distribution, then we will have to limit our basic transfer to about one-fourth of the median family income. Alternatively, we could maintain the basic transfer but reduce the transfer by close to one dollar for each dollar of income. But, of course, if people are not allowed to keep what they earn, they are less likely to work. The shift from cash transfers to tax credits and in-kind transfers is, in part, a response to concerns about the work disincentives caused by AFDC. In the following sections, we examine how work incentives differ for these approaches to poverty.

6. Kathryn Edin and Laura Lein, Making Ends Meet (New York: Russell Sage, 1997), table 2-6. 7. Edin and Lein, table 4-6. 8. http://ferret.bls.census.gov/macro/032002/faminc/new07_000.htm. The calculation is imperfect, because we should adjust for family size. However, the principal point is correct.

68 Chapter 3 3. The Earned Income Tax Credit As we have seen, the EITC has grown to be a significant antipoverty program, surpass- ing TANF as a source of cash transfers to the poor. John Karl Scholz and Kara Levine estimate that the EITC reduces the poverty gap by about 5 percent, and on its own would lift only a little more than 4 percent of the pretransfer poor out of poverty.9 The EITC is an unusual antipoverty measure in that it works through the tax system. The EITC reduces a family’s tax liability by an amount that depends on earnings. For families whose credit exceeds their tax liability, the credit is refundable, which means that families can actually receive a payment from the government rather than pay the government taxes.10 By administering the program through the tax system, the government saves a great deal in administrative costs. Although initially many people who were not entitled to one claimed an EITC, familiarity with the program, refinement of the eligibility criteria, and improved monitor- ing on the part of the Internal Revenue Service (IRS) have reduced false claims to a level comparable to those in the rest of the tax system.11 Nevertheless, it is worth noting that the IRS estimates significant overpayments to EITC recipients that are not recovered through IRS audits or administrative procedures. It estimates that about 27 percent to 32 percent of payments are either made to ineligible recipients or made as excess payments to eligible recipients. It does not provide estimates of payments not made to eligible recipients who fail to claim the credit.12 Although it is possible to receive advance payment of the EITC, relatively few of those who are eligible request advance payment. Instead, most use it as a way of saving and receive the EITC as a tax refund after they file their tax returns. For this reason, many authors believe that it is hard for low-income workers to figure out the true relation between their incomes and the amount of the tax credit.13 The tax credit is not available to people with asset income of $2,650 or more. Be- cause relatively few low-income workers receive significant asset income, in discussing

9. Scholz and Levine. 10. This should not be confused with a tax refund. Over the course of the year, people pay taxes. After the end of the calendar year, they determine how much they should have paid. If they have paid more than they should have, they can obtain a refund of taxes they have already paid. However, typically they are not entitled to a refund that exceeds all of the taxes they have paid. In other words, total taxes paid for the year cannot be negative. The EITC is refundable in the sense that even if workers’ total tax liability for the year is zero, they are still entitled to receive the tax credit and thus receive payments rather than make payments through the tax system. 11. See the discussion of the literature in V. Joseph Hotz, Charles Mullin, and John Karl Scholz, “The Earned Income Tax Credit and Labor Market Participation of Families on Welfare,” in Bruce Meyer and Greg Duncan, eds., The Incentives of Government Programs and the Welfare of Families, http://www.jcpr.org/book/pdf/IncentivesHotzChap3.pdf. 12. Internal Revenue Service, U.S. Department of the Treasury, Compliance Estimates for Earned Income Tax Credit Claimed on 1999 Returns (Washington, DC: Internal Revenue Service, February 28, 2002). 13. Jeffrey Liebman, “The Impact of the Earned Income Tax Credit on Incentives and Income Distribution,” Tax Policy and the Economy 12 (1998): 83–119.

The Evolution of Poverty Policy 69 the incentive effects of the EITC, we will assume that all of the income comes from earnings. For fiscal year 2004, the EITC was as follows:14 • For tax filers with no children, the benefit rate was approximately 7.65 percent. The maximum benefit was $390 at $5,100 and remained at that level until the filer’s income was $6,400. The credit was reduced by approximately 7.65 percent for earnings beyond $6,400 and reached zero at $11,490. In effect, earners with incomes below $5,100 had their social security, disability, and Medicare taxes returned to them. • For tax filers with one child, the benefit rate was approximately 34 percent. The maximum benefit of $2,604 was reached at an income of $7,650 and continued to an income of $14,050. Beyond this level, it was phased out at a rate of approximately 16 percent, hitting $0 at $30,338. • For tax filers with two or more children, the benefit rate was approximately 40 percent. The maximum benefit was $4,306 in the range $10,750 to $14,050, after which the credit was reduced at a rate of approximately 21 percent, hitting $0 at $34,458. • For married couples filing jointly, the range in which the credit was flat and the maximum income at which any credit was received were $1,000 higher. It is important to recognize that earners in the phase-out range of the EITC face very high implicit marginal tax rates. In 2004, a single parent with an earned income of $30,000 would have paid 15 percent federal income tax, 7.65 percent social security and related taxes, and 21 percent lost EITC. This is a higher marginal tax rate than that faced by someone with an earned income of $150,000.

3.1. The EITC and Labor Supply. To understand how the EITC influences labor supply, let us ignore other tax and transfer programs and consider the effect of the EITC in isolation. We will consider the trade-off between leisure and other goods. Let us say that, in the absence of EITC, a woman is paid an hourly wage of $10 per hour. If she does not work at all, she will consume 8,736 hours of leisure (the total number of hours in a year) but will not buy any goods. If she chooses to work one hour, she will consume 8,735 hours of leisure and $10 of goods. At the other extreme, if she chooses to work 8,736 hours, she will consume no leisure and $87,360 of goods. This line is the straight (dashed) line in figure 3.2. Let us now consider what happens in the presence of the EITC. For concreteness, suppose that the woman in our example has two children and no husband. There are four ranges to consider—where the credit is increasing, where it is flat, and where it is decreasing. In addition, there may be effects on those who do not work in the absence of a credit or on those who earn more than the maximum at which they would be eligible for the credit.

14. See IRS publication 596.

70 Chapter 3 Income ($)

With EITC

Without EITC

Hours of Work Figure 3.2 Budget Lines with and without EITC Note: Figure is schematic, not to scale.

Increasing subsidy. For the first 1,075 hours that the woman in our example works, she will receive a subsidy of roughly 40 percent and so will earn $14 per hour instead of $10, and her total income will be higher than in the absence of the EITC. Flat subsidy. From 1,075 hours to 1,405 hours, the woman’s tax credit will neither rise nor fall. She will keep the full $10 for each additional hour she works and will have a total income that is $4,300 higher than she would earn in the absence of the EITC. Declining subsidy. For 1,405 or more hours, her subsidy will go down by $2.10 for every $10.00 she earns, so her net wage will be $7.90 an hour. However, her income will still be higher than it would be in the absence of the EITC. No subsidy. Beyond 3,546 hours, the woman’s earnings will be unaffected by the credit. As we will see, this does not necessarily mean that if she were to work more than 3,546 hours in the absence of the credit, the credit would not affect her labor supply. Figure 3.2 also shows the budget line with the tax credit. From 0 to 1,075 hours, it is steeper than the budget line without the credit because the woman’s after-credit hourly earnings will be higher with the credit. Between 1,075 and 1,405 hours, the new budget line is above the straight budget line but parallel to it. This reflects the fact that the woman will keep the $10.00 she earns each hour in this range and will receive the maximum credit, but the credit will not change as she increases her hours. From 1,405 to 3,546 hours, the budget line with the credit is less steep than the straight line, because for each hour the woman works, the credit will be reduced and her net wage will

The Evolution of Poverty Policy 71 be only $7.90. However, because she will receive some credit until she works the full 3,546 hours, the budget line with the credit is above the budget line without the credit.

3.1.1. Income and Substitution Effects. How will the labor supply of the woman in our examples be affected in each of these ranges? Think about the following question. Suppose you were self-employed and could choose how much to work. Would you work more if you could earn $10 per hour or $500 per hour? Answering the question is not easy, because there are two conflicting effects. If you could earn $500 an hour, you would be much richer. That would make you want to sit back and enjoy more leisure time. On the other hand, at the higher pay rate, an extra hour of work would buy a lot more goods. Economists call these two effects the income and substitution effects. The income effect is how much people would change their work time in response to an increase in their income without any change in their hourly wage rate. When people win the lottery, their wage rate does not change, but typically they work less.15 In general, we believe that the income effect is negative: increases in income, holding the wage rate constant, reduce labor supply. We can also consider the opposite experiment. Suppose you worked 1,000 hours per year at $500 and therefore earned $500,000 per year. Now, suppose we gave you $490,000 per year but reduced your wage rate to $10 per hour. If you kept working 1,000 hours per year, you would still be earning $500,000. But most people would not find it worthwhile to work 1,000 hours in this case. Because the wage rate fell, they would substitute leisure for goods. The substitution effect reflects the fact that the slope of the budget line has changed. Increasing the slope of the budget line (increasing the wage rate) while holding income constant at the hours a person originally worked increases the number of hours worked. Thus, we say that the substitution effect is positive. To summarize, when a worker’s net wage increases, the substitution effect will cause the worker to buy less leisure and thus to work more, because leisure will have become expensive relative to other goods. Thus, we expect that the substitution effect on labor supply will be positive. When a worker’s income goes up, we expect the individual to purchase more of all goods, including leisure, and thus expect the effect on labor supply to be negative.

3.1.2. Income and Substitution Effects of the EITC: Theory. We are now in a position to examine the effect of the EITC on labor supply. We must look at different ranges of the budget line. 1. If, in the absence of the EITC, an individual would be out of the labor force, there would be a pure substitution effect; because workers do not receive

15. Guido W. Imbens, Donald B. Rubin, and Bruce Sacerdote, “Estimating the Effect of Un- earned Income on Labor Earnings, Savings, and Consumption: Evidence from a Survey of Lottery Players,” American Economic Review 91 (September 2001): 778–94.

72 Chapter 3 any extra income if they do not work, there would be no income effect. Because the substitution effect makes people more likely to work, we anticipate that the EITC will increase the fraction of workers who work. 2. If an individual would otherwise work in the range in which the credit is increasing, there would be both an income and a substitution effect; the worker would receive $14 instead of $10 for each hour worked, and the total quantity of goods and leisure the worker could purchase would be higher in the presence of the EITC. Although, in principle, the labor supply effect could be positive or negative, most labor economists would expect some small positive effect on labor supply 3. If without the EITC the worker would choose hours and thus earnings in the range in which the credit was constant, there would be no effect on his net hourly wage, so there would be no substitution effect. Instead, there would be a pure income effect, and thus the effect on labor supply should be negative. 4. If the worker’s choice of hours and earnings in the absence of the EITC would put him in the range in which the credit was declining, the income effect and substitution effect would both reduce labor supply. 5. The EITC could also affect people whose choice of hours and earnings in the absence of the EITC would put them in the range in which they would not receive any credit. Suppose that our worker would, in the absence of the EITC, work 3,547 hours. We know that the leisure he would give up by working a 3,548th hour would be worth more than $10.00 to him.16 Otherwise, he would have chosen to work more. We know that the leisure he would give up by working the 3,547th hour would be worth less than $10.00 to him. Otherwise, he would not work the last hour. It is a good bet that the last hour worked would be worth just about $10.00 and that the last two hours would be worth very close to $20.00. Now suppose that the EITC were established. It would not be worth it for the worker to reduce his hours of work by one hour. He would still give up $10.00 for an hour of leisure worth a little less than $10.00. But if he were to cut back working by two hours, he would give up only $17.90 ($10.00 for hour 3,547 and $7.90 for hour 3,546) instead of $20.00. Because the two hours of leisure would be worth close to $20.00, with the EITC he would work less, even though in the absence of the EITC his earnings and number of work hours would be high enough to make him ineligible for the EITC. The hours any individual must work to be in each particular segment of the EITC depend on the worker’s wage. A worker with two children who earns twenty dollars per hour will be beyond the phase-in range of the EITC if he works as little as ten hours

16. I am ignoring taxes and the costs of working. They can be incorporated into the explanation but only add complexity.

The Evolution of Poverty Policy 73 per week. The same worker must work almost full time for the entire year if he earns only the minimum wage. As a result, the workers who are in the segment of the EITC where we predict that labor supply will be increased tend to be workers with relatively low hourly earnings, while those in the middle section tend to have somewhat higher hourly earnings and those in the phase-out section tend to have even higher earnings. In addition, we must pay for the EITC by raising taxes on workers with yet higher incomes. We saw that this could increase or decrease their labor supply. Most studies suggest that, at least among relatively high-earning male workers, labor supply does not respond much to taxes. However, the increased tax to finance the EITC might reduce married women’s labor supply in high-income families. Thus, the EITC involves a trade-off. It increases the labor supply of the lowest- earning workers while reducing the labor supply of higher-earning workers. The cost to society is greater when a high-productivity worker reduces his hours. Against this, we must weigh the benefits to society of making work economically beneficial to individuals who might otherwise choose not to work.

3.1.3. Income and Substitution Effects of the EITC: Evidence. It is very difficult to tell whether the EITC has its predicted effects. We do not know how much people would have worked in the absence of the EITC and thus do not know if they have changed their behavior. From a research perspective, we would like to make some people randomly eligible for the EITC and others randomly ineligible and compare their labor force participation. One way we might attempt to mimic such an experiment is to compare people who are eligible for the EITC with those who are not. Of course, people with high incomes and households with no workers are not eligible for the EITC, whereas those eligible for the EITC have all worked. This would not be a very informative comparison. However, we might consider using some characteristic that is not directly related to labor force participation but affects EITC eligibility. For example, we might compare two types of households. The first group might consist of households with children that are therefore eligible for the EITC if the adults work but do not earn too much. The second group might consist of households without children and that historically were therefore ineligible for the EITC and even today are eligible only for a very modest tax credit. The problem with this approach is that the labor force participation of the two groups may differ for other reasons. In families with children, women are more likely to stay home and take care of the children. In the absence of the EITC, we would expect the labor force participation of women with children to be lower than the labor force participation of women without children. If we find that even in the presence of the EITC, women with children work less than women without children, we do not want to conclude that the EITC has no effect or even lowers labor force participation. A second approach is to look at changes over time. There were significant increases in the generosity of the tax credit at different times. For example, in the period 1994 to 1997, the benefit became much more generous, especially for households with two

74 Chapter 3 or more children. We might consider looking at whether the labor force participation of individuals in households with two children increased between 1993 and 1998. The difficulty here is similar to the difficulty that arises when we compare two groups. La- bor force participation might have increased (or decreased) over this period for reasons having nothing to do with the change in the tax credit. In this example, the economy came out of a recession and entered a sustained expansion. Labor force participation might have increased because jobs were more available. On the other hand, the boom might have allowed families with two or more children to keep one spouse at home to take care of the children. The most common approach to looking at the effects of the EITC combines the two approaches. It looks at expansions of the EITC that should have affected groups differently and then asks whether their labor force participation changed differently. The critical assumption is that in the absence of the change in EITC, the labor force participation of the two groups would have changed in the same way. Therefore, the difference in the change in behavior between the two groups is due to the additional effect of the EITC expansion. This approach is known as differences-in-differences. Before 1994, having a second child did not affect EITC payments substantially. In 1993, the maximum credit for a recipient with two or more children was $1,511, compared with $1,434 for a recipient with only one child. In the phase-in range, the subsidy was 19.5 percent for the former and 18.5 percent for the latter. Those with two children also had a somewhat higher phase-out rate, so the maximum income a recipient could have and still receive some EITC in either case was $23,050. Starting in 1994, there was a substantial divergence in the generosity of the credit available to recipients with two or more children compared to those with only one child. By 1994, the difference in the maximum credit had risen to $490 and the difference in the phase-in rate to 6.4 percentage points (30.0 percent compared with 23.6 percent). In 1995, the difference in the maximum credit was $1,016, although the difference in the phase-in rate fell to 2 percentage points. In 1996, the difference in the maximum grew to $1,404 and the difference in phase-in rates was 6 percentage points. After 1996, the credit was indexed to inflation. Given this history, we would expect that between 1993 and 1996 the labor force participation of low-wage workers with two or more children would have increased relative to that of low-wage workers with only one child. Joseph Hotz, Charles Mullin, and John Karl Scholz used participants in a welfare-to-work experiment in four sites in California to determine whether this change actually occurred.17 They divided their sample into four subsamples on the basis of whether they were on welfare at the beginning of the experiment (primarily long-term welfare recipients) or entered welfare during the experiment (new entrants) and on the basis of whether there was one parent or two in the household. Table 3.2 summarizes the differences-in-differences estimates from 1993 to 1998 for the four different subsamples. Each estimate should be understood as follows. First

17. Hotz, Mullin, and Scholz.

The Evolution of Poverty Policy 75 Table 3.2 Rates of Labor Force Participation of Families with One versus Two-Plus Children, 1993–1998

Long-Term AFDC Recipients Single-Parent Families Two-Parent Families 1993 1998 Difference 1993 1998 Difference

One Child 33.13 46.08 12.95 38.35 56.55 18.20 Two-Plus Children 24.02 43.21 19.19 29.47 56.61 27.14 Difference –9.11 –2.87 6.24 –8.88 0.06 8.94 (1.97) (3.74)

New Entrants to AFDC Program 1993 1998 Difference 1993 1998 Difference

One Child 42.21 42.06 –0.15 51.75 49.92 –1.83 Two-Plus Children 36.01 39.12 3.11 49.68 53.98 4.30 Difference –6.20 –2.94 3.26 –2.07 4.06 6.13 (2.62) (3.45)

Source: V. Joseph Hotz, Charles Mullin, and John Karl Scholz, “The Earned Income Tax Credit and Labor Market Participation of Families on Welfare,” in Bruce Meyer and Greg Duncan, eds., The Incentives of Government Programs and the Welfare of Families, http://www.jcpr.org/book/, table 6. Note: Standard errors are in parentheses.

the authors calculated the change in labor force participation from 1993 to 1998 for individuals in households with one child. This is shown in the first row of each panel in the third column. The authors assumed that in the absence of an expansion of EITC for households with two children (or more), households with one child and households with two children would have experienced the same change in labor force participation. Then they calculated the change over this period in the labor force participation of individuals in households with two children. This is shown in the second row of each panel in the third column. The authors then subtracted the change in participation for one-child households from the change in participation for two-child households to obtain their estimate of the effect of the EITC expansion on the participation of individuals in two-child households. This is the estimate reported in the third row of each panel in the third column. Let us first examine what would happen if we compared the 1998 labor force participation of individuals in households with one child with those in households with two or more children. In 1998, the labor force participation of single parents with two or more children was lower than the labor force participation of single parents with one child for both the sample of long-term AFDC recipients and new recipients. For two-parent households, participation was almost identical for long-term recipients regardless of size of household, and among the new entrants it was higher for those with

76 Chapter 3 two or more children. If we looked at only this information, we would probably conclude that the earned income credit had no effect on labor force participation. How- ever, we should note that in 1993, when the credit was similar for the two types of households, labor force participation in one-child households was higher for all four samples and much higher for three of the four. Similarly, let us consider what we would conclude if we looked only at the change in labor force participation of individuals in households with two or more children between 1993 and 1998. Long-term AFDC recipients showed very large increases in labor force participation. For single parents, the participation rate grew by over 19 percentage points, and for those in two-parent households it grew by over 27 percentage points. For new entrants to the AFDC program, the changes were much smaller. La- bor force participation grew by 3 and 4 percentage points, respectively, for single- and two-parent households. If we looked only at these results, we would conclude that the expansion of EITC for households with two or more children had dramatically increased the labor force participation of long-term welfare recipients. However, we see that the labor force participation of long-term welfare recipients with only one child also dramatically increased. The differences-in-differences estimates tell a consistent story. For all four samples, labor force participation in 1998 was higher relative to 1993 when two or more children were present than when only one child was present. For long-term welfare recipients (those on AFDC at the start of the experiment), the t-statistic is well above two for both single-parent and two-parent families. Therefore, we are reasonably confident that this is not just the result of random sampling and that the labor force participation of two-child households grew faster than that of one-child households. For new entrants to the AFDC program, the t-statistics are below two, although it is above 1.64 for two-parent families. Therefore, we have less confidence that the difference is not due to random sampling. Taken together, however, the four estimates provide strong evidence of faster growth of labor force participation for two-child households relative to one-child households. Can we really attribute this change to the effect of the increased generosity of the EITC? The fact that the study was based on a sample of welfare recipients is simultaneously an advantage and a disadvantage. It is helpful because the sample consisted of low-wage workers who were likely to be affected by changes in the tax credit. In studies based on national samples, many of the people studied may be ineligible for the EITC. On the other hand, comparing changes among families with two or more children with changes among families with only one child assumes that in the absence of the EITC change, labor force participation would have increased at the same rate for both types of families. We must ask ourselves why labor force participation might have increased more rapidly in families with two or more children than in families with only one child, even in the absence of changes to the EITC. Welfare reform of the type studied in the welfare-to-work experiment was designed to move more potential welfare recipients into the labor market. It is possible that welfare reform increased labor force participation more rapidly among groups with lower initial labor force participation. In 1993, in the sample used by Hotz, Mullin, and Scholz, mothers with two or

The Evolution of Poverty Policy 77 more children were less likely to work than were mothers with only one child. Between 1993 and 1998, the labor force participation of the two groups converged. It is possible that this convergence reflects a differential effect of welfare reform on the two groups rather than the effect of the change in the EITC. Thus, the theoretical expectation that the EITC will encourage labor force participation is strong. The evidence supports the theory, but as we will often conclude about the evidence throughout this book, it stops well short of being conclusive. We have not been able to conduct an experiment that is free of other confounding factors.

4. Cash or In-Kind Transfer: Which Is Better? Our review of the history of assistance programs reveals at least two tendencies. First, there has been a shift toward rewarding work in the form of the EITC, which we discussed in the previous section. Second, the United States relies increasingly on in-kind transfers. In this section, we discuss the benefits of cash and in-kind transfers.

4.1. The Case for Cash Transfers. The standard economics textbook view is that “more choice cannot make you worse off.” If we offer people food or the money it would cost to buy the food, people cannot be worse off taking the money. If they wish, they can buy the food and be in exactly the position they would be in if we just gave them the food. But if we give them money, they have the option of buying something that they prefer to food. In economic jargon, this is known as consumer sovereignty. Government knows less about individuals’ preferences than the individuals themselves know. Therefore, the argument goes, government should just provide money and let families decide for themselves how to spend it. This argument is strengthened by the fact that it is often more expensive to give people goods than to give them money. Government must set up mechanisms for re- imbursing vendors who are paid in food stamps. If the government simply gave money instead of food stamps, it could avoid this expense. Of course, in our earlier discussion we saw that most people spend more on food than the maximum value of food stamps. Therefore, for most people it really does not matter whether we give them food stamps instead of money. But some people would prefer other goods and value the food stamps at less than their face value. By giving money instead of food stamps, the government could reduce the cost of providing support to the poor, and those people who feel that some other need is more pressing than food (perhaps shelter, clothing, or medical care) could spend the money on that need. And many recipients undoubtedly value other in-kind transfers at less than their nominal value. Few Medicaid recipients would buy equally comprehensive health insurance if given the money instead of being provided with free health insurance.

4.2. The Case for In-Kind Transfers. Why, then, do we use in-kind transfers instead of cash? One argument for in-kind transfers is that the public, which pays for the transfers, cares about the way the poor spend the money they receive. The public is prepared to pay for merit goods, such as food, but not for cigarettes. Politically, it may

78 Chapter 3 be easier to garner support for providing food than for providing unrestricted cash. The political argument is reinforced by the support of farmers and the Department of Agriculture for policies that increase the demand for food and by the construction industry’s support for policies that encourage housing consumption.

4.2.1. Targeting Transfers to Children. However, even if we maintain the standard belief (at least among economists) in consumer sovereignty, there are good arguments for in-kind transfers. The first is that there are strong reasons for investing in children, and it is possible to target in-kind transfers to children. Public schools are probably the largest in-kind transfer targeted to children, but so are school lunch programs and some medical care programs. This argument can easily be corrupted to imply that government has to target children because poor people do not know how to take good care of their children or do not care to. However, that is not the nature of the argument. When families, regardless of their economic status, receive additional income, they typically add to the consumption of all members of the family. If society wants to target more consumption to children, it must target transfers so as to favor children either by subsidizing children’s consumption or by providing goods that are used more heavily by children.

4.2.2. Identifying the Poor. A second reason for using in-kind transfers is that they may help us figure out who is poor. We discussed earlier how difficult it is to come up with a definition of who is poor. It is even more difficult for government to come up with a definition that it can use to distinguish between people who need support and those who do not. If government offers money, anyone who is eligible will accept it unless government somehow manages to attach significant stigma to the acceptance of funds. However, if government offers subsidized housing, people who have significant resources will prefer the higher-quality housing available at the market price to the lower-quality housing available at the subsidized price. In this way, government can put fewer resources into identifying who is really poor and may also be more effective at targeting resources to the truly poor. Thus, the young woman in chapter 2 who enjoyed living with her aunt would not give up living in her aunt’s house in return for low-quality public housing. But her counterpart who worked for a wealthy aunt might accept the housing. One way we can determine whether people have more or less resources than we would anticipate on the basis of their income is to seek more extensive information about them. In the 1960s, New York City social workers would engage in surprise visits to AFDC recipients in order to ensure that there was no man living in the home. Such processes are expensive for government and burdensome for the recipients of support. In-kind transfers remove the need for such checks by providing goods that, for the most part, do not appeal to those with high incomes. Typically, anyone is free to eat at a soup kitchen. The providers do not ask for tax returns or evidence of poverty. How- ever, few people with the resources to eat better meals would choose to eat in a soup kitchen. Sometimes such a meal is accompanied by a further requirement, such as attendance at a religious ceremony or lecture. Intentionally or unintentionally, this serves

The Evolution of Poverty Policy 79 to further screen the recipients. Remember that the elderly widow in chapter 2 ate at a soup kitchen. Many people would take this as good evidence that she is, in fact, poor. One difficulty with in-kind transfers is that they may create stigma. By helping the provider identify the poor, in-kind transfers may identify the poor in ways in which they do not wish to be identified. This may reduce the ability of the provider to reach the poor. For instance, poor students may refuse free school lunches because receiving those lunches will reveal that they are poor. The National Health Service in the United Kingdom provides eyeglasses to anyone. However, only a small number of relatively unattractive glasses are available. To obtain a wider variety and more fashionable glasses, individuals must pay the market rate. This prevents wealthy people from obtaining their glasses through the National Health Service, but may deter needy families, particularly fashion-conscious teenagers, from using them as well.

4.3. Taxes, Cash, and In-Kind Transfers. In the case of welfare and the EITC, we saw that government must eventually tax away any cash transfer. Because (almost) everyone is willing to accept money from the government, the government cannot offer it to everybody and must phase it out. In the case of the EITC, we saw that the implicit tax rate was very high in the phase-out range and would be expected to reduce labor supply. Similar issues arise with in-kind transfers. Whether families are eligible for an in- kind transfer often depends on their income. In this case, the in-kind transfer is said to be means-tested. In the most extreme version of means-testing, a family is eligible for the in-kind transfer only if its income is below some cutoff. Thus, in many states families were once eligible for Medicaid only if they met the income requirements for AFDC. Today, in many states families are eligible for special Medicaid programs for children only if their income is less than 133 percent of the poverty line. This creates a strong incentive for families whose incomes would otherwise not be too much greater than the income cutoff to cut back on work in order to be eligible for Medicaid. Government can avoid discouraging people from working if, instead of establishing an income cutoff, it offers an in-kind transfer that is unattractive to higher-income people. This also makes sense because one of the benefits of in-kind transfers is that they help to identify the poor.18 If government offers everyone a free apartment in a housing project, high-income people simply will not take it, and labor supply will be largely unaffected, though there will be a small income effect on those who take the free housing. This does not mean that offering a free apartment is more efficient than giving cash. The free apartment violates consumer sovereignty. If given the cash value of the apartment rental, some people would choose less desirable housing in order to consume more food. Others would choose better housing than the free apartment.

18. See Daniel T. Slesnick, “Consumption and Poverty: How Effective Are In-Kind Transfers?” Economic Journal 106 (November 1996): 1527–45.

80 Chapter 3 Economic theory cannot tell us whether the inefficiency caused by the housing offer is greater or less than that from an equivalent cash transfer. It depends on how housing demand and labor supply respond to changes in their prices, but it is certainly possible that the loss from offering the good is less than the loss from giving money and taxing away the benefit. In practice, in-kind transfer programs typically involve a hybrid of implicit taxes on income and reliance on self-selection by higher-income families. For example, housing programs usually set the rent at a fixed proportion of recipients’ income. If a family’s income goes up, it must pay more rent for the same apartment, which is an implicit tax on income. In addition, because the housing subsidy falls as family income rises, families will also self-select out of the program at a lower income level than they would if their rent were not increased. Again, it is possible that in-kind transfers are less inefficient than are cash transfers.

5. Concluding Remarks We have seen that, as poverty policy in the United States has evolved in recent decades, there has been a shift from cash transfers to in-kind transfers and employment subsidies. Simple-minded economic analysis suggests that the shift in policy must reflect either a lack of understanding of economics or paternalism: either policy makers do not understand that people are better off if they are free to spend their money as they wish or they do not believe it. The shift to in-kind transfers has almost certainly taken place partially in response to the public’s desire to give certain types of goods to the poor. They want to ensure that the poor have enough to eat and have adequate shelter and medical care. Even if some of the poor would prefer to miss some meals in order to have a cell phone, in- kind transfers prevent them from making that decision. The shift to employment subsidies reflects a view that the nonelderly poor should work unless they are disabled or have very young children. However, in analyzing these policies, it is important to recognize that they shift incentives in very different ways from traditional cash transfers. Employment tax credits encourage some people to work more but encourage others to work less. Free or subsidized housing provides some people with housing (or better housing) and helps to target transfers to the poor, but it causes others to accept lower-quality housing. Policies have trade-offs. In-kind transfers and earned income credits are neither clearly worse nor better than cash transfers from the perspective of either the poor or the government.

6. Further Reading Burtless, Gary. “The Economist’s Lament: Public Assistance in America.” Journal of Economic Perspectives 4 (Winter 1990): 57–78. Edin, Kathryn, and Laura Lein. Making Ends Meet. New York: Russell Sage, 1997.

The Evolution of Poverty Policy 81 Hotz, Joseph P., Charles Mullin, and John Karl Scholz. “The Earned Income Tax Credit and Labor Market Participation of Families on Welfare.” In Bruce Meyer and Greg Duncan, eds., The Incentives of Government Programs and the Welfare of Families. http://www.jcpr.org/book/. Liebman, Jeffrey. “The Impact of the Earned Income Tax Credit on Incentives and In- come Distribution.” Tax Policy and the Economy 12 (1998): 83–119. Scholz, John Karl. “Tax Policy and the Working Poor: The Earned Income Tax Credit.” Focus 15 (Winter 1993–1994): 1–12. http://www.ssc.wisc.edu/irp/focus/focus.htm. Scholz, John Karl, and Kara Levine. “The Evolution of Income Support Policy in Recent Decades.” In Sheldon H. Danziger and Robert H. Haveman, eds., Under- standing Poverty. Cambridge, MA: Harvard University Press, 2001, 193–228. Steurele, J. C. Eugene. “Policy Watch: Tax Credit for Low Income Workers with Chil- dren.” Journal of Economic Perspectives 4 (Summer 1990): 201–12.

7. Questions for Discussion 1. Describe the major trends in federal spending on antipoverty programs over the past thirty-forty years. Which programs have grown, and which have declined? How would these changes be expected to affect the official poverty rate? Which programs are most targeted to the poor? 2. What are the economic arguments for preferring cash transfers rather than in-kind transfers to the poor? What are the economic arguments for in-kind transfers? 3. How did AFDC affect incentives to work (legally)? 4. Why might mothers who were eligible for AFDC have had lower levels of consumption if they worked than if they did not work? 5. If the high implicit tax rates associated with cash transfer programs such as AFDC discourage work, why would government impose such high implicit tax rates? 6. Describe schematically how the EITC affects the budget line. Explain how it affects work incentives at different points along the budget line. 7. Evaluate each of the following strategies for measuring the effect of the EITC expansion for families with two children. a. Compare, using data for 1998, the labor force participation of households with two children with the labor force participation of households with only one child. b. Examine the growth in the labor force participation of households with two children between 1993 and 1998. c. Compare the change in labor force participation of households with two children relative to households with one child between 1993 and 1998. 8. Why might the efficiency loss associated with in-kind transfers be higher or lower than the efficiency loss associated with cash transfers?

82 Chapter 3 chapter 4

Trends in Poverty

I           has changed over time. More important, we will try to understand why it has changed. This is interesting in its own right. But it is also important because it helps us understand where policies are likely to be effective. If most of the increase in poverty is due to an increase in the number of female-headed households, we will want to spend some time thinking about whether there are ways to reduce the prevalence of such households. If most of the variation is due to changes in the market for low-skilled workers, we will want to think about whether there are effective policies to improve their labor market. We will also want to know whether these trends are influenced by the way we measure poverty. If a change in the definition of poverty changes poverty trends, we will need to decide which definition provides more useful information.

1. Trends Using the Official Measure Figure 4.1 shows the official U.S. poverty rate from 1959 to 2004. The poverty rate fell dramatically between 1959 and the early 1970s, from about 22 percent to a low of 11.1 percent in 1973. It then fluctuated around 13 percent, rising to 15 percent in 1983 and 1993. From 1993 to 2000 it fell, almost reaching its historical low. With the recession of 2001, it rose again. Although the recession ended, the poverty rate continued to climb in 2004. Note that there is a tendency for the poverty rate to rise during recessions and to fall during expansions, but there are exceptions. The poverty rate continued to rise even after the United States began its recovery from the 1990– 1991 recession, and it rose steadily from 1978 to 1983 during both expansions and recessions.1

1. Poverty data in this chapter are from U.S. Census Bureau, Historical Poverty Tables, Current Population Survey, http://www.census.gov/hhes/income/histinc/histpovtb.html.

83 Poverty Rate

10 1960 1970 1980 1990 2000 Figure 4.1 Official Poverty Rates, 1959–2004

Poverty Rate

35 Younger than 18 Age 18–64 30 Age 65 or Older

1960 1965 1970 1975 1980 1985 1990 1995 2000 2004 Figure 4.2 Poverty Rates by Age, 1959–2004

The most dramatic change was in the poverty rate for the elderly (figure 4.2). In 1959, the poverty rate for the elderly was about 35 percent, considerably above the figure of 27 percent for children and roughly 17 percent for nonelderly adults. By the 1990s, the poverty rates of elderly and nonelderly adults were virtually identical. The rate among the elderly hit record lows of around 10 percent in the late 1990s, while the poverty rate among children had fallen, but only to about 16 percent. The current rate for the elderly remains close to 10 percent, while that for children has risen to about 18 percent. Despite these trends, children as a fraction of the population in poverty actually fell. In 2000, related children under 18 represented a little more than a third of the

84 Chapter 4 Poverty Rate

60 Blacks Hispanics 50 Whites Non-Hispanic Whites 40

0 1960 1965 1970 1975 1980 1985 1990 1995 2000 2004 Figure 4.3 Poverty Rates by Race/Ethnicity, 1959–2004

population in poverty, an all-time low, compared with 44 percent in 1959. Given the change in poverty rates, this can be explained only by a dramatic decline in the proportion of the population consisting of children. Poverty rates also fell dramatically for blacks (see figure 4.3). Over half of blacks were poor in 1959. By 2000, the proportion had fallen to a record low of 22 percent (still above the poverty rate for whites in 1959). Poverty rates for Hispanics were not calculated until the early 1970s, at which time they were significantly below those of blacks but well above those of whites. The gap between blacks and Hispanics has now been substantially closed, largely as a result of a rise in poverty rates among Hispanics in the late 1980s and early 1990s that did not occur among blacks. Figure 4.4 shows poverty rates among families rather than among individuals. All poverty is measured at the family or household level, but the previous figures showed the proportion of people who are poor. If we have three people living in a family that is poor, two people living in a family that is not poor, and an individual living by herself (and therefore not in a family) who is poor, four of the six people are poor, and thus the poverty rate is 4/6 or 67 percent. Among people living in families, three out of five are poor, and thus the poverty rate among people living in families is 60 percent. How- ever, one of the two families is poor, and thus the proportion of families that are poor is 50 percent. As can be seen in the figure, in the middle and late 1990s, the poverty rate among families recovered from its increase in the 1980s and early 1990s, reaching an all-time low of 8.7 percent in 2000. There has also been a very significant decrease in the poverty rate among female-headed families. This stood at 43 percent in 1959 and was at 36 percent as recently as 1993, but now is 28 percent, up from a low of 26 percent. The decline in poverty among female-headed families in which children under 18 years old are present has been even more dramatic, although it, too, has increased in the past few years.

Trends in Poverty 85 Poverty Rate 60

50 Female-Headed Families with Children

30 Female-Headed Families

10 Families

0 1960 1965 1970 1975 1980 1985 1990 1995 2000 2004 Figure 4.4 Poverty Rates of Families, 1959–2004

Taken together, these trends show that poverty decreased dramatically for almost all groups in the 1960s and early 1970s; that the poverty rate fluctuated and tended, if anything, to rise from around 1973 to 1993; and that then, at least for some groups, poverty decreased dramatically after 1993. The recession in 2001 led to an increase in poverty that has not yet reversed itself despite the end of the recession.

2. Trends in Poverty under Alternate Measures Figure 4.5 shows the evolution of poverty rates in the United States since 1979 using four different measures of poverty.2 The first is the official measure of the poverty rate. The second shows the poverty rate if we subtract any cash transfers people received from the government. In an accounting sense, this is the poverty rate that would have existed if there had been no cash social security, disability insurance, or other welfare programs. The third measure is based on a measure of after-tax income that includes both cash and in-kind transfers, while the fourth also takes into account the value of owner-occupied housing. It is important to realize that the accounting approach to measuring the effect of poverty programs on the poverty rate can be very misleading. It assumes that people do not change their behavior in response to the poverty programs. Later we will discuss the work of Charles Murray. Murray has argued that the Great Society antipoverty programs begun under President Lyndon Johnson in the 1960s were responsible for increases in the poverty rate. He argues that welfare programs encouraged people who might have otherwise avoided poverty to be single mothers and not to work.3

2. Data are from U.S. Census Bureau, “Historical Poverty Tables—Poverty by Definition of In- come,” http://www.census.gov/hhes/poverty/histpov/rdp05.html. 3. Charles A. Murray, Losing Ground: American Social Policy, 1950–1980 (New York: Basic Books, 1984).

86 Chapter 4 Poverty Rate 30 Without Government Transfers 25 Official

5 After Taxes, Cash, and In-Kind Transfers After Taxes, Transfers, and Imputed Rents 0 1980 1985 1990 1995 2000 2004 Figure 4.5 Poverty Rates under Alternate Definitions of Income, 1979–2003

Whether or not Murray is right that welfare increased poverty, it is undoubtedly true that poverty programs affect how much people earn. Some poverty programs discourage work because they tax any earnings heavily. On the other hand, Medicaid might encourage work by reducing sickness. With this caveat in mind, we can examine figure 4.5 to see the extent to which cash transfer programs reduced poverty between 1979 and 2003 relative to the poverty rate without cash transfers. It is clear from the comparison that changes in the poverty rate over the period were driven primarily by changes in the pretransfer poverty rate. The fraction of people who would otherwise have been counted as poor but were lifted above the poverty threshold by government transfers fluctuated around 7.5 percent, hitting a low of 6.8 percent in 1990 and a high of 8.4 percent in 1993 and 1994. Increased generosity over this period appears to have reduced the increase in poverty that might otherwise have taken place. Because the proportion of individuals raised above the poverty line by government transfer was fairly constant over this period, the pretransfer and official poverty rates tracked each other closely, although the pretransfer poverty rate was significantly higher. Including the value of cash and in-kind transfers and excluding taxes also does not change the broad pattern. The poverty rate, after accounting for in-kind transfers including medical care, was 2.2 percent to 3.5 percent lower than the official poverty rate. Accounting for imputed rents on owner-occupied housing also has little effect. Although, as we discussed previously, there is considerable support for improving the poverty measure, doing so has only a small effect on recent trends.

3. Accounting for Trends It seems self-evident that in good economic times the poverty rate will fall. When the economy is booming, wages rise, so people with jobs are more likely to earn enough to avoid being poor. In addition, more workers seeking jobs will be able to find employment.

Trends in Poverty 87 In a now classic article, Alan Blinder and Rebecca Blank looked at the relation between economic conditions and the poverty rate during the period from 1959 to 1983.4 It was so obvious that economic conditions and the poverty rate were related that the paper concentrated not on whether there was a relation but on whether the unemployment rate or the inflation rate had a greater effect on poverty. Only the unemployment rate had a statistically significant effect on the poverty rate. They attributed the high poverty rate of the post-1973 period to the very high unemployment rates prevailing over that period. However, later papers, including one by Blank herself, found that the relation between economic performance identified by Blinder and Blank did not hold up after 1983. A series of papers all found that despite rapid economic growth in the latter part of the 1980s and a sharp decline in the unemployment rate, the poverty rate remained stubbornly high. In 1997, Blank wrote of “the ‘death of trickle down economics’” and concluded that “economic growth no longer reduces poverty.”5 The basic puzzle is presented in the graph that shows the relation between real per capita gross domestic product (GDP) and the poverty rate (figure 4.6). Between 1959 and 1973, real GDP per capita6 grew rapidly and the poverty rate plummeted. The economy grew much more slowly between 1973 and 1983, although it continued to grow. If the poverty rate were affected only by average income, the poverty rate should have continued to decline, although at a slower rate than during the economic boom from 1961 to 1973. Indeed, in a review of this literature, Robert Haveman and Jonathan Schwabish take it for granted that the poverty rate will decline as income levels rise. They note that all four studies they review “included the ratio of the poverty line to mean household income for each year. This was to reflect the fact that in times of real economic growth, an absolute poverty line will fall relative to average income; by definition, then, the number of the poor will shrink.”7 Yet, as we can see, the poverty rate was higher in 2004 than in 1973 even though real per capita GDP grew by about 75 percent over the same period. What happened?

4. Rebecca M. Blank and Alan S. Blinder, “Macroeconomics, Income Distribution, and Poverty,” in Sheldon H. Danziger and Daniel H. Weinberg, eds., Fighting Poverty (Cambridge, MA: Harvard University Press, 1986), 180–208. 5. David Culter and Lawrence F. Katz, “Macroeconomic Performance and the Disadvantaged,” Brookings Papers on Economic Activity 2 (1991): 1–74; Rebecca Blank, “Why Were Poverty Rates So High in the 1980s?” in Dimitri Papadimitriou and Edward Wolff, eds., Poverty and Prosperity in the USA in the Late Twentieth Century (New York and London: Macmillan, 1993); Elizabeth T. Powers, “Inflation, Unemployment, and Poverty Revisited,” Federal Reserve Bank of Cleveland Economic Re- view 31 (3rd Quarter 1995): 2–13; Rebecca M. Blank, It Takes a Nation (Princeton, NJ: Princeton University Press, 1997), 53. 6. GDP is a measure of all the goods and services produced in the economy. It is very close to a measure of the entire income earned in the economy, so GDP per capita is very close to per capita income. 7. Robert Haveman and Jonathan Schwabish, “Economic Growth and Poverty: A Return to Nor- malcy?” Focus 20 (Spring 1999): 1–7, quote on p. 3.

88 Chapter 4 Real GDP per Capita Poverty Rate (Thousands) 25 35 Real GDP per Capita ($2000) 30 20 25

20 15

15 Poverty Rate

10 10 1960 1965 1970 1975 1980 1985 1990 1995 2000 2004 Figure 4.6 The Poverty Rate and Real Per Capita GDP, 1959–2004

If per capita GDP increased but the poverty rate did not fall, there are only two mechanisms by which poverty could fail to have decreased. The first is that households became smaller. If a two-adult, two-child family has an income of $20,000, it is not poor. If it splits into two one-adult, one-child families, both families are poor. More realistically, the split would result in a family with a mother and two children; only about one-fifth of single-parent families are headed by men. One explanation for the failure of the poverty rate to fall that draws on this perspective is that there has been an increase in the proportion of female-headed households. The second explanation is that income inequality increased. If all incomes increased at the same rate, holding family composition constant, the proportion falling below some threshold would have to fall, but if income inequality increased, this would not necessarily be the case. Income inequality across families can increase in two ways. First, income inequality can increase even though families form in the same way. This may reflect increasing earnings inequality, increasing inequality in asset income, or growth in asset income relative to earned income. Because asset income is heavily concentrated among families with incomes above the poverty line, increases in asset income will not have a noticeable effect on the poverty rate. The second way that family income inequality can increase is that family formation can change. If high-earning men and women marry each other and low-earning men and women marry each other, there will be more poverty than if high earners marry low earners.

3.1. Does the Labor Market Matter? One explanation for the failure of GDP growth to reduce poverty is that we have reached the point where most of the poverty that could be eliminated by economic growth has already been eliminated. Richard Freeman argues that a substantial fraction of the poor could not be reached by a rising

Trends in Poverty 89 economic tide.8 He calculates that 21 percent of the poor in 1999 were disabled. Furthermore, 15 percent were over age 64, 18 percent had eight or fewer years of schooling, and one-fourth were mostly very low-skilled immigrants. These groups overlap. Still, taken together, over half of the poor were in groups that do not benefit readily from a booming economy, especially one in which skills appear to be increasingly important. Freeman is certainly right that many of the poor are unlikely to be directly affected by a booming labor market. This is not only true today but has been true throughout the period for which we have been measuring the poverty rate. At the beginning of the 1960s a substantial fraction of the poor were retired elderly people and widows. These groups were unlikely to seek and find work even in the economic boom of the 1960s. Along with the economic expansion, however, came an increase in transfers to the elderly, and their poverty rate declined. Still, there are substantial numbers of poor people we would expect to benefit directly from a booming labor market. In 1978, 39 percent of the poor worked at least part-time and for at least part of the year. In 2000, over 40 percent of the poor worked at least part-time and almost one in eight worked full-time year-round. Some of these workers helped to support nonworking family members. Over 60 percent of poor families had at least one worker in 2004.9 We would expect that a booming labor market would translate into higher wages for full-time, year-round workers and into both higher wages and more hours for part-time or part-year workers. In addition, some people who do not work at all when wages are low and jobs are difficult to find will enter the labor market and find employment when times are better. Thus, although we would not expect economic growth to eliminate poverty unless we increased income transfers to the poor, we should expect it to reduce poverty. Moreover, since 1978 there has been no trend in the proportion of the poor who are workers and an upward trend in the proportion of the poor who are full-time, year- round workers (see figure 4.7). The proportion of full-time/year-round workers rose from 7.7 percent in 1978 to 12.6 percent in 1998 before dropping somewhat in recent years. At a superficial level, this suggests that the reason for the rise in poverty in the 1970s and 1980s was not that people stopped working, another hypothesis that has been put forward. More important from our perspective, it is clear that the economy has not reached a point where better labor market conditions are unable to lift workers out of poverty.

3.2. The Role of Earnings. In fact, although the poverty rate has not followed the path that we would have expected based on real per capita GDP, it has followed me-

8. Richard B. Freeman, “The Rising Tide Lifts . . . ?” in Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty (Cambridge, MA: Harvard University Press, 2001), 97–126. 9. U.S. Census Bureau, “Families by Number of Working Family Members and Family Struc- ture: 2004, Below 100 percent of Poverty—All Races,” http://pubdb3.census.gov/macro/032005/ pov/new06_100_01.htm.

90 Chapter 4 Percentage of Poor 45

40 Workers 35

10 Full-Time, Year-Round Workers 5 1978 1985 1990 1995 2000 2004 Figure 4.7 Workers as a Percentage of the Poor, Age 16 and Older, 1978–2004 dian male earnings reasonably well.10 We focus on male earnings not because the earnings of women are unimportant, but because male earnings are generally viewed as a better measure of the state of the labor market. Overall median earnings are heavily affected by the growth of women’s participation in the labor market. Because women tend to earn less than men do, as women enter the labor market, they tend to lower median earnings even if they are responding to increasing earnings opportunities. Changes in median female earnings are also heavily influenced by the rising skill level of women in the labor force. College graduation and labor market experience have increased rapidly among women. These changes can obscure the effects of shifts in the overall state of the labor market. Figure 4.8 shows the relation between median male earnings and the poverty rate from 1960 to 2004. Note that throughout this period, median male earnings exceeded the poverty cutoff for all but extremely large families, so there is no automatic relation between median male earnings and the poverty rate. Yet (using different scales) median male earnings and the poverty rate are reflections of each other. If we ignore calendar year and simply plot the poverty rate against median male earnings (figure 4.9), we observe an almost straight line (shown by the dashed line) with a slope of about a percentage point reduction in poverty for each $1,000 increase in median male earnings. Because the information in figure 4.9 is dominated by two periods, an earlier period with high but rapidly increasing poverty and rising earnings and a later period with lower poverty and slow (if any) earnings growth, it is helpful to restrict ourselves to the later period, which we do in figure 4.10. When we examine figure 4.10, we

10. Earnings data are from U.S. Census Bureau, Historical Income Inequality Tables, http://www. census.gov/hhes/income/histinc/ineqtoc.html and Historical Income Tables—People, http://www.census. gov/hhes/income/histinc/incperdet.html.

Trends in Poverty 91 Median Earnings Poverty Rate (Thousands) 22 34 Real Median 21 Earnings ($2004) 20 19 31 18 17 28 16 15 14 25 13 Poverty 12 Rate 11 22 1960 1965 1970 1975 1980 1985 1990 1995 2000 2004 Figure 4.8 The Poverty Rate and Real Median Earnings of Adult Men, 1960–2004

Poverty Rate 22 1960 1961 21 1962 20 1963 19 1964 18 17 1965 16 1983 1993 15 1982 1994 1966 1992 1984 1981 14 1967 1991 1995 1996 1985 1987 1990 1986 1997 13 1968 1980 1988 2004 1998 1971 1970 1989 2003 12 1975 1969 1972 1999 1976 2002 1977 1979 1978 2001 11 1974 1973 2000

22,00025,000 28,000 31,000 34,000 Real Median Earnings ($2004) Figure 4.9 The Poverty Rate and Real Median Earnings of Adult Men, 1960–2004

observe that there is a tendency for observations between 1968 and 1979 to lie about one and a half points below the observations for the other years. This suggests that something in the 1970s shifted the relation between the poverty rate and median male earnings, but the shift was much less dramatic when viewed from this perspective than from the perspective of the relation between the poverty rate and per capita GDP.

3.3. Rising Earnings Inequality. One explanation for this pattern is growing earnings inequality. Figure 4.11 shows how median and tenth percentile male earnings for

92 Chapter 4 Poverty Rate 1983 1993 15 1982 1966 1992 1984 1994 1967 1991 14 1985 1981 1995 1996 1990 1986 1987 1997 13 1980 1988 1968 1989 2004 1998 1970 1971 2003 1975 1969 2002 12 1972 1999 1976 1979 2001 1977 1978 1974 2000 1973 11 27,000 29,000 31,000 33,000 Real Median Earnings ($2004) Figure 4.10 The Poverty Rate and Real Median Earnings of Adult Men, 1966–2004 full-time, year-round workers have changed over time.11 Note that the scale for median earnings is on the left and the scale for the tenth percentile earnings is on the right. In 1967, median male earnings were about $33,000 in 2003 dollars. This figure rose to about $40,000 in 1973 and fluctuated around this level after that. In contrast, earnings for men at the low end of the earnings distribution never regained their 1973 level and were substantially below that peak throughout the 1980s and 1990s. The ratio of the median (fiftieth percentile) earnings of male full-time, year-round workers to the tenth percentile earnings for this group is a standard measure of earnings inequality. When earnings for the worker with median earnings are much higher than earnings for the tenth percentile worker, inequality is high. In 1973, this ratio was 2.09. In 2003, it was 2.44. Figure 4.12 replicates figure 4.8 except that it uses the 50-10 earnings ratio to explain the poverty rate. We can see that there is a strong relation between earnings inequality and the poverty rate. When earnings inequality is high, so is the poverty rate. Over this period, there was some tendency for the 50-10 earnings ratio to be high when median earnings were high (and vice versa, of course). How do we know whether it is median earnings, the 50-10 earnings ratio, or both that explain changes in the poverty rate? Think about what you might do if you and a friend were discussing whether height or weight or both were important for some activity (maybe playing cornerback in football). You might point out that cornerbacks tend to be tall. Your

11. The Bureau of Labor Statistics does not publish the tenth percentile of earnings for all workers. It does publish the ratio of median to tenth percentile earnings for year-round, full-time workers. I have matched the earlier numbers and extended the series to 2003.

Trends in Poverty 93 Tenth Percentile Median Earnings Earnings ($2003 Thousands) (Thousands) 20 41 Median Earnings

39 18

37 16 35 Tenth Percentile Earnings 33 14 1967 1970 1975 1980 1985 1990 1995 2000 2003 Figure 4.11 Real Median and Tenth Percentile Earnings of Adult Men: Year-Round, Full-Time Workers, 1967–2003

Poverty Rate 1983 1993 15 1982 1992 1994 1984 1967 1991 14 1981 1985 1995 1996 1990 1986 1987 1997 13 1980 1988 1968 1989 1998 1970 2003 1971 1975 1969 2002 12 1972 1976 1999 1979 1977 2001 1978 2000 1974 1973 11 2.052.102.152.202.25 2.30 2.35 2.40 2.45 2.50 2.55 50-10 Ratio Figure 4.12 The Poverty Rate and the 50-10 Earnings Ratio: Year-Round, Full-Time Male Workers, 1967–2003

friend might point out that they tend to weigh a lot. You could both recognize that tall people tend to weigh more than short people, but how would you decide whether height or weight was important? You might ask whether cornerbacks tend to be tall for their weight. If so, height is probably important. And you might ask whether they tend to be heavy for their height. If so, weight is probably important.

94 Chapter 4 Median Earnings ($ Thousands) 1999 2000 2001 1998 2002 33 2003

1973 1978 1988 1989 1987 1972 1979 1997 31 1995 1996 1986 1969 1974 1980 1970 1976 1977 1990 1975 1981 1994 1971 1968 1985 1991 1984

29 1982 1967 1983 1992 & 1993

27 2.05 2.10 2.15 2.20 2.25 2.30 2.35 2.40 2.45 2.50 2.55 50-10 Ratio Figure 4.13 Median Male Earnings and the 50-10 Earnings Ratio, 1967–2003

We can do the same thing with the 50-10 earnings ratio and median earnings. We can plot median earnings and the 50-10 earnings ratio. Figure 4.13 shows how high median earnings are relative to what we would expect on the basis of the 50-10 earnings ratio. The dashed line represents the average relation between median earnings and the 50-10 earnings ratio. We can see that in 1969 and 1974, median earnings were almost exactly what we would predict on the basis of the 50-10 ratio. But in 1999, for example, median earnings were much higher than we would predict, and in 1983, they were much lower. We call the difference between the actual median earnings and the prediction line the unexplained median male earnings. We can also look back at figure 4.12 and ask whether the poverty rate was higher or lower than we would have predicted on the basis of the 50-10 earnings ratio alone. Again we take the difference between the actual poverty rate and the predicted poverty rate and call this the unexplained poverty rate. We can then plot the unexplained poverty rate against unexplained median male earnings and ask whether when median male earnings are unusually high given the 50-10 earnings ratio, the poverty rate is also unusually high given the 50-10 earnings ratio. Because we have removed the effect of the 50-10 earnings ratio on the poverty rate, we say that we are “holding the 50-10 earnings ratio constant.” We do this in figure 4.14. We can see that, even controlling for earnings inequality, there is a strong relation between the poverty rate and median male earnings. The slope of the line implies that a $1,000 increase in median male earnings lowers the poverty rate by about six-tenths of a percentage point holding earnings inequality constant. We can also ask how much of the poverty rate is not accounted for by changes in median male earnings. We can calculate this amount from figure 4.10. And using a

Trends in Poverty 95 Unexplained Poverty Rate 1982 1983

1993 1967 1981 1992 1994 1984 1975 1991 1985 1971 1995 1970 1968 1969 1980 1996 1990 1987 1997 1986 1988 1989 1972 1976 1979 1973 1998 1974 2003 1977 1978 2002 1999 2001

2000 Unexplained Median Male Earnings Figure 4.14 Poverty and Median Male Earnings Holding the 50-10 Earnings Ratio Constant, 1967–2003

figure similar to 4.13, we can calculate how much of the 50-10 earnings ratio is not accounted for by median male earnings. In figure 4.15, we plot the unexplained poverty rate against the unexplained 50-10 male earnings ratio. We see that, even controlling for median male earnings, greater inequality is associated with a higher poverty rate. Comparing the two figures emphasizes that the poverty rate is driven by earnings at the bottom of the distribution, and earnings have not increased at similar rates at different parts of the earnings distribution. The failure of those at the lower end of the earnings distribution to share the earnings gains of those at the top is associated with a less dramatic decline in the poverty rate than would have occurred had inequality remained constant. In fact, together median earnings for men and the 50-10 male earnings ratio at the tenth percentile of the earnings distribution explains over 80 percent of the variation in the poverty rate between 1967 and 2003. The poverty rate is heavily influenced by what happens at the low-wage end of the labor market, and together, these two variables do a good job of capturing the state of this labor market.12 This means that the remaining explanations cannot account for much of the variation. Of course, other explanations may be complementary to the growing inequality story. Growing inequality might have encouraged single motherhood, or attitude changes caused by the availability of welfare might have contributed to the reduced earnings among low-skilled men.

12. In fact, if we divide median male earnings by the 50-10 male earnings ratio, we can explain about 80 percent of the variation in the poverty rate. This ratio has no precise meaning, because the 50-10 ratio applies only to full-time, year-round workers whereas the median is for all male workers, but it seems to do a good job of capturing the state of the low-wage labor market.

96 Chapter 4 Unexplained Poverty Rate 1998 1999 2003 2002 2001 2000 1997 1987 1996 1988 1995 1986 1994 1989 1983 1981 1993 1991 1982 1985 1984 1990 1992 1978 1980 1972 1973 1967 1979 1970 1969 1968 1971 1975 1976 1977 1974 Unexplained 50-10 Male Earnings Ratio Figure 4.15 Poverty and the 50-10 Male Earnings Ratio: Holding Median Male Earnings Constant, 1967–2003

3.4. Other Factors. In the following subsections we will look at whether other explanations can account for variation in the poverty rate that is not explained by the two factors we have discussed so far. We consider four other factors: the unemployment rate, women’s earnings, welfare benefits, and the proportion of households that are female-headed. In each case, the conceptual approach is the same. We take the six potential factors we have identified and drop one of the six. We ask how much higher or lower the poverty rate was compared with what we would have predicted on the basis of the other five factors. We also ask how much higher or lower the left-out factor was compared with what we would have predicted on the basis of the other five factors. We then look at the relation between the “unexplained poverty rate” and the “unexplained part of the left-out factor.” The appendix to this chapter presents a less visual approach and explains the relation between that approach and the visual presentation used in the rest of the chapter.

3.5. The Market for Low-Skilled Women Workers. We have seen that a substantial proportion of the poor live in female-headed households. In the past thirty-five years, the education and work experience of women have risen, and women spend more hours per year in the labor force than they used to. The earnings of women have followed a very different pattern from those of men. They have increased almost steadily throughout the period. To some extent, this reflects the fact that women are working more and not an increase in opportunities. Therefore, we use the median earnings of women working full-time year-round. This is an imperfect way to adjust for changes in the amount that women work, but it is better than ignoring the increased hours worked by women.

Trends in Poverty 97 Unexplained Poverty Rate 1987

1982 1988 1981 1998 1983

1999 1969 1989 1972 1970 1995 1973 1967 1980 1978 1997 1986 1984 1985 1968 1979 1996 1994 1971 1975 2002 1993 1991 2003 1990

1992 2001 2000

1976 1974 1977 Unexplained Median Female Earnings (Full-Time, Year-Round Workers) Figure 4.16 Poverty and Median Female Earnings, 1967–2003

Figure 4.16 shows that increases in the earnings of women are associated with reductions in the poverty rate even after accounting for median male earnings, the male 50-10 earnings ratio, the unemployment rate, average monthly benefits for welfare recipients, and the fraction of households that are female-headed. The estimated effect is that an increase of $1,000 in median earnings for full-time, year-round women workers is associated with about a one-fourth of a percentage point reduction in the poverty rate.

3.6. Unemployment. We began our discussion of the relation between the economy and the poverty rate with the work of Blank and Blinder that shows a strong relation between the unemployment rate and poverty. Perhaps that relation was obscured in later studies that used per capita GDP instead of a measure of labor market earnings. Figure 4.17 attempts to show whether changes in the unemployment rate can account for the part of the poverty rate that cannot be explained by our other explanatory factors. We see that higher unemployment is associated with a higher poverty rate. The estimated effect is that an increase in the unemployment rate of 1 percentage point is associated with a 0.1 percentage point increase in poverty. However, as discussed in greater detail in the appendix, we do not estimate this relation very precisely and therefore do not have great confidence that there is a true relation.

3.7. The Role of Transfers. As we discussed in the previous chapter, in the 1960s and early 1970s, government policy relied to a great extent on cash transfers. Later policy put greater weight on in-kind transfers and tax credits. Because whether a household is poor is determined by its pretax cash income, neither in-kind transfers nor tax credits directly affect the poverty rate.

98 Chapter 4 Unexplained Poverty Rate 1987 1982 1983

1988 1998 1972 1969 1970 1989 1999 1981 1986 1973 1985 2002 1978 1980 2003 1971 1984 1997 1975 1991 1967 1995 1994 1992 1990 1979 1993 1968 2001 1996

2000 1976

1974 1977 Unexplained Unemployment Figure 4.17 Poverty and Unemployment, 1967–2003

Unfortunately, as discussed in the previous chapter, data on the poverty rate after adjusting for taxes and in-kind transfers are available only for the period starting in 1979. To look at the relation between poverty and transfer policy for the period starting in 1967, we use the average monthly welfare benefit per recipient to measure the generosity of transfer programs, although we recognize that it is an imperfect measure. The real monthly benefit peaked in 1978, which is consistent with a shift in the relation between poverty and median earnings after that time. Between 1978 and 2002, the average monthly benefit fell by $81, or about $1,000 per year, which suggests that changes in poverty policy could have accounted for some shift in the poverty rate. As we can see from figure 4.18, there is a negative relation between the generosity of the welfare benefit and that part of the poverty rate not explained by other factors. However, the effect is not large. Once we have taken account of other factors, the difference between the predicted poverty rate when the welfare system was at its peak generosity and the predicted poverty rate when the welfare system was least generous is less than half a percentage point. To some extent, this is not surprising. Even at its most generous, Aid to Families with Dependent Children (AFDC, discussed in the previous chapter) was insufficient to put the average recipient above the poverty line, and because AFDC was reduced sharply when recipients had other income (usually by one dollar for each dollar earned), welfare recipients typically did not have much income beyond AFDC. Therefore, variation in AFDC generosity was not likely to affect their poverty status. This does not mean that AFDC had no effect on their well-being, only that it did not lift them above the poverty cutoff.

Trends in Poverty 99 Unexplained Poverty Rate 1982

1983 1998 1981 1999 1969 1987 1967 1970 1997 1988 1995 1972 2003 1971 1973 1975 1968 2002 1985 1989 1994 1980 1996 1993 1984 1986 2001 1991 1978 1992 1979 2000 1990

1976 1974

1977 Unexplained AFDC/TANF Benefit per Recipient Figure 4.18 Poverty and Average Monthly Welfare Benefit, 1967–2003

3.8. Female-Headed Households. There has been a substantial increase in the prevalence of female-headed families. From 1959 to 2003, the proportion of all families that were female-headed rose from about 10 percent to about 18 percent. The proportion of poor families that were female-headed rose from less than one-fourth to about 50 percent. Because the poverty rate is about 16 percentage points higher for female-headed families than for others, an 8 percentage point shift in the composition of families toward female-headed families would account for an increase of more than 1 percentage point in the poverty rate.13 However, a 1 percentage point increase in the poverty rate due to an increase in the number of female-headed families cannot explain why the relation between GDP per capita and poverty was not maintained after 1973. Indeed, the trend toward more female-headed households was present over the 1959–1973 period but did not obscure the decline in poverty associated with economic growth. Figure 4.19 attempts to answer the question of whether changes in the proportion of households that are female-headed can explain changes in the poverty rate not explained by factors we have already considered. The answer appears to be “Yes.” The slope of the line implies that an increase of 1 percentage point in female-headed families is associated with an increase of one-fourth of a percentage point in the poverty rate, an estimate that is in line with the difference in the poverty rate between female- headed households and others.

13. Of course, this assumes that the “extra” female-headed families have the same probability of being poor as the other female-headed families. This need not be the case.

100 Chapter 4 Unexplained Poverty Rate 1987 1988 1982 1998 1969 1989 1981 1983 1999 1970 1997 1995 1972 1973 1978 1980 1985 1984 1979 1991 1986 1990 1993 1996 1971 1967 1994 2002 1975 1968 2003 1992

2001

2000

1976 1974 1977 Unexplained Female-Headed Households Figure 4.19 Poverty and the Proportion of Female-Headed Households, 1967–2003

3.9. Assortative Mating. When people marry people who are like themselves along some dimension, such as income or education, we say that there is positive assortative mating along that dimension. To understand how the degree of positive assortative mating can increase income inequality, consider the following example. Let us say that we have a society composed of Bob, Carol, Ted, and Alice. Bob earns $40,000, Carol $30,000, Ted $20,000, and Alice $10,000. If Bob marries Alice and Ted marries Carol, each couple has a family income of $50,000. Now suppose the couples divorce and Alice marries Ted and Carol marries Bob. Carol and Bob now have a family income of $70,000, while Ted and Alice have a combined income of $30,000. Without any change in the income of any individual, family income inequality has increased. Over the past several decades, there has been an increase in positive assortative mating along education lines. Traditionally, men with more education tended to marry women with more education. This tendency has increased. However, the relation between education and earnings has not been sufficiently strong and the change not sufficiently large to have generated much of an increase in family income inequality.14 What may have played more of a role has been the increase in the labor force participation of married women, especially married women with husbands whose earnings place the family above the poverty line. Although increased earnings by married

14. Michael Kremer, “How Much Does Sorting Increase Inequality?” Quarterly Journal of Eco- nomics 112 (February 1997): 115–39.

Trends in Poverty 101 women can reduce the poverty rate, married-couple families in which the husband works already have a low poverty rate. In 2004, married-couple families (in which the householder was less than 65 years old) in which the husband worked but the wife did not had a poverty rate of 10.6 percent. When the wife also worked, the poverty rate was only 1.6 percent.15 This does not mean that in couples where the wife worked the poverty rate would have risen to 10.6 percent had she not worked, but it gives us a sense of the potential magnitude of the effect of increased labor force participation by married women on the poverty rate, and it is not large. What is particularly important is that the growth in earnings among married women has been concentrated among wives of men whose earnings were already sufficient to bring the family above the poverty line. Chinhui Juhn and Kevin Murphy show that between 1969 and 1979, earnings among married women grew by about 47 percent. Among women whose husbands’ incomes were in the lowest 20 percent of the earnings distribution for husbands, earnings grew by only 31 percent. In contrast, women whose husbands’ earnings were in the top 20 percent experienced earnings growth of 71 percent.16 This reflects, in part, the growing earnings inequality among women that we have already seen among men. But it also reflects faster employment growth among married women whose husbands have earnings above the poverty threshold. Among those whose husbands were in the bottom 20 percent of the distribution, employment growth was 8 percent, while among those in the top 20 percent of the distribution, it was 14 percent. Juhn and Murphy show that a similar pattern persisted from 1979 to 1989. Among those with husbands at the bottom of the earnings distribution, earnings increased 27 percent, while among those with husbands at the top of the earnings distribution, earnings grew by 50 percent. Again, this partially reflects faster employment growth among women with high-earning husbands.

4. Concluding Remarks In 1992, Bill Clinton’s presidential campaign staff had a mantra: “It’s the economy, stupid.” To a large degree, that is the conclusion of this chapter. Most of the variation in the official poverty rate can be explained by economic factors. In a sense, therefore, this chapter underlines the importance of economic growth for the poor, as well as for the nonpoor. However, it also makes it clear that economic growth does not inevitably reduce poverty even as determined using the official definition. The growth in per capita GDP over the past thirty years has not been accompanied by rising earnings at the bottom of

15. U.S. Census Bureau, Current Population Survey, March 2005, http://pubdb3.census.gov/ macro/032005/pov/new17_100_01.htm. 16. Chinhui Juhn and Kevin M. Murphy, “Wage Inequality and Family Labor Supply,” Journal of Labor Economics 15 (January 1997): 72–97.

102 Chapter 4 the earnings distribution, and therefore poverty has not decreased as we would have expected it to do given the growth in GDP. This lesson is strengthened if we remember the difference between our definition of poverty and the official definition. The official definition remains static as standards of consumption rise. But we have argued that a person is poor if he or she does not have adequate financial resources to participate in society. The growth in per capita income over the past thirty years has undoubtedly increased the cost of participating in society in a way not captured by the official measure. The importance of labor market factors for explaining poverty suggests that increasing the earnings of low-wage workers may be a good place to focus our efforts. In the next chapter, we examine the effectiveness of policies based in the labor market. At the same time, we should not conclude from the analysis in this chapter that “it is only the economy.” As we saw, a substantial fraction of the poor are unlikely to be touched by the economy. Social security undoubtedly played an important role in reducing poverty among the elderly. Other programs, such as AFDC/TANF (Transi- tional Assistance to Needy Families, discussed in the previous chapter), have not had much impact on the poverty rate because they generally do not lift their recipients above the poverty line. This does not necessarily mean that such programs do not help the poor.

5. Further Reading Blank, Rebecca M., and Alan S. Blinder. “Macroeconomics, Income Distribution, and Poverty.” In Sheldon H. Danziger and Daniel H. Weinberg, eds., Fighting Poverty. Cambridge, MA: Harvard University Press, 1986, 180–208. Burtless, Gary, and Timothy Smeeding. “The Level, Trend and Composition of Poverty.” Focus 21 (Fall 2000): 4–8. http://www.ssc.wisc.edu/irp/focus/focus.htm. Freeman, Richard B. “The Rising Tide Lifts . . . ?” In Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty. Cambridge, MA: Harvard University Press, 2001, 97–126. Gottschalk, Peter. “Inequality, Income Growth and Mobility: The Basic Facts.” Jour- nal of Economic Perspectives (Spring 1997): 21–40. Haveman, Robert, and Jonathan Schwabish. “Economic Growth and Poverty: A Re- turn to Normalcy?” Focus 20, no. 2 (Spring 1999): 1–7. http://www.ssc.wisc.edu/ irp/focus/focus.htm. Juhn, Chinhui, and Kevin M. Murphy. “Wage Inequality and Family Labor Supply.” Journal of Labor Economics 15 (January 1997): 72–97. Kremer, Michael. “How Much Does Sorting Increase Inequality?” Quarterly Journal of Economics 112 (February 1997): 115–39. Triest, Robert. “Has Poverty Gotten Worse?” Journal of Economic Perspectives 12 (Win- ter 1998): 98–114. U.S. Census Bureau. “Detailed Historical Tables from the Current Population Survey, Historical Poverty Tables.” http://www.census.gov/hhes/www/poverty.html.

Trends in Poverty 103 6. Questions for Discussion 1. For each of the following, describe the broad trends in poverty rates over the past thirty to forty years, and explain the causes of those trends.

• The entire population • The elderly • Female-headed families • Blacks • Children • Families

2. To what extent do poverty trends differ when we use alternative measures of poverty? 3. Why does Richard Freeman believe that a “rising tide” cannot eliminate a substantial fraction of existing poverty? Are there arguments against this view? 4. Why did the official poverty rate fail to decline after 1973 even though GDP per capita grew rapidly? 5. Why might the median earnings of men be more closely related to the poverty rate than is GDP per capita? 6. Why was the poverty rate 2 percentage points higher in 1994 than in 1971 even though the median earnings of men were almost identical in the two years? 7. Why might a change in the way government makes transfers to the poor change the poverty rate even if there is no change in the amount of money transferred? 8. Did the increase in the fraction of households that are female-headed contribute significantly to the poverty rate? 9. What is meant by assortative mating? How might changes in assortative mating affect the poverty rate? 10. The labor force participation of married women increased dramatically between 1960 and 1990. How did this affect the poverty rate?

7. Appendix: Multivariate Analysis There are many factors that might affect the poverty rate, some of which we can measure, others of which we cannot. Our goal is to find a small set of factors that do a good job of explaining variation in the poverty rate. In our search, we are guided by our beliefs about the factors that are likely to be related to poverty levels. The batting average of the American League batting champion or the average rainfall in northern Scotland might be related in a statistical sense, but most of us would doubt that any such relation was anything but a statistical fluke. If we flip a coin often enough, eventually we will get ten heads in a row. If we try enough variables, something will “explain” the poverty rate.

104 Chapter 4 Table 4.1 Relation between Poverty Rate and Individual Explanatory Variables

Coefficient and Percentage of Variable Standard Error Variance Explained

Median Male Earnings (thousands of dollars) –0.53 43.5 (0.10) 50-10 Male Earnings Ratio 4.61 27.8 (1.26) Median Female Earnings, Year-Round, Full-Time 0.02 0.3 Workers (thousands of dollars) (0.07) Unemployment Rate 0.39 23.7 (0.12) Average Monthly Benefit per Recipient –0.01 2.9 (0.01) Percentage of Female-Headed Households 0.16 11.1 (0.08)

There is no agreed-upon scientific way to search for the possible explanatory variables. One reason policy researchers disagree is that they consider and exclude different variables. Each researcher begins with the factors she thinks are likely to be important. Based on the discussion and the availability of data, we chose six variables: the median earnings of men, the 50-10 male earnings ratio, the median earnings of year-round, full-time women workers, the unemployment rate, the average monthly AFDC/TANF benefit per recipient, and the proportion of households that are female-headed. How should we choose which, if any, of these factors affects the poverty rate? One approach is to look at the relation between each factor and the poverty rate individually. As we did in figure 4.10, we could plot the relation between the variable and the poverty rate and choose the best-fitting line.17 The slope of the line in figure 4.10 is reproduced in the first line of table 4.1 except that in order to use the same set of years for all variables, we have dropped information from 1966 and 2004. The poverty rate decreased by 0.53 point for each $1,000 increase in the median earnings of men. The coefficient measures the effect on the dependent variable (the poverty rate) of a one- unit change in the explanatory variable (the median earnings of men). Because in this case earnings are measured in thousands of dollars, a one-unit change is $1,000. The standard error of the estimate is underneath the slope coefficient. The ratio of the slope coefficient to its standard error is the t-statistic. As discussed in the statistical appendix to chapter 1, if the t-statistic is greater than two in absolute value, we say that it is unlikely (but not impossible) that the relation between the variables is due to pure chance and that it is statistically significant. In the case of median male earnings, the

17. “Best” has a precise sense in the statistical literature.

Trends in Poverty 105 ratio of the coefficient to its standard error is –0.53/0.10, or about five in absolute value. Thus, it is very unlikely that the relation between the poverty rate and median male earnings is due to chance. Moreover, median male earnings explains almost half the variation in the poverty rate. In the last column of table 4.1, I report the percentage of the variance in the poverty rate that is explained by the variable. This is basically a way of asking how much better the best line using this variable is at explaining the poverty rate relative to a flat line. If the best line were flat, the percentage of variance explained would be zero. If all the poverty rates fell exactly on the best line, it would explain 100 percent of the variance. The remaining rows of table 4.1 consider the other five variables, one at a time. As we saw visually earlier in the chapter, there is a strong relation between the 50-10 earnings ratio for men and the poverty rate. The t-statistic is a little less than four. There- fore, it is unlikely that the relation is due to chance. We see that, on their own, median earnings of women do not help us understand the poverty rate. In fact, our best estimate is that increasing the median earnings of women slightly increases the poverty rate. However, the t-statistic is very small, so we are not at all confident that the relation is not truly zero. So we are likely to conclude that there is no relation. Increases in the unemployment rate are associated with increases in the poverty rate. The best-fitting line says that each increase of 1 percentage point in the unemployment rate raises the poverty rate by almost 0.4 points. The t-statistic is over three, and therefore we are quite confident that the relation is not a random statistical event. The unemployment rate explains almost one-fourth of the variance in the poverty rate. The average monthly welfare benefit is not statistically related to the poverty rate. The slope of the best-fitting line is –0.01, which means that a $1 increase in the average monthly benefit lowers the poverty rate by 0.01 point. The $81 difference between the 1973 and 2003 benefits would account for a difference of only about 1 percentage point in the poverty rate. Variation in the average monthly benefit accounts for only a small fraction of the variance in the poverty rate. Finally, the relation between the proportion of households that are female-headed and the poverty rate is strong. The best-fitting line says that each increase of 1 percentage point in the proportion of female-headed households raises the poverty rate by almost 0.2 percentage point. And the t-statistic is two. Therefore, this is not likely to be a random relation. Table 4.2 takes us through the estimates presented graphically earlier in the chapter and helps us to understand the importance of considering factors simultaneously. The first two columns repeat the single-factor estimates for median male earnings and the 50-10 male earnings ratio in table 4.1. The third column shows the effect of considering these two factors simultaneously. We read it in the following way. If we were able to increase the median male earnings by $1,000 without changing the 50-10 male earnings ratio at all, we predict that the poverty rate would fall by 0.62 percentage point. We are quite confident that this relation is not purely random (i.e., that the true effect is not zero) because the absolute value of the t-statistic is about ten. This is

106 Chapter 4 Table 4.2 Regressions Used in This Chapter

Variable (1)* (2)* (3)* (4)*

Median Male Earnings (thousands of dollars) –0.53 — –0.62 –0.55 (0.10) (0.06) (0.10) 50-10 Male Earnings Ratio — 4.62 5.59 3.12 (1.26) (0.60) (1.21) Median Female Earnings, Year-Round, — — — –0.23 Full-Time Workers (thousands of dollars) (0.12) Unemployment Rate — — — 0.10 (0.03) Average Monthly Benefit per Recipient — — — –0.02 (0.01) Percentage of Female-Headed Households — — — 0.24 (0.13) Constant 29.24 2.26 18.94 28.05 (3.09) (2.92) (2.06) (4.25) Percentage of Variance Explained 43.5 27.8 84.1 89.1

*See the text for an explanation of the content of these columns.

the result shown graphically in figure 4.14. This column also tells us that if we were able to hold median male earnings constant while increasing the 50-10 male earnings ratio by 1, the poverty rate would rise by over 5 percentage points. This is the result shown graphically in figure 4.15. Note that combining these two variables greatly increases the fraction of the variance explained. The fourth column shows the remaining results presented visually earlier in the chapter and shows the numerical estimates of the effects. Each coefficient is read as follows. The coefficient is the effect of a one-unit change in the explanatory variable if none of the other explanatory variables changed. Thus, we estimated that a $1,000 increase in median male earnings is associated with a 0.55 reduction in the poverty rate. We are very confident that this is not just a random finding because the absolute value of the t-statistic is over five. Note that some results in table 4.2 look different from the results in table 4.1. In particular, in table 4.2 we find that, controlling for all the other factors, increases in median female earnings are associated with a modest reduction in the poverty rate, while in table 4.1 we find that, not controlling for these other factors, the estimated relation is positive although not significantly different from zero. In interpreting table 4.2, we need to be somewhat cautious about drawing strong conclusions about the relation between female earnings and the poverty rate. The absolute value of the t-statistic is close to two but still short of that number.

Trends in Poverty 107 chapter 5

Labor Market Policies

I          for fluctuations in the poverty rate is the evolution of wages. Poverty tends to decrease when wages rise and when unemployment falls. We have seen that a substantial minority of poor families include a worker. When the labor market is good, workers are able to get better jobs and work longer hours at higher pay. But in the previous chapter we also saw that rising overall wage levels may not reduce poverty if they are accompanied by increased wage inequality. In this case, the wages of low-skilled workers may fall even as the median or higher wages are rising. We therefore begin this chapter by addressing the determinants of wage inequality. We will see that there are two principal explanations: rising demand for skilled workers relative to unskilled workers and changes in institutions that resulted in lower wages for low-skilled workers. Policies designed to reduce the poverty rate by enabling low-skilled workers to get better-paying jobs reflect these factors. We have already discussed the earned income tax credit (EITC), which, at its most generous, increases wages of very low earners by 40 percent. In this chapter we will examine minimum wage laws and the closely related living wage laws (institutional responses to rising wage inequality), job training and related programs designed to increase the skills of individual workers and the overall supply of skilled relative to unskilled workers, and employment subsidies intended to increase the demand for unskilled workers. We will also examine programs designed to move people off welfare and into the work force.

1. Understanding Wage Inequality The study of wage inequality is worthy of an entire book by itself. There is a vast literature just on assessing and explaining the changes in wage inequality over the past forty years. We will barely scratch the surface and will focus on the relative decline in wages among low-wage workers.

108 Nevertheless, it is useful to remember that wage inequality has increased dramatically at the upper end of the income distribution as well as at the lower end of the distribution. Between 1972 and 2002, the average labor income of households in the top 10 percent of the income distribution grew from $105,000 to $152,000 (as measured by the purchasing power of the dollar in the year 2000), while the labor income of those in the top 1 percent of the distribution grew from $216,000 to $481,000 and from $464,000 to $1,657,000 among those in the top 0.1 percent of the distribution.1 The very rapid rise in incomes among those at the top of the income distribution helps explain how the per capita gross domestic product (GDP, the value of all goods produced in the country) could rise rapidly without reducing poverty. However, the increase in inequality at the lower end of the wage distribution is far more relevant for understanding why poverty failed to decline. To understand the reasons for changes in the earnings distribution, we can rely on basic supply and demand. To keep the analysis simple, suppose that there are just two types of workers, skilled and unskilled. We will also assume that a firm’s choice of technology is independent of its size. This is obviously false. Larger firms often use more machinery per worker than smaller firms in the same industry, and there is some evidence that they use more skilled workers. However, the changes we are interested in involve shifts in the number of skilled and unskilled workers at the level of the economy. The technology used in the United States probably would not differ very much if the population were 200 million or 300 million. With these simplifications, we can talk about the demand for skilled labor relative to unskilled labor and its relation to the price (or wage) of skilled labor relative to unskilled labor. Figure 5.1 shows the relation between the relative demand and the relative wage. When skilled wages are high relative to unskilled wages, firms want to hire few skilled workers and lots of unskilled workers. When skilled wages are relatively low, they want to hire mostly skilled workers and relatively few unskilled workers. Thus, the relative demand equation is downward-sloping. Workers also must decide whether to become skilled workers. Although it takes time, workers can stay in school to acquire skills or take jobs in which they will receive training to make them skilled workers. Both approaches are costly. Workers who remain in school give up at least some of their work time, and jobs with a strong training component often pay less than the worker could earn elsewhere. How willing workers are to bear these costs will depend on the relative wages paid to skilled and unskilled workers. If there is little difference in the wage, few workers will be willing to bear the cost. When the wage difference is large, many workers will delay labor market entry and remain in school to acquire the skills they need for a high-wage job. Thus,

1. These data provide the background for Thomas Piketty and Emmanuel Saez, “Income In- equality in the United States, 1913–2002,” in A. B. Atkinson and T. Piketty, eds., Top Incomes over the Twentieth Century: A Contrast between European and English-Speaking Countries (Oxford: Oxford University Press, forthcoming), and are available at http://emlab.berkeley.edu/users/saez/TabFigOUP volume2.xls.

Labor Market Policies 109 Relative Wage

Demand Supply

Excess Supply of Skilled Workers

Workers (Skilled/Unskilled) Figure 5.1 Relative Supply and Demand for Skilled and Unskilled Workers

the relative supply curve in figure 5.1 is upward-sloping. We would expect it to be quite steep in the short run because older workers are less likely to be willing to give up income in order to obtain training. However, in the long run it may be relatively flat as workers adjust their education decisions. Figure 5.1 can be interpreted similarly to a standard supply and demand diagram. If the relative wage of skilled workers is very high, there will be little demand for skilled workers and lots of demand for unskilled workers. On the other hand, there will be lots of supply of skilled workers and little supply of unskilled workers. The excess supply of skilled workers and the excess demand for unskilled workers will cause the relative wage of skilled workers to fall. As the relative wage of skilled workers falls, their relative demand rises and relative supply falls until we reach the relative wage at which the relative supply of skilled workers equals the relative demand. Similarly, if the relative wage of skilled workers is too low, the relative demand for skilled workers will exceed the relative supply. The relative wage will rise until relative supply equals relative demand. We say that the labor market is in equilibrium when relative supply equals relative demand. The figure tells us that there are effectively three types of explanations for changes in wage inequality: • Factors that shift the demand curve • Factors that shift the supply curve • Factors that cause the wage to differ from the equilibrium wage We will consider these in turn.

1.1. Demand Factors: Technological Change. One principal factor that may be responsible for shifting the demand curve for skilled workers relative to unskilled

110 Chapter 5 workers is the way goods are produced. Even casual observers notice that computers (and microchips more generally) have revolutionized the production of goods and services. Before the development of bar codes and scanners, cashiers in supermarkets either had to remember the price of a product or look for the price marked on the product. All products had to be individually priced. If they were not, there was a significant risk that the cashier would not know the price and someone would have to go to the appropriate aisle to check it. Skilled cashiers who knew the prices of many products were much faster than unskilled cashiers. Individually and collectively, skilled cashiers had significant bargaining power with supermarkets because they could not be replaced easily. With the development of scanners, cashiers could be trained much more rapidly. But scanners did much more than that. They gave supermarkets much better control over their inventories. Supermarkets could change prices rapidly in response to changing inventory conditions and resupply stores based on actual need. Thus, scanning technology may have simultaneously reduced the demand for skilled cashiers but increased the demand for people who can analyze the data created by the scanners. Similarly, the development of computers with word processing software transformed the use of secretarial support. To a large extent, the typing pool became a thing of the past. Companies had much less need for large numbers of workers whose sole job was to type material flawlessly. On the other hand, the rise of desktop publishing may have increased the demand for more skilled secretaries who are expected to use computers for tasks far more complex than simple typing. More generally, computers reduce the demand for workers capable of performing only routine mental or physical tasks for which it is possible to follow a simple set of rules or an algorithm. They increase the demand for workers who can perform more complex tasks that involve problem solving or communication.2 It is difficult to establish empirically how important this technological shift really is. David Autor, Frank Levy, and Richard Murnane examined the evolution of the nature of job tasks between 1960 and 1998.3 They found that there was a dramatic shift from routine tasks to nonroutine tasks starting in the 1970s. Moreover, these shifts were concentrated in industries and occupations undergoing rapid computerization regardless of the education level in the occupation or industry. The authors concluded that the shifts were large and favored more educated workers and more skilled workers within education groups and are thus an important part of the explanation for increased wage inequality.

1.2. Demand Factors: International Trade. In 1970, international trade was a relatively small part of the U.S. economy. Exports (and imports) were about 6 percent of GDP. By 2004, exports were 10 percent of GDP and the value of imports equaled 15

2. David H. Autor, Frank Levy, and Richard J. Murnane, “The Skill Content of Recent Techno- logical Change: An Empirical Exploration,” Quarterly Journal of Economics 118 (November 2003): 1279–334. 3. Autor, Levy, and Murnane.

Labor Market Policies 111 percent of GDP. The outsourcing of jobs to other countries had become a common concern among Americans. To understand how the growth of international trade could increase wage inequality, consider the case of two countries that we will call Developed and Low In- come. Each country uses unskilled and skilled labor to produce two goods that we will call hightech and hardwork. Hightech requires a lot of skilled labor and a little unskilled labor to produce. Hardwork is just the opposite, requiring a lot of unskilled labor and a little skilled labor. Developed has a good education system and has lots of skilled workers. Of course, not everyone in Developed takes advantage of the education system, and some are unskilled. Low Income does not have a good education system, but some lucky few manage to acquire the knowledge they need to be skilled workers. People in both countries want to consume a mixture of hightech and hardwork. When there is no international trade, companies in both Developed and Low Income produce some of each good. Because there is lots of skilled labor in Developed, it is relatively inexpensive for companies there to produce hightech. In contrast, in Low In- come there is a shortage of skilled labor, and it is expensive for companies there to produce hightech. Therefore, when there is no trade, the price of hightech relative to the price of hardwork is higher (and the price of hardwork relative to the price of hightech is lower) in Low Income than in Developed. Suppose now that Developed and Low Income decide to allow international trade. People and companies in Developed will buy hardwork from Low Income and sell hightech to it.4 This means that in Developed, companies will begin to produce more hightech, which uses more skilled labor, and to produce less hardwork, which will reduce the demand for unskilled labor. Demand for skilled labor relative to unskilled labor will rise in Developed. The opposite will happen in Low Income. If this explanation were correct, we would expect the increased demand for skilled labor in the United States to be the result of a large shift in employment toward export industries using a great deal of skilled labor and a large shift away from import industries using a great deal of unskilled labor. Because the relative cost of skilled labor has risen, we would actually expect production within industry to shift toward less skilled workers. In our example, demand for skilled labor would increase in Developed because it was producing more hightech and less hardwork, but its production of each good would use more unskilled workers and fewer skilled workers than previously. In fact, production in the United States has shifted toward more skilled workers even within industries.5 Moreover, the decline in employment in import industries and its increase

4. You may be worried that Low Income would be able to produce both goods at a lower price. If so, the people in Low Income would have nothing to buy from Developed and would be unwilling to take money from the people of Developed. So it must be the case that companies in Developed can produce one of the goods less expensively. 5. Eli Berman, John Bound, and Zvi Griliches, “Changes in the Demand for Skilled Labor within U.S. Manufacturing: Evidence from the Annual Survey of Manufactures,” Quarterly Journal of Eco- nomics 104 (May 1994): 367–98.

112 Chapter 5 in export industries are relatively small. As a result, most labor economists do not believe that international trade can explain much of the increase in wage inequality. However, this overlooks the fact that much international trade takes place within industries. In particular, the United States may buy parts of a product from other countries and then assemble the product in the United States. To borrow the terminology from the previous subsection, the United States may import the parts of products whose production involves routine tasks and produce those parts whose production relies on nonroutine tasks in the United States. If so, it will be very hard to distinguish between the technology and international trade explanations.6

1.3. Supply Factors: Education and Immigration. If the supply of skilled labor decreased relative to the supply of unskilled labor, that could explain the rise in wage inequality. In fact, education levels have been rising among Americans of working age. Among Americans 25 years of age or older, in 1959 some 44 percent had four years of high school and 8 percent had graduated from college. By 1999, these figures stood at 83 percent and 25 percent. However, the increase has not been entirely steady. The college graduation rate increased by over 6 percentage points between 1973 and 1983 but by only 3 percentage points between 1983 and 1993. The rate of increase in the proportion of the population with high school educations was very high in the 1960s and early 1970s but started to slow down after that. If there is a constant tendency for demand factors to increase the relative demand for skilled labor, changes in the supply of educated labor may influence the timing of the increase in wage inequality, but cannot, on their own, explain the rise in wage inequality. In 2000, over 10 percent of the U.S. population was foreign born.7 Among those aged 18 to 64, the proportion was 13 percent. Thus, immigrants represent a substantial fraction of the working-age population of the United States. Immigrants are much less likely than native-born Americans to have graduated from high school (or the equivalent) and are much more likely to have less than nine years of education. There- fore, immigrants represent a substantial minority of the very low-skilled labor market and noticeably increase the number of workers in this market. Economic theory strongly suggests that, by increasing the supply of low-skill workers, immigration will lower the wages of unskilled workers. It has been surprisingly difficult to find evidence to support this theory. One approach is to compare wages in cities with high and low proportions of immigrants, but immigrants are probably more mobile than native-born workers and are likely to be drawn to booming cities offering high wages. And when there are large inflows of immigrants for noneconomic reasons, there may be less native-born migration as native- born workers avoid markets where wages are depressed by these inflows.

6. Robert Feenstra and Gordon Hanson, “Global Production Sharing and Rising Inequality: A Survey of Trade and Wages” (Working Paper 8372, National Bureau of Economic Research, Cam- bridge, MA, 2001). 7. Lisa Lollock, The Foreign Born Population in the United States: March 2000, Current Popula- tion Report P20-534 (Washington, DC: U.S. Census Bureau, 2001).

Labor Market Policies 113 George Borjas points out that there is considerable variation in the distribution of posteducation labor market experience among natives and immigrants.8 For example, in 2000 half of high school dropouts with sixteen to twenty years of experience were immigrants, compared with only one-fifth of high school dropouts with less than five years of experience. And between 1960 and 2000, there was substantially more growth in the fraction of immigrants among the former than among the latter. Therefore, we would expect wages to have risen less over this period among high school dropouts with sixteen to twenty years of experience than among high school dropouts with less than five years of experience. Generalizing this argument, Borjas estimates that between 1980 and 2000, immigration lowered the wages of high school dropouts by about 9 percent and those of college graduates by about 5 percent, while it had less impact on high school graduates and those with some college.

1.4. Institutional Factors: Unions and Deregulation. There are a number of reasons that wages may not be fully determined by the forces of supply and demand. The most obvious reason, and one that we will explore in detail later, is that government may set a minimum wage that firms may pay workers. In addition, workers may form unions to negotiate with firms rather than allowing supply and demand to dictate their wages. Firms may be less resistant to unions when their prices are set by government on the basis of cost than when they are set by the market. Even in the absence of unions, firms that can pass cost increases on to customers because of government regulation may choose to pay workers more than the going wage in order to cut down on turnover and gain worker loyalty. In the late 1970s and early 1980s, there was considerable deregulation. Airlines and trucking, both highly unionized industries, were freed from price regulation, and ATT, the monopoly phone service provider, was broken up and subjected to competition. Union membership in the private sector has declined precipitously since its peak in the 1950s. By 2003, only about 8 percent of private sector workers and 12 percent of all workers were union members. Both of these changes have significantly reduced the availability of high-paying jobs for unskilled workers.9

1.5. Rising Wage Inequality: Some Final Remarks. In the end, most economists believe that there is no single explanation for the increase in inequality. Demand side factors are certainly an important part of the story. But it is not clear that we can really distinguish between the effects of trade and the effects of technology. Changes in technology have made outsourcing more feasible. To some extent, routinizing work allows technology to replace native workers, and to some extent, it allows foreign workers to replace them. Moreover, foreign competition is probably part of the explanation for the

8. George J. Borjas, “The Labor Demand Curve Is Downward Sloping: Reexamining the Im- pact of Immigration on the Labor Market,” Quarterly Journal of Economics 118 (November 2003): 1335–74. 9. See Nicole M. Fortin and Thomas Lemieux, “Institutional Changes and Rising Wage In- equality: Is There a Linkage?” Journal of Economic Perspectives 11 (Spring 1997): 75–96.

114 Chapter 5 decline of unions. In some cases, technology weakened unions. Even the evolution of the statutory minimum wage, which we will discuss next, may have been influenced by these factors. Understanding the causes of the decline in the wages of unskilled workers does not necessarily help us determine policy to address it. If technological change or international trade is reducing demand for routinized tasks, the best solution is almost certainly not to try to prevent technological change or international trade. Whatever the source of the growing wage inequality, the serious policy discussions focus on raising the wages of low-skilled workers and on turning low-skilled workers into higher-skilled workers. We therefore turn to two policies, minimum wage laws and living wage laws, designed to raise wages in the low-wage labor market. We will then proceed to a discussion of job training programs.

2. Minimum Wage Laws In the United States, minimum wage laws were initially passed by the states. Many of the early laws were part of a movement to create protective legislation that would limit the involvement of women and children in the labor force. It is striking that initially most state minimum wage laws covered only women and children, and one covered only women and girls. From their outset, minimum wage laws were therefore designed to alter the composition of the labor force. Depending on one’s political perspective, we can say that they were designed to stop employers from exploiting women and children or that they were designed to protect male employment.10 When we analyze the effects of current laws, it is important to remember that this redistributive effect of minimum wage laws persists today, albeit probably in quite different ways. Minimum wage laws affect not only wages and employment levels, but also who obtains the minimum wage jobs. In 1938, the Fair Labor Standards Act (FLSA) established the first national minimum wage. It was 25 cents and covered about 50 percent of nonsupervisory, nonagricultural workers in the private sector.11 Over the years it has expanded; it now stands at $5.15 and covers about 85 percent of private nonagricultural workers as well as government and agricultural workers. There is no mechanism for automatic increases in the minimum wage. Congress sporadically votes to increase the minimum wage, but may not change it for many years. Relative to prevailing wages, the federal minimum wage is currently quite low. Federal law governs when it is higher than the state minimum, but many states have laws that cover additional workers or set higher minimum wages. For example, in Massachusetts the minimum wage is currently $6.75, and the governing law contains

10. See Elisabeth M. Landes, “The Effect of State Maximum-Hours Laws on the Employment of Women in 1920,” Journal of Political Economy 88 (June 1980): 476–94, for a discussion of the effect of maximum hours laws on women’s employment. 11. Charles Brown, “Economics of the Minimum Wage,” in Orley Ashenfelter and David Card, eds., Handbook of Labor Economics, Vol. 3B (Amsterdam: Elsevier, 1999), 2101–64.

Labor Market Policies 115 a provision that the Massachusetts minimum must be at least 10 cents greater than the federal minimum. Note that coverage and impact can be quite different. Just because 85 percent of workers are covered by the federal minimum wage does not mean that 85 percent are affected by it. Many of the uncovered workers are in state or local employment and earn considerably more than the federal minimum wage even though they are exempt from the provisions of the FLSA. Similarly, managers are exempt but are rarely paid less than the minimum wage. And many workers who are covered are also unaffected by the minimum wage because they earn considerably more than the statutory minimum. At the same time, enforcement of the federal minimum wage law is relatively weak, so there is considerable noncompliance.12 The Census Bureau reports that in 1992, when the minimum wage stood at $4.25, 4.3 percent of workers earned less than the minimum wage and 2.8 percent earned exactly the minimum wage.13 The fact that less than 3 percent of workers were earning the minimum wage suggests that the effects of the minimum wage are likely to be small. It is possible that 3 percent underestimates the number of workers affected by the minimum wage law, because firms may adjust the wages of workers earning less than the minimum wage or somewhat above it in response to a minimum wage increase. At the same time, it is clear that the minimum wage law has some effect on wages, because there is a tendency for more workers to receive exactly the minimum wage than a little bit more or a little bit less.14 The minimum wage is particularly important for young workers. The same Cen- sus Bureau report that found less than 3 percent of workers earning the minimum wage reveals that 11.9 percent of teenage workers earned exactly the minimum wage and 14.8 percent earned less than the minimum wage. David Card and Alan Krueger studied the effect of raising the minimum wage from $3.35 to $3.80 in April 1990 and to $4.25 in April 1991.15 They report that almost half of those earning $3.35 to $4.24 per hour prior to the minimum wage increases were 16 to 24 years old, and of those, about 60 percent were enrolled in school. Only 20 percent were living in families with incomes below the poverty line. It is evident that, by itself, the minimum wage cannot have a large effect on poverty. Card and Krueger estimate that fewer than 30 percent of workers in families in the bottom 10 percent of the family income distribution were affected by the minimum wage increase. Still, their results show that for the workers affected, the impact on earnings was not trivial. An average worker affected by the increase, if the wage change did not alter his hours, earned about $25 more per week. The results of Card and Krueger are consistent with what we would expect on the basis of this information.

12. Orley Ashenfelter and Robert S. Smith, “Compliance with the Minimum Wage Law,” Jour- nal of Political Economy 87 (April 1979): 333–50. 13. Cited in Brown, 2143. 14. Brown, 2144. 15. David Card and Alan B. Krueger, Myth and Measurement (Princeton, NJ: Princeton Univer- sity Press, 1995), chapter 9.

116 Chapter 5 The increase in the minimum wage had little effect on the poverty rate but did increase the family incomes of families at the lower end of the income distribution. As usual in economics, there is no free lunch. The minimum wage may reduce family income inequality, but economic theory tells us that this reduction in inequality comes at a cost. By raising the wages of low-skilled workers, the minimum wage induces firms to hire fewer workers with low skills (perhaps substituting automated equipment or more skilled workers), to cut back on nonwage payments such as employee dis- counts or free meals, and to increase the pace and unpleasantness of work. There is considerable controversy regarding the importance of these effects, particularly that of employment loss. Although economic theory tells us that employment must decrease, it does not tell us by how much. It is very difficult to determine the effects of changes in the minimum wage law on employment. This reflects two factors. First, any effects are likely to be modest. Suppose we raise the wage of workers by 45 cents from $3.80. This represents an increase of about 12 percent. Estimates typically suggest that a 12 percent increase in wages should reduce employment by about 6 percent. Because fewer than 10 percent of workers are affected by the minimum wage increase, the total reduction in employment would be only about 0.6 percent. We are looking for the proverbial “needle in a haystack.” To make this problem less severe, we concentrate on groups in which we would expect the effect to be larger. Many studies look at teenagers, but if employers respond to the minimum wage effect by hiring different types of workers, this can be misleading. A higher minimum wage might attract more student applicants and lead employers to use students instead of low-skilled adults, or it might attract low-skilled housewives, whom employers prefer to teenagers. Shulamit Kahn and I find that the former type of substitution occurs in the food industry, though the latter may occur in retail.16 The second factor is that the national minimum wage affects all workers. To find some workers for whom the minimum wage increased and others for whom it did not, we have to look at changes in state minimum wage laws, but these changes affect relatively few workers. The most carefully studied of these changes was an increase in the minimum wage in New Jersey. In 1992, New Jersey raised its minimum wage by 80 cents. Both before and after the minimum wage increase, Card and Krueger surveyed fast-food restaurants near the New Jersey–Pennsylvania border. The authors make a compelling case that, other than the minimum wage increase, there was no reason to expect fast-food employment to evolve differently in the two states. Instead of finding the relative decline in employment expected in New Jersey, they found that employment remained more or less steady in New Jersey while declining in Pennsylvania.17 David Neumark and William Wascher criticized aspects of the Card and Krueger study and used data collected from employment records to replicate the original study. They found that employment

16. Kevin Lang and Shulamit Kahn, “The Effect of Minimum Wage Laws on the Distribution of Employment: Theory and Evidence,” Journal of Public Economics 69 (July 1998): 67–82. 17. Card and Krueger, chapter 2.

Labor Market Policies 117 Tenth Percentile Earnings ($2003 Thousands)

1973 19 1975

1979 1978 18 1974 1972 1969 1976 1980 1971 1977 1970 1981 17 1988 1987 1982 1999 1989 2003 2002 1985 2001 1986 1998 1997 16 2000 1984 1983 1995 1991 1968 1990 1992 1996 1993 15 1994

1967 14 586 7 Minimum Wage ($2003) Figure 5.2 Minimum Wage and Tenth Percentile Earnings of Adult Male Year-Round, Full-Time Workers, 1967–2003

in New Jersey declined somewhat faster than in Pennsylvania, but the effect was not large.18 There is a growing consensus among labor economists that both the negative employment effects of minimum wage laws and the positive income distribution effects are small. At the same time, most labor economists do believe that minimum wage laws reduce earnings inequality.19 Figure 5.2 shows the relation between the real minimum wage and real tenth percentile earnings of adult male year-round, full-time workers. Each $1 increase in the hourly minimum wage is associated with about a $500 increase in tenth percentile earnings. This “effect” should be treated with extreme caution, because it does not control for anything else going on in the economy.20

18. David Neumark and William Wascher, “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania: Comment,” American Economic Review 90 (December 2000): 1362–96. 19. John Dinardo, Nicole Fortin, and Thomas Lemieux, “Labor Market Institutions and the Dis- tribution of Wages, 1973–1992: A Semiparametric Approach,” Econometrica 64 (September 1996): 1001–44. 20. David H. Autor, Lawrence F. Katz, and Melissa S. Kearney, in “Trends in U.S. Wage In- equality: Re-Assessing the Revisionists” (Working Paper 11627, National Bureau of Economic Research, Cambridge, MA, 2005) point out that a higher real minimum wage is associated with a smaller difference between ninetieth and fiftieth percentile earnings. It is hard to believe that this association is causal.

118 Chapter 5 Poverty Rate 1983 1993 15 1982 1992 1994 1984 1991 1967 14 1985 1981 1995 1996 1986 1990 1987 1997 13 1988 1980 1989 1968 2004 1998 1970 2003 1971 1975 12 2002 1972 1969 1999 1976 2001 1977 1979 1978 2000 1974 11 1973 586 7 Minimum Wage ($2003) Figure 5.3 The Minimum Wage and Poverty, 1967–2004

Because we saw in the previous chapter that increases in the earnings of low-earning men and women reduce poverty, it may seem surprising that minimum wage laws should have no measurable effect on poverty. The critical term here is measurable. The minimum wage has a modest effect on the earnings of low-earning workers, and that, in turn, has a modest effect on the poverty rate.21 As shown in figure 5.3, the simple relation between the minimum wage and the poverty rate is consistent with this prediction. Unfortunately, although it is suggestive, figure 5.3 is not compelling. As always, we must be cautious when relying on estimates based on only thirty-seven years of data. There are myriad other factors that may have affected the poverty rate and that may have changed in ways similar to the minimum wage. It is impossible to control for all of them. When we control for the factors addressed in chapter 4, we estimate that a $1 increase in the real minimum wage is associated with a reduction of 0.3 percentage point in the poverty rate. However, the t-statistic is only 1.8, so we cannot be confident that there really is a relation between the minimum wage and the poverty rate.

21. The effect is also sufficiently small that it is unlikely to be detectable in studies that rely on changes in the minimum wage at the state level. Poverty rates calculated at the state level are very imprecise. Recall that at the national level, the standard error of the poverty rate estimate is about 0.2, which means that for the average state it is about 1.4. It is possible to show that if a single state increased its minimum wage by one dollar, the standard error of the estimated effect on the poverty rate would be about 1.2 and the expected t-statistic would be from about 0.5 to 1.2, which would fall well short of statistical significance.

Labor Market Policies 119 3. Living Wage Laws In 1967, a worker earning the federal minimum wage of $1.40 earned enough to lift a family of three out of poverty simply by working full-time. To lift a family of four out of poverty required less than forty-eight hours’ work per week. In 2002, a minimum wage worker in a family of three had to work fifty-four hours per week (with no vacation), and one in a family of four had to work sixty-nine hours per week.22 As the federal minimum wage has declined relative to prevailing wages and the poverty threshold, political activists have shifted some of their focus from trying to raise the federal minimum wage to increasing state minimum wages and enacting legislation at the municipal level. It is generally infeasible for cities and towns to establish local minimum wage laws. Except for retail establishments tied to the locality, low- wage businesses would have a strong incentive to move to nearby communities without a local minimum wage. The establishments that remained in the community would complain about their inability to compete with nearby stores that could hire workers at lower wages. Since the mid-1990s, activists have worked to use the direct economic power of local governments to raise wages through “living wage laws.” Living wage laws come in three forms. First, the law can simply require that no municipal worker be paid less than some wage defined as a living wage. Of the major cities with living wage laws governing municipal workers, the living wage ranged from $7.00 in Dayton, Ohio, to $10.36 in San Jose, California.23 It is not clear why municipalities pass such laws, because they are presumably free to set wages as high as they wish when they are not constrained by col- lective bargaining, and it is unlikely that unions would force them to pay wages lower than the municipality wishes. This may reflect a difference in the position of the legislative branch, which passes the laws, and the executive branch, which is generally responsible for negotiating wages. The second type of living wage law governs payments to employees of municipal contractors. Communities with municipal contractor living wage laws typically require that the contractor pay employees the living wage on any contract performed for the municipality. It is not clear to what extent contractors can get around the law by reducing wages on work not covered by the living wage law to offset the higher wage on municipal contracts or by shifting higher-paid workers to municipal contract work.

22. The reader may recall that poverty thresholds depend not only on the number of people in a household but also the composition of the household. These calculations are based on average thresholds weighted by the frequency of different household compositions given household size. These averages are reported in U.S. Census Bureau, Historical Poverty Tables, table 1: “Weighted Average Poverty Thresholds for Families of Specified Size: 1959 to 2002,” http://www.census.gov/hhes/ poverty/histpov/hstpov1.html. 23. The information in this section comes from Scott Adams and David Neumark, “Living Wage Effects: New and Improved Evidence,” Economic Development Quarterly 19 (February 2005): 80–102, and “The Economic Effects of Living Wage Laws: A Provisional Review,” Urban Affairs Re- view 40 (November 2004): 210–45.

120 Chapter 5 Finally, living wage laws may apply to employees in any firms receiving business assistance, such as tax breaks or grants from the municipal government. It seems likely that this requirement is more difficult for businesses to avoid because it applies to all their workers. A major concern with living wage laws, as with minimum wage laws, is that they will reduce employment. Most of the living wages are considerably higher than the state or federal minimum wage. Consequently, government will find it more expensive to employ low-skilled workers, and therefore will employ fewer. Low-skilled workers will become relatively more expensive on government contracts. Contractors, seeking to be the lowest bidders on government contracts, will shift their labor use to higher- skilled workers, as will employers receiving business assistance. It is not clear how large we would expect these effects to be. The same political forces that encourage municipal governments to pass living wage laws may discourage them from reducing employment of low-skilled workers. All government contractors will be required to pay higher wages so that the competitive pressure to reduce the use of low-skilled workers may be small, and government may give more assistance to employers who promise to hire more low-skilled workers. At this point, living wage laws are new, and we have relatively little information on their impact. Scott Adams and David Neumark estimate that business assistance laws have a larger effect than contractor laws (although this may reflect the tendency for business assistance laws to be more common in areas where neighboring cities also have living wage laws) and that living wage laws raise wages and lower employment (as is the case with minimum wage laws) but that their overall effect is to reduce poverty.24

4. Job Training Programs If much poverty can be attributed to the inability of workers to earn enough to lift themselves out of poverty, and if people with jobs requiring high levels of skill earn more than those in jobs requiring less skill, another natural response is to help people earn higher incomes by training them to work in more skilled jobs. Government training programs with this purpose have been in existence at least since the 1960s. Their roots can be traced to the Works Progress Administration of the depression era and perhaps further. The Manpower Training and Development Act (MTDA) of 1962 focused on workers who had been displaced by changing technology as well as workers whose skills were inadequate for the modern labor force. Over the next eleven years, a variety of job training programs were developed, the most important of which was Job Corps, an extensive program targeted at disadvantaged youths. In 1973, the MTDA was replaced with the Comprehensive Employment and Training Act (CETA), which devolved considerable control of the program to state and local authorities and included a significant job creation program whereby participants

24. Adams and Neumark, “Living Wage Effects” and “The Economic Effects of Living Wage Laws.”

Labor Market Policies 121 could be employed by state and local governments. Although its proponents supported CETA as a mechanism for creating jobs for hard-to-employ workers and giving them work experience, its opponents argued that many CETA jobs were simply “make- work” and that employment in low-skilled jobs in the public sector did not prepare trainees for private sector employment. CETA was, in turn, replaced by the Job Train- ing Partnership Act (JTPA) in 1982. JTPA continued the process of devolution of authority to the states and moved away from job creation, focusing on efforts to involve the private sector. Most recently, Congress passed the Workforce Investment Act (WIA) in 1998. The WIA sought to reduce the overlap of the myriad federal training programs. Under the WIA, the Department of Labor was to provide “one-stop service.” All centers were to provide intake and orientation, assess whether individuals were eligible for services under the WIA, and determine what skills and abilities they required. They were also to provide information about available training programs and information on filing for unemployment insurance. Beyond these core services, WIA provides intensive services largely related to assisting with the job search process (resume preparation, career counseling, job clubs) and also provides training services. The WIA and its predecessors have been primarily voluntary programs. For the most part, they have been available to disadvantaged and other workers needing assistance but have not been mandatory. At the same time, there have been programs that have been run for individuals receiving welfare. Starting in 1967, the Work Incentive Program (WIN) required single recipients of Aid to Families with Dependent Children (AFDC) who did not have preschool children to participate in training. This mandate also applied to married heads of households regardless of whether they had preschool children. However, funding for WIN was quite limited. The consensus is that WIN had little effect.25 In 1988, WIN was replaced with the Job Opportunities and Basic Skills program (JOBS). JOBS was much more extensive than WIN and was designed to create the expectation that welfare recipients who were able to work would either be engaged in employment or receive training. Under JOBS, states were required to offer training, educational programs, community work experience, and job search assistance.26 JOBS was available exclusively to AFDC recipients but still served only about one in eight of the adults on AFDC. Funding was only about $1 billion and required matching funds from states.27

25. Daniel Friedlander, David H. Greenberg, and Philip K. Robins, “Evaluating Government Training Programs for the Economically Disadvantaged,” Journal of Economic Literature 35 (Decem- ber 1997): 1809–55. 26. Ahmad Zargari, “Vocational-Technical Education’s Role in Welfare Reform: Providing Em- ployability Skills for Welfare Recipients,” Journal of Industrial Teacher Education 34 (Spring 1997), http://scholar.lib.vt.edu/ejournals/JITE/v34n3/Comments.html. 27. Demetra Smith Nightingale, “Work-Related Resources and Services: Implications for TANF” (Paper A-7 in the series Assessing New Federalism: Issues and Options for States, Urban In- stitute, Washington, DC, 1997).

122 Chapter 5 Legislation in 1996 and 1997 abolished the JOBS program and replaced it with welfare-to-work (WtW) grants to states and communities. The WtW program is aimed at moving recipients of Transitional Assistance to Needy Families (TANF) into unsubsidized employment. States are required to match federal funds at a rate of one dollar of nonfederal funds for every two dollars of federal WtW funds.

5. Can Job Training Programs Reduce Poverty? For 1999, the Government Accounting Office identified sixteen job training programs (listed in table 5.1) “that required that their participants be economically disadvantaged in order to be eligible for services.” Collectively these programs cost $6.2 billion and served almost three million individuals. With approximately sixteen million poor people aged 18 to 64 in the United States, it might appear that these programs could rapidly eliminate poverty.28 More careful examination suggests that this is unlikely. The largest program, consisting of training programs under TANF, served 875,000 individuals in 1999 at a cost of less than $1,000 per individual served. The second largest program, the Food Stamp Employment and Training Program, served 700,000 individuals at a cost of $266 per person served. To put these numbers into perspective, consider the costs and benefits of education. In 1999, the United States spent an average of about $6,500 per pupil on primary and secondary public education.29 Typical estimates of the return to education are about 8 percent. This is probably a reasonable target for job training programs. On this basis, in order to generate an increase of 10 percent in a trainee’s earnings capacity, we would expect to spend over $8,000. Only three programs, welfare-to-work grants to states and localities, the Job Corps program, and the tiny program of discretionary awards from Community Services Block Grants reach this level of expenditure. To- gether these programs serve just over 200,000 individuals. Many of the lowest-cost programs focus on very low-cost training services in areas such as resume writing, interviewing skills, and job search. These services may provide a large public benefit by reducing the need for financial assistance to jobless workers, but they are unlikely to substantially increase the workers’ long-run earnings capacity. For the low-cost programs, there is also concern that their main effect is not to increase total employment but to juggle the distribution of jobs. Workers whose job search skills have improved may simply displace workers who would otherwise have found a job. Although this is a legitimate concern, there is little evidence to support or

28. U.S. General Accounting Office, “Multiple Employment and Training Programs” (GAO-01- 71, General Accounting Office, Washington, DC, October 2000). An updated version of the report for 2002 found thirteen programs in this category serving about 2.2 million (GAO-03-589, April 2003). Unfortunately, the report does not give information on expenditure on the training components of the programs. For those programs entirely devoted to training, we can compare the 2002 and 1999 costs per trainee. The 1999 costs give a good picture of the intensity of the current programs. 29. http://www.census.gov/Press-Release/www/2002/cb02-69.html.

Labor Market Policies 123 993 563 266 724 — 1,602 5,051 1,815 2,489 4,358 7,047 2,173 1,083 18,505 13,280 10,526 ($1999) Cost per Worker Cost per hington, DC, 2,970 1,600 2,000 2,767 4,136 42,200 70,685 480,000 101,000 700,000 142,500 293,724 875,000 120,000 Unknown 2,838,582 0.9 3 15 68 27 20 871 440 186 731 869 130 1,500 6,168 1,308 Unknown Amount Spent Amount (millions of dollars)Trained Number Training Programs” (GAO-01-71, General Accounting Office, Was Office, General Accounting (GAO-01-71, Programs” Training Federal Training Programs for Disadvantaged Workers, 1999 Workers, for Disadvantaged Programs Training Federal United States General Accounting Office, “Multiple Employment and Employment Office, “Multiple General Accounting States United October 2000). October Welfare-to-Work Grants to States and Localities to States Grants Welfare-to-Work II-B) (Title and Employment Youth Summer JTPA TANF JTPA Adult Training (Title II-A) (Title Training Adult JTPA Community ServicesProgram Senior Employment Table 5.1 Table Community ServicesGrant Block Total–Excluding Source: Job Corps Job Food Stamp Employment and Training Program Training and Employment Stamp Food Awards Community ServicesGrant–Discretionary Block Youth–Youthbuild Opportunities for and Alaska Natives Tribes Recognized to Federally Grants Welfare-to-Work Homeless Veterans Reintegration Project Reintegration Veterans Homeless School Equivalence Program Education–High Migrant Community ServicesGrant Block JTPA Youth Training (Title II-C) (Title Training Youth JTPA Workers Farm and Seasonal Migrant

124 Chapter 5 refute it. If the low-cost programs increase the effectiveness of employment search, in principle, they could increase total employment. At any given time, there are lots of unemployed workers seeking jobs and many job vacancies for which firms would like to find workers. To some extent, the simultaneous existence of unemployed workers and vacancies reflects a mismatch between the location of vacant jobs and the skills required for them and the locations and skills of workers. However, to some extent it simply reflects the inability of workers and firms to find each other. Improving the matching process can reduce both vacancies and unemployment. Unfortunately, there is no convincing evidence on how much job search training raises a worker’s employment prospects by increasing total employment rather than by displacing other job seekers.

6. Evaluating the JTPA During its existence, JTPA was the largest source of federal employment-related funds for the economically disadvantaged generally. About $2.5 billion of its annual $5 billion budget was allocated to job training for low-income adults, youths, and dislocated workers. Roughly one-fourth to one-third of participants received AFDC. However, JTPA served only about 2 percent of adult welfare recipients. In the JTPA adult programs, AFDC recipients accounted for 35 percent of the participants. In the youth programs, AFDC recipients accounted for 27 percent of program participants. Al- though 135,000 AFDC recipients in total were in JTPA activities annually, this represents only a small portion (2 percent) of the adult AFDC population.30 It is too early to evaluate the WIA. However, looking at evaluations of the JTPA can help us understand the effectiveness of training programs. Before we review the research, it is worth spending some time to understand why it is so difficult to evaluate training programs. The simplest review we might do would be to look at the employment and work experience of individuals who completed such a program. If we did, we would find that they had low wages and earnings relative to the general population. After all, the types of people who participate in voluntary training programs are those who have had difficulty obtaining good jobs. Even after training, they are generally less skilled and less educated than the average worker in the labor market. Training was not designed or expected to overcome all of their disadvantages, merely to reduce them. Obviously, we need to compare the performance of graduates of training programs with a group of individuals who are similar except for not having received the training. The trick is to figure out who is similar. Researchers have considered using dropouts from the programs, individuals who applied and were turned down, and people who have similar characteristics (e.g., education) to those who participate in training programs. Robert Lalonde has shown that our assessment of the effectiveness of training programs is very sensitive to our choice of comparison group.31 Yet each of these

30. Nightingale, 2. 31. Robert J. Lalonde, “Evaluating the Econometric Evaluations of Training Programs with Ex- perimental Data,” American Economic Review 76 (September 1986): 604–20.

Labor Market Policies 125 comparison groups is problematic in some way. People who drop out of a training program are likely to be less motivated than those who complete it. Why were some people turned down for the program? Perhaps the program expected to be evaluated on the basis of its success rate and therefore selected those most likely to find employment. Or perhaps the program sought out the most needy applicants. Among people who look alike on paper, it is the less successful who are likely to seek out training programs. Bill Gates may be a college dropout, but he is less likely to use JTPA services than most college dropouts. One solution to our problem is to compare people with themselves before and after training. At first blush, there is no reason to think that the sorts of people who attend job training programs are the sorts of people who would experience unusual wage or employment gains even in the absence of the program. However, Orley Ashenfelter has shown that workers who entered training programs frequently suffered a decline in earnings before they entered the program. If people chose to participate in a training program in response to a temporary decline in their earnings, we may credit the training program with the rebound from the temporary decrease even though they would have experienced the rebound anyway.32 This so-called Ashenfelter dip is important for a wide variety of evaluations. De- spite using techniques (bleeding, leaches) that we now know to be harmful rather than helpful, doctors were once credited with cures using those techniques. Why is this? One explanation is that the human immune system frequently combats disease on its own. People who go to the doctor often are experiencing a temporary decline in health from which they would recover in the absence of medical intervention. They seek medical treatment, note that prior to medical treatment their health was declining and that it improved after medical treatment, and draw the obvious but erroneous conclusion that the medical treatment improved their health. It is equally easy for evaluators to conclude that training programs restored workers to labor market health. The difficulty is distinguishing between natural recovery and truly effective interventions. Given the difficulties with analyzing job training programs, the most convincing evaluations are based on experimental evidence. Under the experimental evaluations, individuals who qualify for the JTPA program are randomly assigned to participate or not participate. We can then follow the two groups and see whether the outcomes were better for the individuals who participated in the JTPA programs. Even here we must be careful in our analysis. Participation in the relevant JTPA program may be randomly assigned, but not all people who apply for and are assigned to a JTPA program actually take part in the program. Some drop out, and some choose not to participate in the first place. Also, some of the “controls” who were not assigned to the JTPA program are accepted into a different program. Assignment to the JTPA program or the control group is random. Actual participation may still be related to factors not observed by the researcher.

32. Orley C. Ashenfelter, “Estimating the Effect of Training Programs on Earnings,” Review of Economics and Statistics 6 (February 1978): 47–57.

126 Chapter 5 Average Quarterly Average Quarterly Earnings ($) Earnings ($) 2,500 2,500 A) Adult Women B) Adult Men 2,000 2,000 Treatment 1,500 1,500 Control 1,000 1,000

500 500 112463 5798 0112463 5798 0

2,500 2,500 C) Female Youth D) Male Youth Nonarrestees 2,000 2,000

1,500 1,500

1,000 1,000

500 500 112463 5798 0112463 5798 0 Quarters after Quarters after Random Assignment Random Assignment Figure 5.4 Earnings of Participants and Controls in the National Job Training Partner- ship Act Study Source: Howard S. Bloom et al., “The Benefits and Costs of JTPA Title II-A Programs: Key Findings from the National Job Training Partnership Act Study,” Journal of Human Resources 32 (Summer 1997): 549–76.

This poses a problem. We are interested in the effect of participating in the program, not the effect of being assigned to the program group rather than the control group. If most people assigned to the program group do not participate, the difference between those assigned to the program group and those assigned to the control group may be small even if program participation has a big effect. Therefore, we need to adjust the difference between the program and control groups for the fraction of the program group actually participating in the program. The appendix to this chapter explains the statistical process of making this adjustment. The JTPA evaluation looked at the effect of training on four groups: adult women, adult men, female youths, and male youths who had not been arrested. Data issues precluded a useful evaluation of the effect of the program on young male arrestees. Within each group, participants were randomly assigned to a JTPA program or to a control group. They were followed for two and a half years following the start of the experiment.33 Figure 5.4 shows the earnings of members of the treatment and control groups. Earnings for all groups rose rapidly in the first quarter following assignment (reflecting

33. Bloom et al. The following results are from this same source.

Labor Market Policies 127 Table 5.2 The Impact of JTPA on Enrollee Earnings as a Percentage of Nonenrollee Earnings

1–6 Months 7–18 Months 12–30 Months Total

Adult Women 8.6* 16.5*** 15.6*** 14.8*** Adult Men 5.8 6.5 10.5** 8.0 Female Youths –0.3 1.3 3.5 2.0 Male Youth Nonarrestees 2.4 –4.2 –8.1 –5.0

Source: Howard S. Bloom et al., “The Benefits and Costs of JTPA Title II-A Programs: Key Find- ings from the National Job Training Partnership Act Study,” Journal of Human Resources 32 (Sum- mer 1997): 549–76, table 3. Notes: Columns reflect number of months since beginning the program for participants and since being assigned to the control group for those in the control group, that is, the number of months from the start of the experiment. *Significant at the .1 level. **Significant at the .05 level. ***Significant at the .01 level.

the Ashenfelter dip) regardless of the group to which individuals were assigned. If we had simply followed the participants and not compared them with a randomly assigned control group, we might have attributed this rise to participation in the JTPA. However, we can see that in the case of youths, this would have been misleading. The earnings of female youths in the control and JTPA groups track each other closely. For the male youths, earnings were, if anything, somewhat higher in the control group. For adult women, and to a lesser extent adult men, earnings did rise more rapidly for people assigned to the program group then for people assigned to the control group. As discussed earlier, not everyone who was assigned to participate in the JTPA followed through on that assignment. Some found jobs before their training began or chose not to participate for some other reason. A few of those assigned to the control group were able to get training at some other center. Taking this into account, how large were the effects on those who actually participated? Table 5.2 shows the results of this exercise. For adult women, we find large and rapid effects, reaching roughly 15 percent starting in the seventh month. For men, the effects developed more slowly and reached only 10 percent but are statistically significant in the long run. For young individuals, the effects were small, sometimes negative, and are not statistically significant. The earnings increases experienced by the adults participating in the program are statistically significant, but were they large enough to make the participants better off and to justify the spending on the program? We can ask this question in three ways: Do participants benefit from the JTPA? Is public expenditure decreased by the JTPA? If participants benefit but public spending is increased, do the benefits to the participants outweigh the increased public expenditures? We will work through this exercise for adult women. The numbers are similar for adult men.

128 Chapter 5 Over the thirty months of the experiment, adult women earned an extra $1,837, of which $154 was a wage subsidy provided as part of the program. In addition, they spent $56 less on training costs. These benefits were offset by an increase of $236 in taxes paid and a reduction of $235 in their welfare benefits. Thus, their net benefits were $1,422. The direct cost of the program was $1,381, which was partially offset over the thirty months by the increase in taxes and the reduction in welfare benefits, leaving an increase in expenditure of $910. Thus, over the thirty months following the start of the experiment, participants gained but government expenditure increased. The gains to the participants exceeded the cost to government by $512. If we value a dollar of gains to the participants at least as much as a dollar of cost to government, the JTPA program had an overall positive benefit over the first thirty months. However, it was costly from the perspective of government. Whether the JTPA paid for itself in the long run depends on how long the effects on earnings, taxes, and welfare receipt last and on how quickly we discount future benefits.34 Given that the wage benefits over the last twelve months of the experiment were at least as large as the wage benefits over the previous twelve months, there is little reason to believe that they would dissipate quickly. Therefore, it is very likely that the JTPA actually paid for itself in terms of reduced government expenditures over the long run.

7. Evaluating the Job Corps and Other Youth Programs The Job Corps program was initiated in 1964. It was integrated into CETA and subsequently into JTPA and WIA. It is primarily a residential training program targeted at disadvantaged youths aged 16 to 24. As noted earlier, it is the most expensive of the training programs on a per trainee basis and costs well over $1 billion per year. In June 2001, Mathematica Policy Research released an evaluation of the Job Corps based on a random assignment experiment.35 The earnings of the program group were initially lower (because most members of the program group were in the program and received only a small stipend) but caught up with those of the control group members after about six quarters. After four years, the program group members earned about $18 more per week on average than did members of the control group. Similarly, the members of the program group initially had lower employment rates. They caught up with the control group after about two years. By the end of four years, the employment rate of the program group was about 2.5 percentage points higher than that of the control group.

34. In general, receiving money today is worth more than receiving the same amount in the future, because one could put the money in the bank and have more money in the future. If government spends less today, it has a smaller debt that it must repay in the future. 35. Sheena McConnell, “National Job Corps Study: The Impacts of Job Corps on Participants’ Employment and Related Outcomes” (Mathematica Policy Research, Princeton, NJ, June 2001), and Sheena McConnell and Steven Glazerman, “National Job Corps Study: The Benefits and Costs of Job Corps” (Mathematica Policy Research, Princeton, NJ, June 2001).

Labor Market Policies 129 One thing to note is that the employment and earnings of the control group grew rapidly over the same period. A before-after design in which the earnings and employment of the program group before and after participation in the program were used to measure the effect of the program would greatly exaggerate the effectiveness of the Job Corps program. A second important point to note is that the program and control groups do not correspond perfectly to the groups “Received Training” and “Did Not Receive Train- ing.” Fully 72 percent of the control group did receive some training over the four years following assignment compared with 92 percent of the program group. However, the training received by program participants was greater. The estimated difference between the amount of training per program participant and the amount of training per control group member was almost 1,000 hours, or roughly the equivalent of a year of schooling. After four years, individuals assigned to the Job Corps program earned an average of $18 per week more, or about 9 percent more, than those assigned to the control group. The estimated impact on weekly earnings per program participant (taking into account that some members of the program group did not participate and some controls group members entered Job Corps elsewhere) was $25, or about 12.5 percent, somewhat higher than typical estimates of the effect of a year of education on earnings. Given that Job Corps is more intensive and expensive than a typical year of education, the effect of Job Corps is consistent with other human capital investments.

7.1. Social Effects of the Job Corps Program. Although in this book we are primarily interested in how job training programs affect poverty, society has an interest in a number of other aspects of how such programs work. In particular, we may be interested in the net cost to the public. By raising earnings, Job Corps reduces the extent to which participants use public assistance. This effect is largest in the first year, when the stipend for Job Corps may substitute for other public funds. Four years after completing the program, participants were using $111 less per year of public funds than would be predicted on the basis of the control group. The study also showed that participation in the program was associated with reduced criminality, although this effect was clearly present only during the first year, perhaps due to the residential nature of the program. The Job Corps program costs about $16,500 per participant, including about $2,400 of payments to participants. Nonparticipants (the rest of society) gain from several sources—increased tax payments over the individual’s lifetime, participants’ reduced use of other programs and services, and their reduced criminality. Measuring the benefits of events that have not yet occurred (earnings and tax payments beyond the four years studied) or of goods not found in the market (safer streets) is quite difficult. Subject to this caveat, the Job Corps study found that from the perspective of the public, Job Corps comes close to paying for itself. Public savings amount to more than four-fifths of the cost. The benefits to the participants are large, so the net benefit to society is also quite large unless one puts very little weight on the benefits to participants.

130 Chapter 5 7.2. JOBSTART. Even before the National Job Corps study, research indicated that Job Corps was a successful program. But as we have already seen, Job Corps is also expensive. Besides, the fact that it is a residential program makes it difficult to expand dramatically. Potential participants do not want to be removed from their friends and families for extended periods. Moreover, evaluations of other programs failed to find positive effects on disadvantaged youths.36 This naturally leads us to ask whether an intense nonresidential program would also be effective. Between 1985 and 1988, the JOBSTART demonstration project was conducted to test this possibility.37 Aside from being nonresidential, JOBSTART differed from Job Corps in important ways. Most significant, JOBSTART participants did not receive a stipend, which may have led to higher attrition from JOBSTART than from Job Corps, and JOBSTART could not provide subsidized employment in a setting that provided training. Thus, JOBSTART focused almost entirely on classroom training. In essence, JOBSTART was designed to be a nonresidential Job Corps with fewer support services. Three of the thirteen experimental JOBSTART sites were also Job Corps sites. The other ten were non–Job Corps training sites that developed additional program elements in order to meet the requirements of JOBSTART but still offered fewer services than the Job Corps sites. To be eligible, applicants had to be 17 to 21 years old, not have a high school diploma or GED, read below the eighth-grade level (with exceptions), and be economically disadvantaged. The program consisted of a minimum of two hundred hours of education and five hundred hours of training developed in cooperation with the private sector and designed to prepare participants for jobs in high-demand occupations. The evaluation study followed randomly assigned participants and controls for four years. For the most part, the results were disappointing. JOBSTART did have positive effects. It increased receipt of a GED and appeared to raise earnings (although the effect fell just short of statistical significance). Women and men with arrest records showed particular benefits. However, because JOBSTART, like Job Corps, was intensive, it was expensive for both participants, who lost earnings during the training period, and for the public because of program costs. Over the four-year period, these costs significantly outweighed the gains, and the high costs for program participants probably contributed to a high rate of attrition. Some subgroups actually lost money over the four-year period. The evaluators concluded that, in order to be effective, similar programs would have to focus on those groups with the worst job prospects (to minimize the cost of lost earnings), provide stipends (to reduce attrition), and provide other support services (also to reduce attrition).

36. See Burt S. Barnow, “The Impact of CETA Programs on Earnings: A Review of the Litera- ture,” Journal of Human Resources 22 (Spring 1987): 157–93, for a review of the literature on CETA training programs. See also Robinson G. Hollister Jr., Peter Kemper, and Rebecca A. Maynard, The National Supported Work Demonstration (Madison: University of Wisconsin Press, 1984). 37. This section draws on George Cave, Hans Bos, Fred Doolittle, and Cyril Toussaint, JOB- START: Final Report on a Program for School Dropouts (New York: Manpower Development Research Corporation, 1993), http://www.mdrc.org/publications/337/full.pdf.

Labor Market Policies 131 The JOBSTART evaluation was almost certainly too negative, because it did not reflect any benefits beyond the four years of the study. If participants were in the labor force for thirty-six years beyond the study and if the benefits depreciated at 5 percent per year, the cost-benefit analysis would have shown the benefits to society as roughly equal to the costs using the 5 percent interest rate used by the researchers. More favorable assumptions or factoring in nonmonetary benefits such as the reduction in criminal activity would tip the scales in favor of JOBSTART, while less favorable assumptions would support the researchers’ original conclusions.

7.3. The Center for Employment Training. In the JOBSTART study, one center, the Center for Employment Training (CET), stood out from the others in terms of its positive effect on employment and earnings.38 This same center also performed substantially better in an evaluation of a program for minority single mothers. This raised the question of what about the CET was so effective and could it be copied elsewhere. The CET model has a number of key elements. The program is full-time. Training takes place in a worklike setting developed and provided with the participation of employers. Participants move at their own speed, depending on their mastery of skills. Basic skills are taught as part of the training rather than being provided in a separate education program. A thirty-month follow-up of twelve sites, six operated by CET and six operated by other organizations with CET’s technical assistance, was somewhat disappointing. Four of the six CET-operated sites were judged to follow the CET model closely. The other two CET-operated cites and four of the others followed it somewhat closely, and two did not follow it closely at all. The sites that followed CET closely had positive effects on young women but small or negative effects on young men. The remaining sites had small or negative effects. The study is ongoing. Thirty months is a short period in which to measure an effect of intensive training, particularly among groups with good initial employment prospects, because participants typically have lower incomes during the period they are in training. This effect may have been particularly important for the CET study, because the economy was very strong during the early part of the study.

7.4. The American Conservation and Youth Services Corps. The National and Community Service Act of 1992 created the American Conservation and Youth Ser- vices Corps.39 A critical element of the Youth Services Corps is that the participants work in teams to provide community service. The program lasts from six to twelve months, during which participants spend about 80 percent of their time on com-

38. This subsection is based on Cynthia Miller et al., Working with Disadvantaged Youth: Thirty- Month Findings from the Evaluation of the Center for Employment Training Replication Sites (New York: Manpower Development Research Center, 2003). 39. This subsection is based on JoAnn Jastrzab et al., Impacts of Service: Final Report on the Evaluation of American Conservation and Youth Services Corps (Cambridge, MA: Abt Associates, 1996), http://www.americorps.org/pdf/ccc_youth_0596.pdf.

132 Chapter 5 munity service projects and 20 percent on educational activities; they receive a stipend while enrolled in the program. Most participants are 18 to 25 years old. In many ways, the Youth Corps resembles the old CETA programs in that it provides public works employment along with a training component. We do not know the long-term effects of participation in the Youth Services Corps. A short-run evaluation found that it was socially beneficial largely because of the high value of the projects completed as part of the program. There was also some evidence of positive attitudinal effects, particularly on African American males.

8. Training Programs and Tagging Early, mostly nonexperimental, studies of training programs suggested that these programs were successful only with adult women and had little or no positive effect on men or teenage workers. With the results of the National JTPA study, it became apparent that job training can be successful for adult men. The National Job Corps study shows that very intensive programs can be successful with young workers. However, even if training programs did not justify themselves on the basis of a cost-benefit analysis, they could be justified on the basis of the arguments we made earlier for in-kind transfers. Training programs are a mechanism for identifying the poor. Individuals who have good living circumstances or good under-the-counter jobs will generally not seek to participate in job training programs. George Akerlof has pointed out that even if we raise workers’ future earnings by less than the cost of such a program, this may be a less expensive way of transferring money to the poor than giving cash transfers.40 Mandatory programs differ from voluntary programs in that they may influence program participation. If TANF recipients must participate in WtW programs, they may prefer not to receive TANF funds. Thus, WtW programs can reduce the cost to the public of TANF even if they do not help participants find jobs. So WtW programs are also a tagging device. As in other cases we have seen, whether this component is good or bad depends on who is deterred from receiving TANF funds, how many people are deterred, and the effect on those individuals and families of not receiving TANF funds. We have little direct evidence on the degree to which individuals who might otherwise receive TANF are deterred from participation because of the WtW component. Mark Berger, Dan Black, and Jeffrey Smith provide an interesting study of a related experience.41 They studied the Worker Profiling and Reemployment Services system, which requires workers to receive employment and training services if they are expected to be unemployed for a long time. They find that the program reduces the amount of

40. George A. Akerlof, “The Economics of ‘Tagging’ as Applied to the Optimal Income Tax, Welfare Programs, and Manpower Planning,” American Economic Review 68 (March 1978): 8–19. 41. Mark S. Berger, Dan A. Black, and Jeffrey Smith, “Evaluating Profiling as a Means of Allo- cating Government Services” (Department of Economics Research Report 2000/18, University of Western Ontario, London, Ontario, September 2000).

Labor Market Policies 133 time during which recipients receive unemployment insurance, reduces the cost to the unemployment insurance system, and raises reemployment earnings. However, these benefits cannot be primarily ascribed to what the participants learn in the program. In- stead they seem to be the result of program avoidance. There is considerable exit from the unemployment insurance program when recipients learn they will have to participate in training.

9. Welfare to Work: Work First The 1996 welfare reform eliminated JOBS and created WtW programs that were designed to move welfare recipients into jobs rather than to provide them with training. In part, the support for this change among moderate Democrats and many Repub- licans reflected the results of a small number of evaluation programs that found that JOBS programs focusing on job placement were more effective than those focusing on providing welfare recipients with training.42 Perhaps the most influential of these was an evaluation of the Greater Avenues to Independence (GAINS) program in California. Researchers examined four urban and two rural counties. In each, eligible welfare recipients were randomly assigned to the GAINS program or to a control group. The program evaluation was most favorable in Riverside County, where the program placed great emphasis on moving participants into jobs quickly. Given the importance of changing the focus from training to work, the evidence in support of the superiority of “work first” was modest. The California study involved six sites that differed along a number of dimensions other than the emphasis on work first versus training. Even at the time, analysts warned that the benefits of training were likely to be longer-term than the benefits of rapid employment. Joseph Hotz, Guido Imbens, and Jacob Klerman analyzed the outcomes for the four urban counties nine years after the initial random assignment.43 They tried to control for differences among the counties in the types of individuals served by the GAINS program. They found that the initial benefits are greater for the work-first approach but that the training approach surpasses the work-first approach after several years. The National Evaluation of Welfare to Work Strategies was designed to obtain better information on the relative value of work-first versus education first. At three sites, participants were randomly assigned to AFDC, a work-first program, or an education- first program. Four sites used education first, and one site used a mixed strategy in which participants were first randomly assigned to AFDC or the experimental group and then members of the experimental group were assigned to work first or education

42. See for example, Lyn A. Hogan, “Jobs, Not JOBS: What It Takes to Put Welfare Recipients to Work,” Progressive Policy Institute Briefing, July 17, 1995. 43. V. Joseph Hotz, Guido W. Imbens, and Jacob A. Klerman, “The Long-Term Gains from GAIN: A Re-Analysis of the Impacts of the California GAIN Program,” Journal of Labor Economics 24 (July 2006): 521–66.

134 Chapter 5 Table 5.3 Differences in Earnings Relative to AFDC: National Evaluation of Welfare-to-Work Strategies

First Three Years Years Four to Five Work Education Work Education City First First Mixed First First Mixed

Atlanta 487 295 318 240 Grand Rapids 392 434 –114 170 Riverside 588 319 359 361 Columbus I 373 263 Columbus II 335 173 Detroit 413 486 Oklahoma 45 131 Portland 1,157 752 Average 489 316 1,157 188 261 752

Source: Gordon L. Berlin, “What Works in Welfare Reform: Evidence and Lessons to Guide TANF Reauthorization” (Manpower Development Research Corporation, New York, 2002). first depending on which was viewed as best for them. The results are summarized in table 5.3.44 The results confirm the findings from the earlier studies. Over the first three years, both at the three sites employing both strategies and on average over the ten sites, participants earned more with work first than with education first. By years four and five, the difference had disappeared. We do not yet know whether the long-run benefit of education first will turn out to be greater than that of work first, but the pattern over the first five years is consistent with the results of the earlier studies. However, the most striking finding is from Portland, the one site that used a targeted strategy in which some participants were enrolled in work first and some in education first. Earnings gains in Portland were substantially higher than with either single approach both over the first three years and over the fourth and fifth years. Although we should be cautious about drawing overly strong conclusions from a single observation (perhaps there was something special about Portland), at this point it is reasonable to conclude that for some people work first is best, but others benefit from some initial education or training, and the people charged with making an assignment have some ability to distinguish between the needs of the two types of people. Although it remains somewhat uncertain whether work first or training provides greater benefits to participants, work first does move recipients off of welfare more rapidly. Thus, the benefits to the public may be greater even if the benefits to the recipients turn out to be smaller.

44. Gordon L. Berlin, “What Works in Welfare Reform: Evidence and Lessons to Guide TANF Reauthorization” (Manpower Development Research Corporation, New York, 2002).

Labor Market Policies 135 10. Employer-Based Subsidies Earlier in this book, we discussed the EITC and saw that, at least in theory, it should increase labor force participation, a prediction that is supported by the evidence. Some programs give tax credits to firms instead of workers. Because we have tax credits for both firms and workers, it is natural to ask which has a bigger effect on employment. Some people would argue that because firms, not workers, create jobs, subsidizing firms to hire workers in target groups is a better way to reduce poverty. The simple answer is that according to the most basic economic theory, it does not matter whether we subsidize workers or firms. Consider the basic model of supply and demand for labor. The intersection between the labor supply curve and the labor demand curve gives the equilibrium wage and equilibrium employment. In figure 5.5, the original equilibrium is shown by the intersection of the demand curve labeled “No Subsidy” with the “Supply” curve. The figure is drawn so that the equilibrium wage is about $4.70. Now suppose that we offer firms a $2.00 wage subsidy for each worker they hire. The number of workers firms are willing to hire at $6.00 now equals the number they had been willing to hire at $4.00, and the number they are willing to hire at $4.00 is the number they were willing to hire at $2.00. This is shown by the demand curve labeled “With Subsidy,” which is parallel to the original “Demand” curve. The figure is drawn so that the new equilibrium is $6.00. Note that although the wage subsidy was $2.00 hour, the firm ended up paying $4.00 after the subsidy, so its cost fell by only 70 cents. Workers, on the other hand, earn $1.30 more because of the subsidy. How would the equilibrium differ if instead we gave the subsidy to workers? If the wage were $4.00, workers would receive $6.00 after the subsidy so they would supply exactly the same amount of labor as if they received $6.00 and no subsidy. Similarly, if the wage were $4.00, firms would demand the same amount of labor as they would if they paid $6.00 but received a $2.00 subsidy. And, we know that labor demand equals labor supply when the wage is $6.00 and firms receive a $2.00 subsidy. Therefore demand also equals supply when the wage is $4.00 and workers receive a $2.00 subsidy. When we account for the subsidy, the outcome is identical. Workers receive $6.00 and firms pay $4.00 regardless of whether the subsidy officially goes to workers or to firms. This simple model is, in fact, misleading in at least two ways. First, it assumes that government can target exactly the workers it wishes to subsidize by giving the money either to the workers or to firms. In fact, government wishes to target workers in low- income families. Given the practical constraints, employee and employer subsidies will target these groups differently. Neither approach is perfect. Under the EITC, there is a risk that secondary earners in high-income families will qualify as primary earners in separate families under the tax law. Employer subsidies are focused on low-wage workers. As we saw in our discussion of the minimum wage, the correspondence between low-wage workers and workers in low-income families is quite imperfect. The second limitation of the basic model is that it assumes that firms have full information about workers. In fact, firms have very imperfect information when selecting workers. Therefore, they rely on information from various sources to screen workers. There is evidence that firms view the fact that a worker is eligible for an employment

136 Chapter 5 Wage before Subsidy ($/hr)

Supply

Demand

With Subsidy

No Subsidy

Employment Figure 5.5 The Effect of a Subsidy on Employment and Wages subsidy as a sign that the worker is not a good worker. Gary Burtless reports on an experiment in which workers were given vouchers that entitled employers to a subsidy if they employed the worker.45 Workers with vouchers were significantly less likely to find employment than workers without vouchers. Burtless speculates that this is because having a voucher stigmatized the worker. The current employment subsidy is the Welfare-to-Work Tax Credit, which gives a tax credit to employers who hire long-term TANF recipients for at least four hundred hours or 180 days. Firms receive 35 percent of wages for the first year of employment and 50 percent for the second year up to $10,000 of wages per year. There is no good evidence as yet on its effectiveness. Lawrence Katz has examined the effect of an earlier program, the Targeted Jobs Tax Credit (TJTC), which existed from 1979 to 1994.46 The TJTC gave a tax credit to employers hiring workers from particular target groups (e.g., economically disadvantaged youths, individuals on public assistance or supplemental social security). Initially the credit paid firms 50 percent of first-year and 25 percent of second-year wages up to $6,000. In 1986, the subsidy for the first year was reduced to 40 percent and the second- year subsidy was eliminated. The cap was not adjusted for inflation over this period.

45. Gary Burtless, “Are Targeted Wage Subsidies Harmful? Evidence from a Wage Voucher Ex- periment,” Industrial and Labor Relations Review 39 (October 1985): 105–14. 46. Lawrence F. Katz, “Wage Subsidies for the Disadvantaged,” in Richard B. Freeman and Peter Gottschalk, eds., Generating Jobs: How to Increase Demand for Less-Skilled Workers (New York: Russell Sage, 1998), 21–53.

Labor Market Policies 137 Eligible workers received vouchers that demonstrated that the employer was entitled to the credit. The process for obtaining the credit seems to have been burdensome, so the TJTC was used primarily by large retail and service sector firms. Katz estimates that in 1991 the TJTC subsidized about 15 percent of the cost of employment for a typical TJTC worker who remained in the job for six months. Of course, the subsidy was higher for lower-wage workers and lower for higher-wage workers. Despite this subsidy, he estimates that firms collected the credit for only about 9 percent of eligible employed youths. Katz attempts to look at the effect of the TJTC by using a change in the law after 1988 that removed 23- and 24-year-olds from the group eligible for the tax credit. His strategy for examining this issue is quite sophisticated. He examines employment rates for 1987 and 1988 and compares them with employment rates for 1989 and 1990. If the TJTC increased the employment of disadvantaged 23- and 24-year-olds, making them ineligible for the TJTC should have caused a drop in their employment relative to disadvantaged 18- to 22-year-olds who remained eligible for the TJTC and 25- to 29-year-olds who were never eligible. It should also have reduced their employment relative to nondisadvantaged 23- and 24-year-olds. As usual, the problem is that perhaps something in the economy affected the employment of 23- and 24-year-olds relative to younger workers, so simply comparing the employment rates of 23- and 24-year-olds with those of younger (or older) workers (even limiting the analysis to disadvantaged workers) would not be convincing. Similarly, something might have changed the relative labor markets of advantaged and disadvantaged young workers over this period, so comparing outcomes for disadvantaged and advantaged 23- and 24-year-olds would also not be convincing. Instead, Katz looks at the relative employment of disadvantaged to advantaged 23- and 24-years-olds and compares that with the relative employment of disadvantaged to advantaged 18- to 22-year-olds and the relative employment of disadvantaged to advantaged 25- to 29-year-olds. As shown in table 5.4, the employment rate of disadvantaged 18- to 22-year-olds was about 28 percentage points lower than the employment rate of nondisadvantaged 18- to 22-year-olds in 1987–1988 but only about 26 percentage points lower in 1989– 1990. In 1987–1988, the employment rate of disadvantaged 25- to 29-year-olds was about 42 percentage points lower than that of nondisadvantaged individuals in this age group. In 1989–1990, this difference was only 40 percentage points. So we see a slight increase in the relative employment of disadvantaged young adults in both the group slightly younger and the group slightly older than 23- and 24-year-olds. In contrast, the employment rate for disadvantaged 23- and 24-year-olds was 37 percentage points below that of their nondisadvantaged counterparts in 1987–1988 but 40 percentage points lower in 1989–1990. Katz finds that the difference between the affected age group, on the one hand, and the two unaffected age groups, on the other, is statistically significant. As usual, there are reasons to be wary of the conclusion. Perhaps most significant, the Katz study cannot distinguish whether the TJTC increased the employment of disadvantaged young adults overall or simply shifted it across age groups. Removing TJTC eligibility for 23- and 24-year-olds might have shifted employment away from

138 Chapter 5 2 5 83 43 –40 25- to 29-Year-Olds 84 42 –42 1987–1988 1989–1990 Generating Jobs: How to Increase Demand for Demand to Increase How Jobs: Generating –3 82 42 –40 81 44 23- and 24-Year-Olds –37 2 5 65 39 d B. Freeman and Peter Gottschalk, eds., and Peter d B. Freeman –26 18- to 22-Year-Olds 65 37 –28 1987–1988 1989–1990 1987–1988 1989–1990 (New York: Russell Sage, 1998), 21–53. Sage, Russell York: (New Employment Rates for Disadvantaged and Nondisadvantaged Young Adults, 1987–1988 and 1989–1990 Adults, Young and Nondisadvantaged Rates for Disadvantaged Employment Lawrence F. Katz, “Wage Subsidies for the Disadvantaged,” in Richar for the Disadvantaged,” Subsidies Katz, “Wage F. Lawrence Change from Earlier to Later Period Table 5.4 Table between 23- and 24-Year-Olds Difference Source: Less-Skilled Workers Less-Skilled Difference Nondisadvantaged Disadvantaged

Labor Market Policies 139 previous eligible workers to workers who remained eligible (18- to 22-year-olds) or to older workers who were never eligible (25- to 29-year-olds).

11. Concluding Remarks The research on labor market policies shows that there are no easy answers. Minimum wage and living wage laws raise the wages of low-earning workers but have some dis- employment effects and are not sufficiently targeted to have a major impact on the poverty rate. Inexpensive training programs have effects compatible with their low costs. Intensive training programs generate returns that are comparable to those from public investment in education. Unfortunately, the number of people involved in intensive programs is modest, so the existing programs are not likely to have much impact on the overall poverty rate. We do not know whether it is feasible to develop a set of effective intensive training programs on a broad scale.

12. Further Reading Adams, Scott, and David Neumark. “The Economic Effects of Living Wage Laws: A Provisional Review.” Urban Affairs Review 40 (November 2004): 210–45. Akerlof, George A. “The Economics of ‘Tagging’ as Applied to the Optimal Income Tax, Welfare Programs, and Manpower Planning.” American Economic Review 68 (March 1978): 8–19. Bloom, Howard S., et al. “The Benefits and Costs of JTPA Title II-A Programs: Key Findings from the National Job Training Partnership Act Study.” Journal of Human Resources 32 (Summer 1997): 549–76. Brown, Charles. “Economics of the Minimum Wage.” In Orley Ashenfelter and David Card, eds., Handbook of Labor Economics, Vol. 3B. Amsterdam: Elsevier, 1999, 2101–64. Burtless, Gary. “Are Targeted Wage Subsidies Harmful? Evidence from a Wage Voucher Experiment.” Industrial and Labor Relations Review 39 (October 1985): 105–14. Card, David, and Alan B. Krueger. Myth and Measurement. Princeton, NJ: Princeton University Press, 1995. Friedlander, Daniel, David H. Greenberg, and Philip K. Robins. “Evaluating Govern- ment Training Programs for the Economically Disadvantaged.” Journal of Economic Literature 35 (December 1997): 1809–55. Lalonde, Robert. “The Promise of Public Sector–Sponsored Training Programs.” Jour- nal of Economic Perspectives 9 (Spring 1995): 149–68. Neumark, David, and William Wascher. “Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania: Comment.” American Economic Review 90 (December 2000): 1362–96.

13. Questions for Discussion 1. How does raising the minimum wage affect the poverty rate? Give a theoretical explanation, and also describe any empirical findings.

140 Chapter 5 2. How do living wage laws differ from minimum wage laws? How do they affect the poverty rate? 3. Government training programs target close to three million people per year. Given all that spending, if government knew anything about training, the only poor adults left would be those unable or unwilling to work. Discuss. 4. Describe the principal findings of the JTPA evaluation study. 5. When evaluating job training programs, what are the problems that arise if we compare people who entered training programs with those who did not? What are the problems with comparing people before they entered the program with their performance after they completed it? 6. Why might the social and private benefits of job training programs differ? 7. Describe the principal finding of the Job Corps evaluation study. 8. The only good job training programs are expensive. Discuss. 9. Suppose that a job training program costs more than the (appropriately discounted) amount by which it increases worker earnings over their lifetimes. Why might it nevertheless be better to give training than to give cash transfers? 10. What is meant by “work first”? What are the arguments for and against work first rather than skill-based training programs? 11. What has been the U.S. experience with employment tax credits for low- skilled workers?

14. Appendix: Adjusting for Program Nonparticipation Suppose that we randomly assign half of some group to participate in a program and half to a control group that does not participate in the program. If everyone participated as assigned, we could ascribe the differences in outcomes to program participation and measure this effect by the difference in average outcomes between those assigned to the program and those assigned to the control group. Suppose, however, that only two-thirds of the program group actually participates. Only two-thirds of those assigned to the program group will derive the benefit of participation. One-third will have the same outcome they would have had if they had been assigned to the control group. Mathematically, the outcome for the group assigned to participate (the experimental group) is 2 outcomeexperimental = outcomecontrol + /3 benefit. Rearranging terms, we have 2 outcomeexperimental – outcomecontrol = /3 benefit or

1.5 (outcomeexperimental – outcomecontrol) = benefit.

Labor Market Policies 141 In order to find the program effect, we would multiply the difference between the program and control groups by 1.5. The problem becomes more complicated if some of the control group members participate in the program even though they were assigned not to participate. Suppose that one-third of the control group participates. We now write 2 outcomeexperimental = outcomenonparticipant + /3 benefit to remind us that being in the control group and being a nonparticipant are not the same. And we write 1 outcomecontrol = outcomenonparticipant + /3 benefit because one-third of the control group is receiving the benefit of participation. Sub- tracting the second equation from the first equation gives us 1 outcomeexperimental – outcomecontrol = /3 benefit or

benefit = 3 * (outcomeexperimental – outcomecontrol). In general, we will multiply by one divided by the difference in the fraction of participants. It may occur to you that the program will not affect everyone in the same way. Per- haps some people will receive no benefit while others will receive a significant benefit. If only people who expect a significant benefit participate when assigned to the program group, this approach will overestimate the effect on a random person assigned to the program group. However, it will still give the average effect on those who participated, which, for most purposes, is what matters. We do not really care about the effect on someone who would not participate anyway.

142 Chapter 5 chapter 6

Family Composition

I  4       female-headed households might partially explain the failure of poverty to decrease despite the growth in gross domestic product per capita over the past thirty years. There has been rapid growth in the proportion of households that are female-headed, and the poverty rate is much higher among these households than among married-couple households. There have also been increases in the proportion of single individuals without children and the proportion of single men with children, both of which have poverty rates above the average, and a decline in married couples with children, who have a poverty rate below the average. The proportion of married couples without children, a low-poverty group, has increased somewhat, but overall the trend in household composition tends to increase the poverty rate. Maria Cancian and Deborah Reed calculate that if the poverty rate for each family type had remained constant at its 1969 rate, by 1998 the shift in household composition would have increased the poverty rate by 3.6 percentage points.1 Using the 1998 poverty rates, the effect would be 2.6 percentage points.2 In chapter 4 we estimated that an increase of 1 percentage point in female-headed households was associated with an increase in the poverty rate of roughly .25 percentage point. This suggests that it is important to understand why the number of female-headed households, particularly with children present, has grown. We also need to assess the importance of the increase of female-headed households for understanding poverty. To what extent do female-headed households cause poverty,

1. Maria Cancian and Deborah Reed, “Changes in Family Structure: Implications for Poverty and Related Policy,” in Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty (Cambridge, MA: Harvard University Press, 2001), 69–96. 2. You may recall that in chapter 4 we calculated that the increase of female-headed households would have increased the poverty rate by 1.8 percentage points between 1959 and 2003. The difference appears to be due to Cancian and Reed’s use of more changes in household composition and to the use of later poverty rates.

143 and to what extent do they reflect poverty? This is particularly important in the case of teen mothers, the great majority of whom are unmarried. Policy makers are particularly concerned about this group of single mothers, because they have very high rates of poverty, even in adulthood. We will see that, in fact, poverty may be a more important cause of teen motherhood than teen motherhood is of poverty. We will also examine the evidence that family composition affects children’s outcomes as adults. If growing up with a single parent hurts children’s outcomes, we will want to consider policies that reduce divorce and encourage marriage. It is clear that children who grow up in poverty are disadvantaged. As a society that believes in equality of opportunity, we have developed policies intended to reduce these disadvantages. Toward the end of the chapter we will examine the effectiveness of these policies.

1. Births to Single Mothers Figure 6.1 shows that there has been a steady and rapid rise in the proportion of births to unmarried women since the 1950s, although the rate of increase has diminished in recent years. The changes are even more dramatic among younger women. In 1960, only 15 percent of births to women aged 15–19 were to unmarried mothers. By 2002, this proportion was 80 percent. Even among 20- to 24-year-old women, over half of births are now to unmarried mothers, compared with fewer than 5 percent in 1960.3 There is a small caveat that serves to remind us of the need to examine carefully how data are collected. Not all states ask a direct question about marital status on birth certificates. Over time, the National Center for Health Statistics has used different methods for imputing a mother’s marital status in these states. Since 1980, marital status data from these states have been based on inferences about the mother’s marital status derived from other information on the birth certificate, such as whether information about the father is missing. One of the factors that was used in six states4 as of 1992 was whether the father and mother had the same last name. Because more married women have been choosing to retain their maiden names, this would have increased the proportion of births to “unmarried mothers.” Today only Michigan and New York do not ask a direct question about marital status on the birth certificate. This change in data collection is too minor to account for the shift toward nonmarital births, but it may affect minor matters, such as the timing of the shift.5

3. All data are from Joyce A. Martin et al., “Births: Final Data for 2002,” National Vital Statistics Reports, vol. 52, no. 10 (Hyattsville, MD: National Center for Health Statistics, 2003), and Stephanie J. Ventura and Christine A. Bachrach, “Nonmarital Childbearing in the United States, 1940–99,” National Vital Statistics Reports, vol. 48, no. 16 (Hyattsville, MD: National Center for Health Statis- tics, 2000). 4. California, Connecticut, Michigan, Nevada, New York, and Texas. 5. Stephanie J. Ventura, “Births to Unmarried Mothers: United States, 1980–92,” National Vi- tal Statistics Reports, vol. 53, no. 21 (Hyattsville, MD: National Center for Health Statistics, 1995), and Martin et al.

144 Chapter 6 Percentage of Births to Unmarried Women 35

1940 1950 1960 1970 1980 1990 2000 Figure 6.1 Percentage of Births to Unmarried Women, 1940–2001

The percentage of births to unmarried mothers is

——————————————————————.births to unmarried mothers births to unmarried mothers + births to married mothers The number of births to unmarried mothers is birth rate per unmarried woman * number of unmarried women, and the number of births to married mothers is birth rate per married woman * number of married women, so the percentage of births to unmarried mothers can be expressed by

———————————————————–————————————.birth rate per unmarried woman * number of unmarried women births to unmarried mothers + birth rate per married woman * number of married women

The percentage of births to unmarried women could rise because 1. The rate at which unmarried women give birth went up. 2. The proportion of unmarried women in the society went up. 3. The rate at which married women give birth went down. Figure 6.2 reveals that both the first and third explanations play a role. The birth rate per one thousand unmarried women was 21.6 in 1960 and had doubled to 43.7 in 2002. If no other change had occurred, the proportion of births to unmarried women would have increased from 5.3 percent to 10.2 percent. We can also see that the birth rate per one thousand married women declined from 156.6 to 86.3. If only the birth rate to married women had declined and neither the birth rate to unmarried women nor the fraction of unmarried women in the

Family Composition 145 Births per 1,000, Ages 15–44

150

125

Married 100

50 Unmarried 25

0 1940 1950 1960 1970 1980 1990 2000 Figure 6.2 Birth Rates for Married and Unmarried Women, 1940–2001

population had changed, the percentage of births to unmarried women would have increased to 9.2 percent. The increase in the birth rate among unmarried women and the decrease among married women together can account for an increase to 17.0 percent, so less than half of the increase can be explained by changes in the birth rates to unmarried and married women. Therefore, most of the increase in the percentage of births to unmarried women is due to an increase in the proportion of women of child-bearing age who are not married. In a sense, even these figures underestimate the importance of declining marriage as an explanation. To some extent, whether to marry in the event of a pregnancy is a choice. Although neither partner may be able to insist upon marriage, they can agree to get married. Some of the increase in the birth rate to unmarried women and the decline in the birth rate to married women result from a decline in the custom of getting married in the event of pregnancy, a topic we will discuss in greater detail in section 3. Hence, in order to understand why the proportion of births to unmarried women rose, we must focus on the decline of marriage.

2. Declining Marriage There are many reasons that people get married, not least of which is the desire for committed companionship. Yet throughout recorded history, marriage has also been an economic decision often governed by strict legal contracts similar to those governing other forms of economic exchange. Both economists and sociologists have focused on economic explanations for the decline in marriage. Their explanations are complementary, although they are expressed in different ways. These explanations can be summarized as follows:

146 Chapter 6 1. Improving economic opportunities for women have caused women to value marriage less. 2. Worsening economic conditions for low-skilled men have made them less desirable as marriage partners. 3. Diminishing differences between the economic opportunities of men and women have reduced the benefits associated with being married. We will consider each of these explanations in turn.

2.1. Increasing Opportunities for Women. Women’s earnings have increased dramatically in the past several decades. Between 1967 and 2001, the real median earnings of women doubled and, for those working full-time, year-round, increased by over 50 percent.6 As a result, women are no longer dependent on men for survival. In the past, the argument continues, women would continue in an otherwise undesirable relationship in order to obtain food and shelter for their children and themselves. How- ever, the fact that women now have better opportunities in the labor market means that they can escape abusive husbands or loveless marriages. This argument leads naturally to a “chicken or egg” question. Did the increased opportunities for women reduce their reliance on marriage? Or did women recognize that the probability of being divorced or not getting married had increased and therefore invest more in labor market skills? Although it is impossible to resolve this question fully, it is worth noting that much of the increase in the earnings of women, particularly among low-wage women, is due to increases in hours rather than in wages. Among women working full-time, year- round, between 1975 and 2000, low-skilled women experienced only modest wage increases. Among those with some high school education but no high school diploma, mean earnings rose by about 6 percent,7 while for those with exactly a high school diploma, mean earnings rose by about 18 percent, and for those with a college degree or more, mean earnings rose by almost 60 percent.8 Francine Blau shows that between 1970 and 1995, marriage rates declined most dramatically among women with the least education (from 78 percent to 56 percent) and somewhat less among those with a high school diploma (82 percent to 67 percent) or some college (79 percent to 65 percent). They declined least among women with college degrees (75 percent to 68 percent), although this is the group for whom opportunities increased the most.9

6. http://www.census.gov/hhes/income/histinc/p41.html. 7. The Census Bureau changed its education measures over this period. For 1975, this refers to those reporting nine to eleven years of education, while for 2000 it includes those reporting twelfth- grade education but no diploma. This change in definition exaggerates the increase in earnings over this period for both dropouts and those receiving their diplomas. 8. http://www.census.gov/hhes/income/histinc/incperdet.html, tables 32–35. 9. Francine Blau, “Trends in the Well Being of American Women, 1970–1995,” Journal of Eco- nomic Literature (March 1998): 141–47.

Family Composition 147 Although it is possible that less skilled women were more affected by their modest increases in opportunities than were highly skilled women by their large increase, it is more plausible that some other factor is the main driving force behind the declining marriage rate.

2.2. Declining Opportunities for Unskilled Men. We also saw earlier that the earnings distribution has become more unequal. Median earnings for men were essentially unchanged between 1967 and the early 1990s but rose thereafter. With increasing earnings inequality, the earnings of male year-round, full-time workers at the tenth percentile in the earnings distribution were lower in 2003 than in 1969, and this ig- nores the increased fraction of men not working full-time or working part-year. Over- all, economic opportunities for low-skilled men have actually declined. Consequently, if women marry men in order to obtain economic security, there are fewer marriageable men available. William Julius Wilson argues that, in particular, black women are unwilling to marry black men because so few are able to obtain steady employment at good wages.10 Kathryn Edin’s interviews with poor unmarried women in Philadelphia support this perspective.11 The women interviewed felt that their lives were too precarious to permit them to live with men who were unable to contribute to the family. If men could not find employment and put food on the table, they could not remain in the household. The women were unwilling to marry a man who might become a drain on household resources. There are several facts that seem to contradict this perspective as a general argument for the decline in marriage. First, the decline in labor market opportunities for low-skilled men dates from sometime in the 1970s, but the decline in the marriage rate precedes that period. David Ellwood and John Crane have examined marriage rates and employment rates for black men from 1960 to 1988.12 In 1960, the number of employed 25- to 34-year-old black men per 100 black women in the same age group was 71. In 1988, it was 67, but the marriage rate among these men fell from 63 percent to 41 percent. The marriage rate of those employed at the time fell from 70.3 percent to 48.1 percent, while the marriage rate of those not employed at the time fell from 32 percent to 23 percent. Second, employment rates among blacks really declined only for high school dropouts. For high school graduates, the employment rate went from 86 percent to 83 percent, while for those with at least some education past high school, it went from 85 percent to 86 percent. If Wilson is right, we would expect highly educated black men to be in particularly high demand as mates. Yet the marriage rate for male high school graduates declined from 64 percent to 42 percent and for those with more education from 61 percent to 44 percent. It is true that the marriage rate declined even

10. William J. Wilson, The Truly Disadvantaged (Chicago: University of Chicago Press, 1987). 11. Kathryn Edin, “A Few Good Men,” American Prospect 11 (January 3, 2000): 1–8. 12. David Ellwood and John Crane, “Family Change among Black Americans: What Do We Know?” Journal of Economic Perspectives (Fall 1990): 64–84.

148 Chapter 6 more dramatically for high school dropouts (from 63 percent to 30 percent), which is consistent with the idea that declining opportunities for unskilled black men are part of the reason for the declining marriage rate. There is one story that suggests that the decline in the number of eligible men could decrease the marriage rate of eligible men. The essential question is whether men must enter into a contract (marriage) in order to obtain the services they hope to receive in marriage. Clearly some aspects of home production are readily available on the market. It is possible to purchase cooking and cleaning. A major reason for marriage is the desire to have sex and to have children, but, of course, it is possible to have both without marriage. If women are in short supply relative to eligible men, they have more power to determine whether marriage is a condition of providing men with services and can hold out for high-quality men, and often this will result in their insisting on marriage. When men are in short supply, if they do not wish to marry, they are able to resist paying the high price of marriage for these services. Joshua Angrist studied the relation between the sex ratio in immigrant groups in the early part of the twentieth century in the United States and the proportion of women in that group who were ever married.13 His results show that when the ratio of men to women was high in an immigrant group, the marriage rate was also high. A shortage of women (surplus of men) increased the marriage rate. It is possible that the shortage of marriageable men, particularly in the African American community, em- powers eligible men not to marry. However, in a similar exercise Robert Wood did not find that differences in marriage rates of black men across metropolitan areas were strongly related to the availability of “marriageable men.”14 He found that metropolitan areas with larger declines in the pool of eligible black men experienced only a slightly larger decline in the black marriage rate. He concluded that the decline in the number of high-earning young black men in the 1970s accounts for 3 or 4 percent of the decline in black marriage rates over that period.

2.3. Diminishing Benefits from Specialization. Both the declining opportunities for men and the increasing opportunities for women explanations rest entirely on the desire of women to get married or not get married. If women’s job opportunities are improving and men’s deteriorating, we should expect men to have a greater desire to get married. The men should be willing to do more in order to get married. While the balance of power would shift within marriage, it is not obvious that the total benefit from marriage would have declined. However, the two arguments can be combined to explain why the benefit from marriage might have declined. According to this argument, the explanation is not so

13. Joshua Angrist, “How Do Sex Ratios Affect Marriage and Labor Markets? Evidence from America’s Second Generation,” Quarterly Journal of Economics 117 (August 2002): 997–1038. 14. Robert G. Wood, “Marriage Rates and Marriageable Men: A Test of the Wilson Hypothesis,” Journal of Human Resources 30 (Fall 1995): 163–93.

Family Composition 149 much that women no longer need men or that they cannot find acceptable men; it is that the gains from marriage decline when women and men have more similar economic opportunities. The benefits from trade depend on specialization, and marriage is a form of trade. Traditionally, men worked in the market and provided goods that could be purchased in the market. Women worked in the home and provided goods that could be produced in the home. When men and women had very different earnings opportunities, it made sense for women to specialize in home production (taking care of the house and children). Now that market opportunities are similar for men and women, there is less benefit from such specialization. The pattern of changes in marriage rates poses difficulties for this explanation as well. If marriage were determined solely by wage rate comparisons, high-wage men should marry low-wage women, and vice versa. Yet this has never been the pattern. High- wage men (or those with high levels of education) tend to marry highly educated women. Presumably this is because some aspects of couples’ backgrounds are complementary. People like people with similar backgrounds, experiences, cultures, and so on. They want to be able to talk about the same books, follow the same sports, dance in similar styles, and listen to music that both enjoy. If these factors push men and women of similar ranking in earnings ability to marry, the economic advantages of marriage have declined the most among highly educated women. Such women are now very likely to be in the labor market, and their increased earnings make it more expensive for them to withdraw from the labor market to specialize in home production. Yet, as we have seen, this is not the group in which the decline in marriage rates is most prevalent.

2.4. Economic Explanations. Although none of the simple economic models fits the data well, it may be that a more complex model can do a better job. One of the benefits of marriage is that a couple pools its expenses. If “two can live as cheaply as one,” a couple will benefit from marriage even if the two partners have the same wages and do not specialize in market or home production. If this benefit increases as income increases, the fact that wages have risen for high-skilled men and women would increase marriage among this group. This offsetting influence could explain why, relative to women with low levels of education, marriage rates have declined less rapidly among highly educated women even though their wages converged more rapidly with those of men. I leave it to the reader to decide whether the advantages of sharing expenses are greater for two people who earn $15,000 apiece or for two who earn $75,000 apiece. Certainly we see more nonromantic sharing of living quarters among lower-income individuals.

3. Changing Social Attitudes

3.1. Increasing Cohabitation. Although we have tried to explain the declining marriage rate, we must consider the possibility that in a real sense there has been no decline in marriage at all. Instead our measurement may have failed to keep up with how society defines marriage. In some countries, there is no civil marriage. A couple that engaged

150 Chapter 6 in an elaborate civil ceremony and executed a “marital agreement’’ with the help of a lawyer would not be considered married. Yet Americans would almost certainly believe that the couple was married. The fraction of births to unmarried women in the United States is not particularly high by international standards. In 1998, the United States (33 percent) ranked behind Iceland (64 percent), Sweden (54 percent), Norway (49 percent), Denmark (45 percent), France (40 percent), the United Kingdom (38 percent), and Finland (37 percent) in the fraction of births to unmarried mothers and its ranking was not much higher than the rankings of Canada and Ireland (both 28 percent). Of the set of comparisons countries, only the Netherlands (21 percent), Portugal (20 percent), Germany (14 percent), Italy (9 percent), Greece (4 percent), and Japan (1 percent) ranked substantially below the United States. Moreover, the rank of the United States among these countries fell between 1980 and 1998. In 1980, only Iceland, Sweden, and Denmark had higher proportions of births to unmarried mothers.15 For the most part, the high fraction of births to unmarried mothers is not viewed as a major social problem in the Scandinavian countries. In part, this reflects the extensive social support system for children, in particular child care. In part, it reflects the fact that many of the unmarried mothers are in long-term relations. The couple simply has not chosen to get married. To what extent is the decline in the marriage rate simply a relabeling of couples— that is, to what extent has cohabitation replaced marriage? Our answer to this question is imperfect, because the Census Bureau has only recently begun asking unmarried people whether they are living with a partner. The 1960 and 1970 censuses and the Current Population Surveys since 1977 have identified who is living together. The Census Bureau has used this information to identify POSSLQs (pronounced pah’-səl- kews)—“persons of the opposite sex sharing living quarters.” A POSSLQ couple is defined by the Census Bureau as two persons of opposite sex who are at least 15 years old and are living together without any other adult in the household. The term POSSLQ has entered common parlance and is often used to refer both to a POSSLQ couple and to one member of a POSSLQ couple, as in “Roses are red. Violets are blue. Would you be my POSSLQ?” The POSSLQ definition is clearly imperfect. A mother and her 16-year-old son living together would be identified as a POSSLQ household, while a couple living with a 16-year-old would not be. Despite these imperfections, the Census Bureau estimates give us some indication of the trends in cohabitation.16 As shown in figure 6.3, there has been a ten-fold increase in cohabitation from 439,000 in 1960 to over five million in 2003. Although the exact numbers are in doubt,

15. S. J. Ventura and C. A. Bachrach, “Nonmarital Childbearing in the United States, 1940–99,” National Vital Statistics Reports, vol. 48, no. 16 (Hyattsville, MD: National Center for Health Statis- tics, 2000). 16. Lynne M. Casper, Philip N. Cohen, and Tavia Simmons, “How Does POSSLQ Measure Up? Historical Estimates of Cohabitation” (Population Division Working Paper 36, U.S. Census Bureau, Washington, DC, 2002).

Family Composition 151 Number (Thousands) 5,000

4,000

3,000 All

2,000

1,000

With Children under 15 0 1960 1970 1980 1990 2000 Figure 6.3 POSSLQ Estimates from the Census and the Current Population Survey, 1960–2001

the trend is not. The number of POSSLQ households with children under 15 has grown from 242,000 to well over three million.17 Although these changes are dramatic, there are good reasons for believing that their impact on our estimates of marriage rates are relatively small. First, as shown in figure 6.4, as a percentage of the population (aged 15 and over) the proportion of women who are cohabiting as measured by the number of POSSLQs is small. The proportion of women who are married or in POSSLQ relations has declined since 1960, although the decline has been somewhat smaller than the decline in those married. The proportion of women over age 15 who were married fell from about 64 percent in 1960 to about 55 percent in 2000, while the proportion measured as cohabiting grew from less than 1 percent to over 4 percent. However, even these figures overestimate the importance of the growth of cohabitation. Without doubt, a substantial fraction of these POSSLQs are women living with their sons, not with a husband-equivalent. This is borne out by more recent and direct Current Population Survey estimates of unmarried couples living together. Since 1996, the Current Population Survey has asked the question eliciting this information directly. In 2000, about 3.8 million women reported that they were part of an unmarried couple. This compares with 4.7 million who are estimated as cohabiting under the POSSLQ definition. In short, undoubtedly some of the decline in marriage recorded is a matter of accounting, but this is unlikely to explain more than one-third of the decline in the marriage rate, and probably explains somewhat less.

17. The data on POSSLQ and on marriage rates are from http://www.census.gov/population/ socdemo/hh-fam/tabUC-1.xls and http://www.census.gov/population/socdemo/hh-fam/tabMS-1.xls.

152 Chapter 6 Percentage 80

70 Women Married or Cohabiting 60

50 Women Unmarried 40

10 Women Cohabiting 0

1960 1970 1980 1990 2000 Figure 6.4 Marriage, Cohabitation, and Nonmarriage Rates for Women over Age 15, 1960–2003

3.2. The Decline of Shotgun Weddings. The increasingly disused expression shotgun wedding came from the threat that if a boy got a girl pregnant, her father would threaten him with a shotgun. Either the boy would marry the girl, or the father would kill him. In fact, the shotgun was probably rarely necessary. The social ostracism that both girl and boy would suffer from their failure to marry was often sufficient to ensure marriage. Society expected the couple to marry. Figure 6.5 shows the proportion of premaritally conceived first children whose parents married before the births.18 It is clear that there was a dramatic change in the prevalence of shotgun weddings in the mid-1970s. Among 15- to 29-year-olds, from the 1930s through the 1960s, over 50 percent of premarital conceptions resulting in birth were followed by a wedding prior to the birth. In the second half of the 1970s, this figure was 32 percent, a dramatic drop from a decade earlier. This rate declined further, so by the first half of the 1990s it stood at 23 percent. Similarly, we see a dramatic decline in the rate at which women marry after giving birth out of wedlock. In the late 1960s, close to 60 percent of women aged 15–29 who gave birth while unmarried were married within five years. In the first half of the 1990s, this figure stood at less than 40 percent (figure 6.6). The decline of shotgun weddings accounts for about half of the increase in the proportion of out-of-wedlock births. In 1960–1964, 26 percent of children born were premaritally conceived. The parents of 60 percent of these married before the birth, so premarital births accounted for 10 percent of all births. In 1990–1994, 53 percent of children born were conceived premaritally. If the parents of 60 percent of these had married before delivery, 21 percent of all births would have been to unmarried mothers.

18. The data are from Amara Bachu, Trends in Premarital Childbearing: 1930 to 1994, Current Population Report P23-197 (Washington, DC: U.S. Census Bureau, 1999).

Family Composition 153 Percentage Marrying after Conception 70

20 1932 1942 1952 1962 1972 1982 1992 Figure 6.5 Marriage between Conception and Birth for Women Not Married at Conception and Aged 15–29 at First Birth, 1932–1992

Percentage Marrying 60

50 Within 5 Years of Birth

30 Within 2 Years of Birth

20 Within 1 Year of Birth

1967 1972 1977 1982 1987 Midpoint of Time Period Figure 6.6 Marriage after Out-of-Wedlock Birth, 1967–1987

In contrast, if the rate of out-of-wedlock conception had remained constant and the postconception marriage rate had declined to 23 percent, 20 percent of all births would have been to unmarried mothers. Thus, both the decline in shotgun marriages and the increase in the proportion of children that were conceived premaritally roughly doubled the proportion of births to unmarried mothers. Together, they quadrupled this proportion.

154 Chapter 6 Why did shotgun weddings decline? George Akerlof, Janet L. Yellen, and Michael L. Katz19 argue that the development of the contraceptive pill and the increased availability of abortion can account for much of the change. Legislative actions and judicial decisions, culminating with Roe v. Wade in 1973, made abortion much more available in the 1970s and allowed them to be performed later in the pregnancy than they had been previously. There were 88,000 abortions in the United States during 1965–1969, 561,000 during 1970–1974, 985,000 during 1975–1979 and 1,271,000 in 1980– 1984.20 In 1980, 59 percent of pregnancies to unmarried women ended in abortion (although by 1995 this figure was down to 41 percent).21 At the same time, contraception became more readily available to unmarried persons. In 1972, in Eisenstadt v. Baird, the Supreme Court struck down a Massachusetts law preventing the distribution of contraceptives to unmarried persons. Use of the pill by unmarried women on first intercourse nearly tripled between 1965–1969 and 1970–1974.22 Akerlof and Janet Yellen write: “Before 1970, the stigma of unwed motherhood was so great that few women were willing to bear children outside of marriage. The only circumstance that would cause women to engage in sexual activity was a promise of marriage in the event of pregnancy. Men were willing to make (and keep) that promise for they knew that in leaving one woman they would be unlikely to find another who would not make the same demand. Even women who would be willing to bear children out-of-wedlock could demand a promise of marriage in the event of pregnancy.” The increased availability of contraception and abortion made shotgun weddings a thing of the past. Women who were willing to have an abortion or who reliably used contraception no longer found it necessary to condition sexual relations on a promise of marriage in the event of pregnancy. But women who wanted children, who did not want an abortion for moral or religious reasons, or who were unreliable in their use of contraception found themselves pressured to participate in premarital sexual relations without being able to exact a promise of marriage in case of pregnancy. These women feared, correctly, that if they refused sexual relations, they would risk losing their partners. Sex- ual activity without commitment was increasingly expected in premarital relationships. The reduced risk of pregnancy probably also contributed to increased sexual activity. Whether improved access to contraception lowers or raises the birth rate depends on the degree to which it increases sexual activity, the effectiveness of contraception when used, and the extent to which contraception is actually used. Among women who were 15 to 44 in 1995 and thus almost all of whom would have had their first sexual experience in the post-pill period, fewer than 60 percent used any type of contraception on

19. George A. Akerlof, Janet L. Yellen, and Michael L. Katz, “An Analysis of Out-of-Wedlock Childbearing in the United States,” Quarterly Journal of Economics 111 (May 1996): 277–317. 20. George Akerlof and Janet Yellen, “An Analysis of Out-of-Wedlock Births in the United States,” Brookings Policy Brief 5, August 1996. 21. Ventura and Bachrach, 11. 22. Akerlof and Yellen.

Family Composition 155 the occasion of their first premarital sexual intercourse. This proportion has increased dramatically in recent years. For women whose first sexual intercourse was before 1985, the proportion was only 50 percent. For 1990–1995, the figure stood at 76 percent, but the entire increase was in condom use, suggesting a response to AIDS rather than to other factors.23 Thus, in the period in which we see a dramatic increase in the proportion of births to single mothers, we observe two phenomena—a decline in the frequency of shotgun marriages and increased sexual activity. The availability of more effective contraceptive methods cannot by itself explain both of these phenomena. Much sexual activity of unmarried women still involves no contraception or relatively unreliable methods. However, it is plausible that the availability of contraception and abortion changed the “market for sex” in such a way that men took less responsibility for pregnancy and were able to demand more sex. The new equilibrium was one with more sexual activity among unmarried partners and fewer marriages subsequent to conception.

4. The Role of Welfare In Losing Ground, a controversial book written two decades ago, Charles Murray argued that the failure of the poverty rate to decline after the early 1970s was the result of the expansion of federal poverty programs.24 He pointed to the dramatically increased generosity of welfare and other programs and claimed that they discouraged marriage and encouraged single motherhood. Murray’s book was written shortly after the generosity of welfare payments peaked (see chapter 3 on the evolution of poverty policy). In retrospect, an obvious response to Murray is that despite the decline in welfare generosity and the growth of programs serving two-parent families, single motherhood has not declined. Murray has noted the difficulty of using the time pattern of welfare payments and the illegitimacy ratio to draw strong conclusions about the effect of welfare payments on single motherhood. We would not necessarily expect single women to have kept close track of changes in welfare payments. It might have taken some time before they would have realized that the generosity of such payments had declined and therefore to respond by reducing their fertility or increasing the frequency with which they married. Moreover, it is possible to argue that increasing welfare generosity spurred the increase in single motherhood and that the increase in single motherhood changed social values. Once attitudes had changed, reductions in generosity were no longer capable of reversing the trend. At best, they might have slowed the trend down.25

23. Joyce C. Abma et al., Fertility, Family Planning, and Women’s Health: New Data from the 1995 National Survey of Family Growth, National Vital Statistics Reports, vol. 23, no. 19 (Hyattsville, MD: National Center for Health Statistics, 1997), table 40. 24. Charles A. Murray, Losing Ground: American Social Policy, 1950–1980 (New York: Basic Books, 1984). 25. Charles Murray, “Does Welfare Bring More Babies?” Public Interest 15 (Spring 1994): 17–30, and “Welfare and the Family: The U.S. Experience,” Journal of Labor Economics 11 (January 1993): S224–62.

156 Chapter 6 The time trends certainly do not support the view that welfare generosity increased single motherhood. However, as we have noted before, by somewhat altering the hypothesis it is possible to reconcile the data and the original theory. Thus, if we allow for persistence of social trends purportedly initiated by welfare and for time lags in adjusting to welfare changes, it is possible that the increased welfare generosity of the 1960s and 1970s contributed to the growth of single motherhood despite the failure of single motherhood to decline as benefits decreased. Are there other ways of testing Murray’s hypothesis? Welfare benefits vary considerably across states. In 1994, the maximum payment for a family of three receiving Aid to Families with Dependent Children (AFDC) was $120 per month in Mississippi and $680 per month in Connecticut.26 Although the cost of living was higher in Connecticut than in Mississippi, the difference in the real benefit was large. Because benefits vary across states, it is possible to look at the relation between the generosity of the welfare benefit and single motherhood. Of course, states differ in a variety ways. Mississippi and Connecticut spend different amounts per student on educating their children, have different fractions of their populations living in rural areas, have different climates, and so on. Researchers attempt to control for these differences by considering other factors that might affect single motherhood. Robert Moffitt reviewed seventeen studies that use differences in the generosity of welfare payments across states to estimate their effect on single motherhood.27 He reports that one found that more generous payments reduced single motherhood, but four found no effect. Of the remaining twelve, nine found a positive effect on single motherhood among whites, and four found a positive effect among African Americans (including one that also found a positive effect among whites). Moffitt reports that most of the estimated effects were small. A reasonable summary of these results is that there probably is a small effect of welfare on single motherhood among whites and that there may be a small effect among blacks, but that is less clear. Murray’s analysis of the data suggests that the results are largely driven by differences between the south, where benefits are generally low, and the rest of the country.28 Southern blacks tend to have higher rates of single motherhood than other blacks. In contrast, southern whites tend to have lower rates of single motherhood than other whites. This analysis raises three concerns. First, we must be cautious about attributing an effect to welfare generosity when the difference may actually be due to other regional differences. Second, because different studies generate conflicting results, our conclusions may be sensitive to the set of state characteristics for which we control in generating our estimates. Third, we need to ask ourselves why different states have different

26. Rebecca M. Blank, It Takes a Nation (Princeton, NJ: Princeton University Press, 1997), 100. 27. Robert A. Moffitt, “The Effect of the Welfare System on Nonmarital Childbearing,” in Re- port to Congress on Out-of-Wedlock Childbearing (Hyattsville, MD: U.S. Department of Health and Human Services, 1995), 167–76. 28. Murray, “Welfare and the Family: The U.S. Experience.”

Family Composition 157 levels of welfare generosity. Erzo Luttmer finds that whites are more supportive of welfare when the recipients are predominantly white.29 It is plausible that there is less support for welfare when more of the recipients are black single mothers and more support when there are more white single mothers. If so, it would not be surprising to find that welfare was more generous when the rate of single motherhood among whites was higher and when the rate among blacks was lower. If so, the causality would run from rates of single motherhood to welfare generosity and not the other way around. One solution to these problems is to look at the relation between changes in welfare generosity and changes in rates of single motherhood. Many characteristics of states do not change or change little over time. Mississippi has always been a southern state and Connecticut a New England state. Mississippi has been more rural than Connecticut for a long time. If single motherhood increases more rapidly in Mis- sissippi than in Connecticut, we know that it is not because Mississippi moved farther south. Thus, in looking at changes we automatically “control for” all factors that do not change over time or change only very slowly. Moffitt reviewed three studies that used this approach. Only one of the three found a positive effect of welfare generosity on single motherhood. Again, the results of all studies taken together point to at most a small effect of welfare generosity. Of course, we must also worry about the proper interpretation of studies based on changes in welfare generosity. Although these studies control for factors within states that do not change or change only very slowly, they do not control for factors that change more rapidly. In particular, if states respond to increases in single motherhood by making their welfare payments less generous, we will again draw the wrong conclusion: we will find that reduced generosity was associated with increased rates of single motherhood. In summary, there is no evidence that differences in welfare generosity account for much of the change in the rate of single motherhood over time. This does not prove that no such relation exists, but it does place a burden of proof on those who advocate reducing welfare because of its impact on single motherhood.

5. Features of Welfare Although welfare generosity in itself appears to have little or no effect on single motherhood, it is possible that aspects of the welfare program do have effects. In particular, for a long time AFDC was limited to single mothers. A woman who married would lose her benefits even if her husband was unable to support her and her children. As a consequence, the 1988 Family Support Act required that all states adopt AFDC-UP (Aid to Families with Dependent Children—Unemployed Parent), a form of welfare available to two-parent families that had previously been adopted by some states, but not all.

29. Erzo F. P. Luttmer, “Group Loyalty and the Taste for Redistribution,” Journal of Political Economy 109 (June 2001): 500–528.

158 Chapter 6 However, even with the presence of AFDC-UP, marriage could be costly. Unlike single mothers eligible for AFDC, married couples were subject to a thirty-day waiting period before being eligible for AFDC-UP. Moreover, the couple was eligible only if it could demonstrate that the husband was connected to the labor force. He had either to be eligible for unemployment insurance or to have worked during six of the previous thirteen quarters. Perhaps most important, the family was ineligible for AFDC-UP if the principal earner worked for more than one hundred hours during a month, regardless of how little he or she earned. In 1994, Minnesota developed the Minnesota Family Investment Program (MFIP) as an experiment in welfare reform.30 Many of the reforms it adopted were also parts of experiments in other states and were incorporated into the 1996 federal welfare reform. The changes in the program were quite extensive, and I will not summarize them all here. There were three very important types of changes. First, a deliberate effort was made to change the benefit formula to “make work pay.” The MFIP grant was equal to AFDC plus the value of food stamps if the recipient did not work. However, recipients were able to keep more of their income. Second, for those who did not obtain work, there was mandatory participation in employment and training services after two years for single parents and six months for two-parent families. Third, the work history and hundred-hour limits for two-parent families were eliminated. As a result, a couple with earned income of $480 per month had a monthly income of $1,259 under MFIP, compared with $1,023 under AFDC. A couple with $960 per month of earned income had a monthly income of $1,455 under MFIP, compared with $1,307 under AFDC. Income under MFIP and AFDC equalized at a monthly earned income of about $1,400.31 Between April 1994 and March 1996, new welfare applicants and long-term welfare recipients in three urban and four rural counties were randomly assigned either to enter the MFIP program or to continue on AFDC. In the three urban counties, single- parent families could also be randomly assigned to MFIP-IO (Incentives Only), which was identical to the full MFIP except for the mandatory employment and training services component. We focus on the MFIP. Table 6.1 shows the effect of the MFIP on marital status.32 The upper panel shows the effect for those participants who were either married or cohabiting before the experiment. We see that three years after entering the program, those in the MFIP were 19.1 percentage points more likely to be married and living together. Almost half of this difference reflects a shift from cohabiting to marriage, because, compared with AFDC recipients, MFIP participants who had been married or cohabiting were 9.3 percentage points less likely to be cohabiting. The rest of the difference was due to the

30. The discussion of MFIP is based on Cynthia Knox et al., Reforming Welfare and Rewarding Work: Final Report on the Minnesota Family Investment Program, Vol. 1: Effects on Adults (New York: Manpower Development Research Corporation, 2000). 31. Knox et al., 158. 32. Knox et al., tables 4.7, 5.6, and 6.6.

Family Composition 159 Table 6.1 Marital Status of Those Receiving MFIP vs. AFDC

Initial Status Current Status MFIP AFDC Difference

Two-Parent Families Married Spouse Present 67.3 48.3 19.1*** (Married or cohabiting) Cohabiting 13.5 22.8 –9.3* Divorced/Separated 8.9 21.5 –12.6*** Single 10.3 7.5 2.8 Single-Parent Recent Applicants Married Spouse Present 17.0 17.2 –0.2 Single-Parent Long-Term Recipients Married Spouse Present 10.6 7.0 3.6**

Source: Cynthia Knox et al., Reforming Welfare and Rewarding Work: Final Report on the Minnesota Family Investment Program, Vol. 1: Effects on Adults (New York: Manpower Development Research Corporation, 2000). *Significant at the .1 level. **Significant at the .05 level. ***Significant at the .01 level.

reduced likelihood of divorce or separation (12.6 percentage points). Thus, the MFIP experiment provides strong evidence that aspects of the AFDC program encouraged marital break-up and cohabitation. The bottom two panels show the effects on participants who were single parents at the start of the experiment. The evidence of an effect on their marital status is weaker. For recent welfare applicants, there was no measurable affect on the probability of being married three years after entering the program. However, for those who were long- term welfare recipients, there was a modest increase (from 7 percent to 10.6 percent) in the probability of being married. Other welfare reform experiments have not found that reform encourages single parents to marry. The most likely reason for such an effect in the MFIP is that the program allowed recipients to keep more of their earnings. However, other studies, which allowed recipients to keep more of their earnings but imposed a time limit on recipi- ency, show that, if anything, reform reduced marriage among single mothers.33 Either time limits offset the effects found in the MFIP study, or the MFIP result was random. Recall that we say that a result is statistically significant if the probability that it occurred by chance is small. If we look at enough subgroups, the probability that at least one unlikely event occurred for one of them is not small. The effect of welfare reform on the marital status of two-parent families has not been widely studied. However, the estimated effect of the MFIP on two-parent families is much larger than on single-parent families and is much more statistically significant. Moreover, a follow-up study of two-parent families seven years after the study ended

33. Lisa Gennetian and Virginia Knox, “Staying Single: The Effects of Welfare Reform Policies on Marriage and Cohabitation” (Working Paper 13, The Next Generation, Manpower Development Research Corporation, New York, 2003).

160 Chapter 6 and after Minnesota had adopted the MFIP statewide continued to show large effects of the experiment on the marriage rates of the MFIP participants.34 The MFIP not only increased the marriage rate but was associated with improved domestic relations. According to Lisa Gennetian, “Single mothers [with children aged 2–9] in MFIP were more likely to marry and were less likely to experience domestic abuse. At the three-year mark, 11.3 percent of MFIP mothers were married, compared with 6.2 percent of AFDC mothers, for a statistically significant increase of 5 percentage points. In addition, MFIP reduced the incidence of domestic abuse among urban long-term recipients; 59.6 of AFDC mothers reported experiencing domestic abuse during the three-year follow-up period, compared with 49.1 percent of MFIP mothers, for a 10.5 percentage point reduction.”35 The MFIP experiment clearly establishes that the design of welfare programs influences marital stability and marriage rates among participants in the program, at least for two-parent families entering the program. Unfortunately, it cannot address the effect of the program on potential participants. We do not know, for example, whether women who do not participate but who believe that they might need to rely on welfare at some point are more or less likely to get married if the welfare program resembles MFIP than if it resembles AFDC. This is an inherent problem with all demonstration studies of this type. However, because AFDC discouraged marriage among participants, it is likely that it discouraged marriage among those people who thought they were likely to make use of the program at some point in the future and that by encouraging divorce or nonmarriage among participants it reduced the marriage rate among those who had used it in the past.

6. Teenage Childbearing Teenage childbearing has become a major issue in the United States. Although the rate of births to teenage girls has declined overall since 1960, the birth rate among unmarried teenage girls increased rapidly from 1940 until the 1990s, although it has declined recently (see figure 6.7). In 2002, 80 percent of births to teenagers were to unmarried mothers. The received wisdom is that this is a problem for the teenagers themselves, their children, and society. In 1992, over half of mothers on AFDC first gave birth as teenagers.36 In this section, we will concentrate on the effect on the mothers themselves. The upper panel of table 6.2 compares outcomes in 2000 for women aged 35–42 who had

34. Lisa Gennetian, “The Long-Term Effects of the Minnesota Family Investment Program on Marriage and Divorce among Two-Parent Families” (Manpower Development Research Corporation, New York, 2003). 35. Gennetian, “Long-Term Effects,” 145. 36. K. Moore, “Background Data on Teenage Fertility” (Child Trends Inc., Washington, DC, 1995), cited in V. Joseph Hotz, Susan Williams McElroy, and Seth Sanders, “The Impacts of Teenage Childbearing on the Mothers and the Consequences of Those Impacts for Government,” in Rebecca Maynard, ed., Kids Having Kids: Economic Costs and Social Consequences of Teen Pregnancy (Washing- ton, DC: Urban Institute, 1998).

Family Composition 161 Birth/Illegitimacy Rate 100 Percentage Births Births to Unmarried 80 per 1,000

Births per 20 1,000 Unmarried

1940 1950 1960 1970 1980 1990 2000 Figure 6.7 Birth and Illegitimacy Rates for Teens, 1940–2001 Source: Vital Health Statistics reports (Hyattsville, MD: National Center for Health Statis- tics), various years.

their first child before 18 with those who had their first child at age 18 or later.37 Teenage mothers ultimately have .8 more children. They are 24 percentage points less likely to graduate high school or pass the high school equivalency exam. Overall, they average two fewer full years of educational attainment. Teen mothers earn one-fourth less than those who have their first child after age 18 and are twice as likely to be in poverty. They receive considerably more in welfare payments, on average, and are substantially less likely to be married. However, these differences should be treated with skepticism. Women who were teenage mothers differ from other women in many ways besides when they first give birth. The lower panel of the table shows that they are more likely to be black or His- panic. They tend to have grown up in families with more children and have substantially less-educated parents. How, then, can we obtain an estimate of the true effect of teenage motherhood on the mother’s outcomes? One approach is to attempt to control for other factors that differ between teenage mothers and those who first gave birth when they were older. A study by Arline Geronimus and Sanders Korenman began with this approach.38 They looked at mothers aged 28 to 38 who, for reasons that will become apparent shortly, have sisters who are also mothers. If we simply compare teen and nonteen mothers in this sample, we observe large differences.

37. These results are based on the National Longitudinal Survey of Youth, 1979, and rely on the month and year of birth of the first child as reported in 1982. A small number of respondents were still 17 at the time of the 1982 interview and could have given birth before their 18th birthday. 38. Arline T. Geronimus and Sanders Korenman, “The Socioeconomic Consequences of Teen Childbearing Reconsidered,” Quarterly Journal of Economics 107 (November 1992): 1187–214.

162 Chapter 6 Table 6.2 Mothers with and without Teen Births: Outcomes and Background Differences

No Teen Birth Teen Birth Outcomes Children 2.2 3.0 Hours Worked 1,485 1,478 High School Diploma/GED 93.3% 69.4% Highest Grade Completed 13.6 11.6 Poor 13.6% 27.0% Earnings $21,389 $15,948 Married 72.4% 51.2% Welfare Receipt $343 $600

Background White 81.8% 57.6% Black 11.9% 32.6% Hispanic 6.4% 9.8% Number of Siblings 3.4 4.3 Father’s Education 11.7 9.8 Mother’s Education 11.6 10.0 Number of Observations 2,885 547

Source: Author’s calculations from the National Longitudinal Survey of Youth, 1979.

Table 6.3 shows the results of Geronimus and Korenman’s further analysis of their data. Before we examine the results, we need to make two quick points about how they should be interpreted. First, when the dependent variable is the log of a variable, the coefficient can be interpreted as the percentage change in the dependent variable from a one-unit change in the explanatory variable. Second, when the explanatory variable takes only the value zero (in this case, when a mother did not give birth as a teenager) or one (when a mother did give birth as a teenager), the coefficient is approximately the percentage difference between teen mothers and nonteen mothers in the outcome, allowing for other factors. The first two outcomes are presented in log form. The remainder are recorded as either true (graduated from high school) or false (did not graduate from high school). The results using true/false dependent variables have been transformed so that the coefficient is approximately the change in the probability of the outcome from a one-unit change in the explanatory variable, and thus the percentage point difference in outcomes between teen mothers and nonteen mothers. The column of table 6.3 labeled “Using Controls” gives the effect of being a teen mother, controlling for age, urban or rural residence, race, mother’s education, father’s education, whether the mother lived in a single-parent family when she was 14, and

Family Composition 163 Table 6.3 Effects of Teen Birth on Adult Outcomes Using Standard Controls and Comparing Sisters (Women Aged 28–38)

Outcomes Using Controls Comparing Sisters

Income/Needs –0.26 –0.08 (0.09) (0.13) Family Income –0.22 –0.13 (0.09) (0.14) In Poverty 0.13 0.13 (0.06) (0.12) On Welfare 0.11 0.04 (0.04) (0.10) Graduated High School –0.19 0.08 (0.05) (0.12) Received Any Postsecondary Schooling –0.25 –0.36 (0.07) (0.17) Currently Married –0.26 –0.43 (0.07) (0.16) Ever Married –0.10 –0.05 (0.04) (0.08) Employed –0.01 0.15 (0.07) (0.12)

Source: Arline T. Geronimus and Sanders Korenman, “The Socioeconomic Consequences of Teen Childbearing Reconsidered,” Quarterly Journal of Economics, 107 (November 1992): 1187–214. Note: Income/needs and family income are measured in natural logarithms. All remaining variables equal one if the sample member has the characteristic and zero if she does not.

the number of siblings she had. Each row represents a different outcome. The table does not show the effect of the other factors controlled for in the estimation. We can see that even adding controls for these factors, the estimated effects of teenage motherhood are large. The expected income difference is about 22 percent. Even in their thirties, women who were teenage mothers are about 13 percentage points more likely to be poor and about 11 percentage points more likely to be on welfare than nonteen mothers. They are about 19 percentage points less likely to have graduated high school, 25 percentage points less likely to have received any postsecondary education, and 10 percentage points less likely to be married, although there does not seem to have been a significant effect on their probability of being employed. Therefore, we conclude that controlling for the measured differences between teen mothers and other women somewhat reduces our estimate of the negative consequences of teenage motherhood, but the reduction is not large. Yet the existence of measured differences suggests that unmeasured differences might be important. What

164 Chapter 6 we would really like to do is compare similar women, one of whom gave birth as a teenager and the other of whom did not. Geronimus and Korenman suggest that the natural approach is to compare sisters. Sisters would be expected to be more similar than two randomly chosen women. Thus, they compared women who gave birth as teens with their sisters. The column labeled “Comparing Sisters” shows the results of this exercise. Now we find that women who gave birth as teens have earnings that are only about 8 percent lower than those of their sisters who did not, and the difference is no longer statistically significant. The estimated effects for most other outcomes also fall and, with the exception of having any postsecondary education, become statistically insignificant. The point estimates suggest that women who gave birth as teens are more likely to have graduated high school and to be employed, although the differences are not statistically significant. We now need to consider why comparing outcomes for sisters might lead to biased conclusions. We should ask, “Why did childbearing outcomes differ for the two sisters?” Although the sisters are more similar than two randomly chosen women, except in the case of some twins they will have grown up in somewhat different families (at the very least they will have different sisters), and except in the case of identical twins they will have different genes. The same differences that caused one sister to have a child as a teenager and the other not to have a child may have caused that first sister to be less successful in life. It is likely that comparing sisters does not control completely for differences that affect life outcomes. For this reason, the comparison of sisters may still overestimate the negative effects of being a teenage mother. On the other hand, there are also reasons that the comparison of sisters might lead us to find an effect that is smaller than the true effect of teenage motherhood on the mother. Families share resources. If one daughter was having problems, the family would have been likely to divert social and economic resources to that daughter. The daughters’ parents may have used funds intended for their daughters’ college education to pay for their grandchild’s daycare. Thus, the sister who did not give birth as a teenager may have been directly affected by the birth. Indeed, in the extreme case where parents tried to equalize the outcomes of their daughters, we would find no effect on outcomes when comparing sisters. We therefore use a different approach that draws on the work of Joseph Hotz, Su- san Williams McElroy, and Seth Sanders.39 We consider a sample of women who first became pregnant before age 18. Some of these carried the pregnancy to term, while others had an abortion and yet others miscarried. If all women who became pregnant either miscarried or gave birth, our problem would be simple. Among women whose first pregnancy occurred before they turned 18, we could compare those who had miscarriages with those who gave birth. Although certain behaviors increase the risk of miscarriage, miscarriage is still largely random. Therefore, comparing women who gave birth as teens with those who miscarried would give a pretty good indication of the

39. Hotz, McElroy, and Sanders.

Family Composition 165 effect of a birth on teens who become pregnant. However, to the extent that alcohol and drug abuse and other risky behaviors both increase the likelihood of miscarriage and are associated with worse adult outcomes, our comparison would understate the harm from teenage motherhood. The problem is further complicated by the fact that some women choose to have abortions, and that decision is nonrandom. Among women who become pregnant as teens, those who have abortions tend to come from more advantaged backgrounds. We need to consider how the availability of abortion affects the comparison of teens who miscarry with those who give birth. We begin with two extreme cases. If all abortions occurred early in pregnancy and all miscarriages occurred late in pregnancy, women who miscarried would be a random sample of those who became pregnant and chose not to have an abortion. By comparing teens who miscarried with those who gave birth, we could measure the effect of a teen birth on women who became pregnant and would not have an abortion. On the other hand, if all abortions occurred late in pregnancy and all miscarriages early, teens who miscarry would be a random sample of all teens who become pregnant. We could therefore estimate the effect of giving birth as a teenager by comparing those who got pregnant and did not have a miscarriage with those who got pregnant and had a miscarriage. As discussed in the appendix to chapter 5, we would have to “blow up” the results of this comparison to account for the fact that some women who do not miscarry have an abortion and thus do not have a live birth. Of course, the reality is between these extremes. Some women who have early abortions would have miscarried, and some women who have early miscarriages would otherwise have had an abortion. What happens to our two approaches in this case? If we compare women who miscarry with women who give birth, the sample of women who miscarry is a mixture of women who would have given birth had they not miscarried with women who would have had an abortion. As we have already noted, teens who have abortions tend to come from more favorable backgrounds than do those who give birth. This means that, on average, teens who miscarry will come from somewhat more favorable backgrounds than do those who give birth. Part of the difference between the two groups will reflect this difference in background. Therefore, if we compare teens who miscarry with teens who give birth, we will tend to overestimate the negative effect of teen motherhood. On the other hand, the sample of teens who miscarry will not be a random sample of all teens who become pregnant, because some of the teens who would have miscarried have an abortion instead. Teens who would abort will be underrepresented among those who miscarry. Therefore, the set of teens who miscarry will tend to come from somewhat less advantaged backgrounds than the set of teens who give birth or have an abortion. A comparison of teens who miscarry with those who either give birth or have an abortion will tend to underestimate the negative effect of a teen birth. Because one comparison gives an estimate that is too large and the other an estimate that is too small, by doing both sets of comparisons we can get a sense of the true effect. Of course, if the two estimates are very different, the exercise will not be very

166 Chapter 6 Table 6.4 Effect of Teen Motherhood on Teens Experiencing Pregnancy

Birth/Abortion Birth vs. Outcomes vs. Miscarriage Miscarriage

Children 0.37 0.54 (0.24) (0.18) Hours Worked –149 –151 (195) (145) High School Diploma/GED (percentage point difference) –4.56 –5.55 (7.61) (5.92) Highest Grade Completed (differences in years) –0.24 –0.28 (0.37) (0.26) Poor (percentage point difference) 6.21 7.27 (8.27) (6.33) Earnings (difference in dollars per year) –6,375 –6,144 (3,266) (2,381) Married (percentage point difference) –4.24 –5.01 (8.85) (6.63) Welfare Receipt (difference in dollars per month) –426 –273 (315) (242)

Source: Author’s calculation from the National Longitudinal Survey of Youth, 1979. informative, but if they are similar, we can get a reasonably precise estimate of the effect of teen motherhood. We carry out this exercise using data from the National Longitudinal Survey of Youth, the same data we looked at in table 6.2, except that here we do not compare all teen mothers with all who became mothers later. Instead we compare teen mothers with women who miscarried as teens, using different approaches to abortion in the two columns. There are 938 women who gave birth as teens, 83 who miscarried or had stillbirths, and 213 who had abortions as teens. Information on some outcomes are not available for all observations. Table 6.4 shows the results. The first thing to notice is that even when we compare births and miscarriages (which we argued would overestimate the adverse effects of teen motherhood), the differences in table 6.4 are substantially smaller than those in 6.2 for most outcomes. The difference in the percentage receiving a high school diploma or GED drops from about 24 percentage points to essentially zero and the difference in educational attainment from about two years to about one-fourth year. The estimated differences in the probability of being poor, the probability of being married, and the total number of children all fall. On the other hand, when we simply compare teen mothers and other mothers, there is no important difference in hours worked

Family Composition 167 when they are 35 to 43, but comparing teen mothers with those who miscarried as teens suggests that the teen mothers may work somewhat less. We observe differences in earnings and welfare receipt that are roughly equal to the differences in table 6.2.40 The second important point about table 6.4 is that the two approaches give quite similar results. In a couple of cases, the differences approach statistical significance using the birth and miscarriage comparison but not the comparison combining births and abortion. However, in all cases the magnitude of the effects is similar, suggesting that neither approach is badly biased.

7. Effects of Growing Up with a Single Parent There is considerable agreement that it is better for a child to grow up in a family with two loving parents than in a family from which one or both parents is absent. There is much less agreement about why that is the case and whether the critical variable is the presence of two adults or of both biological parents. The top panel of table 6.5 compares outcomes for adults who spent their entire childhoods from birth to age 18 with two biological parents with the outcomes for those who spent at least part of their childhoods in some other parental structure. Those who grew up with some other structure show severe deficits. Their scores on the Armed Forces Qualifying Test (AFQT, a test of cognitive ability similar to an IQ test that was administered to all members of the sample regardless of interest in the military) are about 11 points lower (equivalent to roughly 7 IQ points). They are 11 percentage points less likely to have graduated high school, and they have, on average, one year less education. They are 9 percentage points less likely to be married. Their average income is about 20 percent lower, and their poverty rate about 6 percentage points higher. Both their labor income and their net assets are substantially lower. It is evident that it is good to be brought up in a two-parent family. But the important question is: Why do those brought up in two-parent families do better? As the lower panel of the table shows, many of the characteristics associated with growing up in a two-parent family are also characteristics that are generally associated with better outcomes even among those growing up in a similar family structure. Children who grow up with both their biological parents are more likely to be white, to have better-educated parents, and to have been born when their parents were older, and they are considerably less likely to have had a parent who abused alcohol. Table 6.6 reports two types of results. The left-hand columns look at the estimated effect of ten years spent with the biological mother and ten years spent with the biological father without taking account of any other differences among individuals that might affect their adult outcomes. The right-hand columns repeat the exercise but con-

40. Hotz, McElroy, and Sanders found that, if anything, the earnings of teen mothers are higher than those who miscarried as teens. The difference is not due to a difference in sample choice or technique. The data from 2000 show much larger differences than the data from the early 1990s.

168 Chapter 6 Table 6.5 Adult Outcomes for Those Who Grew Up with Two Biological Parents vs. Spending Some Time without Both Biological Parents

Two Biological Parents Other Parental Structure Outcomes Age 28–36 AFQT 51.5 40.8 Highest Grade Completed 13.6 12.6 High School Degree 91.9% 80.5% Married 63.9% 54.5% Income Below Poverty Line 7.2% 13.5% Household’s Labor Income $45,265 $36,038 Own Labor Income $22,860 $17,845 Net Assets $598,834 $67,227

Background Characteristics White 83.7% 71.4% Black 10.6% 21.7% Male 51.5% 48.2% Father’s Education < 12 30.4% 37.2% Mother’s Education < 12 27.2% 39.8% Mother’s Age at Child’s Birth 27.6 25.9 Father’s Age at Child’s Birth 30.6 29.5 Father Abused Alcohol 14.1% 33.8% Mother Abused Alcohol 2.9% 8.3% Number of Siblings 3.3 3.8

Source: Kevin Lang and Jay Zagorsky, “Does Growing Up with an Absent Parent Really Hurt?” Journal of Human Resources 36 (Spring 2000): 253–73, tables 1 and 2. trol for other factors that differ among individuals. For each pair of columns, the one labeled “Mother” shows the effect of spending an additional ten years with one’s biological mother (holding years spent with one’s biological father constant), and the one labeled “Father” shows the effect of spending an additional ten years with one’s biological father (holding years spent with one’s biological mother constant). The top panel shows the effects on sons, while the lower panel shows the effects on daughters. Thus, the first line of the top panel indicates that, when we do not control for other factors, a son’s spending an additional ten years with his biological mother is associated with a trivial reduction in his AFQT score, while spending an additional ten years with his biological father is associated with a large increase in this score. The left-hand columns present a striking and consistent result. For a son, spending more years with his biological father is associated with substantially better adult outcomes. On average, he will have greater cognitive ability (higher AFQT scores), obtain more education, be more likely to be married, and earn more. The presence of his mother, on the other hand, seems to be almost completely unimportant. None of the

Family Composition 169 Table 6.6 Estimated Effect of an Additional Ten Years Spent with One’s Biological Mother and Father

No Controls Controls Outcomes Mother Father Mother Father Sons AFQT –0.35 8.94 0.02 3.26 (1.61) (0.96) (1.35) (0.83) Education 0.14 0.64 0.12 0.21 (0.14) (0.09) (0.10) (0.07) Married 0.03 0.07 0.04 0.01 (0.03) (0.02) (0.05) (0.03) Earnings (thousands of dollars) 0.95 4.52 1.36 0.23 (1.61) (0.90) (1.42) (0.87)

Daughters AFQT 2.78 9.89 4.48 2.56 (1.53) (0.79) (1.23) (0.75) Education 0.44 0.65 0.26 0.15 (0.13) (0.07) (0.09) (0.06) Married –0.01 0.09 0.00 0.02 (0.03) (0.02) (0.02) (0.02) Earnings (thousands of dollars) 1.29 1.99 0.12 –0.04 (0.95) (0.42) (0.82) (0.41)

Source: Kevin Lang and Jay Zagorsky, “Does Growing Up with an Absent Parent Really Hurt?” Journal of Human Resources 36 (Spring 2000): 253–73. Note: Based on regression of outcome on years spent with biological mother and years spent with biological father.

effects of living with his mother is statistically significant, and the effect on his cognitive ability is negative, although statistically insignificant. When we look at the left-hand columns, mothers do seem to be important to their daughters. For daughters, more years with their biological mother is associated with higher cognitive scores (although the t-statistic falls short of two) and more education, although the effects are modest. There is still no statistically significant effect on the two remaining outcomes. Their father’s presence is important for all four outcomes for daughters, and the estimated effects are substantially larger than they are for their mother’s presence. But as we noted in our discussion of table 6.5, parental absence is associated with numerous other background factors that are, in turn, associated with less desirable adult outcomes. The right-hand columns of table 6.6 control for some of these fac-

170 Chapter 6 tors.41 They show that to a large extent the relation between parental presence and the outcomes of their children is due to factors associated with parental absence but do not result from that absence. Once we control for these factors, a father’s presence has no statistically significant effect on his son’s marital status or earnings, and the effect of his presence on his son’s educational attainment and cognitive ability is about one-third of the effect we find without controls. Although sons who spend more years with their father do have somewhat higher cognitive scores and obtain more education, the effects are small. There continues to be no evidence of an important effect of a mother’s presence on her sons. As with sons, the story for daughters changes noticeably when we control for other factors that are correlated with parental absence. The effects of mothers on their daughters’ cognitive ability and education remain statistically significant, although substantively modest. The estimated effects of fathers decline and cease to be statistically significant for marriage and earnings. Moreover, the estimated effects of fathers on their daughters’ cognitive ability and educational attainment are now smaller than the modest effects of mothers. The results suggest that a mother’s and father’s presence has only a modest effect on their children’s outcomes. However, we must, as always, treat such results with caution. Our controls for other factors that affect children’s outcomes are quite limited. It is possible that with better controls, the entire effect of family composition could disappear. For example, we have no controls for mother’s and father’s cognitive ability. Be- cause cognitive ability is at least partially inherited, controlling for measures of parents’ cognitive ability might completely eliminate the effects of family composition on a child’s cognitive ability and schooling attainment. Earlier we saw that controlling for observed differences reduced the estimated effects of giving birth as a teenager, but that finding a good comparison group (sisters, teens who had miscarriages) reduced the effects even further. One natural response is to compare siblings who spent different amounts of time with both parents. This is similar to comparing sisters when we look at teen births. The problem is that divorce may have different effects on children of different ages. Should we attribute the difference in outcomes between the sibling whose parents divorced when she was 15 and the sibling who was 5 at the time to the ten years’ difference in time spent in a two-parent family or to the difference in the effect of experiencing a divorce at age 5 instead of 15? One way to try to resolve this problem is to find a source of parental absence that is at most weakly related to other factors that affect child outcomes. Although parental death is not entirely unrelated to other background characteristics that affect child outcomes, it is not strongly related to either favorable or unfavorable characteristics. Parents who are older when their children are born are more likely to die before their children reach 18, as are less-educated parents and those who abuse alcohol. Some of

41. Number of siblings, race/ethnicity, age, mother’s and father’s ages when the child was born, alcohol abuse by the mother and father, the mother’s and father’s education, whether the child was born in the United States, and region of residence.

Family Composition 171 the factors that increase the probability of parental death are associated with better outcomes for children, while others are associated with worse outcomes. Overall, parental death is associated with factors that tend to worsen child outcomes, but this relation is much weaker than between divorce or separation and these factors. Thus, we expect that if we look at the effect of parental death on child outcomes, we will overestimate the negative impact but that this overestimation will be smaller than in the case of divorce and separation. Therefore, we compare the estimated effect of parental death and the estimated effect of other sources of parental absence on child outcomes. If the effects are similar, the easiest explanation is that the true effects are similar and that the biases are small. It turns out that most of the effects are similar. There is no difference between the effect of death and the effect of other causes of parental absence on cognitive ability, educational attainment, or earnings. However, sons whose fathers die are less likely to be married than sons whose parents are absent for some other reason or whose fathers are not absent. Except for this one inconsistency, the results suggest that the causal effect of parental absence on child outcomes is small.42

8. Intergenerational Transmission of Poverty The previous discussion shows that growing up with a single parent does not have a large effect on adult outcomes. However, it does suggest that growing up in disadvantaged circumstances has long-term consequences. In fact, although there is considerable intergenerational mobility in the United States, there is also considerable persistence of economic status. A variety of studies suggest that a son whose father earns half the average earnings will himself earn, on average, three-quarters of the average earnings. A son whose father earns twice the average will tend to earn about 50 percent more than the average. Overall, a son is likely to be halfway between the average and his father. There are fewer studies for daughters, but they suggest a similar relation between a daughter’s family income as a child and her family income as an adult and between her earnings and her father’s earnings.43 David Zimmerman has estimated that 40 percent of sons whose fathers are in the bottom quartile of the earnings distribution will themselves be in the bottom quartile.44 At the other extreme, 41 percent of sons whose fathers are in the top quartile will also be in the top quartile. Only 12 percent of sons whose fathers were at the bottom will make it to the top, and only 9 percent whose fathers were at the top will fall to the bottom.

42. Kevin Lang and Jay Zagorsky, “Does Growing Up with an Absent Parent Really Hurt?” Jour- nal of Human Resources 36 (Spring 2000): 253–73. 43. For a review of this literature, see Gary Solon, “Intergenerational Mobility in the Labor Mar- ket,” in Orley Ashenfelter and David Card, eds., Handbook of Labor Economics, Vol. 3A (Amsterdam, New York, and Oxford: Elsevier North-Holland, 1999), 1761–800. 44. David J. Zimmerman, “Regression toward Mediocrity in Economic Stature,” American Eco- nomic Review 82 (June 1992): 409–29.

172 Chapter 6 Mary Corcoran estimates that of children whose family income averaged below the poverty line when they were growing up, 24 percent will be poor from age 25 to 27.45 In contrast, only 4 percent of those who were not poor (by this measure) when they were growing up will be poor. Overall, she finds that growing up poor is associated with substantially worse outcomes with respect to earnings and income; modest effects on education, employment, and fertility; and little or no effect on cognitive ability, health, or psychological well-being. The effects of childhood poverty may depend on the age at which the child experiences poverty. A review of the literature by the National Academy of Sciences emphasizes the importance of the first five years of life for development.46 Some studies find stronger associations between early childhood poverty and outcomes.47 It is important to note that these associations do not demonstrate that poverty itself is the cause of children’s poorer outcomes as adults. One explanation, not widely held, is that many traits are inherited. Children are likely to inherit the characteristics that cause their parents to be poor. Susan Mayer argues that much of the difference in outcomes can be attributed to differences in parenting practices and other behavioral differences between poor parents and those who are better off.48 The strongest evidence that money has a direct influence on the outcomes of poor children comes from a synthesis of welfare reform experiments.49 In the 1990s, a number of evaluation studies assessed the effects of proposed reforms. All of these reforms were intended to increase work among actual or potential welfare recipients. In some of the experiments, poor families increased their incomes as well as the amount they worked. In others, the effect was only on work and not on income. Young children saw improvements in their outcomes in the experiments that raised income but not in the experiments that only increased work.50 Among adolescents, however, a similar review does not reveal a positive effect of income but shows a negative effect of maternal work.51 To some extent, our theories about the intergenerational transmission of poverty will guide the policies we consider. If we think that poor children suffer from cognitive deficits because their parents cannot provide them with adequate nutrition, we are

45. Mary Corcoran, “Mobility, Persistence, and the Consequences of Poverty for Children: Child and Adult Outcomes,” in Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty (Cambridge, MA: Harvard University Press, 2001), 162–89. 46. Jack P. Shonkoff and Deborah A. Phillips, eds., From Neurons to Neighborhoods: The Science of Early Childhood Development (Washington, DC: National Academy Press, 2000). 47. See Greg J. Duncan and Jeanne Brooks-Gunn, eds., Consequence of Growing Up Poor (New York: Russell Sage, 1997). 48. Susan E. Mayer, What Money Can’t Buy: Family Income and Children’s Life Chances (Cam- bridge, MA: Harvard University Press, 1997). 49. See chapter 9 for a discussion of welfare reform. 50. Pamela A. Morris et al., How Welfare and Work Policies Affect Children (New York: Manpower Development Research Corporation, 2001). 51. Lisa Gennetian et al., How Welfare and Work Policies Affect Adolescents (New York: Manpower Development Research Corporation, 2002).

Family Composition 173 more likely to focus on programs that give families more income or provide them with nutritious food. If we think that the problem is parenting practices, we will consider programs that educate parents about such practices or look for ways to provide high- quality daycare programs. However, to some extent, the channel through which a program works may be less important than knowing that it works. A program that promotes better parenting practices may be effective because it puts the parent in contact with a medically knowledgeable nurse, or a preschool program may provide wholesome meals and allow the parent to work longer hours. In the following sections we will review policies aimed at children, categorized imperfectly by the age group they are intended to serve. For the most part, even if the evidence suggests that they work, we will not know whether they work through the intended channel.

9. Policies Aimed at Infants and Toddlers The three likely areas for intervention aimed at infants are nutrition, parenting practices, and medical care. Note that these interventions are likely to begin before birth, because prenatal medical care and nutrition are likely to affect the health of newborns.

9.1. Nutrition: WIC. The Special Supplemental Nutrition Program for Women, In- fants, and Children (WIC) is intended to improve nutrition in pregnant women and those who have recently given birth as well as in children aged 5 and under. Recipients must have incomes below 185 percent of the poverty line. In many ways, WIC is similar to the food stamps program except that it is more targeted both with respect to the population that is eligible and the foods for which WIC can be used. In particular, WIC is limited to foods that are high in nutrients, particularly vitamins and minerals.52 Although WIC has been heavily studied, and most studies have found positive effects on child outcomes, the research has also been criticized because few studies have had good control groups.53 The work of Lori Kowaleski-Jones and Greg Duncan is a recent exception.54 These researchers compared siblings where the mother participated with one sibling but not the other. Their sample size was small, and we have the usual concerns about whether the siblings differed in ways that affected both their outcomes and their probability of participating. Nevertheless, Kowaleski-Jones and Duncan found that prenatal participation has an important effect on birthweight. Marianne Bitler and Janet Currie examined outcomes for women whose deliveries were paid for by Medic-

52. Commission on Behavioral and Social Sciences and Education, Evaluating Food Assistance Programs in an Era of Welfare Reform: Summary of a Workshop (Washington, DC: National Academies Press, 1999). 53. See Leighton Ku, “Debating WIC,” Public Interest 135 (Spring 1999): 108–12, and Douglas Besharov and Peter Germanis, “A Reply,” Public Interest 135 (Spring 1999): 112–17. 54. Lori Kowaleski-Jones and Greg J. Duncan, “Effects of Participation in the WIC Program on Birthweight: Evidence from the National Longitudinal Survey of Youth: Special Supplemental Nu- trition Program for Women, Infants, and Children,” American Journal of Public Health 92 (May 2002): 799–804.

174 Chapter 6 aid and who therefore were automatically eligible for WIC.55 Consistent with the literature on take-up rates for poverty programs,56 the women from this group who actually enrolled in WIC appeared more disadvantaged than those who did not. Yet the outcomes in terms of higher birthweight, longer gestational period, and fewer nights in the hospital for both mother and child were more favorable for those enrolled in WIC.

9.2. Parenting Practices. There have been many small-scale programs designed to improve parenting practices among disadvantaged parents, and several of them have been evaluated in careful randomized experiments. The typical program includes home visits by a professional or paraprofessional who provides counseling, conducts developmental (and sometimes physical) screenings, and may also provide links to a broader range of services. The consensus in the literature is that these programs can increase children’s cognitive performance provided that the people conducting the home visits are sufficiently qualified.57

9.3. Health. Later in this chapter we will discuss Medicaid and the State Children’s Health Insurance Program (SCHIP), government health programs aimed at the poor in general and at poor children in particular. Here we will discuss briefly some small- scale interventions that were the subject of randomized evaluations. Poor children are noticeably less likely to be immunized against communicable diseases than are other children.58 This might reflect less information about the availability and importance of immunizations or difficulties in gaining access to them. In a randomized trial, families whose children did not have all their immunizations were given one of six treatments: (1) a general letter about the types of immunization and when they were supposed to be given, (2) a letter giving information about the immunization situation of the particular child, (3) the same letter along with increased access to immunizations, (4) the same letter plus a monetary incentive, (5) a letter unrelated to immunizations, or (6) no letter. Relative to the control group, the monetary incentive had the largest effect on immunizations, followed by increased access and the specific letter. The results suggest that both information and other factors play a role in the relatively low rate of immunizations among poor children.

9.4. Programs with Multiple Components and Longer Durations. The Abecedar- ian Project is probably the best-known intervention program that began when the children were infants. The study followed 111 children of single disadvantaged women,

55. Marianne P. Bitler and Janet Currie, “Does WIC Work? The Effects of WIC on Pregnancy and Birth Outcomes,” Journal of Policy Analysis and Management 23 (Winter 2005): 73–91. 56. Janet Currie, “The Take-Up of Social Benefits” (Working Paper 10488, National Bureau of Economic Research, Cambridge, MA, 2004). 57. Tamara Halle et al., Background for Community-Level Work on School Readiness: A Review of Definitions, Assessments, and Investment Strategies, Part II: Reviewing the Literature on Contributing Factors to School Readiness (Child Trends Inc., Washington, DC, 2000), www.childtrends.org. 58. The following is drawn from Tamara Halle et al.

Family Composition 175 almost all of whom were African American. The experimental group consisted of 57 children who were randomly assigned to be given high-quality full-day daycare until they entered kindergarten. Both the control and the treatment groups received nutritional supplements, social work services, and medical care, so we should interpret any difference between the two groups as the effect of early education in the presence of these services. Half of each group (again randomly assigned) received additional support in the form of a home-school resource teacher during the first three years of school. The teacher provided parents with additional materials aimed at their child’s needs. Some 105 of the students have now been followed through age 21. The results show that the early education had large effects. The treatment group had greater cognitive skills at 18 months, had higher test scores in grades seven and ten, and were less likely to have been retained in grade and less likely to have been in special education. At age 21, the preschool group had IQs that averaged 4.5 points higher than the controls and reading and math levels that were two and one years higher, respectively. In addition, the preschool group were more likely to attend a four-year college, more likely to hold skilled jobs, less likely to have been teenage parents, and less likely to have used marijuana in the past month or to be regular smokers. By age 21, no effects were found from the school-age treatment.59 However, there were positive effects on the teenage mothers. By the time their children entered kindergarten, almost half of the treatment group teen mothers compared with one-eighth of those in the control group had graduated high school. By the time their children were 15, 80 percent of the treatment group teen mothers had graduated high school and 90 percent held jobs, compared with 28 percent of the control group who had graduated high school and 66 percent who held jobs.60 The Brookline Early Education Project (BEEP) also shows positive results from an intensive and extended intervention, but its evaluation is less compelling because the long-term evaluation did not make use of the random assignment component of the study design. The program was made available to parents in Boston and Brookline, a suburb of Boston. Participation began three months before the mother’s due date. The original design randomly assigned the children to one of three groups. The first received regular health and development screenings and participated in a play group starting at age 2 and in a daily preschool program from age 3 until kindergarten. Boston participants were permitted to attend the Brookline schools after they finished the program. The second group received the services received by the first group plus ten to twelve home visits, and their parents participated in five parent sessions over the first two years. The third group received the same services as the first group and fourteen to eighteen home visits, and their parents participated in eight parent sessions.

59. Frank Porter Graham Child Development Center, Early Learning, Later Success: The Abecedar- ian Study, http://www.fpg.unc.edu/~abc/summary.cfm, and Frances A. Campbell et al., “Early Child- hood Education: Young Adult Outcomes from the Abecedarian Project,” Applied Developmental Science 6 (January 2002): 42–57. 60. Craig T. Ramey et al., “Persistent Effects of Early Childhood Education on High-Risk Chil- dren and Their Mothers,” Applied Developmental Science 4 (October 2000): 2–14.

176 Chapter 6 In the early years (kindergarten and second grade), the children receiving the more intensive treatments had higher-level cognitive skills and social development, and all participants did better than a matched sample from the same classrooms and similar socioeconomic backgrounds. The participants were studied again in their mid-20s. Unfortunately, the published report of the adult outcomes reports only on the results from comparing the participants with apparently similar nonparticipants who also attended Brookline High School. Despite the caveat this suggests, BEEP showed very positive outcomes for the Boston participants as adults but little effect on the Brook- line participants. BEEP participants had more education, higher incomes, better health, and less depression than did the urban comparison group.61 The promising results from early intervention programs encouraged policy makers to extend Head Start (a national preschool program discussed later) to younger children in the form of Early Head Start. Like BEEP, Early Head Start combines a preschool component with a parental education component. It is too soon to ascertain the long-term effects of Early Head Start, but at least we have the results of the short-term effects from a random assignment experiment.62 Be- tween July 1996 and September 1998, 3,001 applicants in seventeen programs were randomly assigned to either Early Head Start or a control group in roughly equal numbers. Samples at the seventeen sites ranged from 141 to 225. The study covered a very large number of outcomes, which makes it difficult to summarize. The study found statistically significant beneficial effects of participation on all four measures of child cognitive and language development. The proportion of 3-year-olds scoring below 85 on IQ tests fell by 4.7 percentage points on one test and by 6.0 percentage points on the other. The study also found positive effects of participation on children’s social- emotional development, including reduction in aggressive behavior, and also on parental behavior. Parents of children in the program were more likely to read to their children every day (a 4.9 percentage point difference) and less likely to spank their children (7.1 percentage points lower) than were parents in the control group. Parents in the program were also more likely to obtain job training themselves and to find employment and less likely to have another child than were parents in the control group.

10. Preschool Programs The Abecedarian Project and BEEP both included preschool programs as important components. However, they were unusual in that they began essentially at birth. A

61. Judith S. Palfrey et al., “The Brookline Early Education Project: A 25-Year Follow-up Study of a Family-Centered Early Health and Development Intervention,” Pediatrics 116 (July 2005): 144–52. 62. John M. Love et al., “Making a Difference in the Lives of Infants and Toddlers and Their Families: The Impacts of Early Head Start” (Child Outcomes Research and Evaluation, Office of Planning, Research, and Evaluation, Administration for Children and Families and the Head Start Bureau Administration on Children, Youth and Families, Department of Health and Human Services, Washington, DC, June 2002).

Family Composition 177 number of other initiatives with a strong preschool focus have begun their interventions at the preschool stage.

10.1. Head Start. Head Start is a program aimed at providing disadvantaged children with preschool education. In 2002, its budget was about $6.5 billion and it served over 900,000 children, mostly 3- and 4-year–olds, at an average cost of $6,934 per child. The children in the program were roughly equally divided among whites, His- panics, and blacks, with a small proportion of other ethnic/racial groups. Over three- fourths had family incomes below $15,000.63 Head Start began as a summer program in 1965 and rapidly expanded to a year- round program. Enrollment declined to about half of its peak level by 1977. Since the early 1980s, the program has expanded steadily under both Republican and Democratic administrations. Starting in 1995, Early Head Start expanded the program to families with infants and toddlers.64 The goal of all the Head Start programs is to promote school readiness among disadvantaged children. However, because Head Start requires immunizations and provides meals, it may have effects on child health and nutrition as well. As with many of the programs we have already examined, it is difficult to evaluate Head Start because it is aimed at a selected sample of the disadvantaged population. If we compare Head Start participants with children in the society as a whole, the Head Start children are likely to do worse because they are disadvantaged. However, because not all disadvantaged children can get into Head Start, we do not know if we can simply compare disadvantaged Head Start participants with disadvantaged nonparticipants. Disadvantaged parents who enroll their children in Head Start may be more motivated and more focused on their children’s educational success. On the other hand, when program administrators must choose among applicants, they may choose those they think will benefit most from the program. They could give preference to the most or the least disadvantaged. Janet Currie and Duncan Thomas have evaluated Head Start by examining differences between Head Start participants and nonparticipants.65 Table 6.7 shows the relation between Head Start participation and four outcomes: performance on the Pic- ture Peabody Vocabulary Test (PPVT, a measure of cognitive ability), whether the child (at least 10 years old) ever repeated a grade, whether the child has been immunized against measles, and the child’s height relative to the median height for his or her age (a measure of long-term nutrition). The first column of the table shows that, compared with children who did not go to preschool, white Head Start participants did worse on the PPVT, did about the same on grade repetition, were more likely to be immunized, and were about the same height as nonparticipants. Among African Americans, the Head Start participants were more likely to be immunized and were taller but did not differ from nonparticipants with respect to the PPVT and grade repetition.

63. http://www.acf.hhs.gov/programs/hsb/research/2003.htm. 64. http://www.acf.hhs.gov/programs/hsb/research/2003.htm. 65. Janet Currie and Duncan Thomas, “Does Head Start Make a Difference?” American Eco- nomic Review 85 (June 1995): 341–64.

178 Chapter 6 Table 6.7 The Effect of Head Start Relative to No Preschool

Outcomes No Controls With Controls Between Siblings Whites Picture Peabody Vocabulary Test (percentile) –5.621 –0.383 5.875 (1.570) (1.453) (1.520) Never Repeated a Grade –0.035 0.004 0.473 (0.058) (0.061) (0.112) Received Measles Immunization 0.152 0.030 0.082 (0.025) (0.019) (0.030) Height for Age –0.171 –0.207 0.084 (0.330) (0.328) (0.399)

African Americans Picture Peabody Vocabulary Test (percentile) 1.037 0.739 0.247 (1.223) (1.135) (1.358) Never Repeated a Grade –0.001 0.000 0.008 (0.067) (0.064) (0.098) Received Measles Immunization 0.167 0.072 0.094 (0.026) (0.020) (0.034) Height for Age 1.024 0.452 0.549 (0.382) (0.364) (0.540)

Source: Janet Currie and Duncan Thomas, “Does Head Start Make a Difference?” American Economic Review 85 (June 1995): 341–64. Note: Based on regression of outcome on Head Start participation.

The second column controls for a number of factors that vary across individuals— age, sex, birth order, a measure of family income, mother’s and grandmother’s education and mother’s cognitive ability, height, and number of sibling. Controlling for these factors eliminates all significant differences between white participants and nonparticipants and between African American participants and nonparticipants except for immunization, which continues to be more common among African American participants than among nonparticipants. If we were to limit ourselves to the results in the middle column, we would conclude that Head Start had little effect except for increasing immunizations among African American children. However, as we have seen previously, controlling for measured differences is imperfect. In particular, even measurably similar participants and nonparticipants may differ with respect to motivation. The third column compares siblings who participated with those who did not go to preschool. Among whites, we now observe positive effects of Head Start on PPVT (almost 6 percentage points higher), avoiding grade repetition (almost 50 percentage points less likely), and immunization

Family Composition 179 Table 6.8 The Effect of Head Start Relative to No Preschool

Outcomes No Controls With Controls Between Siblings

Completed High School –0.075 0.006 0.037 (0.035) (0.034) (0.053) Attended Some College –0.016 0.075 0.092 (0.033) (0.033) (0.056) log Earnings (Ages 23–25) 0.053 0.170 0.194 (0.116) (0.117) (0.257) Booked or Charged with a Crime 0.041 0.012 –0.053 (0.026) (0.026) (0.039)

Between Siblings Mother’s Education ≤ 12 African American White African American White

Completed High School –0.025 0.203 0.000 0.283 (0.065) (0.098) (0.071) (0.119) Attended Some College 0.023 0.281 0.031 0.276 (0.066) (0.067) (0.067) (0.120) log Earnings (Ages 23–25) 0.073 0.566 0.051 1.004 (0.321) (0.459) (0.357) (0.516) Booked or Charged with a Crime –0.116 0.122 –0.126 0.058 (0.045) (0.077) (0.050) (0.095)

Source: Eliana Garces, Janet Currie, and Duncan Thomas, “Longer-Term Effects of Head Start,” American Economic Review 92 (September 2002): 999–1012. Note: Based on regression of outcome on Head Start participation.

(8 percentage points more likely). Among African Americans, we continue to find an effect only on immunization (9 percentage points more likely). The difference in the results for whites and African Americans is somewhat surprising. Currie and Thomas argue that the failure to find a sustained effect among African Americans is due to the tendency of Head Start participants to attend worse schools.66 Among African Americans in the same school, they find better outcomes for those who attended Head Start. Table 6.8 looks at longer-term effects of Head Start. Again we see that with minimal controls for age, sex, and race (labeled “No Controls”), Head Start participants appear to have worse outcomes than do adults who did not attend preschool as chil-

66. Janet Currie and Duncan Thomas, “School Quality and the Longer-Term Effects of Head Start,” Journal of Human Resources 35 (Fall 2000): 755–74.

180 Chapter 6 dren. In three of the four cases, the average outcomes are worse for those who participated in Head Start, and in the case of high school completion, the difference is statistically significant. When we add controls for mother’s and father’s education, living in a female- headed household at age 4, family income in early childhood, birthweight, and birth order (labeled “With Controls”), three of the four outcomes are better for participants, with the difference in college attendance statistically significant. When we look at differences between siblings who did and did not participate in Head Start, all the differences are positive. Although none is statistically significant, the effects are larger than estimated when we measure the effects controlling only for measured differences between participants and nonparticipants. The lower panel looks at differences between siblings but breaks the samples into smaller but more homogeneous groups. The results suggest a long-term beneficial effect of Head Start, reducing criminal activity among blacks and increasing educational attainment and earnings among whites, especially those with mothers who had no more than a high school education. We have already discussed the difficulties of relying on studies that compare siblings. Siblings who do not participate in the program may be affected by siblings who do, in which case such studies will underestimate the benefits of the program. On the other hand, siblings who participate may be different from those who do not participate in ways that are not caused by Head Start. If parents are more likely to enroll their children in Head Start when their own lives are more stable, we will overestimate the benefits of the program. If enrollment reflects household problems, the sibling who enrolls may have faced more difficulties than the sibling who does not. A randomized evaluation of Head Start is under way, but no results are currently available, and it will be a long time before we can assess the long-term effects.

10.2. Other Preschool Programs. Although Head Start is the best-known public preschool program, there have been other programs designed to provide disadvantaged preschool children with early education. Because these programs are significantly smaller than Head Start, evaluation of the programs is generally based on very small samples. On the other hand, many were designed with evaluation in mind. Probably the best known is the High/Scope Perry Preschool Project, which provided a combination of preschool education and home visits by teachers starting at age 3 or 4.67 The preschool program was two and a half hours per day during October through May and was run by trained teachers. Class size was small (five or six students per teacher). Each day, teachers conducted home visits that lasted approximately ninety minutes. The program was limited to children whose IQs prior to entry were

67. The following two paragraphs are based on High/Scope Educational Research Foundation, The High/Scope Perry Preschool Project through Age 40, http://www.highscope.org, and Lawrence J. Schweinhart, “Benefits, Costs, and Explanation of the High/Scope Perry Preschool Program” (paper presented at the meeting of the Society for Research in Child Development, Tampa, Florida, April 26, 2003).

Family Composition 181 between 70 and 85. A sample of 123 African American children was divided randomly into treatment (58 children) and control (65 children) groups. The control group did not receive any treatment. The participants and control group have now been followed through age 40. The study showed large education effects on girls. Female participants were more likely to graduate from high school and were less likely to have been in special education or retained in grade. Overall, achievement test scores in school were higher for treatment group members than for controls. At age 40, both men and women in the program group were more likely to be employed than were the controls. This is reflected in higher probabilities of home and car ownership. Only 71 percent of the program group compared with 86 percent of the control group had received social services (e.g., welfare) at some point. Participants were less likely to have been arrested five or more times (36 percent vs. 55 percent) and to have been arrested for a violent crime (32 percent vs. 48 percent), a property crime (36 percent vs. 58 percent), or a drug crime (14 percent vs. 34 percent). Only 28 percent of the program group compared with 52 percent of the control group had spent time in prison or jail. Program males were also more likely to be raising their own children. The researchers estimate that every $1.00 invested in the preschool program saved the public $12.90. Most of this benefit has been due to reduced losses from crime. Measuring these benefits is somewhat problematic. Still, there are clear reductions in education costs. Even though participants received more education, this is more than offset by reductions in grade repetition and special education. Participants also pay more taxes and receive less welfare. These three factors alone generate a return of about $1.60 for each dollar invested. In a review, Lynn Karoly concludes that the effect on cognitive ability of such programs is transitory.68 However, there are longer-lasting effects in other areas. In particular, preschool programs can reduce the likelihood that a child will end up in special education and increase the probability of high school graduation. Some studies also show effects on criminal behavior. She speculates that duration of program may be extremely important. It may be that programs must last a few years beyond preschool to be effective.

11. Programs for School-Age Children There is a vast array of programs aimed at children in school and at teenage dropouts. In this section we will discuss programs that are provided primarily outside of a school setting. In some cases, the program may be conducted in a school building but is not operated by the school department. Programs that are run primarily through the school system will be discussed in chapter 8 on education. Job training and employment programs were discussed in chapter 5.

68. Lynn A. Karoly, “Investing in the Future: Reducing Poverty through Human Capital Invest- ments,” in Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty (Cambridge, MA: Harvard University Press, 2001), 314–56.

182 Chapter 6 Many programs have numerous components. A single program might provide food when teens meet together in the afternoon, one-on-one tutoring, a mentoring relationship, and assistance with part-time employment. If the program works, it is difficult to know which of its elements is important or whether having all the elements together is essential. With this caveat in mind, I nevertheless divide the discussion into parts: nutrition, academic achievement, mentoring, civic engagement, and pregnancy prevention.

11.1. School Nutrition Programs. The National School Lunch Program serves about twenty-seven million children, while the School Breakfast Program serves around seven million.69 The programs reimburse schools for the cost of school meals and also distribute surplus foods to these schools. Under these programs, children from families with incomes of less than 130 percent of the poverty line receive free meals, while those from families with incomes of between 130 and 185 percent of the poverty line receive their meals at a subsidized price.70 It is widely believed that having a good breakfast improves student performance. However, the chief dietary problem among the poor in the United States is increasingly obesity rather than a shortage of calories. Because school meals tend to be high in fat, there is some concern that the school nutrition program might be worsening the problem rather than improving student nutrition. The School Lunch Program is difficult to evaluate because at this point it is almost universally available. There is no compelling control group. Because the School Break- fast Program is less universal and because coverage has been growing, this program is a better target for evaluation. Alan Meyers, Amy Sampson, and Michael Weitzman took advantage of the growth of the program.71 They studied the effect of introducing the School Breakfast Program in Lawrence, Massachusetts, and found that the program increased performance on standardized tests and reduced tardiness. Jayanta Bhattacharya, Janet Currie, and Steven Haider take advantage of the fact that not all schools participate in the School Breakfast Program.72 One possibility would be to compare income-eligible children in participating schools with income- eligible children in other schools. However, participating schools tend to be poorer than other schools. Poor children in middle-class areas are likely to differ from poor children in other areas for reasons unrelated to the School Breakfast Program. Another possibility would be to compare students during the school year with themselves during

69. “Federal Statistics: USDA Food and Nutrition Programs,” Family Economics and Nutrition Review 11 (Winter 1998): 66–67. 70. Commission on Behavioral and Social Sciences and Education, 5, 8. 71. A. Meyers, A. Sampson, and M. Weitzman, “Nutrition and Academic Performance in School Children,” Clinics in Applied Nutrition 1 (April 1991): 13–25. 72. Jayanta Bhattacharya, Janet Currie, and Steven Haider, “Breakfast of Champions? The School Breakfast Program and the Nutrition of Children and Families,” Journal of Human Resources 41 (Summer 2006): 445–66.

Family Composition 183 school vacations. However, most vacations are during the summer, and nutrition may differ between the summer and the rest of the year for reasons that are also unrelated to the nutrition program. The researchers therefore ask whether there is a greater difference between school- year and summer behavior among students in participating schools than among students in nonparticipating schools. They find no difference in eating breakfast and calories consumed. However, the decline from school year to summer in an overall index of healthy eating is greater among students in participating schools than among students in nonparticipating schools. The same is true for a number of individual nutrients. Therefore, it appears than the School Breakfast Program significantly improves nutrition, although it has no effect on the probability of eating breakfast or on calories consumed. The study finds that the effects of the program on the preschool and adult members of households are similar to those on the students themselves. There are two ways to interpret this result. The first is that the household benefits both from the increased nutritional awareness from the School Breakfast Program and from the increased resources, because funds that would be used to provide breakfast for the student can be diverted to other uses. Neither mechanism seems plausible. Nutritional awareness should not depend on whether school is in session. If participants and their families learn better dietary habits as a result of the School Breakfast Program, we would not expect this to be a short-term effect that disappears in the summer and reemerges each school year. The subsidy provided by the program is modest, on the order of twenty to twenty-five dollars per month during the school year. If the household has the sophistication to redistribute resources within the household during the school year, it should also be able to redistribute these resources from the school year to the summer in order to keep nutritional quality relatively constant. The second interpretation is that even in the absence of the School Breakfast Pro- gram, households with children in schools that participate would change their consumption from school year to summer differently from households with children in schools that do not participate. As in all studies that rely on the differences-in-differences approach, there is an important and untestable assumption that in the absence of the policy intervention, the differences would be the same for the two groups. If we do not accept this assumption, we will not, in general, be able to conclude anything from the study. In this case, the fact that the rest of the family also changes its behavior differently depending on whether the child is in a school with a breakfast program casts doubt on the assumption that the child’s behavior would have changed in the same way regardless of whether he or she was in a school with a breakfast program.

11.2. Academic Achievement Programs. Zakia Redd and her coauthors reviewed evaluations of a dozen academic achievement programs. Six of the evaluations were experimental.73 The other six relied on various statistical approaches to try to control for

73. Zakia Redd et al., Academic Achievement Programs and Youth Development: A Synthesis (Wash- ington, DC: Child Trends, 2002). The programs with experimental evaluations were the Boys and Girls

184 Chapter 6 differences between youths participating and not participating in the programs. They concluded that the results of the programs were mixed. Only one out of five (not all outcomes were reported for all twelve studies) had a positive effect on student grades. The studies had consistently positive effects on an outcome when only one study examined that outcome. Despite this variation in results, most of the programs had a favorable effect on at least one academic outcome (test scores, graduation, retention in grade, attendance, aspirations), and some had positive effects in other dimensions, such as social and sexual behavior. The results tend to suggest that academic achievement programs can be successful. We know less about what aspects of the programs contribute to success. More important, we do not know how the success of the programs compares with that of similar programs run by school departments. Providing the services outside a school setting may make at-risk students more open to the programs. On the other hand, school systems are more aware of the precise academic needs of their students.

11.3. Mentoring Programs. Mentoring programs provide at-risk children with car- ing adult relationships outside their homes. There is considerable variation in mentoring programs in terms of the frequency with which the mentor and mentee meet, the purpose of the relationship (e.g., academic support, reducing risk behaviors), and the amount of guidance and supervision given to mentors. Big Brothers Big Sisters (BBBS) is one of the largest and best-known mentoring programs. In 1992 and 1993, eight sites with waiting lists were selected for a national evaluation of the program.74 Applicants who met the local chapter’s criteria (typically based on residence and willingness to accept the program rules, with preference given to those with only one parent active in their lives) were randomly assigned to the top or bottom of the waiting list. Those at the bottom of the waiting list were not placed until after the completion of the eighteen-month study. The sample consisted of 1,138 youths between 10 and 16 years old. Of these, 487 members of the treatment group and 472 members of the control group actually participated in the study. Not all of the treatment group received a mentor. Some became ineligible because the parent re- married. Others withdrew, and in some cases, no suitable match could be found. It took an average of five months to find a match, and by the time of the eighteen-month follow-up, the average match had lasted a little less than a year.

Clubs of America Educational Enhancement Program, Children at Risk, the Howard Street Tutoring Program, the Quantum Opportunities Program, the Summer Training and Education Program, and Upward Bound. The other programs were Foundations, LA’s Best, Sponsor-a-Scholar, the Texas Parks and Recreation Department After-School Programs, Fifth Dimension, and an unnamed program using university student athletes as tutors. 74. The discussion of the Big Brothers Big Sisters evaluation is based on Joseph P. Tierney and Jean Baldwin Grossman, Making a Difference: An Impact Study of Big Brothers Big Sisters (Philadelphia: Public/Private Ventures, 2000), with additional material from Jean Baldwin Grossman and Joseph P. Tierney, “Does Mentoring Work? An Impact Study of the Big Brothers/Big Sisters Program,” Evalu- ation Review 22 (June 1998): 403–26.

Family Composition 185 Mentors and mentees in BBBS are required to meet at least twice per month at all locations and a least weekly at most of the study sites. Meetings last two to six hours. Mentors receive training before taking on their role, and the staff supervises the relationship on a regular basis. In the study, most of the pairs met three or four times per month for an average of four hours per meeting. The study found favorable effects on drug and alcohol use, hitting behavior, homework completion, GPA, school attendance, and relations with parents. In general, these effects were strongest for those with initially low achievement levels and those with the weakest family support. One striking result is that the strength of the mentoring relationship was closely related to its success. The study found negative effects from mentoring relationships that lasted less than three months and stronger effects for those with more frequent contact and those rated more positively by the mentee.75 We must be cautious about assuming that these ratings reflect the effect of bad mentoring relations on the mentee. Deterio- rating mentee behavior might lead to a bad mentoring relation. However, it is possible that short-term and weak mentoring relationships are harmful to the mentee, suggesting that successful mentoring programs should have significant training and supervision components and careful matching of mentors and mentees. In their review of studies of nine mentoring programs, Susan Jekielek, Kristin A. Moore, and Elizabeth Hair conclude that mentoring or mentoring in combination with other program elements reduces school absences, increases college attendance, and improves behavior, including reducing illegal behavior and drug use. It does not have a clear effect on grades.76

11.4. Civic Engagement Programs. Civic engagement programs are intended to help children by involving them in community service. The programs can provide them with social and technical skills and increase their sense of self-worth and attachment to the community and the people with whom they work. Again there is considerable variation across programs, and many programs have multiple elements. Subject to the large caveat that the community service elements may not have been key to the outcomes, a review of evaluations of nine programs found that they had positive effects on behavior and education outcomes and increased community service activity and attention to the news.77

11.5. Teen Pregnancy Prevention Programs. Several programs have undergone rigorous evaluations that suggest they have been effective in reducing teen pregnancy, but

75. Jean Baldwin Grossman and Amy Johnson, “Assessing the Effectiveness of Mentoring Pro- grams,” in Jean Baldwin Grossman, ed., Contemporary Issues in Mentoring (Philadelphia: Public/Private Ventures, 1999). 76. Susan Jekielek, Kristin A. Moore, and Elizabeth Hair, Mentoring Programs and Youth Devel- opment: A Synthesis (Washington, DC: Child Trends, 2002). 77. Erik Michelsen, Jonathan F. Zaff, and Elizabeth C. Hair, Civic Engagement Programs and Youth Development: A Synthesis (Washington, DC: Child Trends, 2002).

186 Chapter 6 most have not been used extensively. In such cases, we are often concerned that the success of the program depends on factors that cannot be copied easily. In chapter 5 we saw that the Center for Employment Training proved its effectiveness in two separate evaluations but did not do well when the program was extended to other sites. The Teen Outreach Program (TOP) stands out as a program that has been extended to a large number of sites and subjected to rigorous evaluation. Because it is primarily school- based, TOP differs from the other programs discussed in this chapter. However, because of its focus on social behavior, I include it here rather than in the chapter on education.78 TOP is offered in high schools as either a health or a social studies class. Partici- pants are required to do at least twenty hours of community service over the year, and they participate at least weekly in class discussions of their experiences in community service and in discussions and activities centered on adolescent development. It is notable that little of the curriculum is explicitly about sex or teenage pregnancy, and many sites do not use that part of the curriculum because it overlaps with the sex education curriculum used elsewhere at the school. Twenty-five of the forty-five sites that existed at the time agreed to participate in an evaluation in 1997. In most schools, there was excess demand for the class, which allowed the investigators to randomly assign some students who wanted to take the class to a control group. In some schools, entire classes were randomly assigned to TOP or to an alternate curriculum. There were 342 students in the participant sample and 353 in the control sample. The researchers found that participants were about 40 percent less likely to fail a class, be suspended, or (in the case of girls) become pregnant. A second evaluation used randomization for 660 students but supplemented both the experimental and the nonexperimental groups with another 2,617 students who were similar along observed characteristics but were not selected by random assignment. The purpose of the nonrandom sample was to allow the researchers to ask questions about relatively small subgroups. By showing that the overall results are similar for the two samples, the researchers can make a more convincing case that their matching process worked. For the full sample, the second study showed large reductions in class failure, suspension, and pregnancy. The program was broadly effective for all groups but was particularly effective in reducing class failure among students who had experienced a suspension and in preventing pregnancy among girls who were already teen mothers.

11.6. The Quantum Opportunities Program. Although small, the Quantum Oppor- tunities Program (QOP) is a particularly interesting intervention, because its approach resembles those envisaged under the Workforce Investment Act, and we do not yet

78. This section is based on Joseph P. Allen et al., “Preventing Teen Pregnancy and Academic Failure: Experimental Evaluation of a Developmentally Based Approach,” Child Development 64 (Au- gust 1997): 729–42, and Joseph P. Allen, “Who Benefits Most from a Broadly Targeted Prevention Program? Differential Efficacy across Populations in the Teen Outreach Program,” Journal of Com- munity Psychology 29 (November 2001): 637–55.

Family Composition 187 have evaluations of programs under that act.79 It is a combination of academic achievement, mentoring, and civic engagement programs. Participants engage in up to 250 hours per year in each of education support activities, community service activities, and development activities. The program recruited highly disadvantaged youths and paid them a modest stipend for participating in the program. The initial program, and the one on which much of the assessment of QOP is based, was designed to determine whether an intensive program could attract, retain, and work successfully with highly disadvantaged high school students. In contrast with most evaluations, the one for QOP was not set up to randomize participation among a self-selected group. Instead, in each of five locations, the researchers randomly selected fifty students from local housing projects and assigned half to QOP eligibility, while the other half served as controls. Each local agency was required to seek out the selected participants and encourage them to participate. In order to encourage their involvement, participants received an hourly stipend beginning at $1.00 and rising to $1.33. In addition, after completing one hundred hours in the program, they received a $100 bonus and an additional $100 in an interest-bearing “opportunities account” to be used for postsecondary education or training. For frequent participants, this represented a significant amount of money. The average participant earned about $5,000 over the four years, with about $2,300 of that accrued in the opportunities account. Students could be dropped from the sample and replaced only if they were ineligible for the program (e.g., they had moved). One of the five locations was unsuccessful and was dropped from the evaluation. There was very little attrition from two of the sites over four years. After the program ended, researchers were able to reach all fifty members of the sample in Philadelphia and all but one participant and two controls in Oklahoma City. In Saginaw, five participants and seven controls could not be reached, while in San Antonio these numbers were eleven and fifteen. Based on a 1994 evaluation of the experiment, QOP is widely regarded as a model program.80 The American Youth Policy Forum summarizes its policy forum on QOP by citing the statement that QOP is “the most successful youth development program ever evaluated.”81 And in many ways, the initial evaluation of the program was very promising. As shown in table 6.9, members of the participant group were more likely

79. The material in this section draws on Andrew Hahn, Evaluation of the Quantum Opportuni- ties Program (QOP): Did the Program Work? (Waltham, MA: Center for Human Resources, Brandeis University, 1994), and Myles Maxfield, Allen Schirm, and Nuria Rodriguez-Planas, The Quantum Opportunity Program Demonstration: Implementation and Short-Term Impacts (Washington, DC: Math- ematica Policy Research, 2003). 80. It is highlighted in a National Academy of Science study (Jacquelynne Eccles and Jennifer Appleton Gootman, Community Programs to Promote Youth Development [Washington, DC: National Academy Press, 2002], chapter 6) as one of three programs shown to be effective after rigorous evaluation. Child Trends describes it as an effective intervention in its review of both academic achievement programs and civic engagement programs (Redd et al. and Michelsen et al.). Other organizations listing it as a model program include the Eisenhower Foundation. 81. http://www.aypf.org/forumbriefs/1994/fb102894.htm.

188 Chapter 6 Table 6.9 Outcomes from the Quantum Opportunities Program Evaluation (Fall after the Normal Year of High School Graduation)

First Evaluation (%) Second Evaluation (%) Outcomes Participants Controls Participants Controls

High School Graduate 63 42 46 40 Receive Postsecondary Schooling* 42 16 42 34 Attend Four-Year College 18 5 15 12 Attend Two-Year College 19 9 16 13 Has a Child 24 38 23 26 Received an Honors Award 34 12 Not Working and Not in School 30 50 12 18 In Trouble with the Police in the Last 6 13 25 29 12 Months**

Sources: Andrew Hahn, Evaluation of the Quantum Opportunities Program (QOP): Did the Program Work? (Waltham, MA: Center for Human Resources, Brandeis University, 1994), and Myles Max- field et al., The Quantum Opportunity Program Demonstration: Implementation and Short-Term Impacts (Washington, DC: Mathematica Policy Research, 2003). *In second evaluation, includes armed services. **Second evaluation asks whether ever arrested or charged.

to graduate high school, more likely to obtain some postsecondary education, including being more likely to attend a four-year college and more likely to attend a two-year college; were less likely to have had a child; were more likely to have won an honors award; were less likely to have been in trouble with the police; and were more likely to be economically active by either attending school or working. The study also showed that participants had more positive views of their lives. However, the study also provided reasons for being cautious in this assessment. There was clearly variation in the success of the program across sites. Philadelphia was the only site that successfully created a group identity, which should be an important component of the development aspect of the program. The others followed more of a case management approach with, when successful, strong ties between individual case managers and participants. One of the sites essentially failed completely and was not included in the evaluation, and there was considerable attrition at two of the sites. One of the sites (San Antonio) showed positive effects, but, given the attrition, the sample size was too small to allow the researchers to see statistically significant effects. Another (Saginaw) had a positive effect only on participants’ views of their lives. Given the high attrition rate in Saginaw and the evidence that short-term mentoring relations can have negative effects on mentees, this result must be regarded with caution. In sum, QOP

Family Composition 189 was very successful overall, but much of this overall success was driven by one very successful site (Philadelphia) and one moderately successful site (Oklahoma City). The success of the initial QOP encouraged the Department of Labor and the Ford Foundation to conduct a second study from 1995 to 2001. The second study involved eight sites serving a total of 680 youths. The initial results of this second study are much more mixed. Participation was disappointing, averaging 247 hours in the first year and 89 hours in the fourth year, compared with a target of 750 hours per year. By the fourth year, 36 percent of the participant group did not attend at all. The cost per member of the participant group was around $25,000 and in Philadelphia was $49,000. The study also found that the sites had difficulty following the QOP model. Although five of the sites made minor changes, two deviated substantially. Despite these concerns, the study has found positive effects. The participant group was more likely to have graduated high school, more likely to be receiving postsecondary education, and more likely to be economically active (see table 6.9). However, these effects were much smaller than in the original study. Moreover, in contrast with that study, the more recent evaluation finds no significant effects on school performance or sexual activity and unfavorable effects on substance abuse. Perhaps most disturbing of all were the differences in outcomes across sites. As in the first evaluation, Philadelphia stood out as an unusually successful site. The positive effects of the program were generally large in Philadelphia. However, four of the eight sites had no positive significant effects, and of those, three had significant negative effects in some dimension. Given the small sample sizes and the large number of outcomes tested, not too much should be read into either the absence of significance or the small number of negative effects at individual sites. What is disturbing is that the directions of the estimated effects differ across sites. The only outcome on which QOP had a favorable (though not necessarily statistically significant) effect at all eight sites was high school graduation. At seven of eight sites, the estimated effect on receiving postsecondary training was also positive. At this point, it seems safe to say that we are confident only that QOP affects high school graduation and possibly postsecondary training and that these effects are more modest than previously believed. As in the case of the Center for Employment Train- ing, much of the benefit of QOP seems to be site-specific.

12. Medicaid and SCHIP Medicaid, the federal health insurance program for the poor, is not limited to children. However, we discuss it here because significant elements are targeted toward them as well as to the impoverished elderly and the disabled. The eligibility requirements for Medicaid are complex and depend in part on the decisions of states that share the cost of the program with the federal government. All states are required to provide Medicaid to low-income families with children who would be eligible for AFDC under the old AFDC rules and to pregnant women and children under age 6 with family incomes below 133 percent of the poverty line. They must also provide Medicaid to certain recip-

190 Chapter 6 ients of Medicare (the federal health insurance program for the elderly) and to individuals covered by supplemental security income.82 There has been a dramatic expansion of Medicaid over the past twenty years. In 1998, over twenty million children were enrolled and almost nineteen million children were served by Medicaid. Medicaid covers over one-third of all births as well as 25 percent of children under age 3, 22.9 percent between the ages of 3 and 5, and 15.5 percent between the ages of 12 and 17.83 The expansion of Medicaid has been partially offset by a decline in employer provision of health insurance. Between 1988 and 1993, Medicaid coverage of children grew by 8 percentage points, while employer provision for children fell by 7 percentage points. Much of the decline in coverage occurred among families who had incomes that were too high to allow them to qualify for Medicaid. The State Children’s Health Insurance Program (SCHIP) was designed to address this problem. It is targeted at children in families with incomes below 200 percent of the poverty line. SCHIP allowed states to expand Medicaid to children in higher-income families or to establish a separate program. Many states do both. The expansion and new programs increased coverage by about two million children.84 There are a number of important issues surrounding Medicaid and SCHIP. The first is the extent to which the availability of Medicaid affects labor supply behavior. This was particularly important historically when Medicaid and welfare receipt were linked. Prior to the 1980s, families generally needed to receive AFDC payments in order to qualify for Medicaid. Women who found jobs not only lost their AFDC payments but also lost health insurance for their families unless they were lucky enough to find a job that provided them with insurance.85 Needless to say, this made many women reluctant to find employment. Over time, the relation between welfare and Medicaid receipt has been severed, so under welfare reform there is no longer a direct relation.86 Nevertheless, Medicaid may still affect labor supply, because when family income passes a critical threshold, the family loses access to Medicaid. A second issue is the extent to which Medicaid “crowds out” private insurance. If people who receive Medicaid would otherwise have purchased health insurance,

82. Centers for Medicare and Medicaid Services, Medicaid Eligibility, http://www.cms.hhs.gov/ medicaid/eligibility/criteria.asp. 83. Health Care Financing Administration, “A Profile of Medicaid: Chartbook 2000” (U.S. De- partment of Health and Human Services, Washington, DC, 2000). 84. Health Care Financing Administration, 72. 85. John Mullahy and Barbara L. Wolfe, “Health Policies for the Non-Elderly Poor,” in Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty (Cambridge, MA: Harvard Uni- versity Press, 2001), 314–58. 86. For differing views on the importance of these changes, see Aaron Yelowitz, “The Medicaid Notch, Labor Supply, and Welfare Participation: Evidence from Eligibility Expansions,” Quarterly Jour- nal of Economics 110 (November 1995): 909–40, and John C. Ham and Lara D. Shore-Sheppard, “Did Expanding Medicaid Affect Welfare Participation?” Industrial and Labor Relations Review 58 (April 2005): 452–70.

Family Composition 191 Medicaid does not increase insurance coverage. Of course, even in this case Medicaid might increase the quality of insurance coverage. And it still makes recipients better off, because they can use the money they would have used for health insurance on something else. David Cutler and Jonathan Gruber estimate that roughly half of the increases in Medicaid coverage between 1987 and 1992 were offset by reduced purchases of insurance or by reduced provision of insurance by employers.87 Of course, ultimately we are interested in whether public provision of health insurance improves health outcomes. Although the availability of Medicaid has expanded steadily since 1984, it expanded at different times and at different rates in different states. Janet Currie and Jon Gruber have taken advantage of this variation to look at the effect of Medicaid availability on children’s use of the medical system and health.88 In essence, they ask whether, when eligibility expands in a state, medical system usage goes up and health improves. They find that the answer to both questions is yes. They estimate that making a child from a low-income family eligible for Medicaid increases the probability that she will have a check-up by 15 percent and increases the probability of her seeing a doctor in the course of a year by 8 percent. Moreover, the use of the medical system is effective in the sense that it reduces child mortality.

13. Concluding Remarks Changes in family structure are undoubtedly part of the reason that poverty has not declined as much as we might have expected over the past thirty years. Although it is not true that “two can live as cheaply as one,” it is true that two people living together can live a lot more cheaply than two living separately. The poverty thresholds reflect this reality. The reasons behind the decline in marriage are complex. It is likely that many, if not all, of the factors we have discussed played some role, and in some cases they may have interacted. “Economic” factors may have encouraged some men and women not to marry. And the increased prevalence of nonmarriage may have encouraged others not to marry. Yet it is important not to exaggerate the role of changing family structure in explaining poverty. We saw earlier that much of the variation in poverty can be explained by labor market factors. Of course, changes in the labor market may have contributed to the decline of marriage, but, if so, it may be easier to address labor market issues than to change family structure directly. We have also seen that the consequences of the change in family structure are less dire than they might seem at first glance. Single motherhood and teen motherhood are as much symptoms of poverty as they are its cause. Simple comparisons of outcomes

87. David M. Cutler and Jonathan Gruber, “Does Public Insurance Crowd Out Private Insur- ance?” Quarterly Journal of Economics 111 (May 1996): 391–430. 88. Janey Currie and Jonathan Gruber, “Health Insurance Eligibility, Utilization of Medical Care, and Child Health,” Quarterly Journal of Economics 111 (May 1996): 431–66.

192 Chapter 6 for teens who give birth and those who do not dramatically exaggerate its consequences for the mothers. Similarly, comparing outcomes for children who grow up in two-parent households with those who grow up without one parent is misleading. The best research says that the causal effects of teen motherhood on adult outcomes and of single- parent families on child outcomes are modest. On the other hand, we do know that there is considerable intergenerational correlation in economic status. The children of high-income parents are likely to have high incomes as adults. The children of the poor are at much greater risk of being poor themselves. The fact that children who grow up in poverty have worse adult outcomes suggests a need for policies that focus on poor children. Although we need more research both to ensure that our current assessments are correct and to determine the mechanisms by which they work, early education programs, mentoring and development programs, academic achievement programs, food programs, and health care have all shown some success at addressing the issues associated with child poverty. Much of what we know about what works and what does not comes out of increased attention to program evaluation. This is a tremendously positive development. If we are going to develop effective interventions, we must not rely solely on participants’ or even observers’ impressions of whether a program was successful. However, the studies reviewed in this chapter also suggest the need for caution when relying on evaluation studies. There are two principal concerns. The first is that some programs may be successful for reasons that are not easy to copy. In the last chapter, we saw that the Center for Employment Training was found to be successful in two separate evaluations but could not be successfully replicated elsewhere. To a lesser extent, this is also true of the Quantum Opportunities Program. Other sites have had some success but not nearly to the same degree as the Philadelphia site. The other concern is that the programs for which results are published and for which continued research funding is available are not selected randomly. Information about small experimental programs that work is generally far more interesting to edi- tors than is information about those that do not work. Consider how likely it would be for a funding agency to support a follow-up study at age 40 of students who participated in a preschool program for which no positive effects were found at age 6. Lawrence Schweinhart, the lead researcher on the High/Scope Perry Preschool study, writes: “We did not set out to conduct such an extraordinarily long study. Each set of findings along the way served as an impetus for the next round of data collection. The critical findings pertained to intellectual performance at the end of the preschool and grade placement at age 10.”89 Recall that a result is statistically significant if it is unlikely to occur by chance. But if we try enough programs, some outcomes that are very unlikely to occur by chance on a single occasion are quite likely to occur by chance on some occasion. A poker

89. Lawrence Schweinhart, “How the High/Scope Perry Preschool Study Grew: A Researcher’s Tale,” Research Bulletin, Phi Delta Kappa Center for Evaluation, Development and Research 32 (June 2002), available online at http://www.pdkintl.org/edres/resbul32.htm.

Family Composition 193 player knows that the odds of being dealt a flush are low, but if he plays long enough, the odds that he will be dealt a flush at some time are high. The same problem arises in research. When we randomly assign some children to participate and others not to participate in some program, the odds that the children assigned to participate in the program are favorably selected are low. But if we conduct enough studies, the odds are high that some program will appear to be effective, even over a very long period, because its participants were accidentally favorably selected. Despite these caveats, the research does give us grounds for hope. There is evidence that some programs are effective and that they can be successfully copied and established elsewhere. The case for expanding early interventions (and evaluating them) is strong, as is the case for some interventions aimed at older children. Finally, it is important to be realistic. The results of the second QOP evaluation are disappointing in part because the results of the first evaluation were so startling. Even in the second evaluation, QOP increased the fraction of participants obtaining postsecondary education from 34 percent to 42 percent. We must bear in mind that a baseball player who raised his batting average from .340 to .420 would be suspected of using steroids.

14. Further Reading Akerlof, George, and Janet Yellen. “An Analysis of Out-of-Wedlock Births in the United States.” Brookings Policy Brief 5, August 1996. Angrist, Joshua. “How Do Sex Ratios Affect Marriage and Labor Markets? Evidence from America’s Second Generation.” Quarterly Journal of Economics 117 (August 2002): 997–1038. Blau, Francine. “Trends in the Well Being of American Women, 1970–1995.” Journal of Economic Literature (March 1998): 141–47. Cancian, Maria, and Deborah Reed. “Changes in Family Structure: Implications for Poverty and Related Policy.” In Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty. Cambridge, MA: Harvard University Press, 2001, 69–96. Corcoran, Mary. “Mobility, Persistence, and the Consequences of Poverty for Chil- dren: Child and Adult Outcomes.” In Sheldon H. Danziger and Robert H. Have- man, eds., Understanding Poverty. Cambridge, MA: Harvard University Press, 2001, 162–89. Currie, Janet. “The Effect of Welfare on Child Outcomes.” In Robert Moffitt, ed., Welfare, the Family and Reproductive Behavior. Washington, DC: National Academy Press, 1998. Currie, Janet, and Jonathan Gruber. “Health Insurance Eligibility, Utilization of Med- ical Care, and Child Health.” Quarterly Journal of Economics 111 (May 1996): 431–66. Currie, Janet, and Duncan Thomas. “Does Head Start Make a Difference?” American Economic Review 85 (June 1995): 341–64. Currie, Janet, and Duncan Thomas. “School Quality and the Longer-Term Effects of Head Start.” Journal of Human Resources 35 (Fall 2000): 755–74.

194 Chapter 6 Cutler, David M., and Jonathan Gruber. “Does Public Insurance Crowd Out Private Insurance?” Quarterly Journal of Economics 111 (May 1996): 391–430. Duncan, Greg J., and Jeanne Brooks-Gunn. Consequences of Growing Up Poor. New York: Russell Sage, 1997. Edin, Kathryn. “A Few Good Men.” American Prospect 11 (January 3, 2000): 26–31. Ellwood, David, and John Crane. “Family Change among Black Americans: What Do We Know?” Journal of Economic Perspectives 4 (Fall 1990): 64–84. Garces, Eliana, Janet Currie, and Duncan Thomas. “Longer-Term Effects of Head Start.” American Economic Review 92 (September 2002): 999–1012. Geronimus, Arline T., and Sanders Korenman. “The Socioeconomic Consequences of Teen Childbearing Reconsidered.” Quarterly Journal of Economics 107 (November 1992): 1187–214. Horn, Wade F. “Wedding Bell Blues: Marriage and Welfare Reform.” Brookings Review (Summer 2001): 39–42. Hotz, V. Joseph, Susan Williams McElroy, and Seth Sanders. “The Impacts of Teenage Childbearing on the Mothers and the Consequences of Those Impacts for Gov- ernment.” In Rebecca Maynard, ed., Kids Having Kids: Economic Costs and Social Consequences of Teen Pregnancy. Washington, DC: Urban Institute, 1998. Karoly, Lynn A. “Investing in the Future: Reducing Poverty through Human Capital Investments.” In Sheldon H. Danziger and Robert H. Haveman, eds., Under- standing Poverty. Cambridge, MA: Harvard University Press, 2001, 314–56. Lang, Kevin, and Jay Zagorsky. “Does Growing Up with an Absent Parent Really Hurt?” Journal of Human Resources 36 (Spring 2000): 253–73. Mayer, Susan E. What Money Can’t Buy: Family Income and Children’s Life Chances. Cambridge, MA: Harvard University Press, 1997. McLanahan, Sara. “The Consequences of Single Motherhood.” American Prospect 18 (June 23, 1993): 48–58. McLanahan, Sara, and Gary Sandefur. Growing Up with a Single Parent. Cambridge, MA: Harvard University Press, 1994. Mullahy, John, and Barbara L. Wolfe. “Health Policies for the Non-elderly Poor.” In Sheldon H. Danziger and Robert H. Haveman, eds., Understanding Poverty. Cam- bridge, MA: Harvard University Press, 2001, 314–58. Murray, Charles. “Does Welfare Bring More Babies?” Public Interest 115 (Spring 1994): 17–30. Murray, Charles. “The Underclass Revisited.” American Enterprise Institute for Public Policy Research Papers and Studies. http://www.aei.org/ps/psmurray.htm. Sawhill, Isabel V. One Percent for the Kids. Washington, DC: Brookings Institution Press, 2003. Solon, Gary. “Intergenerational Income Mobility in the United States.” American Eco- nomic Review 82 (June 1992): 393–408. Wilson, William J. The Truly Disadvantaged. Chicago: University of Chicago Press, 1987. Zimmerman, David J. “Regression towards Mediocrity in Economic Stature.” Ameri- can Economic Review 82 (June 1992): 409–29.

Family Composition 195 15. Questions for Discussion 1. What has been the trend since 1940 in the fraction of children born to married women? 2. Describe the trend in the birth rate for • Teenagers • Married women • Single women 3. How important is each of the following factors in increasing the fraction of births to unmarried women? a. Increased birth rates among unmarried women b. Reduced birth rates among married women c. Decline in the marriage rate 4. What is the evidence for and against the following views: a. Increasing labor market opportunities for women reduced the marriage rate. b. Worsening job opportunities for low-skilled men reduced the marriage rate. c. The increasing similarity of job opportunities for men and women reduced the marriage rate. d. The availability of effective contraception and abortion reduced the marriage rate. e. Welfare generosity contributed to increasing the fraction of births to unmarried mothers. f. Welfare rules governing payments to one- and two-parent families reduced the marriage rate. 5. What is the evidence for and against each of the following views: a. Being a teenage mother leads to worse outcomes for the mother. b. Growing up in a household without both parents leads to worse outcomes as an adult. 6. Does Head Start improve outcomes for disadvantaged children? 7. In the absence of evidence from a random assignment experiment, what are the difficulties involved in evaluating Head Start? 8. Describe the principal results of the Early Head Start evaluation study.

196 Chapter 6 chapter 7

Concentrated Poverty

C   . Slums are the neighborhoods of the city that most of the nonpoor and many of the poor avoid. They suffer high crime rates. Many people suspect that the concentration of poverty is particularly harmful to children. Regardless of his own family circumstances, a poor child growing up in a middle-class neighborhood has many more opportunities to interact with successful adults and to be influenced by middle-class peers than does a poor child growing up in a high-poverty area. When we read about failing schools, we are often reading about schools in which the poverty rate is very high. This raises important issues: • Do neighborhoods matter? Is it true that otherwise similar children do worse in life if they grow up in a white slum, ghetto, or barrio? Does living in such a neighborhood make it harder for adults to escape poverty? • What sorts of policies can reduce the prevalence of neighborhoods with concentrated poverty? As we discussed earlier, Mary Jo Bane and Paul Jargowsky found that defining ghetto neighborhoods as urban census tracts with poverty rates over 40 percent corresponds well to people’s conceptions of high-poverty neighborhoods. By this definition, in 1990 some 18 percent of the urban poor and 34 percent of the black poor lived in high- poverty areas.1 Of course, the definition is imperfect, in part because our measure of poverty is imperfect. Jargowsky reports visiting a Boston “slum” that turned out to be largely populated by Boston University students. Jargowsky himself has pointed out that the term ghetto is misleading, because historically ghetto was associated not with poverty but with ethnic concentration, primarily

1. Paul A. Jargowsky, Poverty and Place: Ghettos, Barrios, and the American City (New York: Russell Sage, 1997).

197 of Jews. Today, many people assume that ghettos are primarily African American neighborhoods. In fact, many high-poverty neighborhoods are primarily Hispanic, and others are primarily white. Jargowsky has proposed the terms ghettos to refer to high-poverty areas that are at least two-thirds black, barrios for those that are at least two-thirds Hispanic, and white slums for those that are at least two-thirds white. The remainder of the high-poverty areas do not have a single group that is prevalent and are called mixed slums. We will follow Jargowsky in using these terms when we wish to identify the racial composition of a high-poverty area.

1. Life in High-Poverty Neighborhoods Television shows and movies often suggest that slums are primarily inhabited by welfare mothers and their children, who continually struggle against drug dealers and their clients. It is easy to come away with the impression that in such neighborhoods there are few people who work or who live middle-class lives. But American neighborhoods contain cross sections of people. There are two-parent and single-parent families in both middle-class and high-poverty neighborhoods, although there are more single- parent families and fewer two-parent families in the latter. By definition, the fraction of people in high-poverty neighborhoods who are poor is higher than in other neighborhoods, but there are still many people who are not poor. After all, a neighborhood is still counted as a high-poverty neighborhood if almost 60 percent of its residents are not poor. Table 7.1 shows that, compared with urban residents as a whole, residents of high- poverty neighborhoods are less likely to be employed, more likely to be out of the labor force, less likely to be employed year-round and full-time, and more likely to receive public assistance and to live in female-headed households. They are also less educated and less likely to own their homes. There is little doubt that, on average, people who live in slums are in more frequent contact with people whose behavior does not con- form to the ideals of American life. On the other hand, it should be equally clear that there are many slum residents who are working, do not receive public assistance, and have reasonable levels of education. Elijah Anderson’s ethnography of Philadelphia describes the conflict between decent behavior and street behavior that is exhibited in ghetto neighborhoods, a conflict that may be internal to individuals as well as between different elements of the community.2 There are role models of both types.

2. Do Neighborhoods Matter? In a classic psychological experiment, the subject is brought into a room with other “subjects.” These other subjects have been prepped to give the wrong answer to questions, and the room is set up so that the unprepped subject is the last to answer each question. Depending on the setting, many or most of the real subjects choose to con-

2. Elijah Anderson, Code of the Street (New York: W. W. Norton, 1999).

198 Chapter 7 Table 7.1 Characteristics of Metropolitan and Slum Areas (Percentages)

Characteristics Metropolitan Areas High-Poverty Areas

Men Employed 70.8 46.3 Men Employed Full-Time, Year-Round 51.1 23.7 Men Not in Labor Force 22.6 43.3 Women Employed 54.6 36.0 Women Not in Labor Force 41.9 56.4 Women Employed Full-Time, Year-Round 30.0 15.4 Household Income from Public Assistance 0.8 6.9 Female-Headed Households 19.4 56.6 Poor Female-Headed Households 12.2 23.9 High School Dropouts 23.0 51.7 People with Four-Year College Degrees or More 22.2 8.2 Owner-Occupied Housing Units 56.8 24.1 Vacant Housing Units 8.3 13.6

Source: Paul A. Jargowsky, Poverty and Place: Ghettos, Barrios, and the American City (New York: Russell Sage, 1997), various tables in chapter 4. form and give the wrong answer. It is clear that people are influenced by the beliefs and behaviors of people around them. In this psychological experiment, everyone else acts in the same way. But we have already seen that actual neighborhoods are diverse. People can choose to associate with the drug dealers or those in legal employment, with high school dropouts or the more educated individuals in the community. It is not obvious that neighborhoods actually influence behavior. It is even less obvious that it is better for a child in a poor family to live in a middle- class neighborhood than in a poor neighborhood. In a middle-class neighborhood, she may be more likely to suffer from social rejection because of her inability to consume at the level of her peers, while in a poor neighborhood, she may build a group of friends with good values. Put differently, is it better to be a big fish in a small pond or a small fish in a big pond? James Davis argued that many students did better going to a less competitive college because they received higher grades than they would have at a more competitive college.3 A star student in a high school with few college-bound students may receive more attention and more benefit from the school than he would as a B student in a competitive suburban high school. Of course, it is equally easy to argue the advantages of being in the competitive suburban high school. Students are exposed to more highly talented fellow students, have the opportunity to take more advanced classes, and learn about options that are not discussed in schools where few students go on to college.

3. James A. Davis, “The Campus as a Frog Pond: An Application of the Theory of Relative Dep- rivation to Career Decisions of College Men,” American Journal of Sociology 72 (July 1966): 17–31.

Concentrated Poverty 199 One natural way to address this question is to look at children who look similar but who live in different neighborhoods or attend different schools. We might find that students who attend schools with a higher graduation rate are themselves more likely to graduate than similar students at a school with a lower graduation rate. This approach has been tried. In general, we find that people who live in an area in which a higher fraction of people display a given behavior are themselves more likely to display that behavior than are otherwise similar people. Youth who live in high-crime areas are, everything else equal, more likely to commit crimes. The problem is that where a person lives may tell us a lot about that person even if where he lives does not affect how he behaves. Students who attend schools with a higher proportion of African Americans are more likely to be African American themselves. Given information on the income, parental education, and other background factors of a student but not her race, knowing the fraction of students in the school who were African American would help us predict more accurately whether the student was African American herself. However, attending a school with lots of African Amer- icans is not what made her African American. This is not to say that there are no neighborhood effects, only that they are difficult to measure. It is very difficult to untangle the effects of characteristics that members of neighborhood have in common from the effects that they have on each other. Ethnographic studies do suggest an important neighborhood effect. Susan Eaton has interviewed former participants in Boston’s METCO (Metropolitan Council for Educational Opportunity) program.4 The METCO program is an unusual desegregation program that sends mostly black students from Boston to mostly upper-middle- class suburban schools. Participation in the program is voluntary on the part of both the students and the school districts. It is evident that, on average, METCO students have better adult outcomes than students who remain in the Boston schools. They are more likely to attend four-year colleges and attend more prestigious colleges when they do. The participants credit their success to their increased familiarity with the upper- middle-class white world. Many emphasize that they became aware of aspects of the college application process that they would not have known about had they remained in Boston. The middle-class students understood the importance of choosing the right courses and preparing for the SAT exams and were familiar with many more colleges than were commonly known in the Boston neighborhoods from which the students were drawn. The former METCO students perceive their experience in the suburban schools as important to their success, and their perception is plausible. However, we have no statistical basis for determining whether they are correct in their assessment. In order to participate in METCO, students had to endure a lengthy bus ride to and from school each day and to handle the social complexities of leaving their neighborhood on a daily basis and adapting to two sets of social rules. Their parents had to be sufficiently

4. Susan E. Eaton, The Other Boston Busing Story: What’s Won and Lost across the Boundary Line (New Haven, CT: Yale University Press, 2001).

200 Chapter 7 ambitious and organized to get them registered for the program. Perhaps the METCO students would, in general, have been successful wherever they went to school. In order to assess the importance of neighborhoods, we must look for cases in which otherwise similar people find themselves in different neighborhoods. We look at two cases, an accidental experiment (the Gautreaux Program) and a deliberate experiment (Moving to Opportunity).

3. The Gautreaux Program Because of the difficulties just described, there are very few convincing studies of the effects of neighborhoods on children. The most widely cited study involves the Gau- treaux Program in Chicago, named after the lead plaintiff in a 1976 lawsuit against the federal Department of Housing and Urban Development.5 As a result of a court order, black residents of public housing and applicants for that housing could apply for a housing voucher that would allow them to live elsewhere. In practice, the agency administering the program found the housing. The agency offered the housing to those who received a voucher. It was simply a matter of luck whether they were given housing in some part of the city or in a middle-class white suburb. Although families could refuse their placement, there was a long waiting list for the vouchers, so few refused. Therefore, we can compare the effect of living in the city with the effect of living in the suburbs by comparing outcomes for Gautreaux participants who were assigned to housing in the city with those for participants who were assigned to suburban housing. James Rosenbaum and his colleagues interviewed 114 participant children when they were between 8 and 15 and again when they were between 15 and 22. The evidence suggests that the children benefited dramatically from being assigned to suburban housing. Table 7.2 shows that those placed in suburban communities were less likely to drop out of school, more likely to be in a college-track program, more likely to attend college, more likely to attend a four-year college, more likely to be employed full-time if not in college, and more likely to have a higher hourly wage and receive benefits. One concern was that the black students in the suburban communities would be socially isolated. In fact, suburban students were more likely to interact with white students than were urban Chicago students, but they were no less likely to interact with black students. There were no differences in their tendencies to view themselves as popular, fitting in, and part of the in-crowd or as misfits. On the other hand, the suburban students were more likely to report harassment by white students. However, because those who suffered from harassment were also, if anything, also more likely to report having white friends, this did not translate into a feeling of social isolation. Although the main focus of this section is on children, it is worth noting that the Gautreaux Program also shows positive effects on participants as adults. Those placed

5. This section is based on James E. Rosenbaum, “Changing the Geography of Opportunity by Expanding Residential Choice: Lessons from the Gautreaux Program,” Housing Policy Debate 6 (1995): 231–69.

Concentrated Poverty 201 Table 7.2 Outcomes for Gautreaux Program Youths Placed in the City versus the Suburbs (Percentages)

Outcomes City Suburbs

Dropped Out of School 20 5* In a College-Track Program 24 40** Attend College 21 54*** Attend Four-Year College 4 27*** Employed Full-Time (if not in college) 41 75*** Paid under $3.50 per Hour 43 9*** Paid over $6.50 per Hour 5 21*** Receive Job Benefits 23 55***

Source: James E. Rosenbaum, “Changing the Geography of Opportunity by Expanding Residential Choice: Lessons from the Gautreaux Program,” Housing Policy Debate 6 (1995): 231–69, table 5. *Significant at the .1 level. **Significant at the .05 level. ***Significant at the .025 level.

in the suburbs were more likely to be employed. In-depth interviews revealed that Gautreaux participants found jobs more readily available, felt safer being away from their children, and were motivated by the fact that others around them were working, all factors that we have discussed earlier.

4. Moving to Opportunity In response to the apparent success of the Gautreaux Program, the federal Department of Housing and Urban Development (HUD) decided to initiate a trial program in Bal- timore, Boston, Chicago, Los Angeles, and New York called Moving to Opportunity (MTO). Families with children living in public housing projects could apply to participate in the program. They were randomly assigned to receive a section 8 rental certificate that entitled them to a government subsidy in private rental housing. Some of those given section 8 certificates were randomly assigned to a group that could use them only in low-poverty (below 10 percent) areas, while others received a traditional section 8 certificate usable anywhere. Those who did not receive a certificate were in the control group. In addition, those receiving the experimental restricted vouchers were given counseling and assistance in finding housing. The other two groups received whatever support was typically available through section 8 or the projects. Participants were overwhelmingly black and Hispanic, and almost all heads of households were women.6

6. Office of Policy Development and Research, “Expanding Housing Choices for HUD-Assisted Families: First Biennial Report to Congress: Moving to Opportunity Fair Housing Demonstration” (U.S. Department of Housing and Urban Development, Washington, DC, April 1996).

202 Chapter 7 Participants were interviewed in 2002, five to eight years after their initial placement.7 For adults, relative to the control group, there were improvements on measures of economic self-sufficiency and physical and mental health for both the standard section 8 and the MTO participants. However, only the overall improvement in mental health (as well as some individual components) and reduction in obesity were statistically significant, and then only for the MTO participants. For girls, there were favorable effects on physical and mental health, behavior, and education. The effects in all areas except physical health were significant for MTO participants, and the effects on mental health and behavior were significant for section 8 participants. In particular, moving out of projects reduced girls’ drug and alcohol abuse. For boys the results were very disappointing. The outcomes for boys were worse in all cases except for mental health among section 8 participants, and this difference fell well short of statistical significance. The boys in the MTO and section 8 groups did significantly worse in terms of physical health and education. A review of administrative data revealed similar results with respect to criminal activities.8 Girls in the MTO group had significantly fewer arrests for both property and violent crimes, and those in the section 8 group had significantly fewer arrests for violent crimes. The results for other measures of criminal behavior were generally favorable for girls but not statistically significant. In contrast, boys in the MTO group were significantly more likely to have been arrested for property crimes, and the effects on other measures of criminal behavior were generally unfavorable but not statistically significant. On all measures, girls in the section 8 and MTO groups benefited more than boys, often significantly so. To some extent, the increased arrest rate for boys may reflect the fact that people who commit property crimes are more likely to be arrested if they commit the crime in a low-crime neighborhood. This is not easily reconciled with the improvement for girls. In addition, the same deterioration in behavior is found in the results of the survey discussed in the previous paragraph, which suggests that the difference does not simply reflect better policing in suburban neighborhoods.

5. Public Housing In 1981, Chicago’s then-mayor Jane Byrne moved into the Cabrini-Green housing project on a temporary basis. The housing project, which served 13,500 people in seventy-eight buildings, suffered from a high crime rate, gang activity, and violence.9 The mayor’s time there provoked a temporary clean-up of the project but left an en- during negative image of high-rise housing projects.

7. The following is drawn from Jeffrey R. Kling, Jeffrey B. Liebman, and Lawrence F. Katz, “Ex- perimental Analysis of Neighborhood Effects,” Econometrica (forthcoming). 8. Jeffrey R. Kling, Jens Ludwig, and Lawrence F. Katz, “Neighborhood Effects on Crime for Fe- male and Male Youth: Evidence from a Randomized Housing Voucher Experiment” (Working Paper 10777, National Bureau of Economic Research, Cambridge, MA, 2004). 9. Nathaniel Sheppard, “Chicago Mayor’s Proposed Move to Housing Project Touches Off Cleanup Blitz,” New York Times, March 30, 1981, Late City Final Edition, A14.

Concentrated Poverty 203 Journalistic accounts of life in housing projects reinforced this view. Alex Kot- lowitz’s There Are No Children Here presents a riveting and discouraging picture of life in the Robert Taylor Homes, another Chicago project.10 As a consequence, there has been a sharp shift in government policy, away from housing projects in general and high-rise projects in particular and toward more diffuse approaches, such as section 8 rental vouchers. The Chicago Housing Authority (CHA) is in the process of tearing down its high-rise developments. There is no doubt that Cabrini-Green, the Robert Taylor Homes, and many other projects were dangerous. Outcomes for families that live in projects are bad. But as we have seen with regard to similar issues, it is difficult to determine whether these outcomes are bad because the families live in the projects or because they have characteristics that cause them to live in the projects. Moreover, these housing projects became notorious in part because they were atypical. Many housing projects look very different from these massive high-rise projects. Brian Jacob has used the process by which the CHA is tearing down the projects to examine their effect on children’s educational outcomes.11 He argues that with some exceptions, which he excludes from his study, the order in which buildings within a project are torn down is random. He therefore compares outcomes for tenants who were displaced by a building’s being torn down with tenants in those same projects who were not displaced. Displaced tenants received a section 8 voucher, so his study effectively measures the difference between being in a project and receiving a voucher. There was no effect on the academic performance of younger students. Older students who were displaced were more likely to drop out of school. Jacobs attributes this to the direct effect of displacement rather than to a difference between voucher-subsidized housing and public housing. The absence of any significant difference probably reflects the fact that voucher recipients do not move to neighborhoods that are substantially different from the ones from which they were displaced. Instead, they tend to remain in the same neighborhood either to retain social contacts or because that is where they are most likely to be able to locate a landlord willing to accept a tenant with a section 8 voucher. Phillip Oreopoulos has looked at variation in outcomes across different types of public housing.12 He argues that assignment to individual projects in Toronto was random but that projects differed substantially in size and in location. Yet he was unable to find any difference in adult outcomes among projects. Adults who as children lived in projects in relatively wealthy neighborhoods did no better than those who had grown up in projects in poor neighborhoods (which were more commonly high- rise projects).

10. Alex Kotlowitz, There Are No Children Here (New York: Doubleday, 1991). 11. Brian A. Jacob, “Public Housing, Housing Vouchers, and Student Achievement: Evidence from Public Housing Demolitions in Chicago,” American Economic Review 94 (March 2004): 233–58. 12. Phillip Oreopoulos, “The Long-Run Consequences of Living in a Poor Neighborhood,” Quarterly Journal of Economics 118 (November 2003): 1533–75.

204 Chapter 7 Thus, the recent evidence does not suggest a large effect of neighborhood on outcomes except for outcomes very closely tied to the neighborhood characteristics, such as exposure to violence and health.

6. Gangs In this section we will discuss the role of gangs in the economic life of high-poverty neighborhoods. This is not to deny the social importance of gangs. Clearly gangs exist for mutual protection and socialization. Gangs can provide members with a sense of membership and protection. Elijah Anderson eloquently describes the importance to a young man named Tyree of getting in with bols (the young men who ran his neighborhood).13 It is clear that Tyree’s primary concerns are not economic, but involve safety and status. Still, gangs have become an important part of the economic life of high-poverty neighborhoods, where they are viewed as a significant source of employment. Gangs have become major corporations involved in the marketing of drugs. Relatively little is known about pay, employment levels, and profitability of street gangs because they do not respond to the types of government surveys from which we obtain most of our data. However, Steven Levitt and Sudhir Venkatesh have analyzed the accounts kept by a former local gang leader to cast light on this business.14 Levitt and Venkatesh liken the gang they studied to a franchise. The local gang (franchise) makes payments to the larger organization (a regional or national gang) in return for various services, including protection if its territory is attacked. Like a franchise owner, the (local) gang leader keeps any profits the gang makes after making payments to his workers and to the higher levels of the gang. The authors write: The next tier in the organization is a group of “gang leaders” with specific terri- torial responsibility for one or more localized gangs. In the organization we study, there are roughly 100 of such gang leaders. Reporting in to each gang leader are three “officers.” The “enforcer” is responsible for ensuring the safety of group members, the “treasurer” is responsible for the liquid assets of the group, and the “run- ner” performs the risky task of transporting large quantities of drugs and money to and from the supplier. Reporting to the enforcer are the “foot-soldiers” who serve as street-level drug sellers and from whose ranks future officers and leaders arise. Foot- soldiers are typically 16–22 years of age, although potentially much older. On the periphery of the gang is a group of “rank and file” spanning all ages (the age range in the group we study is 14 to 40), who have little formal responsibility for drug selling. Rank and file, unlike foot-soldiers and higher gang members, pay dues to the gang, in return receiving protection, status, and a reliable supply of drugs for those who deal independently. The structure of the overall organization is similar

13. Anderson, 80–87. 14. Steven D. Levitt and Sudhir Alladi Venkatesh, “An Economic Analysis of a Drug-Selling Gang’s Finances,” Quarterly Journal of Economics 115 (August 2000): 755–89.

Concentrated Poverty 205 to that of a franchised company. Gang leaders pay a fee to the franchisers (central gang leadership), but are the residual claimants on the profits accruing to their franchise. In return for those tribute payments, local gangs are provided with protection (both on their turf and in prison), access to reliable sources of wholesale drugs, and the possibility of some day rising up the organization into the upper echelon. The individual gang units, like separate franchise owners, have relatively little interaction with one another.

What is striking about Levitt and Venkatesh’s analysis is the low rates of pay received by foot-soldiers. In periods when there is no gang war, the wage received by the street-level dealers is less than minimum wage, even if we make a reasonable allowance for some skimming by dealers who steal and resell some of the drugs or who overcharge clients and keep the extra earnings. Even allowing for some additional benefits (for example, the gang pays up to $5,000 for funeral costs and compensation to the family when a foot-soldier dies), employment in legal sector jobs offers compensation at a level that is competitive with dealing. In fact, many of the foot-soldiers also hold legal jobs, and, consistent with their low earnings, many live at home. Working as a mercenary (a non–gang member who fights gang wars) is more lu- crative. Mercenaries earned about $2,000 month in 1995 dollars, and the gang leader’s profits were generally between $4,200 and $10,900 per month, for an annual wage of $50,000 to $130,000. The three officers below him earned about $1,000 per month, so officers and the leader earned more than the likely alternative, the minimum wage. This is well above what most leaders, given their education and work experience, could hope to earn in the legitimate sector. The study by Levitt and Venkatesh suggests two important points. First, gang employment seems attractive only because of the potential for very high earnings for those who rise in the gang hierarchy. Second, expected earnings for gang members are sufficiently low that strategies to create jobs and to create better legal opportunities for potential gang members, if successful, should attract potential gang members to the legal sector.

7. Community Development One way to attack neighborhood poverty is to move poor people out of high-poverty areas. Another is to revitalize the neighborhood to create more jobs. To this effect, the federal and state governments have undertaken a variety of initiatives to improve housing and employment opportunities in high-poverty areas.

7.1. The Community Reinvestment Act of 1977. In 1977, Congress passed the Com- munity Reinvestment Act (CRA). The CRA was passed in response to concerns about redlining, a practice under which financial institutions refused to give loans for businesses or homes located in certain neighborhoods. Without access to credit, businesses found it difficult to open or expand. Housing fell in value, because potential buyers were unable to obtain mortgage loans.

206 Chapter 7 The CRA requires federally chartered financial institutions to invest in the communities in which they operate. If institutions receive deposits from a neighborhood, they must also lend in that neighborhood.

7.2. The Empowerment Zones and Enterprise Communities Act of 1993. The Empowerment Zones and Enterprise Communities Act, passed in 1993, uses a variety of approaches to foster economic growth and social development in both urban and rural areas. It provides a combination of grants and tax relief to encourage both business investment and community projects. In 1994, the federal government designated three rural and six urban empowerment zones and thirty rural and sixty-five urban enterprise communities. For use over a ten-year period, each rural zone was awarded $40 million, each urban zone $100 million, and each enterprise community $2.95 million. These grants can be used for a wide variety of community infrastructure, from daycare to education to sewers. In 1999, the federal government added a further five rural and fifteen urban empowerment zones, but the grants were smaller and shorter-term.15 In addition, businesses located in empowerment zones can claim a tax credit of up to $3,000 per year for each resident of the zone they employ. They may also be entitled to additional tax deductions and to use tax-free bonds to finance their borrowing, lowering the interest that they have to pay on loans.

7.3. State Enterprise Zones. Individual states initiated enterprise zones well before the passage of the federal legislation. A large number of these programs were created in the 1980s. Currently about two-thirds of the states have some version of an enterprise zone program, and over three thousand enterprise zones currently exist.16 It is difficult to summarize the effects of the zones, because their characteristics vary significantly from program to program. Some states provide business tax breaks to businesses that locate in a zone. Even among those offering business tax breaks, there is variation in the extent and duration of the tax breaks. Others zones provide investment in infrastructure. Some are targeted at particular industries or small geographic areas where there are particular development issues, such as brownfields (areas with significant pollution problems). How much effect do the enterprise zones have on the availability of jobs? This question does not have an easy answer. Certainly some of the businesses that locate in enterprise zones would have located there anyway. In surveys, about one-third of businesses say that they would not have located in the enterprise zone without the incentives for locating there. Even if we accept that businesses answer the question honestly,

15. Administration for Children and Families, U.S. Department of Health and Human Services, “The Empowerment Zone and Enterprise Community Program,” http://www.acf.hhs.gov/programs/ ocs/ez-ec/info.htm. 16. Alan H. Peters and Peter S. Fisher, State Enterprise Zone Programs: Have They Worked? (Kala- mazoo, MI: W. E. Upjohn Institute for Employment Research, 2002), and Timothy Bartik on “Mid- morning,” Minnesota Public Radio, January 24, 2003.

Concentrated Poverty 207 this does not mean that the enterprise zones dramatically increase the availability of jobs to zone residents. Only about one-fifth of jobs in a zone are held by zone residents, and only about one-tenth of zone residents work in the zone.17 Moreover, we must be concerned that many of the jobs created in the zone were merely diverted from nearby locations and thus did not increase the availability of jobs to zone residents.

7.4. Community Development Corporations. There are a wide variety of neighborhood-based groups that seek to encourage community development. Most commonly these are referred to as community development corporations (CDCs), although sometimes this term is reserved for formal corporate groups that invest in affordable housing and provide support for private businesses. Some proponents of community development felt that CDCs that focused on affordable housing and business investment were too narrow. They urged CDCs to take on a broader mandate of helping to develop social services and infrastructure. These broader CDCs are referred to as comprehensive community initiatives. The federal government also supports community development financial institutions that provide credit to businesses in high-poverty areas. The key element of all these programs is that they are community based and, in principle, tailored to the needs of the community and the nature of the leadership. Some community development programs are widely viewed as successes, though others have failed. Because such programs are by their nature idiosyncratic, there is no compelling research on whether they are, in general, successful and what distinguishes successful from unsuccessful programs.

7.5. Programs Oriented Toward Youths. In the past two decades there have been a number of intensive neighborhood-based initiatives aimed at developing a comprehensive set of services for in-school and out-of-school youths in poor neighborhoods. The goal of these initiatives has been to transform neighborhoods by keeping large numbers of adolescents from dropping out of school, encouraging postsecondary education, and helping out-of-school youths find jobs. These programs, with names like Youth Op- portunities Unlimited, Youth Fair Chance, and the latest program, Youth Opportunities Grants (YOG), have tended to be short-lived, perhaps because they ultimately support only a small number of communities and therefore lose political support. YOG supported 14- to 21-year-old youths in thirty-six poor urban and rural areas. The funding provided was supposed to be over $1 billion over four years beginning in 2000, but was subsequently cut. Funding was further reduced in the later years, and it currently appears unlikely that the program will be repeated. Unfortunately, there is no thorough evaluation of this program or any of the others.

8. Concluding Remarks There is no question that being poor and living in a high-poverty neighborhood is different from being poor and living in a middle-class neighborhood. A high-poverty

17. Bartik.

208 Chapter 7 neighborhood is more dangerous, leads to more stress, and generally entails more health risks from pollution and other sources. There is also no doubt that we are influenced by our peers and our environment. Children who grow up in neighborhoods where most people speak English are likely to speak English even if their parents do not. Fashion and fads could not exist without peer influence. However, the experimental and quasi-experimental evidence suggests that neighborhood effects are smaller than we might have anticipated. This may reflect the fact that all neighborhoods are diverse and that, especially after reaching adolescence, children can travel to other neighborhoods. Thus, to some extent, children and adults choose their peers from a diverse pool. But we should also be aware of the limitations of even the experiments that have been undertaken. The studies of both the Gautreaux Program and MTO rely on samples of residents of housing projects who would like to leave those housing projects. Brian Jacob’s work on the effects of mobility on families displaced from housing projects tells us something about what might happen if we reduced the availability of projects. We do not know the effect of increasing neighborhood mobility on people who live in poor neighborhoods but not in housing projects. And, as is usually the case with experiments, we do not know the effect of a large-scale change in policy on eligibility and demand for assistance. We would expect that many people who would not be willing to live in a project would be willing to accept a section 8 voucher. If the number of families served did not change, a switch from projects to vouchers would undoubtedly increase competition for that assistance. The depopulation of some central cities is both a challenge and an opportunity. The loss of high-skilled labor and the high crime rates in some central cities make them unattractive for development. But the existence of large plots of unused land allows for planned redevelopment. Often central cities find themselves in a vicious circle. The loss of industry and the middle class leads to a declining tax base, which in turn makes it difficult to provide the kinds of services and infrastructure that attract jobs and the middle class. Tax incentives may attract jobs, but they worsen a city’s fiscal problem, which, if it undercuts the city’s ability to provide infrastructure, can also undermine its ability to attract jobs. But there are hopeful signs. After at least twenty years in which poverty became more concentrated, the 1990s saw a reversal of the trend.18 Concentrated poverty was less common in 2000 in both rural and urban areas (although not in suburban areas) than it was in 1990. And many urban neighborhoods have been revitalized.

9. Further Reading Ferguson, Ronald F., and William T. Dickens. Urban Problems and Community Devel- opment. Washington, DC: Brookings Institution Press, 1999.

18. Jargowsky, Poverty and Place, especially chapter 2, and Paul A. Jargowsky, Stunning Progress, Hidden Problems: The Dramatic Decline of Concentrated Poverty in the 1990s (Washington, DC: Brook- ings Institution, 2003).

Concentrated Poverty 209 Jacob, Brian A. “Public Housing, Housing Vouchers, and Student Achievement: Evi- dence from Public Housing Demolitions in Chicago.” American Economic Review 94 (March 2004): 233–58. Jargowsky, Paul A. Poverty and Place: Ghettos, Barrios, and the American City. New York: Russell Sage, 1997. Jencks, Christopher, and Susan Mayer. “The Social Consequences of Growing Up in a Poor Neighborhood.” In Laurence E. Lynn and Michael G. H. McGeary, eds., Inner-City Poverty in the United States. Washington, DC: National Academy Press, 1990. Kotlowitz, Alex. There Are No Children Here. New York: Doubleday, 1991. Levitt, Steven D., and Sudhir Alladi Venkatesh. “An Economic Analysis of a Drug- Selling Gang’s Finances.” Quarterly Journal of Economics 115 (August 2000): 755–89. Oreopoulos, Phillip. “The Long-Run Consequences of Living in a Poor Neighbor- hood.” Quarterly Journal of Economics 118 (November 2003): 1533–75. Peters, Alan H., and Peter S. Fisher. State Enterprise Zone Programs: Have They Worked? Kalamazoo, MI: W. E. Upjohn Institute for Employment Research, 2002. Rosenbaum, James E. “Changing the Geography of Opportunity by Expanding Resi- dential Choice: Lessons from the Gautreaux Program.” Housing Policy Debate 6 (1995): 231–69. Wilson, William Julius. The Truly Disadvantaged. Chicago: Chicago University Press, 1987, chapter 2. Zax, Jeffrey S., and John F. Kain. “Moving to the Suburbs: Do Relocating Companies Leave Their Black Employees Behind?” Journal of Labor Economics 14 (July 1996): 472–504.

10. Questions for Discussion 1. Poor people in high-poverty areas are more likely to commit crimes and less likely to graduate high school or work than people with the same income in low-poverty areas. Why? 2. People often exhibit behaviors similar to those of the people with whom they interact a great deal. Why is this not convincing evidence that they are influenced by the people with whom they interact? 3. What have we learned from the Gautreaux and Moving to Opportunity Programs? 4. Does growing up in a housing project worsen outcomes for poor children? 5. Discuss the role of gangs as a source of employment.

210 Chapter 7 chapter 8

Education and Education Reform

E        to social mobility. By getting a good education, disadvantaged children can make up for their disadvantages. Yet one of the ways in which high-poverty areas typically differ from other parts of the metropolitan area is that they have lower-quality schools. Children in inner-city poverty neighborhoods often attend schools that are markedly inferior to those attended by children in well-to-do suburban communities. It is widely believed that many of the inner-city schools are broken. Critics claim that students are rewarded for “seat time,” that children who simply stay in school are rewarded with meaningless diplomas even if they cannot read or add and subtract. If one of the reasons that children from high-poverty neighborhoods do poorly in life is that they attend low-quality schools, it is natural to ask whether there are ways to reform schools so that poor children do better. There is general agreement that although money can be helpful, increasing spending on schools is not sufficient to ensure reform.1 Many struggling urban school systems spend more per pupil than successful suburban school systems. The suburban school systems are able to create environments in which it is less costly to educate children, perhaps because much of the social education of the children has been done at home. As a consequence, there have been a number of movements to reform the public schools in the United States. Many of these concern curriculum or school management issues that are designed to substitute for the advantages of suburban school systems. Some experts believe that highly structured school reform programs such as Success for All can be used at the level of the individual school to achieve reform.

1. Eric A. Hanushek, Making Schools Work: Improving Performance and Controlling Costs (Wash- ington, DC: Brookings Institution, 1994).

211 Discussion of curricular, administrative, or structural reform of schools is beyond the scope of this chapter. Instead, we will focus on reforms that are being considered or implemented at the district, state, or federal level. In particular, we look at • Mandatory testing • Vouchers • School choice • Charter schools However, before we do so, we will discuss the relation between earnings and education. Is it really true that getting an education is the path to upward mobility?

1. Education and Earnings People with college educations, on average, earn more than those with only high school diplomas, who, in turn, earn more than high school dropouts. By this point in the book, I hope you are saying, “But that does not mean that education increases earnings. Maybe the type of person who gets a lot of education also tends to be the type of person who would earn a lot anyway.” Economists have grappled with this question. It is not easy to determine whether education increases earnings, but there have been a large number of attempts to address this question, and almost all have found that at least part of the relation between education and earnings is causal, that is, that getting more education increases earnings. Rather than attempting a full review of this vast literature, I will take it for granted that the relation is at least partially causal and discuss why education might increase wages. There are three principal explanations: human capital, sorting, and credentialism. The human capital story is simple. School teaches students valuable skills: mathematics, reading, and how to get along with other people. Employers observe these skills and are willing to pay more to workers who have them.2 In its extreme form, the sorting model says that education does not provide skills, but more able workers obtain more education to signal that they are more able. Em- ployers cannot determine directly whether a worker has the skills they want, but knowing that more able workers obtain more education, they use education as a screen, hiring more educated workers for jobs requiring more ability. For example, firms are willing to pay a premium to hire smarter workers. If smarter workers obtain more education, the firms will pay a premium to more educated workers if they cannot observe their “smarts” directly. All workers will recognize that more educated workers are paid more, but those who find education more costly will obtain less education than those who find it relatively low cost. If smarter people find education less painful (and thus less costly), they will obtain more education. In equilibrium, more able workers get more education because it is rewarding to them

2. The classic book in this area is Gary S. Becker, Human Capital, 2nd ed. (Chicago: Chicago University Press, 1975).

212 Chapter 8 and less costly. Firms pay more for more educated workers because more educated workers are more able.3 Clearly both mechanisms are at work. Firms can use interviews, recommendations, and other information to learn about worker productivity without relying solely on education, but it is likely that they also infer information from a worker’s educational attainment. Firms care about some, but not all, of the skills learned in school. Econo- mists have developed hybrid models that account for both roles of education.4 Education also provides credentials. Credentialism is superficially similar to sorting in that it maintains that employers use education to sort workers. However, the mechanism is somewhat different from that of sorting. Credentialism maintains that the actual productivity differences associated with education are often small. However, employers are often forced to pay a wage that is independent of a worker’s quality because the wage is set by bargaining or government regulation (especially in government jobs) or is advertised. In such cases, the firm will hire the best applicant. But figuring out who the best applicant is can be difficult. So the firm uses a shorthand, education, to narrow the pool of candidates from which it chooses.5

2. Testing One of the greatest changes that has occurred in U.S. education has been a movement to create standardized tests at the state level. Many states use the tests as a partial basis for grade promotion or graduation. Students who do not do sufficiently well on the required test are not eligible for promotion or graduation. Tests that have consequences for students, schools, or school districts are referred to as high-stakes tests. This section examines the effects of high-stakes testing on students. Later in this chapter, we will examine the effects on schools or school districts of making the tests high-stakes for them. Supporters of high-stakes testing for students usually make two arguments: • Standards for students are very low in many schools. Students graduate on the basis of “seat time” rather than accomplishment. A student who shows up at school and behaves will graduate regardless of whether he can read or add and subtract. By requiring a minimum standard of achievement, testing ensures that students will put in not only seat time, but also effort. • Potential employers have little information about the performance of high school graduates. They cannot rely on the fact that a job applicant has graduated from high school as proof that she has acquired basic skills. By making

3. Michael Spence, “Job Market Signaling,” Quarterly Journal of Economics 87 (1973): 355–374. 4. Kevin Lang, “Does the Human Capital / Sorting Debate Matter for Development Policy?” American Economics Review 84 (March 1994), and Andrew Weiss, “Human Capital vs. Signalling Explanations of Wages,” Journal of Economic Perspectives 9 (Fall 1995): 133–54. 5. Credentialism is usually presented as irrational, as in Ivar E. Berg, Education and Jobs: The Great Training Robbery (New York: Praeger, 1970). The discussion here develops it in the context in which it is profit-maximizing for the firm.

Education and Education Reform 213 students pass a test in order to graduate, we ensure that potential employers will know that high school graduates have at least mastered some set of skills measured by the test. It is important to recognize that, while somewhat different, both of these arguments are based on the idea that it is important to raise standards, at least at some schools. The theory we will now discuss focuses on the idea that testing raises standards.

2.1. Theory. If employers can quickly figure out how good a worker is, the results of a test will provide them with no additional information, and there is no reason for them to pay any attention to the test. Therefore, high-stakes testing makes sense only if we are in a world described by either the sorting model or a hybrid model in which employers use educational attainment to infer information about worker productivity. We will begin by distinguishing two extreme cases and then discuss what happens if we combine them. In the first case, workers do not really care about the skills tested on the exam. They want employees who can answer algebra questions because these students will generally be good at arithmetic and other reasoning skills, but they do not expect their employees to use algebra in their jobs. We call this the pure sorting model. In the second case, employers care about the content of the test. Workers who have acquired the skills assessed by the test are more productive than those who have not acquired the skills, regardless of any general reasoning or mathematical skills. We call this the human capital case, although, because employers use the test to learn about productivity, it also contains an element of sorting.6

2.2. Pure Sorting. In our first, extreme, case, education does not increase human capital. Instead, employers use information from the test to screen workers for characteristics that will make them more productive, not to ensure that the workers have the actual tested skills. For example, in Iran at least part of the national exam consists of fill-in-the-blank questions where the only correct answer is the word used in the textbook. The ability to memorize the textbook sufficiently to use the exact word found in the text may be indicative of other cognitive and behavioral traits that are useful to employers, but it is unlikely that memorizing the textbook is directly useful on the job. Thus, the actual skills tested are of no productive use, but the types of students who can graduate high school with less effort are also the types who will do well in the job market. In reality, the relation is not perfect. Many of us know high school dropouts who went on to be very successful in the job market. Again, however, just to keep the story simple, we will assume that the relation is perfect. Firms will pay workers who pass the test more than they will pay workers who do not because they expect workers who pass the test to be better workers. If, on average, high school dropouts who apply for jobs are worth $7.00 per hour, in the absence of

6. These models draw on Julian R. Betts and Robert M. Costrell, “Incentives and Equity under Standards-Based Reform,” in Diane Ravitch, ed., Brookings Papers on Education Policy: 2001 (Wash- ington, DC: Brookings Institution, 2001).

214 Chapter 8 other information, employers will be willing to pay them that amount. On average, high school graduate applicants might be worth $10.00 per hour and will be paid that even if they are somewhat better or worse than the average applicant. Of course, over time, employers will learn about which high school graduates are particularly good workers and which are not, and will adjust these wages, but just to keep the story simple, let us assume that this learning will take so long that most starting workers care only about what they will earn initially. Both the average productivity of students who pass the test and the average productivity of workers who do not pass the test will depend on the ability cutoff at which workers pass the test. The higher this ability cutoff, the higher the average ability of both students who pass the test and those who do not. In case this is not clear, consider an example with six workers who produce 1, 2, 3, 4, 5, and 6, respectively. If only the person who produces 6 passes the test, the average of those who pass the test is 6 and the average of those who do not is 3. If the three most productive people pass the test, their average productivity is 5, and the average productivity of the remainder is 2. If everyone except the person producing 1 passes the test, the average productivity of those passing the test is 4 and the average productivity of the one person who does not is 1. In this example, the wage differential between those who pass the test and those who do not is always 3. That is the result of the specific example. In general, increasing the proportion of people who graduate high school (pass the test) can either increase or decrease the differential, but to keep things simple, we will ignore this effect in the discussion that follows. What happens if we raise the standard for high school graduation by adding a high- stakes test? Two groups have no reason to change their behavior. Those who would have dropped out anyway will find it even more difficult to pass the test and will have no reason to work harder in order to graduate. Those who already meet the new standard will also have no reason to work harder. However, those who met the old standard but not the new standard must decide whether to increase their effort and meet the new standard or drop out. Those close to meeting the new standard will increase their effort. Those who had difficulty meeting the old standard will drop out. To summarize, there are four groups who will respond differently to the increased standard. Low-skilled students who would have dropped out anyway will not change their behavior. Low- to moderate-skilled students will respond by putting in less effort and dropping out. High- to moderate-skilled students will respond by increasing their effort in order to graduate, and high-skilled students who would have met the standard anyway will not change their behavior. Perhaps surprisingly, the groups that will most clearly be helped by the increased standards are the two that did not change their behavior. Because they would have dropped out anyway, these low-skilled students will not change their effort. However, because the set of dropouts now includes the low- to moderate-skilled students, employers will be more willing to take a chance on dropouts, and their wages will increase. Wages for high school graduates will also increase, because they will now be a more select group. For those who do not have to work harder to meet the new standard, this will clearly be a benefit.

Education and Education Reform 215 Table 8.1 Testing in a Pure Sorting Model: An Example

Original Standard Higher Standard Required Actual Net Required Actual Net Productivity Effort Effort Wage Wage Effort Effort Wage Wage

1 3.2 0.0 1 1.0 6.4 0.0 2 2.0 2 2.4 2.4 4 1.6 4.8 0.0 2 2.0 3 1.8 1.8 4 2.2 3.6 0.0 2 2.0 4 1.2 1.2 4 2.8 2.4 2.4 5 2.6 5 0.6 0.6 4 3.4 1.2 1.2 5 3.8 6 0.0 0.0 4 4.0 0.0 0.0 5 5.0

Note: See section 2.2 of the text for an explanation of the numbers used.

The effect on the other two groups will be more complex. For some of the individuals who work harder to meet the new standard, the higher wage for high school graduates will offset the added cost of meeting the higher standards. For others it will not. Perhaps surprisingly, some of the new dropouts will be helped and some hurt. Those who were almost indifferent between meeting the old standard and dropping out will be better off, because the opportunities available for dropouts have improved. Those who are almost indifferent between meeting the new standard and dropping out will probably be worse off. Table 8.1 continues our example. The left-hand panel shows the equilibrium with the “original” standard. I have chosen the effort required of the students at each level of productivity so that all except those at the lowest level choose to meet the standard. As discussed earlier, the wage for those meeting the standard in this case will be 4 and the wage for those not meeting it will be 1. Because it costs those at level 1 an effort of 3.2 to meet the standard, they will choose not to. For all other types, the cost will be less than 3, and they will choose to meet the standard. The fourth column in the table shows the net wage after subtracting the cost of meeting the original standard. The right-hand panel of table 8.1 shows the equilibrium with a higher standard. Now, workers at productivity level 1, 2, or 3 will choose not to meet the standard, while those at productivity level 4, 5, or 6 will choose to meet it. Wages for both those meeting the standard and those not meeting the standard will be higher under the higher standard. However, as discussed more generally earlier, there will be winners and losers when we take into account changes in the decision to meet the standard and changes in effort. Comparing the last column in each of the two panels, we see that the person who would not meet either standard will be better off, because he will not change his level of effort but will benefit from being grouped with better workers. The same will be true for the person who meets the new standard effortlessly. Among those who choose not to meet the higher standard but would have chosen to meet the lower standard, the less productive worker will be better off and the more productive worker worse off.

216 Chapter 8 The opposite will be true among the workers who choose to increase their effort to meet the new standard.

2.3. Human Capital. Now, in contrast with the previous subsection, we will assume that firms care about the content of what is taught in school. We call this the human capital model because education makes workers more productive. It is, in fact, a hybrid model, because a test is required to determine whether the worker has acquired the human capital. The first thing to note about this case is that (ignoring for the moment students going on to college), students will either put in the effort they need to pass the test or not put in any effort at all. There is no benefit to learning some of the material covered in the test even if employers value it, because the employers will know only if the student has passed the test.7 Each type of student has to decide whether it is worth making the effort to graduate high school in order to earn the additional money. Some will find graduating high school very easy and will put in the small amount of effort required. Others will find high school so difficult that they could not pass with any reasonable level of effort and will drop out. Most will weigh the costs and benefits. Some will decide that they would have to put in too much effort and will drop out, while others will decide that the effort is worthwhile. Employers know that students who pass the test put in exactly the effort needed to pass and are worth whatever firms are willing to pay for workers with that level of knowledge. Workers who have not passed the test have put in no effort and will be paid at the level of workers lacking those skills. Suppose now that the standards for the diploma are increased. Because the skills measured by the test have increased, firms will be willing to pay more for workers with a high school diploma, and the benefit will increase. However, it is now also more costly to acquire the skills needed to graduate. The ability level at which individuals decide to acquire the necessary skills to graduate may increase or decrease. Put differently, the graduation rate can rise or fall. If the additional cost of acquiring the new skills is low and the benefit in terms of market productivity high, more students will put in the effort to get a high school diploma. In this scenario, all students will be better off or at least no worse off with the higher standards. Those who would have graduated anyway will have increased earnings that more than outweigh their additional costs. Those who would not graduate under either standard are unaffected by the change.8 Finally, those who now choose to graduate but would not under the lower standard will be better off, because they will now be choosing

7. If passing the test is to some extent a random event, or if there is some information about performance beyond merely whether the student passed or failed, the argument becomes more complicated but is similar. 8. Because the higher standards increase the supply of skilled relative to unskilled labor, this may change their relative prices. We ignore this effect as small, at least in the short run, because the stock of unskilled and skilled labor is large relative to the flow.

Education and Education Reform 217 graduation when they could have dropped out and had the same earnings as under the old standard. On the other hand, if the additional market productivity from the increased standard is low and the added cost of acquiring the skills high, the situation is more complicated. Again, those who would not graduate under either standard are unaffected. Those who would have graduated under the old standard but will not under the new standard are worse off. In addition, at least some of those who continue to graduate under the new standard will be worse off. They will benefit little from the increased standard but must pay a high cost to meet it.

2.4. More General Models. It is possible to make apparently minor modifications to the models that change some of the conclusions about the effect of raising graduation requirements by requiring high-stakes tests. However, these simple models capture the basis for much of the debate around high-stakes testing. In general, there are likely to be winners and losers. Some of the debate around testing concerns how large each of these groups is and how much it wins or loses. Some of it reflects differences in the level of concern about the different groups. Although it would be comforting to say that there should be no winners and losers in education, the reality is that we make choices, whatever policies we choose. There is a second source of debate not covered by this model. High-stakes tests raise standards only in those areas that are actually tested. If a test has only multiple- choice questions, students may divert their efforts from learning to write good essays to acquiring the information needed to answer multiple-choice questions. If the test favors five-paragraph essays, students will not learn as much about writing term papers. In the jargon of economics, this is known as the multitask principal agent problem.9 Management scientists refer to it as “the folly of paying for A while hoping for B.”10 Teachers call it “What you test is what you get.” In the economics version of the problem, let us say that I want you to perform a job for me, and that job requires two tasks, one of which I can observe and the other of which I cannot. Depending on the exact nature of the technology, it may or may not be a good idea for me to pay you on the basis of the task that I can observe. For example, suppose I want you to produce as many units of high-quality output as possible but cannot easily determine whether the quality is high. If I pay you at a piece rate (i.e., on the basis of the quantity you produce), I will be encouraging you to work fast and produce a lot of output. If the quality of what you produce is very sensitive to the speed at which you work, you will produce a lot of low-quality output, which is not what I want. If the quality of what you produce is relatively insensitive to the speed at which you work, paying you at a piece rate may be a good idea.

9. Bengt Holmstrom and Paul Milgrom, “Multitask Principal-Agent Analyses: Incentive Con- tracts, Asset Ownership, and Job Design,” Journal of Law, Economics, and Organization 7 (Special Issue 1991): 24–52. 10. Steven Kerr, “On the Folly of Rewarding A, while Hoping for B,” Academy of Management Journal 18 (December 1975): 769–83.

218 Chapter 8 In the case of testing, the issue is whether schools “teach to the test.” Whether teaching to the test is good will depend a great deal on the degree to which the test does a good job of covering the full range of skills that we want students to develop. If we test only English and math, schools may neglect science and social studies. If we test science, they may not teach how to conduct experiments if that is not easily tested and does not prepare students for the exam. On the other hand, testing can promote good teaching if the best way to prepare students for the exam is by having them perform experiments, even if the ability to do experiments is not tested directly.

2.5. Evidence. It is difficult to obtain good evidence of the effects of testing. John Bishop and Ferran Mane found that individuals from states with high-stakes tests (in 1992) were less likely to graduate high school and more likely to have passed a high school equivalency exam (GED).11 The increase in the number of students passing the GED was about equal to the decrease in the number graduating high school. However, because the GED is generally viewed as inferior to a high school diploma,12 this suggests that the high-stakes tests decrease educational attainment at the lower end of the distribution. Bishop and Mane also found that individuals in states with high-stakes testing have higher earnings for a given level of education, as predicted by the theoretical models. However, this effect operates through individuals’ working more rather than through their being paid higher hourly wages, which is not predicted by the model. The main difficulty with the Bishop and Mane study should be familiar by now. Why do some states but not others choose to use high-stakes tests? If states with major dropout problems are less likely to use high-stakes tests, the effect of testing on the dropout rate will appear to be smaller than it really is. On the other hand, if high-stakes tests are more politically acceptable when high school graduation is less common, we might falsely attribute the higher dropout rate in these states to the high-stakes tests. Therefore, we will focus on the case of Texas, which has implemented a series of reforms since the mid-1980s. We choose Texas for two reasons. First, it has been studied extensively. Second, it is a large state, which makes it easier to obtain data by which Texas can be compared to other states. However, although the Texas reforms included high-stakes testing, Texas also initiated a number of other reforms, from requiring a minimum GPA for eligibility to play sports to establishing an accountability system with rewards and punishments for schools and school districts. It is difficult if not impossible to separate the effects of the high-stakes testing from the effects of the other reforms. In particular, because the graduation rate is one element of the accountability system, schools may try to reduce the dropout rate even if the direct effect of high- stakes testing increases it.

11. John H. Bishop and Ferran Mane, “The Impacts of Minimum Competency Exam Graduation Requirements on High School Graduation, College Attendance and Early Labor Market Success,” Labour Economics (May 2001): 203–22. 12. Stephen V. Cameron and James J. Heckman, “The Nonequivalence of High School Equiva- lents,” Journal of Labor Economics 11 (January 1993): 1–47.

Education and Education Reform 219 The most recent reform in Texas involved the creation of the Texas Assessment of Knowledge and Skills (TAKS). This replaced the Texas Assessment of Academic Skills (TAAS), which was the subject of much study and controversy. Beginning with the class of 2005, all students were required to pass eleventh-grade TAKS exams in English language arts (ELA, or reading and writing), mathematics, science, and social studies. Because achievement generally increases rapidly in the first few years that a test is administered, Texas will use lower requirements for attaining a passing grade in 2005 and 2006.13 The TAKS mathematics exam now has a very small number of open-response questions (there was only one such question in spring 2004), while the ELA test has a small number of questions requiring a brief essay response. We will focus on the TAAS, because the TAKS has been implemented too recently to permit any evaluation. Although we will treat high-stakes testing as beginning with the TAAS, Texas had a statewide test prior to the TAAS. However, the test was viewed as easy and not a real hurdle for graduation. The TAAS was quite different. It was viewed as difficult. The fall 1990 pretest of the TAAS revealed that the passing rates would be in the range of 40 to 60 percent, with pass rates for black and Hispanic students on the math portion of the TAAS falling to the range of 27 to 33 percent. The test was first administered as a high-stakes test to eleventh graders in fall 1991. Starting in spring 1993, the test was given to tenth graders. Students had up to ten opportunities to pass the tenth-grade exam, which they were required to pass in order to graduate. Starting in 1994, the TAAS reading, mathematics, and writing tests were administered to students in the fourth, eighth, and tenth grades in the spring of each year. The focus of test-based accountability in Texas was on the TAAS tests of reading, mathematics, and writing (there are also TAAS tests of social studies and science and end-of- course tests in some high school subjects). The TAAS tests were mostly multiple-choice in format with the exception of the writing exam. Following implementation of the TAAS, scores on the test improved dramatically. Stephen Klein and his coauthors at the Rand Corporation compared the dramatic improvement on the TAAS with improvements on the National Assessment of Educational Progress (NAEP).14 While TAAS scores shot up, fourth graders’ scores on the NAEP rose only somewhat, not dramatically, faster in Texas than elsewhere, and gains at the eighth-grade level were similar to those in the rest of the country. The authors’ findings raise serious questions about the meaning of these test score improvements. It appears that the high-stakes nature of the test creates strong incentives for improvement on the TAAS that are not reflected in other measures of performance. The authors’ other major conclusion is that while the performance gap between minority and white students has narrowed on the TAAS, it increased, if anything, on the NAEP.

13. Texas Education Agency, Student Guide to Graduation, http://www.tea.state.tx.us/student. assessment/resources/grad/grad_broch.pdf. 14. Stephen P. Klein et al., “What Do Test Scores in Texas Tell Us?” (Issue Paper 202, Rand Cor- poration, Santa Monica, CA, 2000).

220Chapter 8 David Grissmer and his colleagues, also at Rand, reached a somewhat different conclusion.15 They also studied the results of the NAEP and determined that the test scores increased more rapidly in Texas than in the United States as a whole. They suggest that the Texas accountability program, including TAAS, is a plausible source for this faster growth. On the basis of these studies, it is reasonable to conclude that the Texas education reform did increase performance on standardized tests. It increased performance on the assessment instrument (TAAS) significantly more than on other standardized tests, but because some of what is measured on the TAAS is similar to that measured on the NAEP, some of the improvement carried over to that test. What is not at all clear is whether the reform improved performance on the types of skills that are not measured on such tests.

3. Decentralization and School Quality Although so far we have focused on high-stakes testing for students, much of the focus of education policy is on how to make schools better. Some of the resulting policy initiatives, notably charter schools, vouchers, and school choice, are designed to change the link between where students reside and where they attend school. Before we examine these policies, we must examine the theory that supports the traditional decentralization of the U.S. public school system. Historically, education in the United States has been mostly the responsibility of local government. State government has played a supporting role, and the federal government has been even less involved. Most of the funding for education has come from the local community and usually has been raised via taxes on property. By international standards, this is a highly unusual way to fund education. It has advantages and drawbacks. The major advantage is that, because communities pay for education, a situation not found in most developed countries, they are free to offer education in the way that they favor. The state may require a minimum number of days of school, but the community is free to offer more days. It can choose class size, teacher pay, and many aspects of the curriculum. It can choose to put money into the football program or into performing arts. The economist Charles Tiebout showed that, under some restrictive assumptions, the competition among communities worked to create a market for goods such as education.16 To see how this works, suppose that all families have exactly one child who goes to school. Suppose that there is just one varying element of school quality, for example, class size. Once the community has chosen class size, it has established how much it will cost to educate each student. A class size of thirty might involve a cost of $4,500 per student, while a class size of twenty might cost $6,000 per student.

15. David W. Grissmer et al., Improving Student Achievement: What State NAEP Test Scores Tell Us (Santa Monica, CA: Rand Corporation, 2000). 16. Charles M. Tiebout, “A Pure Theory of Local Expenditures,” Journal of Political Economy 64 (October 1956): 416–24.

Education and Education Reform 221 In order to enroll their child in a school, parents must pay two prices. They must pay a fee to become part of the community. This fee is the cost of a house (or the value of the land on which the house is located). The other price is the taxes that residents must pay. Let us consider the taxes first. Because people prefer not to pay taxes, the community will never set the tax above the cost of providing the education. On the other hand, if it sets the tax below the cost of providing the education, the community will not be able to pay the teachers and pay other costs. Therefore, the tax will be exactly the cost of education per child. Now let us think about how much people will have to pay to buy a house. If two communities offer the same school quality, but one charges higher taxes, people will prefer the community with the lower taxes. Therefore, the price of housing will rise in the low-tax community and the price of housing will fall in the high-tax community, until people are indifferent between paying the higher cost of housing and lower taxes, on the one hand, and paying less for housing but higher taxes, on the other. Members of a community know that if their community is inefficient and pays too much for a given level of school quality, their housing values will fall. Therefore, competition among communities creates an incentive for communities to produce school quality efficiently. But the competition among communities does more than provide an incentive for efficient production. It also pushes communities to provide the levels of school quality that people want. Suppose that everyone is willing to pay an extra $2,000 per year to have twenty students per class instead of thirty students per class. If one community offers classes of thirty and taxes of $4,500 and another offers classes of twenty and taxes of $6,000, people will prefer to live in the community with smaller classes. They will bid up the price of houses in the community with the smaller classes and bid down the price of houses in the community with the larger classes. Because they care about the value of their houses, owners will pressure their communities to have classes of twenty rather than thirty. Of course, not everyone has the same tastes. Some will prefer to pay $4,500 and have large classes, while others will prefer to pay $6,000 and have small classes. If too many communities offer large classes, people who want small classes will have difficulty finding houses in the communities they like and will bid up their price. On the other hand, if too many communities offer small classes, prices will be bid down in those communities. Residents, interested in maximizing the value of their homes, will respond to the market incentives. Some communities will offer high-quality education and require high tax payments, while others will offer low-quality education but require only low tax payments. In contrast with the systems in countries in which decisions about education are made by the central government, the U.S. system allows people to choose among communities with differing levels of school quality. Those who value school quality a great deal will live in communities with good schools, while those who place less value on education will live in communities with weaker school systems. Moreover, because school quality is not one-dimensional, families can choose the types of schools that they like. One family may choose to move to a school with a strong traditional curriculum, while another may prefer the school with more experimental programs.

222 Chapter 8 The real world is much more complicated than this simple story suggests. The taxes that people pay in their communities are not some fixed amount but typically depend on the value of their houses. In the same community, people with expensive houses pay more taxes than people with inexpensive houses. Some people have more than one child, while others have none. Because we base taxes on the value of property, not on the cost of educating a family’s children, some people more than pay for the cost of educating their children, while others do not cover this cost. Still, local control of education has two distinct advantages: 1. It creates competition among communities so that they attempt to provide education efficiently and in a manner that responds to the preferences of the community. Communities that do this effectively will have higher housing values. 2. It maintains public support for funding education; even people who do not have children in the public schools may support funding for public education because better schools increase property values. Does the market for school quality work the way that the theory suggests? We will examine the relations among school quality, competition, and housing values. The question of political support and school funding is complex and is not central to the study of poverty.

3.1. Competition and the Efficiency of the Provision of School Quality. Caroline Hoxby has studied the relation between the extent of choice among school districts in a metropolitan area and the performance of the schools in that area.17 More district choice has a mixed relation to student performance, raising student performance along some measures and lowering it along others. However, the number of districts in a metropolitan area is, to some extent, a matter of choice. Politicians might respond to dissatisfaction with school districts by merging them or by breaking them up. In the former case, poor performance would lead to fewer districts. Therefore, Hoxby looked for physical features of the metropolitan area that influence the number of districts. Historically, rivers were major barriers to travel, and in many cases they continue to slow movement between areas. Hoxby finds that metropolitan areas with more rivers have more school districts. She predicts the number of school districts based on the number of rivers and other factors. She then looks at the relation between this predicted school district concentration and student performance. She finds that measures of student performance are consistently higher when there is more competition among districts. This does not by itself prove that the schools are more efficient. It could be that 60 percent of residents prefer low taxes and low school quality and 40 percent prefer high

17. Caroline M. Hoxby, “Does Competition among Public Schools Benefit Students and Tax- payers?” American Economic Review 90 (December 2000): 1209–38.

Education and Education Reform 223 taxes and high quality. When there is only one district, the majority vote for low taxes and low quality. When there are two districts, those who want low taxes and low quality live in one district and vote for low quality and low taxes as before, but those who want high quality and high taxes live in the other district and vote to provide these. The move from one district to two districts improves students’ performance but does not necessarily increase the efficiency of the school districts. However, Hoxby goes on to look at school spending. She finds that increased competition among districts lowers school spending while reducing class size. Thus, the improved student performance cannot be attributed to increased spending. More- over, she finds that increased choice is associated with fewer students’ attending private school. Hoxby’s results are certainly consistent with the expectation that more choice will push schools to be more efficient and more responsive to the preferences of the community. However, more choice may also be associated with more community control and involvement. When school districts are more compact, schools may be more effective, regardless of competition.

3.2. School Quality and Housing Prices. There is an extensive literature on the relation between house prices and school quality. In general, neighborhoods with better schools also have higher house prices. However, this relation is not informative. People with higher incomes will tend to spend more on housing. Because of the advantages associated with being in a richer family, their children will also tend to do better in school. It is not surprising that wealthy areas have both schools with high test scores and expensive homes. Put differently, a neighborhood with expensive single-family homes will have children who will, on average, do better in school compared with children in a neighborhood with small rental apartments even if the children in the two neighborhoods attend the same school. Sandra Black reduces this problem by looking at districts with houses located on the boundaries between school attendance districts.18 Within school districts, such boundaries are often drawn in very arbitrary ways and divide neighborhoods. The houses on one side of the street are likely to be similar to those across the street. The houses and neighborhoods are therefore very similar except that the children on different sides of the street attend different schools. She finds that a 5 percent increase in test scores increases the price of housing by 2.5 percent. Katharine Bradbury, Christopher J. Mayer, and Karl E. Case used a somewhat different approach.19 They looked at the effect of a tax limitation measure on housing 1 prices. Massachusetts Proposition 2 /2 constrained some but not all communities to spend less on education than they would otherwise have chosen to do. The authors

18. Sandra E. Black, “Do Better Schools Matter? Parental Valuation of Elementary Education,” Quarterly Journal of Economics 114 (May 1999): 577–99. 19. Katharine L. Bradbury, Christopher J. Mayer, and Karl E. Case, “Property Tax Limits, Local 1 Fiscal Behavior, and Property Values: Evidence from Massachusetts under Proposition 2 /2,” Journal of Public Economics 80 (May 2001): 287–311.

224 Chapter 8 found that property values rose less rapidly in communities that were forced to restrain their spending, particularly their education spending.

3.3. School Choice. People who can afford to live in the suburbs often have considerable choice about where to send their children to school. If they are unhappy with their children’s current school, they can move them. For the middle class, the Tiebout model may apply well. Moreover, voters in suburban school districts may recognize a strong link between their housing values and school spending. In contrast, poor people living in high-poverty areas often have little choice. The school choice movement aims to give poor people a range of choices. By promoting competition among schools, the movement hopes to imitate the benefits of competition seen by Tiebout. Of course, such choice will not replicate the link between housing values and school quality. Indeed, as we will see, it may weaken or sever the link. There are a number of proposals for enhancing school choice: • Residential neighborhoods can be separated from school districts. Within a district, students may be able to apply to more than one school rather than simply attending a school assigned to them on the basis of neighborhood boundaries. The district then uses parents’ preferences when assigning students to schools. Neighborhood may or may not play a central role in this assignment. The school choice may also be extended beyond districts. • Vouchers provide more choices by allowing students to attend private (and in some cases, parochial) schools. In a typical program, if a student is accepted by a private school and chooses to attend, the school accepts the amount of the voucher as full payment. • Charter schools are privately run public schools. They may be for-profit or nonprofit organizations. Because they are privately run, they are typically exempt from some state regulations and are expected to be more experimental than regular public schools. Usually a charter school receives a direct payment from the state for the students it enrolls and is not permitted to accept private tuition-paying students. We examine each of these approaches in turn.

3.4. Interdistrict and Intradistrict School Choice. Intradistrict school choice has been associated primarily with racial desegregation of schools rather than with an attempt to increase competition among schools. There are some exceptions, including a mixed neighborhood/choice program in New York City. We will discuss it only briefly, because the major focus of more recent reforms is on choice across districts. In a typical intradistrict school choice plan, parents are asked to list in order the schools in the district that they would like their children to attend. The district then assigns the children to schools on some basis that partially reflects parent preferences. For example, the district might decide that the first criterion is that children have the right to attend schools where they have siblings. After assigning all children whose first-choice school is attended by a sibling, the district might then look at the racial,

Education and Education Reform 225 sex, and ethnic balance of the schools. It would then assign students to their first- choice school if it improved the balance along these lines. Students are not guaranteed that they will be placed in their preferred school or even in one of their top choices. Because public schools are required to serve all the children in a district, intradistrict school choice has little effect on competition across schools. School districts typically try to avoid having excess capacity. If the number of children attending public school in the district is approximately equal to the number of openings in the district, all schools will be full. Even if a school is placed last by every parent, the school district will still assign some students to that school. Of course, the district may respond by changing the principal, giving the school more resources, or in some other way, but the staff does not automatically suffer if their school is not highly regarded by parents. We can see that intradistrict school choice gives some people more choice and others less. When school assignment is based on neighborhood, parents with sufficient resources can choose their child’s school by buying or renting housing in the neighborhood. School choice eliminates the guarantee that parents can choose that school. On the other hand, when school assignment is neighborhood-based, housing will generally be more expensive in the neighborhoods with schools that many people find attractive. For poor people, living in such neighborhoods may not be a realistic option or may involve considerable sacrifice. Under school choice, poor parents living in one neighborhood can select a school in a different neighborhood and have some chance that their child will be enrolled in that school. Whether this increased choice helps poor children depends on a number of factors, including how much the quality of a school depends on the social background of the students attending the school and how much is due to other factors. Upper-income parents are likely to respond to their failure to achieve their child’s enrollment in their first-choice school by leaving the district, either by moving or sending their child to private school. This not only removes more advantaged students from the schools, which will be detrimental to the other students if there are significant peer effects, but may also weaken the public schools politically. When there are already lots of alternatives for upper-income families, intradistrict school choice is therefore less likely to be effective in helping poor children. Intradistrict school choice is much more attractive in large school districts, such as that of New York City, where many of the children of the well-to-do already attend private schools.

3.4.1. Intradistrict Choice in New York City. Community School District (CSD) 4 is one of thirty-two districts within the New York City public school system, the largest in the United States.20 It is located in East Harlem, a ghetto and barrio neighborhood. In the 1970s, the district initiated a reform program that gave schools considerable

20. This section is drawn heavily from Peter W. Cookson Jr. and Sonali M. Shroff, School Choice and Urban School Reform (New York: Teachers College, Columbia University, December 1997, http://eric-web.tc.columbia.edu/monographs/uds110/), and their “Recent Experience with Urban School Choice Plans,” Eric Clearinghouse on Urban Education Digest 127 (October 1997), http:// eric-web.tc.columbia.edu/digests/dig127.html.

226 Chapter 8 power to reform themselves and to open new schools. Schools were distinguished from school buildings, and one junior high school building could house more than one independent junior high school. There was considerable expansion of the number of separate schools. Starting at the junior high school level, parents were allowed to select which school their child attended. They ranked their top choices, and their child was considered for admission by the first school on their list. If the child was turned down, they were considered by the second, and then, if necessary, by the third. If the child was not admitted to one of the top three choices (5 percent of applicants in 1992), the parents met with administrators to agree on a placement. During this period, CSD 4 students’ performance soared. More students gained placement in the city’s elite high schools, and standardized test scores for district children went from being the lowest of the thirty-two districts to being well above average. After the initial gains, there was some slippage. As of 2002, CSD 4 children’s scores on the state’s eighth-grade math and English tests placed it twenty-ninth in math and twenty-second in English among the system’s thirty-two regular districts as measured by the proportion scoring a 3 or 4 (proficiency or better) on the tests.21 It is apparent that the reform of CSD 4 was a success, although perhaps less dramatic than is sometimes claimed in political circles. What is less obvious is how much of a role choice played and how much of the improvement must be attributed to the role of inspired education reformers and the transformation of the programs, particularly by shifting to smaller schools. At the same time, it is questionable whether the spirit of reform could have been fostered without the opportunity for choice.

3.4.2. Intradistrict Choice in Chicago. Chicago’s school district is probably the largest district to fully embrace intradistrict choice.22 Students have the right to attend their neighborhood school but may also apply to any other school in the district. If there is excess demand for a school, the district uses a lottery to fill the spaces not used by neighborhood children. Almost half of students apply to a school other than their neighborhood school. Julie Berry Cullen, Brian A. Jacob, and Steven Levitt find that student outcomes are better for students who have chosen a school outside their neighborhood. This seems to suggest that school choice helps these students. However, as opponents of school choice suggest, it may be that more motivated students take advantage of school choice and that the students who remain behind are the less motivated students. If so, the difference between those who use school choice and those who do not might have nothing to do with choice. Cullen, Jacob, and Levitt use two approaches to get around this problem. First, they note that most students who do not attend their neighborhood school still attend

21. http://www.nycenet.edu/daa/test_results/. 22. This section is based on Julie Berry Cullen, Brian A. Jacob, and Steven Levitt, “The Impact of School Choice on Student Outcomes: An Analysis of the Chicago Public Schools,” Journal of Pub- lic Economics 89 (June 2005): 729–60, and, by the same authors, “The Effect of School Choice on Student Outcomes: Evidence from Randomized Lotteries” (unpublished paper, 2005).

Education and Education Reform 227 a nearby school. Students who live close to more schools are more likely to benefit from school choice. They ask whether students who live in neighborhoods with more nearby schools do better than those with few nearby schools. They find no difference in outcomes with the exception of those for students who attend career academies, smaller schools with a somewhat more vocational emphasis. This first approach depends critically on the assumption that whether a student lives in a neighborhood with lots of nearby schools is unrelated to unmeasured differences among students. To get around this problem, the researchers also look at the outcomes of school lotteries and ask if lottery winners do better than lottery losers. For the most part, the answer is again no, although there is some evidence of beneficial effects on behavior. Although the results of the Chicago studies are consistent with the conclusion that competition increased the quality of all schools, it is not easy to tell a story in which all schools respond to competition but students who are able to attend the most popular schools receive no benefit from doing so. Similarly, the Chicago studies do not support the concerns of opponents of school choice, who argue that peer effects are important and that those left behind will be hurt by the departure of the more motivated students. Again, this is not consistent with similar outcomes for lottery winners and losers. The simplest explanation is that intradistrict school choice does not have large effects on student outcomes.

3.5. Interdistrict School Choice. Interdistrict school choice differs from intradistrict school choice in two important ways: 1. There is a home district that has responsibility for educating all of the students in the district. Districts may accept additional students on a space- available basis. Students are not guaranteed a spot in a district other than their own. 2. In all such programs, funds flow to the district that receives out-of-district students, and in most programs, funds are subtracted from the district from which those students are sent. As a result, it is possible for the receiving district to benefit financially and the sending district to be harmed. Interdistrict school choice simultaneously increases and decreases competition among school districts. On the one hand, if a school district provides high-quality education, it can attract students from other districts. If it costs less to educate those students than the district receives for educating them, the district receives a fiscal windfall. It can provide more services to its residents or reduce taxes. On the other hand, the Tiebout model we discussed earlier relied on the fact that people have to obtain housing in a district in order to send their children to its schools. If parents can send their children to a school without living in the district, this will reduce demand for housing in districts with good schools and increase demand in districts with poorer schools. If the relation between housing prices and school quality is weakened, this may reduce support for funding local schools. Minnesota was the first state to initiate statewide school choice. Under the Min- nesota Open Enrollment Option, any student can apply to any public school in the

228 Chapter 8 state. If the district believes that, after serving its own students, it has sufficient space, it accepts the student. If it has sufficient space for some but not all of the out-of-district students who apply, it must choose among them using a lottery. The receiving district is responsible for transportation from the district boundary to the school, but parents (with some state support for poor families) are responsible for transportation between home and the district boundary. State funding for the pupil is transferred from the student’s home district to the district where he attends school. State funding is considerably less than the average cost of education but may or may not be greater than the marginal cost of educating an additional student. Unfortunately, although the program has been in effect for more than a decade, there is no clear evidence of its effectiveness. However, there have been two studies of the effect of the program on property values.23 Both find that property values increased in districts that lost students while they decreased in districts that accepted students from outside. In fact, Randall Reback calculates that the increased property tax from the higher property values was about what would be needed to offset the lost revenue in communities losing students. Thus, the interdistrict choice program seems, if anything, to have reduced the incentive of school districts to compete efficiently for students.

3.6. Vouchers. Voucher programs provide students with vouchers that can be used to pay for tuition at private and, in some cases, parochial schools. The voucher programs that have been tried in the United States have been oriented toward poor children. In principle, every student could receive a voucher. We will discuss universal voucher systems later, because no such system exists at present, so any such discussion is largely speculative. William Howell and Paul Peterson have evaluated voucher experiments in New York, Dayton, and Washington, DC.24 All three were privately funded voucher programs that provided subsidies for poor children to attend private or parochial schools. In each case, more children applied to the program than could be served, and vouchers were allocated on the basis of a lottery. The researchers found that, after two years, African American students offered a voucher did significantly better on the Iowa Test of Basic Skills than did African American students who were not offered a voucher. The implied effect of using a voucher was about 6 national percentile points. In contrast, there was no significant difference between members of other racial or ethnic groups who did and did not receive a voucher.

23. Randall Reback, “Capitalization under School Choice Programs: Are the Winners Really the Losers?” (Occasional Paper 66, National Center for the Study of Privatization in Education Teachers College, Columbia University, New York, 2002), and William E. Hoyt, “Public School Choice, Prop- erty Values, and Investment in Public Education” (unpublished paper, University of Kentucky, Lex- ington, 1997). 24. William G. Howell, Patrick J. Wolf, Paul E. Peterson, and David E. Campbell, “Test-Score Effects of School Vouchers in Dayton, Ohio, New York City, and Washington, DC: Evidence from Randomized Field Trials” (paper presented at the annual meetings of the American Political Science Association, Washington, DC, September 2000). See also William G. Howell and Paul E. Peterson, The Education Gap: Vouchers and Urban Schools (Washington, DC: Brookings Institution, 2002).

Education and Education Reform 229 The New York City experience has been studied most extensively25 and has been the basis for considerable controversy. For this program, a private foundation offered vouchers worth $1,400 per year to each of more than 1,200 students selected randomly from more than 20,000 applicants. Of those offered a voucher, 74 percent used it the first year, 62 percent the second, and 53 percent the third. Inevitably, experiments are imperfect. Some families offered vouchers do not make use of them. Researchers lose track of some of these families and of some families that did not receive vouchers. Or they are unable to collect some information for some families. Moreover, there is no agreed-upon definition of race. The results of the New York City experience are somewhat sensitive to decisions about how to deal with these difficulties. A fair summary of the debate is that most choices of research strategies lead to the conclusion that the vouchers had a modest positive effect on the recipients. However, some choices lead to the conclusion that there were no statistically significant effects, and others lead one to conclude that the effects were large. For the most part, the voucher programs that have been tried in the United States have been quite modest both in terms of the level of the voucher and the number of students involved. The major exception is the Milwaukee Parental Choice Program, which in 2002–2003 served more than 11,000 students. Participating schools received up to $5,783 per student in the program.26 Participation is limited to 15 percent of enrollment in the Milwaukee public schools, and only families with incomes no higher than 1.75 times the poverty rate are eligible. Approximately half of the participating schools are parochial. Cecilia Rouse tried to evaluate the effect of the Milwaukee program on the participants.27 Her evaluation was based on the early years of the program, when spaces were very limited. Perhaps the most striking result of her evaluation was the difficulty she had in finding a comparison group even early in the program, when participation was much more limited. Although places were allocated on the basis of a lottery, most of those who failed to win the lottery subsequently left the Milwaukee public schools anyway because they won a subsequent lottery, used their own resources to pay for tuition, or left Mil- waukee. This raises the question as to the degree to which vouchers are largely a subsidy to families that would not have attended the inner-city public schools in any event. Thus, there is room for those with strong feelings on each side to find data to support their views. However, in some sense the debate over the New York City voucher

25. Paul E. Peterson and William G. Howell, “Efficiency, Bias, and Classification Schemes: Esti- mating Private-School Impacts on Test Scores in the New York City Voucher Experiment: Latest Re- sults from New York City School Voucher Research—African Americans in Private Schools Score Higher (Working Paper PEPG/02-15-22, Program on Education Policy and Governance, Harvard University, Cambridge, MA, 2002); Paul E. Petersen et al., “School Vouchers: Results from Ran- domized Experiments,” in Caroline M. Hoxby, ed., The Economics of School Choice (Chicago: Uni- versity of Chicago Press, 2003); and Alan B. Krueger and Pei Zhu, “Another Look at the New York City School Voucher Experiment,” American Behavioral Scientist 47 (January 2004): 658–98. 26. http://www.dpi.state.wi.us/dpi/dfm/sms/choice.html. 27. Cecilia E. Rouse, “Private School Vouchers and Student Achievement: An Evaluation of the Milwaukee Parental Choice Program,” Quarterly Journal of Economics 113 (May 1998): 553–602.

230Chapter 8 experiment misses the point. The arguments for the traditional system of local control in the United States and for new types of competition are both based on the assumption that parents can make intelligent choices on behalf of their children. It would be surprising to both groups if large numbers of parents chose to send their children to schools that did not improve their educational outcomes. If they did make such choices, we would have to conclude either that parents cared about different outcomes from the ones measured by the tests used to evaluate the voucher program or that they could not assess school quality. The real issue is whether the competition generated by the existence of vouchers improves education for students who do not use vouchers. On this point there is almost no evidence.

3.7. Charter Schools. Charter schools are public schools that are permitted to operate with considerably less regulation and direct supervision than is typical for public schools. Although the precise rules vary among states, in general, a charter school must establish a clearly defined set of objectives and a set of measures that will help determine whether it has met those objectives. Schools that meet their objectives may continue; those that fail to meet their objectives will be closed. Charter schools are generally free from having to hire unionized teachers and may have exemptions from teacher certification requirements. In principle, the greater flexibility afforded charter schools should lead to more innovation. Setting up charter schools is different from giving students vouchers, because the state (or other authority) has the power to close the school if it is not proving effective. The school cannot accept private tuition-paying students. In other ways, the charter and voucher movements are similar. Each is designed to create more choice for parents and their children. And each is designed to foster competition that is intended to improve education in the regular public schools. There are two questions that we ask about charter schools: 1. Do charter schools improve the performance of the students who attend? 2. Does competition from charter schools improve the performance of the regular public schools?

3.7.1. Charter Schools and Student Performance. One of the states that has made the most extensive use of charter schools is Michigan. Eric Bettinger compared 18 Michi- gan charter schools with 546 regular public elementary schools and 178 regular public middle schools located within five miles of one of the charter schools.28 He used results from Michigan’s statewide tests to determine student performance. How can we tell whether the Michigan charter schools improved student performance? One way we might consider answering this question is by comparing students who attend charter schools with those who attend regular public schools. Two years

28. Eric P. Bettinger, “The Effect of Charter Schools on Charter Students and Public Schools” (unpublished paper, Case Western Reserve University, Cleveland, OH, 2002).

Education and Education Reform 231 Table 8.2 Differences between Charter and Other Public Schools in the Proportion of Students Scoring “Satisfactory” and “Low” on the Michigan Educational Assessment Program

Math Reading

Fourth Grade Seventh Grade Fourth Grade Seventh Grade Levels Satisfactory –17 –18 –12 –13 Low12171011

Changes Satisfactory 1.9 –0.2 –4.1 * (5.2) (6.7) (5.0) (5.9) Low –7.0 –3.4 0.1 –2.0 (4.5) (6.9) (4.7) (5.7)

Changes Adjusted for Initial Scores Satisfactory –10.5 –0.6 –8.9 0.3 (3.4) (4.2) (4.1) (2.9) Low 7.4 6.2 6.6 –0.2 (4.3) (4.3) (3.2) (2.7)

Source: Eric P. Bettinger, “The Effect of Charter Schools on Charter Students and Public Schools” (unpublished paper, Case Western Reserve University, Cleveland, OH, 2002). *Less than 0.05.

after the founding of the charter schools, student scores were substantially worse at the charter schools than at the regular public schools. The proportion of fourth graders scoring “satisfactory” was 17 percentage points lower on the math exam and 12 percentage points lower on the reading exam. Among seventh graders, the proportions were 18 percentage points and 13 percentage points lower in math and reading, respectively. There are symmetric differences among the proportions receiving low scores. Students in charter schools were much more likely to receive low scores two years after the charter school was founded (see table 8.2). However, charter schools are likely to attract students from low-performing public schools. These students will be disproportionately disadvantaged, and it will not be surprising if they do not perform, on average, as well as the average regular public school student. On the other hand, students in charter schools may come from particularly motivated families. If we do a good job of controlling for differences in family background, we may find that students in charter schools do better simply because we cannot measure parental motivation.

232 Chapter 8 Therefore, Bettinger looked at whether the test scores of charter school students improved more or less than the test scores of students in comparison schools. When he looked at changes in scores, there was much less evidence of a difference, and the results, if anything, suggest that charter schools may outperform regular public schools. None of the differences is statistically significant, but, for three of the four tests, charter schools saw a greater decline in the fraction scoring low, and there was essentially no difference between charter schools and other public schools on the fourth test. Al- though we would not want to draw strong conclusions from statistically insignificant results, these findings certainly push us to explore further the possibility that charter schools improve the performance of the lowest-performing students. However, recall that when we discussed training programs, we raised the concern that the people who enter such programs often have had unusually low earnings before entering. People who have had decent jobs and then suffered long-term unemployment may be more likely to enter training programs and appear to “improve” as a result of the training program. Just as sick people often get better on their own, people who are “down on their luck” often recover through their own efforts and changes in their luck. This same phenomenon may apply to charter schools. Students whose performance in the public schools is below what they anticipate may apply to a charter school. Good students who have had a few years of bad teaching or bad behavior may apply. Many of these students would have had improved experiences in regular public schools. Bettinger attempts to address this by looking at changes while controlling for initial scores. His results suggest that, at least among fourth graders, the charter schools did worse than the regular public schools. Controlling for students’ poor initial level of performance, over two years the charter school fourth graders showed less improvement than did fourth graders in regular public schools. Caroline Hoxby and Jonah Rockoff studied seven charter schools in Chicago.29 Because the schools are oversubscribed, students are admitted by lottery. Overall the researchers found no difference in outcomes between lottery winners and lottery losers. When they broke the students down by the grade for which they applied, among those applying for kindergarten or first grade, lottery winners did somewhat better in reading but not in math. Among those applying for second or third grade, lottery winners did better in math but not in reading. Hoxby and Rockoff found no difference between lottery winners and losers applying for the fourth and fifth grades or the sixth through eighth grades. We should not read too much into these studies. After all, together Bettinger’s and Hoxby and Rockoff’s samples are limited to twenty-five charter schools in two states. Charter schools may have performed better in some states with different charter school regulations. Moreover, it is undoubtedly true that just as there are good and bad regular

29. Caroline M. Hoxby and Jonah E. Rockoff, “The Impact of Charter Schools on Student Achievement” (unpublished paper, Harvard University, 2004).

Education and Education Reform 233 public schools, there are good and bad charter schools. Some charter schools are founded by talented and creative leaders. Others are led by principals who lack the skill to run a high-quality school. Two years is a short period in which to create a first-rate school. What this research does show is that we should not equate charter school with good school. This conclusion is reinforced by a Rand study of charter schools in California.30 The study found similar levels of student performance and improvement in performance between charter schools and regular public schools once they controlled for differences in student backgrounds. However, it found considerable variation in performance among charter schools. In particular, it found that charter schools with a significant out-of- school instructional component performed less well. The authors are careful to note that such schools may attract very different students from those who attend traditional public schools and other charter schools and that therefore the poorer performance may not reflect worse school quality.

3.7.2. Effects on Regular Public Schools. Perhaps the most innovative aspect of Bet- tinger’s research on Michigan charter schools was his attempt to find their effect on other public schools.31 This is a difficult task. People are likely to want to found charter schools, and charters are likely to be granted in areas where regular public schools are viewed as weak. If we compared public schools in areas with lots of charters with those in areas with no charters, we would almost definitely find that public schools were weaker where there were charters. This would prove not that charters hurt public schools but that charters were founded where public schools were weak. Bettinger recounts that the Michigan governor put great pressure on the state’s public colleges and universities to found charter schools. As of 1999, 150 of 170 charter schools in Michigan had been founded by colleges and universities. Because institutions generally open charter schools near their own location, in Michigan charter schools were, on average, located in areas where school test scores were slightly better than average. Bettinger looked at changes in the performance of public schools and found no consistent evidence that public schools located near charter schools exhibited less or more improvement than schools located farther from charter schools. He was able to establish that any effect, whether positive or negative, was not large.

3.8. Universal Vouchers. In principle, a system of vouchers could replace the present system of neighborhood schools. Under such a system, schools would play much the same role in public education that health maintenance organizations (HMOs) do in our public health programs. Just as HMOs receive a payment from Medicare or

30. Richard Buddin and Ron Zimmer, “Academic Outcomes,” in Ron Zimmer et al., eds., Char- ter School Operations and Performance: Evidence from California (Santa Monica, CA: Rand Corporation, 2003). 31. Bettinger, section IV.

234 Chapter 8 Medicaid for each eligible person they enroll, the school would receive a fixed voucher payment for each pupil that might be based on pupil characteristics. In some cases, the fixed payment might be supplemented for particularly expensive special education, just as health care providers currently receive supplemental payments for certain expensive procedures. We cannot draw on our experience with vouchers to determine how such a system would work. No voucher program in the United States has been sufficiently extensive to approximate what would happen if vouchers replaced public schools. We can spec- ulate that a voucher system would have many of the strengths and weaknesses of a public health system in which people are free to choose their HMO. On the positive side, there would be considerable competition among schools for profitable students. On the negative side, if they were free to select students, schools would turn away un- profitable students. If not, they would tailor their programs to attract the students who were most profitable. Without knowing the details of the voucher system, it is impossible to know in advance precisely how this would work. One possibility is that all regular education students would received the same voucher, while special education students would receive a voucher for a greater amount that depended on the precise nature of their special education status. If poor children were more expensive to educate, schools could follow policies designed to discourage poor students from applying—locating far from high- poverty neighborhoods, refusing to participate in the school lunch program, and so on. On the other hand, if the voucher payment for poor students were sufficiently high relative to other students, schools would try to attract poor students. More generally, within any group for which the voucher payment was constant, schools would try harder to attract the students least expensive to educate.32 This concern is supported by research on managed care in the U.S. Medicare system. Under Medicare, HMOs receive a fixed payment per enrollee (with some modest adjustments for personal characteristics associated with the cost of treatment). Re- searchers have found evidence that HMOs choose the services they offer in order to attract healthy patients and deter patients with chronic illnesses from enrolling.33

3.9. Evidence from New Zealand. Beginning in 1989 and culminating with additional reforms in 1991, New Zealand moved from a highly centralized education system under the control of the Department of Education to one with locally managed schools competing for students.34 The reforms can most easily be described as a system

32. For a discussion of the health care case, see Richard Frank, Jacob Glazer, and Thomas McGuire, “Measuring Adverse Selection in Managed Health Care,” Journal of Health Economics 19 (November 2000): 829–54. 33. Zhun Cao and Thomas McGuire, “Service-Level Selection in Medicare,” Journal of Health Economics 22 (November 2003): 915–31. 34. This section draws heavily on Edward B. Fiske and Helen F. Ladd, When Schools Compete (Washington, DC: Brookings Institution, 2000).

Education and Education Reform 235 of universal vouchers and universal charter schools under the governance of parent boards. Schools were given wide latitude to run themselves and received a fixed payment from the central government for each student enrolled, with a supplement for students from disadvantaged backgrounds. In contrast with most charter schools in the United States, oversubscribed schools were free to select the students they preferred from among their applicants. By 1995, almost half of the schools in the two largest metropolitan areas were oversubscribed. In the United States, most oversubscribed charter schools must select their students on the basis of a lottery. This difference should not be exaggerated. Charter schools in the United States have the potential to attract the types of students they want by choosing policies that make them attractive to those types of students. They can expel students for minor discipline violations or set high academic standards and flunk those who fail to meet them. The evidence from New Zealand provides a mixed message. On the one hand, although there are no national exams that can be used to quantify changes in student performance, most teachers and principals believe that the program has improved education,35 and most boards are pleased with their autonomy. On the other hand, there has been a substantial increase in social segregation across schools. Successful schools are able to select the brightest, most motivated students and draw their students from advantaged families. A significant minority of schools would be classified as failing. Their boards are overwhelmed by the administrative responsibilities that come with their autonomy. They end up attracting only those students who are unable to obtain places in the oversubscribed schools. Their students are overwhelmingly drawn from disadvantaged populations. Although, in principle, such schools should be allowed to fail, there is not a large population of education entrepreneurs who are eager to open schools that target disadvantaged students. Indeed, because most parents want their children educated with more advantaged, brighter, more motivated students, creating a school targeted at difficult-to-educate children is not a natural pathway to commercial success. The New Zealand experience reinforces the view that competition among middle- class schools is an effective mechanism for generating high-quality education. It does not, however, support the view that creating charter schools and using vouchers on a large scale are likely to address the lack of competition and choice for disadvantaged urban children.

4. Using Tests to Increase School and District Accountability Both federal law and many state laws now require schools and school districts to ad- minister state tests. Although, as we have discussed, many of these state tests are high- stakes for students, some are used primarily to evaluate schools.

35. Fiske and Ladd, 243–44.

236 Chapter 8 The system used in Texas has served as a model for other states and for the federal legislation. The Texas Board of Education is mandated to rate the performance of schools and school districts according to a set of “academic excellence indicators,” including TAAS or TAKS results, dropout rates, and student attendance rates.36 State law also prescribes that student performance data be disaggregated by ethnicity and socioeconomic status. The performance rating system holds that school performance is not acceptable if the performance of any subgroup is not acceptable. Based primarily on percentage of students passing each of the state tests, the more than six thousand schools in Texas have been rated since 1994 as “exemplary,” “recognized,” “acceptable,” or “unacceptable.” Schools are eligible for cash awards for high ratings; if they are rated as low-performing twice in a row, they are subject to sanctions from the Texas Educa- tion Agency, including possible closure. Under the No Child Left Behind (NCLB) Act of 2001,37 each state must establish standards in reading, math, and science and must test children annually in grades 3 through 8 and once in high school to determine whether they are meeting those standards. Currently schools’ performances are calculated separately for math and English / language arts and separately for each test by ethnicity and socioeconomic status. Schools are supposed either to meet a standard that is higher each year or to make adequate progress toward that standard every year. Schools must publish a “report card” on their performance. A school that fails to meet the standard or improvement requirement on any test for any subgroup is considered to have failed to make adequate yearly progress. Schools that fail to make adequate yearly progress two years in a row are provided with support and are subject to sanctions. In particular, school districts are required to offer parents the choice of another school in the district if their current school is deemed to be failing to make adequate progress. Ultimately such schools can be closed down or their staffs can be replaced. Even the simple act of publishing test results may have important consequences. Realtors use information about school performance when they show houses to clients. If housing prices and rents rise in districts with high scores, homeowners will put pressure on school authorities to ensure that their schools do well on the tests. Therefore, publishing test results is a natural extension of the argument of the Tiebout model: if people do not know the quality of the schools in different neighborhoods, how can their choice of where to live respond to school quality? However, there are three cautionary scenarios with respect to the use of tests to ensure accountability. The first is one we have already discussed. Schools will respond to the test. If too much weight is placed on math relative to science, schools will ignore science and focus on math. If the test measures calculation and not calculus, schools may not teach calculus.

36. Texas Education Agency, Accountability Manual, 2004, http://www.tea.state.tx.us/perfreport/ account/2004/manual/index.html. 37. NCLB reauthorizes the Elementary and Secondary Education Act and is therefore also known as ESEA.

Education and Education Reform 237 Angela Dills has addressed this question indirectly by looking at the relation between housing prices and changes in student performance on the TAAS.38 She reasons that if parents value the improvements in education measured by the TAAS, they should bid up housing prices in communities where performance on the TAAS improves. She finds no evidence of such a relation. Thus, it appears that, at least in the eyes of parents, the benefits of improved performance on the state exams are offset by reduced performance in other dimensions. The second concern is that, even along those dimensions that the test is designed to measure, tests measure year-to-year changes very imprecisely. Changes in school performance from year to year reflect at least three factors—real changes in the way the school prepares students for the test, random variation in the ability of the students from year to year, and random variation in the preparation of students or other random factors affecting performance. In this last category, one school might have administered the test during a bad thunderstorm that distracted many students, or the grade might have had a special project on some topic that was addressed on the test one year but not the next. Thomas Kane and Doug Staiger show that randomness is an important element in gains, especially for those schools experiencing the largest gains or largest losses.39 When states such as North Carolina give rewards to the schools showing the greatest improvement, the overwhelming majority of those awards go to small schools. A school with a single class of twenty students is much more likely to have large year-to- year changes in student quality than a school with five hundred students in a grade. Figure 8.1 plots the change in average score on the Massachusetts tenth-grade math test for schools that administered the exam all three years from 2000 to 2002. The vertical axis shows the change from 2001 to 2002, while the horizontal axis shows the change from 2000 to 2001. The size of each circle is proportional to the size of the school’s tenth grade in 2001. It is apparent from looking at the figure that the points are arrayed along a downward- sloping line; schools that improved a lot from 2001 to 2002 tend to be those schools that improved little the previous year. Relatively few schools stand out from the mass two years in a row in either direction. Moreover, most of those that do tend to be small. This does not mean that we cannot learn anything from changes in test scores, just that we need to be careful about putting too much weight on annual changes. Changes that persist over a period of years are likely to be real rather than statistical flukes. How- ever, many short-term “successes” will quickly disappear. In addition, among schools of the same size, more diverse schools will have greater difficulty making adequate yearly progress. A school with 250 students, all of the same race, needs to meet the standards only for the group as a whole. A school with 250 students equally divided among blacks, whites, Hispanics, Asians, and native Americans

38. Angela K. Dills, “Do Parents Value Changes in Test Scores? High Stakes Testing in Texas,” Con- tributions to Economic Analysis and Policy 3, no. 1 (2004), article 10, http://www.bepress.com/bejeap/ contributions/vol3/iss1/art10. 39. Thomas J. Kane and Douglas O. Staiger, “The Promise and Pitfalls of Using Imprecise School Accountability Measures,” Journal of Economic Perspectives 16 (Fall 2002): 91–114.

238 Chapter 8 Change in Mean Math Score, 2001–2002 20

–10

–20 –10 0 10 20 30 Change in Mean Math Score, 2000–2001 Figure 8.1 Massachusetts Tenth-Grade Math Score Changes in Consecutive Years, 2000–2002 has to meet the standard for each group. Because each group is small, there is more randomness than for the group as a whole. And of course, the school risks failing with any one of the five groups rather than with just one group. The third concern about putting considerable emphasis on high-stakes exams is the strong pressure on educators to cheat. There are high-profile newspaper stories of teachers giving students the right answers or erasing answers and filling in the correct answers, but it is difficult to know how common such actions are. Brian Jacob and Steven Levitt used a clever technique to uncover improbable patterns in students’ answers.40 If a group of students got easy questions wrong and then got many hard questions right, it was unlikely that this occurred naturally. They estimated that such cheating occurred in about 4 to 5 percent of Chicago classrooms and that the incidence of cheating rose 75 percent when sanctions for low-performing schools were introduced. Moreover, the increase in cheating occurred in schools in the bottom third of the performance distribution. The authors’ work turned up evidence that led to the disciplining of some teachers, but in many cases, results were unlikely but there was insufficient evidence to prove that any individual teacher cheated.

5. Concluding Remarks There is good reason to believe that increasing the educational attainment of poor children and improving the quality of schools they attend would improve their outcomes as adults. Although proving causality is difficult, higher school quality is associated with

40. Brian A. Jacob and Steven D. Levitt, “Rotten Apples: An Investigation of the Prevalence and Predictors of Teacher Cheating,” Quarterly Journal of Economics 118 (August 2003): 843–77.

Education and Education Reform 239 children’s completing more years of school. So any improvement in school quality is likely to increase educational attainment as well as to ensure that students learn more while in school. There is some evidence that competition among schools raises school quality. In many areas, the competition among suburban school districts and the importance of the schools for the community put pressure on political and educational leaders to ensure that the schools provide high-quality education efficiently. Many, perhaps most, urban schools are immune from these pressures. Education reform movements try to improve urban schools by substituting other mechanisms for the competition among districts found in suburban areas. There is little evidence that any of these mechanisms has been effective. Texas, often held up as an example of tremendous success, has primarily succeeded in improving performance on its own test, with some much more limited spillover to performance on other tests. This may have come at some cost in terms of an increased dropout rate, increased energy ex- pended by administrators to look good on the assessment (for example, by recording students as transferring rather than dropping out), and reduced emphasis on aspects of education not covered by the assessment system. Just as there are good and bad regular public schools, there are also good and bad charter schools. If anything, charter schools appear to be, on average, worse than public schools serving comparable populations, probably because they are newer. Charter schools that survive an initial start-up period seem to have student outcomes similar to those of other public schools. There is little evidence that increased competition from charter schools has the desired effect on the regular public schools with which they compete. Although vouchers undoubtedly help some of the students they serve, the evidence is weak that voucher recipients generally achieve large gains from the voucher. Our limited experience with vouchers, charter schools, and both intra- and interdistrict school choice does not suggest that revamping our education system to create a system of universal vouchers (or charters) would solve the problem of inner-city education. The experience in New Zealand indicates the opposite. This does not mean that such reforms have no role to play in improving education. Because of the multitask principal agent problem, putting too much emphasis on state tests is problematic, but it is possible to use the tests while putting less weight on them. State tests could play the same role for students leaving school after high school that the SAT and ACT play for those applying to college. Charter schools and pilot schools within urban school systems could be selected more rigorously in terms of both leadership and innovation and then carefully monitored and evaluated to determine their efficacy. The experience with CSD 4 in New York suggests that inspired leadership and innovation can have a large impact on schools even in high-poverty neighborhoods.

6. Further Reading Betts, Julian R., and Robert M. Costrell. “Incentives and Equity under Standards- Based Reform.” In Diane Ravitch, ed., Brookings Papers on Education Policy: 2001 (Washington, DC: Brookings Institution, 2001).

240Chapter 8 Black, Sandra E. “Do Better Schools Matter? Parental Valuation of Elementary Educa- tion.” Quarterly Journal of Economics 114 (May 1999): 577–99. Buddin, Richard, and Ron Zimmer. “Academic Outcomes.” In Ron Zimmer et al., eds., Charter School Operations and Performance: Evidence from California. Santa Monica, CA: Rand Corporation, 2003. Cookson, Peter W., Jr., and Sonali M. Shroff. School Choice and Urban School Reform. New York: Teachers College, Columbia University, December 1997. http:// eric-web.tc.columbia.edu/monographs/uds110/. Dills, Angela K. “Do Parents Value Changes in Test Scores? High Stakes Testing in Texas.” Contributions to Economic Analysis and Policy 3, no. 1 (2004): article 10. http://www.bepress.com/bejeap/contributions/vol3/iss1/art10. Fiske, Edward B., and Helen F. Ladd. When Schools Compete. Washington, DC: Brook- ings Institution, 2000. Grissmer, David W., et al. Improving Student Achievement: What State NAEP Test Scores Tell Us. Santa Monica, CA: Rand Corporation, 2000. Hanushek, Eric A. Making Schools Work: Improving Performance and Controlling Costs. Washington, DC: Brookings Institution, 1994. Hoxby, Caroline M. “Does Competition among Public Schools Benefit Students and Taxpayers?” American Economic Review 90 (December 2000): 1209–38. Jacob, Brian A., and Steven D. Levitt. “Rotten Apples: An Investigation of the Preva- lence and Predictors of Teacher Cheating.” Quarterly Journal of Economics 118 (Au- gust 2003): 843–77. Kane, Thomas J., and Douglas O. Staiger. “The Promise and Pitfalls of Using Imprecise School Accountability Measures.” Journal of Economic Perspectives 16 (Fall 2002): 91–114. Klein, Stephen P., et al. “What Do Test Scores in Texas Tell Us?” Issue Paper 202, Rand Corporation, Santa Monica, CA, 2000.

7. Questions for Discussion 1. What are the principal reasons that education may have a causal effect on earnings? 2. If education is primarily a sorting device, and if high-stakes testing raises standards, who is helped and who is hurt by the tests? 3. If students learn marketable skills in school, will raising standards increase or decrease the dropout rate? 4. What is the multitask principal agent problem? How is it relevant to the testing debate? 5. Explain how competition among school districts generates efficient provision of education in the Tiebout model. What are the weaknesses of the model for representing the real world? 6. What are the theoretical advantages and disadvantages of decentralized funding for education? Does empirical research support these predictions?

Education and Education Reform 241 7. Discuss the theoretical and empirical effects of intradistrict and interdistrict school choice on school quality. 8. Do voucher programs improve educational outcomes for participants? 9. What is the evidence regarding the effects of charter schools on student performance? 10. What is meant by each of the following: intradistrict school choice, interdistrict school choice, vouchers, charter schools? 11. What does the experience of CSD 4 in New York City tell us about the effectiveness of intradistrict school choice programs?

242 Chapter 8 chapter 9

Welfare Reform

E         poverty rate stalled between the mid-1970s and the early 1990s. We discussed a wide variety of explanations for the failure of poverty to decline. These included • Increasing inequality both within the labor force and through the increasingly positive relation between husbands’ and wives’ earnings • Growth of single-parent families • Increased reliance on in-kind transfers and transfers through the tax system By the early 1990s, a substantial fraction of Americans had become convinced that part of the problem was the welfare system itself. They believed that welfare discouraged marriage, encouraged single motherhood, reduced people’s inclination to work, and made people dependent on the welfare system. In this chapter we briefly review the arguments for reform, most of which we have discussed previously. We then summarize the major components of welfare reform and discuss the limited evidence on its effects.

1. The Case for Reform As we discussed in chapter 6, conservatives like Charles Murray argued that the welfare system encouraged single motherhood. Liberals and conservatives alike recognized that providing welfare only to single mothers discouraged marriage. Partially in response to this concern, the Family Support Act of 1988 required states to participate in the Aid to Families with Dependent Children–Unemployed Parent (AFDC-UP) program. However, AFDC-UP was less generous than the standard AFDC program. As we saw in the chapter on family composition, the Minnesota Family Investment Program study showed that making welfare more generous for families encouraged marriage, although subsequent studies did not confirm this.

243 1.1. Work Disincentives. Liberals and conservatives also recognized that AFDC discouraged work. As we discussed earlier, the implicit tax rate on AFDC was often 100 percent. Recipients who left welfare were worse off than those who remained unless they were able to find a job paying considerably more than most welfare recipients could expect to earn.

1.2. Welfare Dependence. This, in turn, raised the concern that welfare might prevent recipients from establishing themselves in the labor market and thus become financially independent. Conservatives, in particular, argued that we had created a culture of welfare dependence. They noted that the longer a woman had been on welfare, the less likely she was to get off welfare. Recall that when we discussed poverty duration, we observed that the longer someone had been in poverty, the less likely he was to get out of poverty. We saw that this relation could reflect either or both of two factors. First, time spent in poverty could make it more difficult for a particular individual to get out. He might lose labor market contacts or good work habits or be viewed as a bad employment risk by potential employers. We referred to this as “state dependence.” On the other hand, even if a person’s probability of getting out of poverty was unaffected by how long she had been poor, the people who had been poor longer would tend to be those who had difficulty getting out of poverty. Someone who is poor because of a temporary illness will probably exit poverty quickly. Someone with very poor labor market skills probably will not. The population of those who have been poor for ten years consists almost entirely of people who have a hard time getting out of poverty. We referred to this as “individual heterogeneity.” The same argument applies to welfare dependence. We observe that people who have spent a lot of time on welfare tend to remain on welfare. That does not tell us whether the reason is state dependence—using welfare makes people more likely to continue using welfare—or individual heterogeneity—the people who use welfare a lot have difficulty finding employment or a spouse. Undoubtedly both factors are important. In studies of women receiving AFDC, Rebecca Blank has found that most of the variation in welfare duration can be explained by measured differences.1 Once we take account of the factors that lead some women to have more difficulty exiting welfare, there does not seem to be much evidence of an increased probability of welfare use due to having experienced welfare.

1.3. Passing Welfare from Generation to Generation. Observers were also concerned that welfare seemed to spread from generation to generation. They noted that children tend to adopt their parents’ occupations. Although most children of physicians do not themselves become physicians, they are much more likely to do so than are the children of parents with jobs outside of medicine. In the same way, most

1. Rebecca M. Blank, “Analyzing the Length of Welfare Spells,” Journal of Public Economics 39 (August 1989): 245–73. The focus is on women because most households receiving welfare are headed by single mothers.

244 Chapter 9 daughters of mothers who receive welfare do not end up receiving welfare themselves, but they are considerably more likely to do so than are other women.2 Recall that when we discussed neighborhood effects, we saw that it is very difficult to determine whether people are influenced by others around them or whether they tend to be with people who are similar to them. People who are interested in committing crimes are more likely to hang out with people who commit crimes than are people who want to stick to a legal path. The same concern applies to families. Daughters share many characteristics with their mothers. It is well beyond the scope of this book to explore the nature versus nur- ture debate. Daughters of welfare mothers may resemble their mothers because they imitate the behavior they observe or because they share genetic traits. These daughters are more likely than other women to share the physical, psychological, or cognitive disabilities that lead their mothers to be on welfare. On the other hand, they are also more likely to know how to work with the welfare bureaucracy and may be less likely to feel stigma attached to welfare receipt. We have already noted that distinguishing between shared characteristics and the influence of one person on another is extraordinarily difficult. In Blank’s review of the literature, she concludes that most but not all of the intergenerational correlation in welfare use can be attributed to the characteristics of daughters of welfare recipients rather than to their mothers’ welfare use.3 However, such an evaluation depends heavily on assumptions about which of a daughter’s characteristics are caused by her mother’s welfare receipt and which are not.

2. The Welfare Reform Act of 1996 Bill Clinton ran for president in 1992 promising to “end welfare as we know it.” In part, this promise reflected the issues discussed earlier. Clinton and many Americans wanted to end welfare as an entitlement and ensure that welfare recipients who were able to work actually did so. Despite concerns raised by liberals within his own administration, in 1996 he signed the Personal Responsibility and Work Opportunity Reconciliation Act (PRWORA), promising to amend some aspects after the 1996 election. However, it would be a mistake to believe that welfare reform began with the passage of PRWORA. Beginning in 1992, states began to apply for waivers from federal welfare regulations in order to experiment with alternatives to the standard AFDC program. Twenty-nine states implemented some form of waiver before adopting Temporary Assistance to Needy Families (TANF).4 Below we discuss some of the key features of PRWORA and the debates surrounding them before and since the passage of the act. The debate has continued in part

2. Peter Gottschalk, “Is the Correlation in Welfare Participation across Generations Spurious?” Journal of Public Economics 63 (December 1996): 1–25. 3. Rebecca M. Blank, It Takes a Nation (Princeton, NJ: Princeton University Press, 1997), 157. 4. Robert Schoeni and Rebecca Blank, “What Has Welfare Reform Accomplished? Impacts on Welfare Participation, Employment, Income, Poverty, and Family Structure” (Working Paper 7627, National Bureau of Economic Research, Cambridge, MA, 2000).

Welfare Reform 245 because many of the provisions were scheduled to end in 2002. In order to continue those provisions, Congress had to reauthorize them. Instead, until 2006 Congress merely extended the act without reauthorizing TANF.

2.1. TANF versus AFDC. PRWORA abolished AFDC and replaced it with TANF. Other than the new name, there are two major components to the change. First, welfare is no longer what is known as an “entitlement.” Under AFDC, anyone who met the federal eligibility criteria was entitled (had the right) to received welfare. The states provided funding for AFDC under federal guidelines and received matching funds from the federal government. Under TANF, states receive a fixed amount of money (a block grant) from the federal government to be used subject to considerably less federal regulation. Thus, states can design their own welfare programs subject to some limitations. In particular, states are free to set requirements for eligibility. They may require recipients to work or participate in training programs, or they may set other requirements, such as requiring teen mothers to live at home or stay in school. TANF comes in the form of a fixed grant that does not depend on how much the state actually spends on welfare. States therefore have a stronger incentive to try to reduce welfare spending. Any reductions in spending can be used for other programs or returned to taxpayers. PRWORA does require states to maintain a certain level of spending based on their historical level of spending but does not require that it increase with inflation. Most important, federal spending on TANF does not vary with the state of the economy or the number of people who are eligible for welfare. Under AFDC, when the economy was performing poorly, more households would be eligible for welfare. As states added families to their welfare rolls, they would receive more support from the federal government. Under the block grant structure of TANF, no such relation exists. If the economy is doing well and the number of families receiving TANF falls, states have more money to spend on each recipient. Conversely, in bad times, states will have to draw on their own resources, cut benefits, or restrict the number of recipients.

2.2. Work Requirements. We have seen that the high implicit tax rates under AFDC discouraged work. One of the reasons for bipartisan support for the earned income tax credit (EITC) is that it encourages work. However, even with EITC, the disincentive to work under AFDC was strong. Therefore, PRWORA made working a requirement for receiving welfare. Under AFDC, mothers with children under three years old were exempt from any work requirement. PRWORA lowered this to one year, and states may choose a lower age. After two years of receiving TANF, unmarried recipients must work at least thirty hours per week. In two-parent families, the parents must work a total of thirty-five hours per week. At least 50 percent of recipients (and 90 percent of two-parent family recipients) in a state must meet these work requirements, although states can get credit toward this percentage by reducing the number of recipients in the state. States that fail to meet this requirement will be penalized by having their block grant reduced.

246 Chapter 9 Because caseloads have fallen dramatically, as we will discuss in detail later, the 50 percent requirement has not been a significant concern for states. Prior to the 2006 reauthorization, the 50 percent was reduced by 1 percentage point for each percentage point reduction in the caseload relative to 1995. Because caseloads have fallen, on average, by 50 percent, many states did not need any recipients to meet the work requirement in order to avoid the reduction in their block grant.5 Nationally, in 1996, 8.2 percent of mothers were on AFDC/TANF. By 2000, this proportion had fallen to 3.9 percent. Of these, over half worked at some point during the calendar year.6 For many of these workers, employment was intermittent and part-time, so they were not working enough hours to meet the federal work requirement. Still, it is evident that the reduction in caseloads ensured that most states had no difficulty meeting the federal mandate. States also have some freedom in how they define the work requirement. Work can include up to twelve months of vocational training and, for teenagers, full-time high school education. However, recipients in vocational training and teenagers in high school cannot account for more than 20 percentage points toward the 50 percent requirement. Employment may be in the private or the public sector, may be subsidized or unsubsidized, and may include community service. States determine the penalties families face if they refuse to work. The majority of states initially reduce benefits, but nineteen stop benefits for some period. Continued noncompliance results in heavier penalties. In seven states, family benefits can be eliminated permanently.7 In 1998, on average, over one hundred thousand families per month were penalized by having their benefit reduced, and about twenty-five thousand lost their entire benefit.8 Defining what constitutes work has been one of the most controversial aspects of the law. Some people support expanding the use of education as a form of work so that everyone, not just teens in full-time high school education, could count education hours toward the work requirement. Similarly, there are both advocates and opponents of expanding the use of vocational training.9 The debate over vocational training is closely related to the debate over work first versus training that we discussed in the section on training programs. Advocates of reducing the use of vocational training programs as a form of work believe that the

5. Ron Haskins and Paul Offner, “Achieving Compromise on Welfare Reform Reauthorization” (Policy Brief, Welfare Reform and Beyond 25, Brookings Institution, Washington, DC, 2003). 6. Brian J. O’Hara, Work and Work-Related Activities of Mothers Receiving Temporary Assistance to Needy Families: 1996, 1998, and 2000, Current Population Report P70-85 (Washington, DC: U.S. Census Bureau, 2002). 7. Vee Burke and Gene Falk, “TANF Sanctions—Brief Summary” (Congressional Research Ser- vice, Washington, DC, 2001). 8. General Accounting Office, “Welfare Reform: State Sanction Policies and Number of Families Affected” (General Accounting Office, Washington, DC, 2000). 9. For a good summary of the debate, see Ron Haskins and Rebecca M. Blank, “Five Years after Welfare Reform: An Agenda for Reauthorization,” in Rebecca M. Blank and Ron Haskins, eds., The New World of Welfare (Washington, DC: Brookings Institution, 2003).

Welfare Reform 247 programs are not effective and that we should move welfare recipients into jobs where they can obtain real work experience. Supporters of expanded use of vocational training programs argue that the longer-term programs have proved cost-effective and, in the long run, have a better chance of permanently removing recipients from the welfare rolls. AFDC created strong work disincentives, because benefits were reduced sharply when respondents had other income. As caseloads fell (partially because of the effects of PRWORA and partially because of the booming economy), many states chose to allow TANF recipients to keep more of their income (i.e., reduce TANF benefits less). While this helped states meet the objective of encouraging work, it also encouraged recipients to stay on TANF.10 The reauthorization of TANF in 2006 came with surprisingly few changes. As before, states must meet the 50 percent (90 percent for two-parent families) work requirements but can count caseload reductions toward the work requirements. The major change is that the base against which the reduction is calculated has been reset to 2005. Because the proportion of recipients who meet the work requirement is well below 50 percent, unless there is a further surprising reduction in caseloads, states will need to make a considerable effort to meet the 50 percent requirement.

2.3. Time Limits. PRWORA limited TANF receipt to a total of sixty months, although states could provide noncash support to people whose eligibility for TANF had expired. In addition, five years was a maximum. States could choose a shorter time limit, and some chose to do so. Finally, the law recognized that there might be a case for exceptions. States may exempt up to 20 percent of their caseloads from the time limit. Twenty-three states applied the sixty-month limit. Seventeen states chose shorter limits, while eight use state funds to provide reduced support beyond the sixty months and three have effectively no limit.11 The time limit was and remains one of the most controversial elements of PRWORA. Advocates view it as essential to ensure that recipients will work to get off welfare. They argue that, knowing that their benefits will run out, recipients will not look for ways to avoid the work requirements. Instead, they will look for ways to make themselves self-sufficient. Opponents of time limits raise a number of concerns. At the time the bill was passed, critics said that it would make “a million kids go hungry.” When parents lose their welfare benefits, children lose benefits as well. This may happen either because the state disqualifies the entire family or because families share the entire benefit. When the parents lose their benefits, everyone in the family has less on which to live. Many opponents of time limits also argue that anyone who has received TANF for five years will also have been working. They believe that if people continue to work,

10. Rebecca M. Blank, “Evaluating Welfare Reform in the United States,” Journal of Economic Literature 40 (December 2002): 1105–66. 11. Gordon L. Berlin, “What Works in Welfare Reform: Evidence and Lessons to Guide TANF Reauthorization” (Manpower Development Research Corporation, New York, 2002).

248 Chapter 9 they should also continue to receive support if they need it because of either low pay or short-term joblessness. Time limits should reduce welfare use in two ways. The direct route is that people who previously would have remained on welfare beyond the term limit are no longer able to do so. The indirect route is that people give up welfare in order to have the option of using it later should the need arise. A family receiving only a small TANF payment might choose to remove itself from TANF so that it would have access to TANF if its income fell. Or a family member might accept a work opportunity that she would otherwise have rejected. Unfortunately, there are no experiments in which the sole intervention was to establish time limits. However, Jeffrey Grogger and Charles Michalopoulos point out that time limits affect families differently depending on the age of the youngest child in the family.12 TANF is available only to families with a child under age 18. If the youngest child in the family is over 13 years old, a family with sixty months of TANF eligibility need never worry about exhausting its time limit (assuming that it expects not to have any more children). In contrast, a family with younger children must be concerned about the time limit, and the younger the child, the more important this concern should be. Florida’s Family Transition Program (FTP) randomly assigned AFDC recipients to a program with a work requirement, more generous benefits, and a time limit of two or three years depending on the circumstances. The control group remained on traditional AFDC. The direct effect of the changes except for the term limits should have increased welfare use. Grogger and Michalopoulos show that among participants whose youngest child would turn 18 before the end of the time limit, the experimental group was on welfare more frequently than the control group. This difference was smaller among participants with younger children who therefore might exhaust their benefits before the youngest child turned 18. This is consistent with the idea that families with younger children but not those with only older children will exit welfare in order to maintain future eligibility. Perhaps poor families with younger children respond differently to welfare reforms than do poor families with older children. Grogger and Michalopoulos study other experiments that were similar but did not have time limits and do not find that the age of the youngest child is related to the effect of the reforms. Their best estimate it that the time limits reduced welfare use by 16 percent the first two years, even though no participant hit the time limit in that period. This is likely to be an underestimate, because participants knew that they were in an experiment. Some of the families assigned to the experimental group probably did not believe that the two- or three-year limit would actually be applied after the end of the experiment. As we will discuss in greater detail later, it is difficult to determine the overall effects of welfare reform, let alone separate the effects of its various elements. Nevertheless, we

12. Jeffrey Grogger and Charles Michalopoulos, “Welfare Dynamics under Time Limits,” Jour- nal of Political Economy 111 (June 2003): 530–54.

Welfare Reform 249 can derive some indication of the magnitude of the effect of time limits by examining welfare participation rates for families unaffected by the five-year limit (those whose youngest child is at least 13 years old) and those with younger children. Moreover, we can also use variation across states in the timing and implementation of time limits. Six states adopted time limits early (one in 1993 and five in 1995), while two used state funds to get around the federal five-year limit.13 Consistent with the results from the FTP, relative to states without time limits, welfare participation in states with time limits dropped more rapidly among households with younger children than among those with older children. One estimate suggests that time limits accounted for about one-eighth of the decline in welfare use between 1993 and 1999, but this estimate is very imprecise.14

2.4. Child Support. PRWORA increased federal involvement in enforcing payment of child support by absent parents. The law set up mechanisms to make it easier to track parents across state lines and enforce payment of interstate child support. The law encouraged states to set up better tracking systems and to establish paternity at birth. Under AFDC, the custodial parent was able to keep the first fifty dollars per month of child support, but the rest effectively went to the state. No such pass-through is required under PRWORA. If a woman leaves welfare and succeeds in collecting overdue child support, half of what she collects goes to the state.

2.5. Child Care. PRWORA dramatically increased funding for child care and folded programs into a Child Care Development Block Grant. Moreover, states could use their TANF block grant to support child care. These provisions are relatively uncontroversial. However, there is considerable controversy over whether and how much they should be increased.15

2.6. Teen Pregnancy and Marriage. Under PRWORA, minors are eligible for welfare only if they live at home and participate in a training program or attend school. There is also funding for programs that discourage teen pregnancy and promote marriage. These programs are politically popular, because it is hard to oppose reducing teen pregnancy and encouraging marriage. Despite the evidence presented earlier, most people believe that delaying pregnancy until adulthood is good and that if it is not accompanied by spousal abuse, marriage is beneficial. The main problem is that even those who believe that single-parent households are a major problem are at a loss to come up with programs that reduce out-of-wedlock births and encourage marriage.16

13. Jeffrey Grogger, “Time Limits and Welfare Use,” Journal of Human Resources 39 (Spring 2004): 405–22. 14. Jeffrey Grogger, “Time Limits and Welfare Use” and “The Effects of Time Limits, the EITC, and Other Policy Changes on Welfare Use, Work, and Income among Female-Headed Families,” Re- view of Economics and Statistics 85 (May 2003): 394–408. 15. Haskins and Offner. 16. Charles Murray, “Family Formation Issues in Welfare Reform,” in Rebecca Blank and Ron Haskins, eds., The New World of Welfare (Washington, DC: Brookings Institution, 2001). See also chapter 6.

250 Chapter 9 There is little evidence that PRWORA affected nonmarital fertility. In chapter 6 we saw that the nonmarital fertility rate stopped increasing around 1990, well before the passage of PRWORA, and it has demonstrated no tendency to decline from its high rate. As we will discuss later, to some extent this lack of overall change obscures much more complex changes and differing effects on various groups in the population.

2.7. Immigrants. Perhaps the most controversial element of PRWORA was that it denied public assistance to many legal immigrants. Immigrants who had been in the United States for fewer than five years and who were not citizens were ineligible for TANF and many other forms of aid. Sponsors of immigrants became legally liable for the immigrants if they needed financial support. Refugees continued to be eligible for most forms of public assistance. Immigrants who arrived before PRWORA was passed could be eligible for assistance in some cases. In fact, every state except Alabama made these immigrants eligible for TANF, and every state except Wyoming made them eligible for Medicaid. There was considerably more interstate variation in the provision of other forms of assistance and in the treatment of immigrants who arrived later.17 The fact that states treated immigrants differently and that different types of immigrants were treated differently allows us to measure the effect of assistance on recipients. George Borjas found that food insecurity increased more rapidly for immigrants in states that treated immigrants less generously than in those that treated them more generously.18 This was true even though food security among natives fell more rapidly in these states. Borjas estimates that about half of all immigrant families that lost public assistance as a result of welfare reform became food insecure.

3. Assessing the Effects of Welfare Reform Figuring out the effects of welfare reform is difficult, because it was enacted at a national level. If we observe (as we do) that the number of people on welfare fell after welfare reform was passed, we cannot be sure whether the change would have been the same, larger, or smaller without welfare reform. Perhaps the decline was due to the booming economy or to changes in the EITC. Of course, we can try to look at groups that should have been affected by welfare reform and those that should not have been, but it is not easy to find good comparison groups. To some extent, we can get around this problem, because states enacted welfare reform at different times. We have already noted that the majority of states had implemented some major welfare waiver before they enacted TANF. There were also smaller differences in when states implemented TANF. So one approach might be to ask

17. Wendy Zimmermann and Karen C. Tumlin, “Patchwork Policies: State Assistance for Immi- grants under Welfare Reform” (Occasional Paper 24, Urban Institute, Washington, DC, 1999). 18. George J. Borjas, “Food Insecurity and Public Assistance,” Journal of Public Economics 88 (July 2004): 1421–43.

Welfare Reform 251 whether welfare receipt declined earlier in states that used waivers and in states that were quicker to implement TANF. There are two major drawbacks to this approach. The first is that differences in timing were modest. We might expect that a five-year time limit would not have its full effect for five or more years. Comparing welfare loads in 1996 for states adopting term limits in 1995 with those adopting them in 1997 may not tell us very much about the long-term effects of such limits. The other problem is that state laws are not assigned randomly. Just as workers who have been having trouble obtaining a good job are more likely to seek training, states that have experienced a sharp increase in their welfare caseloads may be more likely to seek waivers and enact welfare reform rapidly. The Ashenfelter dip would make it look as if welfare reform reduced their caseloads. A third approach is to rely on actual randomized experiments from the period leading up to the passage of PRWORA. We have discussed some of these experiments in earlier chapters. As we have discussed, experiments are very valuable, but the effect of a small-scale temporary experiment may be different from the effect of a long-term nation- wide policy change. Although each of these approaches is problematic, when a variety of approaches lead to the same conclusion, the simplest explanation is that the conclusion is correct. However, because I generally cannot point to a single compelling study, in what follows, I will generally present a very simple analysis of the data and then provide a summary of the weight of the evidence in the literature.

4. Effect on Welfare Receipt Figure 9.1 shows the evolution of the fraction of the population on AFDC/TANF over the past forty years. It also plots the poverty rate over that period. Even a quick glance reveals that the proportion of the population on welfare is heavily influenced by the welfare rules. From the early 1960s until the mid 1970s, the percentage of Americans on welfare grew rapidly, from less than 2 percent to over 5 percent even though this was a period when the poverty rate was falling rapidly. To a large degree, this reflects the expansion of AFDC under the Great Society programs. AFDC receipt then gradually declined from 1973 to 1989, when it again increased rapidly. It is tempting to ascribe the post- 1989 increase to the expansion of the AFDC-UP program, but although the number of two-parent families receiving AFDC increased dramatically over this period, two- parent families remained a small minority of AFDC recipients. Single-parent families accounted for most of the increase. Looking at the figure also makes it difficult to argue that PRWORA was not at least partially responsible for the decline in caseloads. It is certainly true that the expansion of the economy contributed to the decline in caseloads. Although TANF did not begin until July 1997, caseloads peaked in 1993–1994. However, by 2000, the caseload as a fraction of the population had returned to its 1962 level and was well below its 1973

252 Chapter 9 Percentage of Poverty Rate People on Welfare 6 20 Percentage of People on Welfare 5

4 15

Poverty 3 Rate

10 2 1960 1965 1970 1975 1980 1985 1990 1995 2000 Figure 9.1 AFDC/TANF Caseloads (1962–2003) Source (for percentage on welfare only): Office of the Assistant Secretary for Planning and Evaluation, Indicators of Welfare Dependence: Annual Report to Congress, 2005 (Washington, DC: U.S. Department of Health and Human Services, 2005), http://aspe.hhs.gov/hsp/ indicators05/apa.htm#tbt1. peak, despite the fact that the official poverty rates in 1973 and 2000 were similar. Al- though the poverty rate increased after 2000, the caseload continued to decline. Figure 9.2 suggests that part of the decline in caseloads was driven by the economy. From 1967 to 1973, because AFDC was liberalized, caseloads increased even though women’s earnings increased. After 1973, caseloads tended to decline as the earnings of low-earning women increased. Thus, we would expect the 1990s boom to have lowered caseloads. However, this is not the entire story. Even if one projects welfare rates based on the post-1973 experience, recent very low welfare rates must be attributed to some factor other than the economy, of which welfare reform is the most plausible. This simple analysis is consistent with the general findings in the literature, which shows some evidence, though it is not overwhelming, that welfare reform reduced welfare use.19 Of course, welfare is not the only social assistance program. In chapter 3 we discussed the fact that use of disability insurance (supplemental security income, or SSI, for the blind and disabled) grew rapidly following liberalization of eligibility requirements in 1984. Growth also accelerated in the early 1990s before slowing in the mid- to late 1990s. Shifting recipients from the welfare rolls to SSI is attractive to states, because SSI is paid for entirely by the federal government, whereas AFDC payments were only partially federally funded and any savings under TANF can be used by the

19. Jeffrey Grogger, Lynn A. Karoly, and Jacob Alex Klerman, Consequences of Welfare Reform: A Synthesis (Santa Monica, CA: Rand Corporation, 2002).

Welfare Reform 253 Percentage on Welfare 6

1993 1994 1976 1992 1973 1975 1995 1972 1977 5 1974 1981 1991 1978 1971 1980 1984 1986 1996 1979 1987 1990 1983 1985 1989 1982 1988 4 1997

1970

1998 3 1969 1968 1999 1967 2000 2001 2 2002

1,0001,500 2,000 2,500 3,000 Tenth Percentile Earnings Figure 9.2 Welfare Rate and Earnings of Low-Earning Women (1973–2002)

state. After welfare reform, recipients may prefer SSI because it is not subject to the time limit or to work requirements. There is some casual evidence that the early 1990s spurt in disability receipt was related to welfare reform. The growth in the use of disability insurance in this period was concentrated among single mothers, particularly in states with major welfare waivers.20 However, the fact that the growth slowed after PRWORA suggests that some other force was probably at work as well. If states that adopted welfare waivers were most concerned about the growth of welfare receipt, they may also have worked harder to shift recipients from welfare to disability insurance. It is important to recognize that reducing the welfare rolls is not the same thing as reducing poverty. We could eliminate everyone from the welfare rolls simply by eliminating welfare. Therefore, in the next section we look at the effect of welfare reform on well-being.

5. Welfare Reform and Well-Being As we have already discussed, it is difficult to know what the effects of PRWORA have been beyond reducing caseloads. Most of the outcomes that might be affected by welfare reform are also heavily affected by other trends in society. Thus, proponents of welfare reform hoped that it would encourage adolescents to remain in school, while its opponents feared that reduced parental supervision would increase the dropout rate.

20. Lucie Schmidt and Purvi Sevak, “AFDC, SSI, and Welfare Reform Aggressiveness: Caseload Reductions versus Caseload Shifting,” Journal of Human Resources 39 (Summer 2004): 792–812.

254 Chapter 9 Labor Force Participation Rate 90

80 Widowed, Divorced, or Separated

70 Married

50 Never Married

1980 1985 1990 1995 2000 2003 Figure 9.3 Labor Force Participation of Women with Children (by Marital Status) (1980–2003) Source: U.S. Census Bureau, Statistical Abstract of the United States (Washington, DC: U.S. Census Bureau, various years). Note: Data are for every five years until 1990, for every year thereafter.

However, the period of welfare reform has coincided with considerable education reform in the United States. It is difficult, if not impossible, to disentangle the effects of welfare reform, movements in the economy, and education reform on high school graduation rates. We can examine changes in who receives welfare and the behavior of those on welfare. From 1996 to 2000, the proportion of adult recipients who were employed increased from 11 percent to 26 percent, suggesting that the law has had some effect on increasing employment among recipients, but it has since declined to 23 percent (as of 2003).21 The decline may reflect the weaker economy, or those individuals who have not managed to get off welfare may also be those who have particular difficulty finding employment. Further evidence of welfare reform’s effect on employment comes from the experience of single mothers. Overall employment of single mothers and of never-married single mothers has increased dramatically.22 Figure 9.3 shows the labor force participation rates (LFPRs) of women aged 18 years and over who have children, by marital status. The LFPR of married women with children increased dramatically in the 1980s (continuing an earlier trend) and leveled off in the second half of the 1990s. In contrast, the LFPR of never-married mothers was essentially constant until the early 1990s and exploded from 1995 to 1998, when it surpassed the LFPR of married mothers. The LFPR of widowed, divorced, and separated

21. Office of the Assistant Secretary for Planning and Evaluation, table TANF 7. 22. Robert A. Moffitt, “From Welfare to Work: What the Evidence Shows” (Welfare and Beyond Poverty Brief 13, Brookings Institution, Washington, DC, 2002).

Welfare Reform 255 mothers is higher throughout this period than the LFPR of married or never-married mothers. Still, it started to increase significantly in the mid-1990s. It is difficult to examine figure 9.3 without attributing some of the increase in work activity to PRWORA. Never-married mothers are much more likely and formerly married mothers somewhat more likely than married mothers to be on TANF. The timing and the magnitude of the increase in the LFPR of single and previously married women is consistent with a large effect of welfare reform. Of course, some of this increase probably reflects the booming economy and the greater attractiveness of work because of the increase in the EITC. However, the LFPR of never-married mothers continued to increase even after the boom. Historically, women who left AFDC had employment rates of between 48 percent and 65 percent one year after leaving. For women leaving TANF, early studies showed that the rate has been 60 percent. Of course, the booming economy contributes to this high employment rate.23 However, because TANF involves some pushing of people into work, we would expect that more of the women leaving TANF would be less qualified and thus have low employment rates. Again, this simple analysis is consistent with the general literature, which contains some evidence that welfare reform increased employment among the target population.24 Of course, we are concerned not only with employment rates but also with income. Women who left welfare report an average hourly income of less than $7.00. Less than one-fourth of those working have health insurance. About three in ten work thirty-five hours per week or less.25 Overall income gains for those working are probably around 20 percent relative to TANF, much of it due to higher earnings among other family members and the EITC.26 However, given the small absolute gains, much of the increased income is probably offset by increased expenses. Figure 9.4 supports the view that there were probably modest gains in income for families affected by welfare reform. The upper line compares twenty-fifth percentile incomes for female-headed households with children under the age of 18 with the tenth percentile incomes for female-headed households with no children under the age of 18. Note that these households generally have incomes about the threshold at which they would no longer receive TANF, but households without children are not eligible for TANF regardless of their incomes. We see from the figure that after 1997, twenty-fifth percentile income rose rapidly for female-headed households with children relative to the comparison group without children. It is difficult to determine how much of the improvement in the relative earnings of female-headed households with children is due to the economy and whether any of it reflects welfare reform, but the timing is suggestive.

23. Moffit, loc cit. 24. Grogger, Karoly, and Klerman, 77–98. 25. Pamela Loprest, “Families Who Left Welfare: Who Are They and How Are They Doing?” (Assessing the New Federalism Discussion Paper 99-02, Urban Institute, Washington, DC, 1999). 26. Moffitt.

256 Chapter 9 Income Relative to Tenth Percentile, Female-Headed Families with No Children Present

1.0 Twenty-fifth Percentile

0.8

0.6 Tenth Percentile

0.4

Fifth Percentile 0.2

1988 1990 1992 1994 1996 1998 2000 2002 Figure 9.4 Family Incomes of Female-Headed Families with Children Relative to Tenth Percentile Female-Headed Families without Children (1987–2003) Source: Author’s calculations from the Current Population Surveys, http://www.bls .gov/cps/.

The bottom two lines in figure 9.4 show the incomes of tenth and fifth percentile female-headed families with children relative to tenth percentile female-headed families without children. The tenth percentile line is basically flat over this period. The fifth percentile line shows some downward movement, but it is small. Most studies of the effect of welfare reform are based on data collected before the time when many families would be hitting their time limit. Subject to that qualification, such studies have tended to find evidence that welfare reform increased incomes.27 The bottom line in figure 9.4 raises the concern that welfare reform may have affected some families adversely. If 60 percent of welfare leavers are working, 40 percent are not. Many of these women report that they are unable to work for health reasons or are unable to find a job. Overall, there are signs of significant economic distress among those who have left welfare. A third reported having had to skip or cut down the size of meals in the last year because of a lack of money, six in ten were worried that food would run out, and about half actually experienced running out of food before the end of the month. Four in ten were unable to pay rent, mortgage, or utility bills sometime during the year.28 More recent studies suggest that some of the positive results reflected the booming economy. Employment among welfare leavers fell to 42 percent in 2002, and 14 percent reported no source of income.29

27. Grogger, Karoly, and Klerman, 149–82. 28. Loprest, 20–21. 29. Pamela Loprest, “Fewer Welfare Leavers Employed in Weak Economy” (Snapshots of Amer- ica’s Families 5, Urban Institute, Washington, DC, 2003).

Welfare Reform 257 There is evidence that welfare receipt is increasingly concentrated among the types of people who will have great difficulty finding employment. From 1996 to 2003, the percentage of welfare recipients living in public housing grew from 9 percent to 19 percent.30 Welfare recipients are increasingly concentrated among blacks and Hispanics and among inner-city residents.31 Perhaps of greatest concern is the dramatic increase in the proportion of families receiving welfare that are child-only families. This proportion almost doubled to 41 percent over the seven-year period. Child-only families are either children in some form of foster care or families in which the adults have been disqualified from support. The increase in the number of such families raises the possibility that families are giving up children or that many children are living in families with greatly diminished support. In fact, however, much of the increase in the proportion of child-only families can be attributed to the decline in parent-present families on the rolls. The absolute number of child-only family cases has actually fallen by about one-fifth.32

6. Living Arrangements The primary goals of PRWORA included encouraging marriage and increasing the number of children growing up in two-parent families. At this point, it is difficult to draw a firm conclusion as to whether there are important effects that vary across subgroups of the population or whether welfare reform simply had no consistent effects on marriage. For example, one study that uses differences in the timing of welfare reform across states finds that the effect of TANF was quite complex for black inner-city children.33 TANF reduced the probability that they lived with an unmarried parent or with a parent and grandparent but increased the probability that they lived with neither parent. Yet the same study found no significant effects for Hispanic and non-Hispanic white children. Moreover, the study found that the shift from AFDC to TANF was associated with a higher rate of having never been married among black inner-city women but also with their being more likely to be divorced, with no effect on the overall marriage rate. It would not be surprising if welfare reform changed people’s preferred living arrangements so that we would simultaneously see more marriages dissolving but also more marriages forming. This is consistent with the result described in the previous paragraph. Although TANF had no overall effect on the long-term marriage rate, it encouraged marriage among some of those who would not marry in pre-TANF times and thus reduced the fraction never married. At the same time, some of the women who preferred to be married under AFDC preferred not to be married under TANF, which increased the fraction of women who were divorced.

30. Office of the Assistant Secretary for Planning and Evaluation, table TANF 7. 31. R. Kent Weaver, Ending Welfare as We Know It (Washington, DC: Brookings Institution, 2000). 32. Office of the Assistant Secretary for Planning and Evaluation, table TANF 1. 33. Marianne P. Bitler, Jonah B. Gelbach, and Hilary W. Hoynes, “The Impact of Welfare Re- form on Living Arrangements,” Journal of Human Resources 41 (Winter 2006): 1–27.

258 Chapter 9 Surprisingly, one study found the opposite. Welfare reform was associated with both less divorce and less marriage.34 It may be that the uncertainty surrounding welfare reform discouraged people from taking steps that might prove costly to reverse. But rec- onciling the two studies is not straightforward. At this point, it is safe to conclude that welfare reform did not have large and consistent effects on living arrangements.

7. Effects on Children and Adolescents One of the major debates over welfare reform concerned the potential effect on children and adolescents. Advocates of welfare reform believed that children would benefit from better role models, increased family income, and perhaps even better daycare. Opponents maintained that if their mothers found work, children would lack adequate support and supervision from parents who were either absent or tired. If their mothers did not find work, the children would suffer from the family’s loss of income. We have seen that employment rose among women who would be likely to have received AFDC. On the other hand, the proportion of poor children who received AFDC/TANF fell from about 60 percent in 1996 to just over 30 percent in 2003.35 Together these facts suggest that there was an element of truth to the positions of those on both sides of the debate. The evidence so far is mixed, probably in part because some aspects of reform tend to have positive effects and some negative, and in part because reforms affect children of different ages differently. Thus, mandating work has inconsistent effects across studies.36 One review of seven experimental studies found that for school-age children, the effect of welfare reform tended to be positive when it improved the economic status of the family by increasing income and that it tended to be negative when it did not. Moreover, in some cases, programs that provided mothers with increased skills through training and education also improved the cognitive and educational outcomes for school-age children. In contrast to the positive effect of reforms on school-age children, the reforms were more likely to have negative effects on adolescents.37 Welfare reform and, in particular, mandated work-related activities, are associated with worse outcomes for adolescents with respect to both behavior and education.38

8. Concluding Thoughts Welfare reform has been neither the disaster that its opponents feared nor the success that some of its proponents claim. Beyond this, it is too early to evaluate its impact

34. Marianne P. Bitler et al., “The Impact of Welfare Reform on Marriage and Divorce,” De- mography 41 (May 2004): 213–36. 35. Office of the Assistant Secretary for Planning and Evaluation, table TANF 2. 36. Grogger, Karoly, and Klerman, xviii. 37. Martha J. Zaslow et al., “Experimental Studies of Welfare Reform and Children,” The Future of Children 12 (Winter/Spring 2002): 79–95. 38. Grogger, Karoly, and Klerman, xiv.

Welfare Reform 259 fully. We are only beginning to know what happened to TANF recipients in the post- boom era. Nevertheless, there are a few points worth making. First, the implementation of PRWORA was greatly facilitated by the boom of the 1990s. This made it far easier for states to reduce their TANF rolls. It also encouraged them to be more generous with payments, so TANF recipients were able to keep more of the income they earned. This undoubtedly allowed them to substitute goods for activities they would normally have undertaken themselves (e.g., buying prepared meals instead of cooking meals). Second, welfare reform involved trade-offs. There are without doubt some people who are hurt by the time limits and other elements of welfare reform. There are also without doubt people who are pushed to find jobs and end up dramatically improving their economic circumstances. Even if we were able to quantify fully the numbers of people benefiting and being hurt, policy analysts would continue to disagree about welfare reform because of the relative importance they assign to different effects. Finally, although this chapter has focused on PRWORA, welfare reform is a process that predates PRWORA and is ongoing. One of the great benefits of the welfare reform process has been that when states were given the opportunity to experiment, they did so. And they were required to evaluate the outcomes of their experiments. We know a great deal more about welfare than we did fifteen years ago because of these experiments.

9. Further Reading Berlin, Gordon L. “What Works in Welfare Reform: Evidence and Lessons to Guide TANF Reauthorization.” Manpower Development Research Corporation, New York, 2002. Blank, Rebecca M. “Evaluating Welfare Reform in the United States.” Journal of Eco- nomic Literature 40 (December 2002): 1105–66. Borjas, George J. “Food Insecurity and Public Assistance.” Journal of Public Economics 88 (July 2004): 1421–43. Gottschalk, Peter. “Is the Correlation in Welfare Participation across Generations Spu- rious?” Journal of Public Economics 63 (December 1996): 1–25. Grogger, Jeffrey, and Charles Michalopoulos. “Welfare Dynamics under Time Limits.” Journal of Political Economy 11 (June 2003): 530–54. Grogger, Jeffrey, Lynn A. Karoly, and Jacob Alex Klerman. “Consequences of Welfare Reform: A Research Synthesis.” Report of the Rand Corporation, Santa Monica, CA, 2002. Haskins, Ron, and Rebecca M. Blank. “Five Years after Welfare Reform: An Agenda for Reauthorization.” In Rebecca M. Blank and Ron Haskins, eds., The New World of Welfare. Washington, DC: Brookings Institution Press, 2003. Haskins, Ron, and Paul Offner. “Achieving Compromise on Welfare Reform Re- authorization.” Policy Brief, Welfare Reform and Beyond 25, Brookings Institution, Washington, DC, 2003.

260 Chapter 9 Loprest, Pamela. “Families Who Left Welfare: Who Are They and How Are They Doing?” Assessing the New Federalism Discussion Paper 99-02, Urban Institute, Washington, DC, 1999. Moffitt, Robert A. “From Welfare to Work: What the Evidence Shows.” Welfare and Beyond Poverty Brief 13, Brookings Institution, Washington, DC, 2002.

10. Questions for Discussion 1. In the early 1990s, why did many Americans believe that welfare reform was necessary? 2. On average, of two people on welfare, the one who has been on welfare longer is less likely to exit over the next several months (unless benefits are exhausted). Why? 3. The daughter of a woman who spent some time on welfare is more likely to receive welfare as an adult than is a woman whose mother did not. Why? 4. How does the change in the way that TANF is funded (relative to AFDC) change states’ incentives to provide welfare? 5. What are the issues surrounding the work requirement under PRWORA? Based on what you have learned so far in the book and elsewhere, what should the work requirement be? 6. Why might a lifetime limit on how much time someone can spend on welfare cause people to leave welfare well before they hit the time limit? 7. What were the principal elements of PRWORA? 8. Why did welfare caseloads decline in the second half of the 1990s? 9. Did welfare reform increase labor force participation among low-skilled women with children? 10. Did welfare reform help or hurt women who would have been eligible for AFDC? 11. Did welfare reform make “a million kids go hungry”?

Welfare Reform 261 This page intentionally left blank part 2 DISCRIMINATION This page intentionally left blank chapter 10

Discrimination: Theory

I          . It will become evident that much of what we discuss could, with only minor changes, apply to sex discrimination or discrimination against members of other groups. We focus on race in this chapter because we will address theories and issues that are specific to sex discrimination later. You may wonder why we worry about theories of discrimination. There are a number of reasons. The first is that these theories help us interpret the data on discrimination. Certain theories imply that even if individual employers discriminate, black and white workers will have similar wages. If these theories are correct, evidence of discrimination by individual employers is insufficient to prove the existence of discrimination in labor market outcomes. The second is that if we do find evidence of discrimination, theory can suggest ways to address it. Some theories suggest that affirmative action will be helpful, and others suggest that it will worsen discrimination. Before we look at the evidence, it is helpful to understand how discrimination can arise. We will not engage in an extended discussion of the definition of race. It is generally (although not universally) recognized that race is socially rather than biologically determined. In the United States, under the so-called one drop of blood rule, people with identifiable African ancestry are generally viewed as black. Those with only Euro- pean ancestry are white. In the nineteenth century, fine gradations of race were commonly recognized. As recently as 1920, the U.S. Census distinguished among whites, blacks, and mulattos rather than simply between blacks and whites. At that time, people of southern European ancestry were not universally recognized as white. Although we will use the term race freely in the next few chapters, you should remember that con- cepts of race can and do change over time.

1. What Is Discrimination? Before we study discrimination, we must begin by defining it. This is a surprisingly difficult task. We will need to distinguish between discriminatory attitudes, which we will

265 call prejudice, and discriminatory outcomes, which we will call discrimination. And we will need to distinguish between segregation and discrimination.

1.1. Discrimination and Prejudice. Consider the case of a taxi driver who is driving down a one-way street late at night in a dangerous part of town and is hailed simultaneously by people on each side of the street. It is equally easy for her to pick up either customer. On the left side of the street is a little old lady. On the right side is a tall African American teenage boy wearing a hood. The taxi driver unhesitatingly chooses the little old lady over the teenager. Did she discriminate? This question often makes my students uncomfortable. They want to avoid it by giving the taxi driver some additional reason for picking up one person or the other. “Maybe it was easier to pull to the left than to the right,” they want to suggest, but the question does not allow that option. So many students are torn. On the one hand, it seems that the taxi driver did discriminate. After all, she knew nothing about the little old lady or the teenager. The little old lady might have had a gun and intended to rob the taxi driver. The teenager might have been a nice boy who did community service in the poor part of town and was dressed the way he was in order to fit in with the people he was helping. But it is statistically more probable that the taxi driver would have been attacked by the teenager than by the little old lady. The taxi driver was clearly engaging in racial (or perhaps age) profiling, which is a form of discrimination. On the other hand, even the young black males in my classes typically admit that they would pick up the little old lady as well. The racial profiling makes sense. Provided that the taxi driver would change her decision if additional information suggested that picking up the teenager was really safer or that his need was greater (he was in more danger or physically weaker), most students feel uncomfortable labeling her a racist. What we should learn from the example is that discrimination need not imply racism, sexism, or, to use milder language, prejudice. Treating people differently on the basis of their membership in a class is discrimination, but it can arise from rational responses to imperfect information. We will thus distinguish between prejudice and discrimination. We will refer to prejudice as dislike, distaste, or misperception based on innate characteristics such as race or sex. Discrimination occurs when there is an impact on the individual so that the individual has greater difficulty getting a taxi, earns a lower wage, or is otherwise hurt by the action. Prejudice does not necessarily generate discrimination, because people may not be able to act on their prejudice. Conversely, discrimination may occur even when people are not prejudiced.

1.2. Racial Profiling and Discrimination. The distinction between prejudice and discrimination is particularly important when we consider racial profiling in the justice system. It is clear that blacks are more likely to be stopped, questioned, searched, arrested, and imprisoned for crimes than are whites. These differences may be justified on the basis of differences in rates of criminality and effective policing. But it is almost definitely the case that blacks who are innocent of a crime are more likely to be searched for that crime than are innocent whites.

266 Chapter 10 Let us take an extreme example. Suppose the police receive a report of a shooting and that a black teenage male wearing a baseball cap and a blue shirt was seen fleeing the scene. The police are likely to question any young black male wearing a baseball cap and a blue shirt whom they discover in the neighborhood. They are also likely to question some young black males who are not wearing a baseball cap or a blue shirt and may question some young white males or young black women wearing a baseball cap and blue shirt. Few of us would question the appropriateness of this behavior or claim that the police were questioning black teenagers because of their prejudice. Given the available information, the police seem to be acting in an appropriate manner (assuming that their behavior is otherwise appropriate). However, we should recognize that even this is profiling. The police, at this point, do not have direct information linking any individual to the crime. On the basis of the information they have so far, relative to questioning a random person, it is statistically more likely that if they question a young black male wearing a baseball cap and a blue shirt they will be questioning someone with information about the case. They are making a reasonable statistical inference (i.e., profiling) on the basis of race, age, sex, and clothing. Most of the people questioned will be innocent, and innocent young black men will almost certainly be much more likely to be questioned than innocent young white men. Depending on who the guilty party really is, guilty black men may or may not be more likely to be questioned than other guilty parties. The important point is that, whether or not prejudice is present in this example, discrimination is. Equals are not being treated equally. Instead they are being treated unequally on the basis of group membership.

1.3. Further Definition of Discrimination. To continue our discussion of discrimination, we turn to an example in which four firms face the same market realities but determine wages using different rules. Let us consider the case of a firm that sells cosmetics. Black sales representatives sell more of the good when selling in primarily black neighborhoods, while white sales representatives sell more of the good when selling in primarily white neighborhoods, perhaps because they feel more comfortable or because they have better contacts. De- mand for the product is stronger in white neighborhoods, so that white representatives working in white neighborhoods sell more of the product than do black representatives working in black neighborhoods, even though the two groups work equally hard. Now consider the following information about four different firms, and decide whether each discriminates.

• Firm A, knowing that blacks sell more in black neighborhoods and whites in white neighborhoods, assigns white representatives to white neighborhoods and blacks to black neighborhoods. It observes that its white representatives sell more and pays them more than it does its black sales representatives. • Firm B is identical to firm A except that its representatives are free to choose their own sales areas. Of course, because they can sell more in areas primarily inhabited by members of their own race, the representatives choose

Discrimination: Theory 267 pretty much the same areas assigned to them in the first scenario. Black representatives sell less and are paid less than are white representatives. • Firm C does not pay its sales representatives a salary but instead pays a 20 percent commission on their sales. Black representatives choose to sell in primarily black neighborhoods and white representatives in primarily white neighborhoods. Although the commission is the same for all workers, blacks on average earn less because they sell less on average. • Firm D does not hire any sales representatives at all. Instead, any person can choose to become an independent distributor of the firm’s products. The firm sets retail and wholesale prices for its products. The retail prices are marked up 25 percent from the wholesale prices. Independent distributors meet with customers, who choose the products from a catalogue. The distributor earns the 25 percent mark-up on each sale. Because blacks’ opportunities in the labor market are not as good, on average, as those of whites, blacks become independent distributors in sufficiently large numbers relative to whites that the black distributors earn less than do white distributors. The courts would almost certainly rule that Firm A was discriminating, and most people would agree. The two groups are putting in equal effort and doing the same job. Experience suggests that most people believe that blacks and whites should receive the same salary, especially because the company assigned the sales area for which each representative is responsible. To some people, the fact that Firm B does not assign the neighborhoods makes it different from Firm A, because it establishes that the lower sales by black representatives cannot be blamed on the company’s decision about sales areas. However, in fact, the company’s decision did not constrain representatives. Firm A chose the sales areas that each representative preferred. Generally speaking, people and the courts are much more comfortable with Firms C and D. In each case, the firms are clearly following a rule that is independent of race. However, it is easy to construct an example in which, except for the words used to describe company policy, the four firms are behaving identically. Suppose that retail sales for all four firms are $2,500 for whites and $2,250 for blacks. At Firm C, whites receive a commission of $500, while blacks receive $450. At firm D, whites buy $2,000 worth of goods that they “mark up” to $2,500, for a profit of $500, while blacks buy the goods for $1,800 and sell them for $2,250, for a profit of $450. If Firms A and B both offer salaries of $500 to whites and $450 to blacks, there really is no difference in how the firms are behaving. If one firm is discriminating, all four must be viewed as discriminating. Therefore, we will use discrimination to refer to any situation in which blacks and whites with identical observed characteristics have systematically different outcomes. Note that, in contrast with what frequently matters in legal cases, intent is unimportant in this definition. It is the outcome that matters.

1.4. Segregation. We will also distinguish between discrimination and segregation. Segregation is the separation of people on the basis of race. If blacks and whites work in different firms or types of job or live in different neighborhoods, they are segregated.

268 Chapter 10 Segregation and discrimination may go together, but there can be discrimination without segregation and, at least conceptually, segregation without discrimination. If blacks and whites lived in different neighborhoods because all whites preferred one type of neighborhood and all blacks preferred another, we would observe segregation but not discrimination. Unless we believe that living in an all-black neighborhood is inherently better or worse than living in an all-white neighborhood, segregation need not entail discrimination. If blacks are forced to live in neighborhoods that are demonstrably worse than white neighborhoods, the segregation is discriminatory. On the other hand, if we have black and white slums and black and white middle-class neighborhoods, we cannot automatically conclude that the segregation is discriminatory.

2. Theories of Discrimination: Prejudice The models of discrimination we will study are highly stylized, but each captures an element of the real world. They allow us to investigate whether explanations for discrimination that appear plausible can stand up to the discipline of the market. Gary Becker won the Nobel Prize in economics in part for his work on discrimination.1 As one of the pioneers in the field, it was natural for him to start with explanations of discrimination that were based on prejudice, and we use those as our starting point. Throughout the discussion we will assume that blacks and whites are equally productive innately. We will review what happens when each of three groups, employers, white workers, and consumers, is prejudiced against black workers. In the first two cases, we will see that if the labor market works smoothly otherwise, prejudice leads to segregation but no wage differential (discrimination). When customers are prejudiced, blacks with particular skills in areas requiring their visibility may suffer from discrimination, although most blacks will not. When labor markets do not work smoothly, prejudice can cause discrimination.

2.1. Employer Discrimination. We begin by assuming that it is employers who are prejudiced. Even though black and white workers are equally productive, employers dislike hiring blacks. Normally we assume that firms seek to make as much profit as they can. To take account of prejudice, we assume that firms care not only about profits but also how many black workers they employ. The easiest case, and one that captures all of the important intuition, is one in which firms dislike hiring black workers, so they require some fixed profit per worker in order to be willing to hire black workers. Suppose, for concreteness, that employers are sufficiently prejudiced against black workers that they will hire a black worker only if hiring that worker increases their profit by one dollar per hour. Remember that they would hire a white worker if doing so increased their profit at all. Recall that black and white workers are equally productive. So exchanging a black worker for a white worker or vice versa has no effect on output or revenues. Because

1. Gary S. Becker, The Economics of Discrimination, 2nd ed. (Chicago: Chicago University Press, 1971).

Discrimination: Theory 269 the firm could always switch to hiring a white worker instead of a black worker, the only way that the firm can make an additional dollar an hour by hiring a black worker is if black workers earn a dollar an hour less than white workers. If the wage differential between black and white workers is less than one dollar per hour, firms will not want to hire black workers. There will be lots of unemployed black workers, and their wages will fall. There will be too much demand for white workers, and their wages will rise until the wage differential is one dollar per hour. On the other hand, because firms require an increase in profit of only a dollar an hour in order to be willing to hire blacks, if the wage differential exceeds one dollar per hour, firms will want to hire only black workers. There will be unemployment among white workers and their wages will fall, while firms will bid up the wages of black workers. This process will stop only when the wage differential has been reduced to one dollar per hour. Thus, the only possible solution is for the wages of black and white workers to differ by exactly the amount firms require to compensate them for hiring black workers. So far we have assumed that all firms share the same prejudice against black workers. But this is implausible. Even if all owners were prejudiced against blacks, some owners have very little interaction with their workers and are unlikely to care about the race of their employees. What happens in this case? For any black-white wage differential, there will be a prejudice threshold above which firms hire only whites and below which firms hire only blacks. If the wage differential is one dollar, firms that require a differential of more than a dollar will hire only whites. Firms requiring a differential less than one dollar will hire only blacks. Those requiring a differential of exactly one dollar will be indifferent between hiring blacks and hiring whites and might hire both or either race of worker. Thus, differences in the degree of prejudice would cause firms to segregate themselves. Because they can pay lower wages, firms that hire blacks will tend to be more profitable. As they expand, they will make firms that hire the more expensive whites less profitable and drive them out of business. As this process continues, the demand for white workers will fall and the demand for black workers will rise. With only a very modest proportion of unprejudiced firms, the wages of black and white workers will adjust until they are equal. If blacks are 10 percent of the work force, we need only 10 percent of workers to be hired by unprejudiced firms, and this can be achieved with far fewer than 10 percent of firms. Even if employers are generally prejudiced against blacks, there are likely to be firms that choose not to discriminate. Stockholders of large companies or managers at a distant head office are largely unaffected by the race of their employees. Of course, if the wages of white and black workers are equal, prejudiced firms will not hire any blacks. So some firms will be completely white. Others may be all black or integrated, but there will be no difference in the wages received by black and white workers. Equilibrium entails segregation but not wage discrimination.

2.2. Employee Prejudice. If employer prejudice cannot explain sustained discrimination, perhaps employee prejudice can. Suppose that instead of employers not want-

270 Chapter 10 ing to hire blacks, white workers do not like working with blacks. Unless the wage at a firm hiring some black workers is sufficiently high relative to the wage at an all-white firm, white workers will refuse to work at the integrated firm. Therefore, if the firm hires black workers, it has to pay a premium to white workers. Because a firm that has white workers has to increase the wage it pays its white workers when it hires black workers, the firm will not hire a black worker unless it can pay the black worker a wage that is less than the wage it would have to pay a white worker. If the wage differential between the white worker and the black worker is sufficiently large, the firm can raise the wage it pays its white workers to compensate them for their distaste for working with the black worker and still be better off than if it hired the white worker. So if the wage differential between white and black workers were sufficiently large (but not too large), we could get integrated firms. However, because black workers will now earn less than white workers (and the two types of worker are equally productive), the most profitable decision for the firm is simply to hire an all-black work force. As they seek to make a profit based on the difference between black and white wages, firms will bid up the wages of black workers, and the wages of white workers will fall until there is no wage differential. As in the case of employer prejudice, employee prejudice causes segregation but not discrimination.

2.3. Adjustment Costs. We must offer one word of caution about the equilibrium with employee prejudice. Firms cannot change their work force overnight. They will have trained workers with specialized skills. If workers are prejudiced against black workers and require a premium to work with them, as the firm hires blacks, it will have to pay higher wages to its white workers. Therefore, the transition period will be expensive. In this case, if we start in a situation in which there is discrimination, it may be difficult to move to a nondiscriminatory equilibrium.

2.4. Customer Prejudice. Now let us think about what happens if customers are prejudiced. They do not like being served by black sales attendants or having blacks come into their homes to service appliances. In this case, black workers will clearly earn less if they work where they are in contact with customers. However, blacks can avoid this disadvantage by working in jobs where they do not have customer contact. There- fore, if customer discrimination is important, we should observe blacks working in less visible jobs. If there are enough jobs where the race of workers is not observed by consumers, we should get segregation, not wage discrimination. For some workers, there will be enough invisible jobs. Computer technicians can work in the repair shop while someone else deals directly with the customer, or they can provide phone support rather than in-house support. However, some workers will have particular talents that would cause them to work, in the absence of prejudice, in jobs where they are visible. The most obvious example is sports, which also provides a natural basis for testing the theory. The evidence across sports is mixed. The clearest evidence of discrimination is in basketball, where white players draw larger crowds and earn higher salaries than

Discrimination: Theory 271 do black players. There is also evidence that basketball teams have more white players when the local audience market has a higher proportion of whites.2

3. Prejudice in Imperfect Labor Markets The previous discussion suggests that although prejudice can lead to segregation, competition in the labor market will eliminate discrimination except possibly for workers with specialized skills that require customer contact. However, that conclusion is based on some very strong assumptions about how smoothly labor markets work. In the story we have told so far, firms act quickly to take advantage of wage differentials. If there is even a small difference between the wages of black and white workers, nondiscriminating firms will replace white workers with black workers. The firms have no difficulty locating suitable black workers, and black workers have no difficulty finding suitable jobs. In reality, firms do not fill vacancies instantly, and workers need time to find jobs. There are many reasons that labor markets do not work as smoothly as the markets in textbooks written for courses in economic principles. First, workers are not instantly aware of which firms have vacancies, and firms do not know which workers are seeking jobs. Even when workers learn of a vacancy, applying is time-consuming. Workers have to decide whether to apply for a particular job, knowing that applying for one job may make it difficult for them to apply for another. If everything else is equal, workers will prefer to apply to firms with high wages and a low number of applicants. Firms, on the other hand, will want to attract many good candidates to ensure that they will hire a qualified applicant. Once faced with a pool of candidates, they must decide which applicants they will evaluate, because the process of interviewing and checking references is also time-consuming, and while a firm is busy considering one applicant, it may lose the opportunity to hire another. Moreover, firms do not instantly adjust wages in response to the number of applications they receive. Setting a salary or wage is often part of a recruiting strategy. Ad- vertising a high wage or being known as a high-paying firm attracts more applicants. As a result, many jobs have wages that are set before anybody applies for the job. Writing down formal models that take these realistic features of the labor market into account is difficult and would involve a level of technical complexity beyond what is appropriate for this book. Still, a number of results in the literature make intuitive sense. If at least some firms discriminate against blacks, any of the following may occur: 1. Knowing that they are at a competitive disadvantage against whites, blacks will avoid applying for jobs that attract many white applicants. Because whites will be attracted to high-wage jobs, blacks may be discouraged from applying for these high-wage jobs.

2. Lawrence M. Kahn, “Discrimination in Professional Sports: A Survey of the Literature,” In- dustrial and Labor Relations Review 44 (April 1991): 395–418, and Richard C. Burdekin and Todd L. Idson, “Customer Preferences, Attendance and the Racial Structure of Professional Basketball Teams,” Applied Economics 23 (January 1991): 179–86.

272 Chapter 10 2. When blacks and whites apply for the same jobs, blacks will experience more unemployment. 3. Blacks will be willing to accept lower wages because they expect that it will take them longer to find employment. 4. If blacks are more willing than comparable whites to accept jobs with which they are poorly matched, firms may be less inclined to hire blacks, making the discrimination by some firms reinforcing.3

4. Transaction Costs Models There is a set of models, similar to taste- or prejudice-based models, that relies on the idea that social distance may make it difficult for people of different backgrounds to communicate effectively. However, it is important for people from different social groups to interact either to sell to each other or to work together. At a minimum, groups must be able to communicate in order to trade.4 At first blush, the idea that blacks and whites or men and women might have difficulty communicating in the workplace may be surprising. Men and women from the same region share the same regional accent, and for the most part, regional differences in American English are sufficiently small that they do not significantly interfere with understanding. Even the differences between black English and standard American are sufficiently small that two people who are making a good-faith effort to understand each other will only rarely encounter difficulties. However, language is more than just a set of grammar rules and vocabulary. The expression “Call me some time” can mean “I really hope you will call me” or “Stop bothering me, and please do not call me.” The manner in which the phrase is said con- veys a great deal of information. Early e-mailers and instant messagers often learned this the hard way. Comments meant as jokes that would easily have been understood as such in ordinary conversation were misunderstood in writing. Hence, e-mailers learned to add the symbol :-) after jokes, while instant messagers added “jk” (just kidding). Differences in nonverbal signals between blacks and whites or other subgroups in the population can be important and can lead to miscommunication and significant misunderstandings. A northerner may think that a southerner who uses ain’t is uneducated, while the southerner may view the northerner’s “refusal” to switch to ain’t as evidence of priggishness. Whites in the United States signal that they have heard what

3. See Dan A. Black, “Discrimination in an Equilibrium Search Model,” Journal of Labor Economics 13 (April 1995): 309–33; Asa Rosen, “An Equilibrium Search–Matching Model of Discrimination,” European Economic Review 41 (August 1997): 1589–613; Kevin Lang, Michael Manove, and William T. Dickens, “Racial Discrimination in Markets with Announced Wages,” Amer- ican Economic Review 95 (September 2005): 1327–40; and Kevin Lang and Michael Manove, “Wage Announcements with a Continuum of Types,” Annales d’économie et de statistique 71–72 (July/ December 2003): 223–44. 4. See Kevin Lang, “Language and Economists’ Theories of Discrimination,” International Jour- nal of the Sociology of Language 103 (July/August 1993): 165–83, and “A Language Theory of Dis- crimination,” Quarterly Journal of Economics 101 (May 1986): 363–82.

Discrimination: Theory 273 the speaker is saying by a pronounced shake of the head, often accompanies by a guttural response (“uh-huh,” “right”). Blacks, on the other hand, when not code-switching to communicate with whites, use a much less pronounced shake of the head, generally without a guttural response. Whites often miss the listening response that tells them that the black listener has received the message. In the absence of a listening response, the speaker is likely to repeat the message. The listener who has already indicated understanding may perceive repeating the message as a slight.5 If it is costly to mix black and white workers, the result will be very similar to the case in which white workers do not like to work with black workers. However, to the extent that employers are disproportionately white, they will require translators to communicate with their black workers. This works very much like the situation in which employers do not like hiring black workers. Thus, if social distance makes it costly for blacks and whites to work together, there will tend to be segregation as in the taste models, but there can also be persistent discrimination, because blacks must share the cost of the premium paid to workers who can code-switch and thus act as translators between whites and blacks. The language of work is not inherently the language of white middle-class men. However, as long as this language is the dominant language of business, people who understand and speak the language fluently will have an advantage in business, which will, in turn, reinforce the role of the language. Thus, French remained the language of diplomacy long after France had ceased to be the dominant world power. Most diplomats knew French. Therefore, people going into diplomacy needed to know French. But just as French was supplanted by English as the dominant language of diplomacy, as other groups succeed in business, the language of business may change. This provides one justification for affirmative action. If promoting blacks and women to leadership positions in business changes the language of business, it may be easier for other women and blacks to succeed. However, the social distance argument also reinforces the position of those who insist that schools must teach standard English and middle-class rules of social interaction to all students for their students to be successful.

5. Statistical Discrimination In contrast with models based on prejudice, models of statistical discrimination assume that firms seek to maximize profit and are not prejudiced against blacks. However, they do use race to draw inferences about workers that are, on average, correct. The earliest models assumed that, on average, because blacks came from more disadvantaged backgrounds, they were less productive than whites. Although some blacks were more productive than some whites in those models, firms could not distinguish within the

5. Frederick Erickson and Jeffrey Shultz, in The Counselor as Gatekeeper (New York: Academic Press, 1982), describe an interaction between a white guidance counselor and a black student. The guidance session breaks down because the counselor fails to recognize the listening response given by the student.

274 Chapter 10 two groups and thus paid low wages to all blacks and high wages to all whites.6 It is not clear that this is discrimination by the definition we developed earlier. However, we now understand that, with statistical discrimination, blacks and whites can face different outcomes even when, ex ante, the two groups are identical, which is the case that we more readily recognize as discrimination.

5.1. The Basic Statistical Discrimination Model. The basic statistical discrimination model assumes that employers can observe the productivity of whites more readily than they can observe the productivity of blacks.7 In general, the authors of such models have not explained why this difference in observability exists. Presumably it reflects the language differences discussed earlier or differences in social networks. Whites are more likely to have connections to a firm or to have attended schools with which the employer is familiar. To give the flavor of such models, we will assume that employers can observe the productivity of individual whites but treat all blacks as “the average black.” Yet there are two types of worker, good and bad, present in each racial group in equal numbers. We will begin by assuming that half of workers are good and produce 10, and half are bad and produce 4. Because firms can observe which white workers are good and which are bad, half of white workers will earn 10 and half will earn 4, so whites will have an average wage of 7. In contrast, because firms cannot observe the productivity of each black worker, they will treat all black workers as if they are average. The average productivity of a black worker is 7, so all black workers will be paid 7. The equilibrium involves the same average wage for black workers and white workers. Good white workers earn the most; bad white workers earn the least. A worker who did not know whether he was good or bad and who was risk-averse would prefer to be black.

5.2. Job Matching with Comparative Advantage. To generate discrimination from this model, the lack of information about individuals must end up reducing their productivity. One way that this can happen is if the lack of information interferes with the appropriate matching of workers to jobs. Suppose that good workers are much more productive in skilled jobs and only slightly more productive in unskilled jobs. Bad workers produce nothing in skilled jobs but 4 in unskilled jobs. Good workers produce 10 in skilled jobs and 5 in unskilled jobs. Because firms can tell which whites are good workers and which are bad, good white workers are assigned to the skilled jobs and earn 10, while bad white workers are assigned to the unskilled jobs and earn 4.

6. Edmund S. Phelps, “The Statistical Theory of Racism and Sexism,” American Economic Review 62 (September 1972): 659–61. 7. The basic statistical discrimination model described here is derived from Dennis J. Aigner and Glen G. Cain, “Statistical Theories of Discrimination in Labor Markets,” Industrial and Labor Rela- tions Review 30 (Fall 1977): 175–87.

Discrimination: Theory 275 Table 10.1 Productivity of Workers with Different Characteristics in Different Jobs

Patient Active

Fishing 7 1 Hunting 2 10

Note: See the text for an explanation of the numbers used.

In contrast, firms cannot assign all black workers to the appropriate job. The average black worker will produce 5 in a skilled job, because half will produce 10 and half will produce 0. If assigned to an unskilled job, the average black worker will produce 4.5, because half will produce 4 and half will produce 5. Firms will hire blacks at a wage of 5 and assign them to the skilled job. Note that all black workers earn 5, while, on average, white workers earn 7. Thus, black workers face discrimination in the sense that workers who are, on average, similar receive different average wages. However, the discrimination still affects good black workers adversely and bad black workers positively. If workers did not know whether they were good or bad workers and if they were sufficiently risk-averse, they would prefer to receive 5 with certainty rather than 10 or 4 with equal probability.

5.3. Job Matching with Absolute Advantage. In the last example, bad workers were only slightly less productive than good workers in unskilled jobs but much worse than good workers in skilled jobs. Suppose instead that some workers are better at some jobs, but other workers are better at others; for instance, patient people are better at fishing, while active people are better at hunting. The amount produced by each type in each job is given in table 10.1. The table says that active people produce 10 if they are assigned to hunt but only 1 if they are assigned to fish. This means that if firms know that a worker is active, those that have jobs for hunters will be willing to pay up to 10 to hire that worker. Because there is competition, they will bid the wage of the worker up to 10. Whites who are patient are assigned to fish and earn 7. Whites who are active are assigned to hunt and earn 10. In contrast, employers cannot tell if blacks are patient or active. Yet, as in our earlier examples, half of each racial group is of each type. If employers assign blacks to fish, they will produce an average of 4. If they assign blacks to hunt, they will produce an average of 6. Therefore, blacks are hired as hunters and earn 6. Because of this mismatching, all blacks earn less than all whites.

5.4. Unobserved Investments. Shelly Lundberg and Richard Startz have suggested a different route by which statistical discrimination may affect the relative wages of blacks and whites.8 If firms have more difficulty assessing the productivity of

8. Shelly J. Lundberg and Richard Startz, “Private Discrimination and Social Intervention in Competitive Labor Market,” American Economic Review 73 (June 1983): 340–47.

276 Chapter 10 blacks, blacks will have less reason to make unobservable investments that increase their productivity. Suppose, for example, that whites typically attend schools that are well connected to the labor market. When a white student applies for a job, employers understand transcripts from the student’s high school or are able to call the guidance counselor. So when white students work hard at school, they will be rewarded in the labor market for that extra effort, and they will therefore have more interest in working hard at school. If, in contrast, blacks typically attend schools where no such connections exist, they will receive little immediate benefit from putting in extra effort. Therefore, blacks will be less likely to put in effort in school. They will end up being less productive, and, on average, will be paid less than whites.

6. Self-Confirming Expectations Although statistical discrimination can explain earnings differences, it depends critically on the assumption that information barriers are greater between black workers and employers. If all employers were white, this might be a reasonable assumption. How- ever, we would expect black entrepreneurs who could assess their black workers’ productivity to take advantage of the depressed wages for blacks and make more profit. The theory of self-confirming expectations builds on the theory of statistical discrimination and is similar to the unobserved investment story but does not rely on differences in the observability of productivity. Instead it shows that stereotypes can affect behavior in a way that results in the stereotypes’ being true. In the self-confirming expectations model, employers have equal difficulty observing the productivity of blacks and whites.9 However, they have very different expectations about whites and blacks. They think it is less likely that blacks invest in themselves than that whites do so. Therefore, they require a higher signal of investment in order to believe that a black has invested. Given that a higher signal is required, blacks are less likely to invest. Therefore, employers’ expectations are correct. To see how this works, we will develop a simple example. There are two types of jobs: skilled and unskilled. There are also two types of workers: trained and untrained. All workers produce 10 in unskilled jobs. Trained workers produce 20 in skilled jobs, while untrained workers produce 0 in skilled jobs. It costs workers 5 to obtain training. Firms cannot observe whether the worker has gone through training, but they can observe the worker for some signal of whether the worker is trained. Perhaps they can interview the worker and ask questions that indicate (though imperfectly) whether the worker was trained. The signal comes in three forms. The signal may by high (H), medium (M), or low (L). The relation between training and the signal is summarized in the first three columns of table 10.2. If a worker is trained, four-tenths of the time she gives the high signal and six-tenths of the time she gives the medium signal. She never gives the

9. Stephen Coate and Glenn C. Loury, “Will Affirmative-Action Policies Eliminate Negative Stereotypes?” American Economic Review 83 (December 1993): 1220–40.

Discrimination: Theory 277 Table 10.2 Probability of Each Signal and Wages for Trained and Untrained Workers

Signal Untrained Trained Wage in Good Equilibrium Wage in Bad Equilibrium

H 0 .4 20 20 M.4.62010 L .6 0 10 10

Note: See the text for an explanation of the numbers used.

low signal. If she is untrained, she never gives off the high signal, but four-tenths of the time gives off the medium signal and six-tenths of the time gives off the low signal. Note that in this example, H is possible only if the worker is trained; L is possible only if she is untrained. Thus, if employers observe that the worker’s signal is H, they will know that she has been trained, while if they observe that her signal is L, they will know that she has not been trained. However, if the employers observe a signal of M, what they will conclude depends on what they believe about the fraction of workers who obtain training. If they believe that all workers obtain training, observing an M will not contradict that belief, and they will believe that the worker has been trained and gave a signal of M. If they believe that no workers obtain training, again the M will not contradict that belief, and they will conclude that the worker is an untrained worker who gave a signal of M. We can figure out their conclusion for more complicated beliefs, but that will not be necessary. Note that if a workers gives a signal of H, she will be assigned to the skilled job and paid 20. If she gives a signal of L, she will be assigned to the unskilled job and paid 10. What happens if she gives a signal of M will depend on the employers’ beliefs. The question we want to ask is “What beliefs will be consistent with the actual behavior of workers.” If, for example, firms believe that half of workers are trained and half untrained, will half of workers choose to obtain training and half choose not to do so? If the answer is that with these beliefs, no worker obtains training, firms will quickly learn that workers who give M signals are less productive on average than they had expected. So firms will change their beliefs. If the answer is that with these beliefs exactly half of workers become trained, firms will observe that the productivity of workers who give M signals is (on average) exactly what they expected. In this case, firms will not have to change their beliefs. When firms’ expectations are confirmed by the productivity they observe in the labor market, we have an equilibrium. Firms do not have to change their expectations. It turns out that there are at least two equilibria in this example. For reasons that will become obvious, we will refer to one equilibrium as the good equilibrium and to the other as the bad equilibrium. In the good equilibrium, employers believe that everyone obtains training. There- fore, if they observe an M signal, they believe that the individual is trained. They conclude that the individual is not trained only if they observe an L (in which case they know that their original belief was incorrect). Therefore, whenever they observe an M

278 Chapter 10 or an H, they pay the worker a wage of 20 and put him in the skilled job. If they observed an L, which is something they do not expect, they would pay 10 and put the worker in the unskilled job. This is shown in the fourth column of table 10.2. Given these wages for the signals they give, what should workers do? If they obtain training, they will give a signal of M or H. In both cases, they will be paid 20. If they do not obtain training, 40 percent of the time they will give an M and still be paid 20 because firms mistakenly believe they are trained, and 60 percent of the time they will give an L and be paid 10 because firms realize that they are not trained and put them in the unskilled job. On average, they will earn 14. Receiving training increases their average wage by 6 and costs 5. Workers will find the investment worthwhile. Therefore, if firms believe that all workers obtain training, training will be worthwhile and all workers will obtain training. Thus, the situation in which firms believe that all workers obtain training and in which all workers actually do obtain training is an equilibrium in the sense that firms have no reason to change their beliefs given workers’ behavior and workers have no reason to change their behavior given firms’ beliefs. In the bad equilibrium, employers believe that nobody obtains training. Therefore, if they observe a signal of M, they conclude that the individual is untrained. If they observe an L or an M, they treat the worker as untrained and pay 10. Only if they observe an H do they conclude that their initial belief was wrong and pay the worker 20 as a trained worker. This is shown in the last column of table 10.2. In this case, the worker knows that if he obtains the training, 40 percent of the time he will give a signal of H and thus a wage of 20, while 60 percent of the time he will give an M and thus receive a wage of 10. Therefore, on average, he will earn 14. With- out training, he will always give an L or an M and thus be paid 10. The difference between his average wage with training and without training is 4. This is not sufficient to induce him to spend 5 on training. Therefore, if firms believe that no worker obtains training, training will not be worthwhile and no worker will obtain training. Thus, the situation in which firms believe that no worker obtains training and in which no worker actually obtains training is an equilibrium. How does all this relate to discrimination? Suppose that employers believe that all whites receive training and that no blacks do. Then these beliefs will be self-confirming. Whites will find it worthwhile to receive training, while blacks will not. Whites will obtain training, be hired into skilled jobs, and earn 20. Blacks will not obtain training, be hired into unskilled jobs, and earn 10. If we believe that self-enforcing stereotypes generate discrimination, how should we combat it? One policy that would work in the previous example is to subsidize training. If we could reduce the cost of training below 4, the bad equilibrium would not exist. However, it may not be feasible to subsidize training; if training is not observable by employers, it is probably not observable by government. Another solution is to try to force employers to act as if they had the same beliefs about black workers as they have about white workers. Because we cannot observe their beliefs or the signals they observe, we can force them to act similarly only in a very simple way. We can require them to hire whites and blacks into skilled jobs in the same

Discrimination: Theory 279 proportions and pay them the same wage in each job. We will refer to this as affirmative action. What would happen in this situation? Obviously, either the good equilibrium for all workers or the bad equilibrium for all workers would be a possible equilibrium, because in both cases black and white workers would obtain the skilled jobs at the same rates—100 percent in the good equilibrium and 0 percent in the bad equilibrium. But there can be a third equilibrium. Suppose that whites are 95 percent of the work force. Suppose that employers continue to believe that whites always obtain training and that blacks never do. We know that in order for this to be an equilibrium, it must be the case that, given these beliefs, whites all prefer to obtain training and all blacks prefer not to do so. We will show that this can occur if, in order to meet the affirmative action requirement, employers assign all workers to the skilled job even though none of the black workers are trained. Because all workers are put into skilled jobs and must be paid the same regardless of race, the pay for skilled jobs will be the average productivity of all workers on skilled jobs. Because 95 percent of workers are trained white workers who produce 20 and 5 percent are untrained black workers who produce 0, the average productivity will be 19. A black worker who does not obtain training is given the skilled job anyway and so has no incentive to obtain training. A white worker who did not obtain training would give a signal of M with a probability of 0.4, in which case, he would be paid 19. With a probability of 0.6, he would give an L and be revealed (contrary to expectations) to be untrained, be assigned to the unskilled job, and be paid 10. Thus, his average wage if untrained would be 13.6. By obtaining training, the white worker would, on average, earn 5.4 more than without training, which would be sufficient to cover the cost of training. Thus, white workers will still obtain training. In this last case, affirmative action eliminates discrimination, but it does so by making black workers less productive. It does not eliminate the negative stereotypes about blacks. If anything, it exacerbates them by contributing to a mismatch between their skills and their job assignments.

7. Concluding Remarks One view of the labor market is that it is competitive and generally works smoothly. Firms are good judges of a worker’s value. Firms and workers have little difficulty finding each other. Lots of firms compete with each other for workers and bid their wages up until they are paid what they are worth. In such a world, it is difficult for discrimination to persist except in niches where customer discrimination may hurt blacks. The other view of the world emphasizes imperfections in the labor market. Such imperfections may arise because it is difficult to assess worker productivity or because firms and workers have difficulty finding each other. In such a world, discrimination may persist because firms are prejudiced and can act on that prejudice, because blacks believe that firms are prejudiced and take steps to diminish the effect of that prejudice, or because of firms’ and workers’ rational responses to imperfect information.

280 Chapter 10 It is important to remember that our focus has been on labor market discrimination. Social scientists who believe that labor markets work sufficiently smoothly to largely eliminate discrimination may believe that discrimination is important in settings that are not governed by markets. In particular, government may discriminate in its provision of education, justice, or purchases of goods and services. It is also worth noting that the statistical discrimination and self-enforcing expectations theories are closely related to labeling and stereotyping theories in sociology.10 The difference is that the theories discussed in this chapter were largely developed by economists who assume that people will change their beliefs in response to contradictory information. Thus, if a police officer consistently finds drugs in 2 percent of the cars he searches that are owned by blacks, the models we have discussed require that he eventually come to believe that, in similar searches, he has a 2 percent chance of finding drugs. In the next chapter we will review the evidence on the existence of discrimination. We will see that we can use experiments to test for the existence of prejudice. But the theory developed in this chapter shows that prejudice does not necessarily cause discrimination, and it is much harder to determine whether discrimination exists in the labor market. When services are provided by government rather than by a competitive market, the theory does not imply that the market overcomes prejudice. Therefore, in chapters 12 and 13 we will also look at discrimination in contexts besides the labor market.

8. Further Reading Altonji, Joseph G., and Rebecca M. Blank. “Race and Gender in the Labor Market.” In Orley Ashenfelter and David Card, eds., Handbook of Labor Economics, Vol. 3C. Amsterdam, New York, and Oxford: Elsevier, North-Holland, 1999, 3143–259. Becker, Gary S. The Economics of Discrimination, 2nd ed. Chicago: Chicago University Press, 1971. Burdekin, Richard C., and Todd L. Idson. “Customer Preferences, Attendance and the Racial Structure of Professional Basketball Teams.” Applied Economics 23 (January 1991): 179–86. Coate, Stephen, and Glenn C. Loury. “Will Affirmative-Action Policies Eliminate Negative Stereotypes?” American Economic Review 83 (December 1993): 1220–40. Holzer, Harry J., and Keith R. Ihlanfeldt. “Customer Discrimination and Employment Outcomes for Minority Workers.” Quarterly Journal of Economics 113 (August 1998): 835–68. Kahn, Lawrence M. “Discrimination in Professional Sports: A Survey of the Literature.” Industrial and Labor Relations Review 44 (April 1991): 395–418. Lang, Kevin. “Language and Economists’ Theories of Discrimination.” International Journal of the Sociology of Language 103 (July/August 1993): 165–83.

10. See Howard S. Becker, The Outsiders (New York: Free Press, 1997).

Discrimination: Theory 281 Loury, Glenn C. The Anatomy of Racial Inequality. Cambridge, MA: Harvard Univer- sity Press, 2002.

9. Questions for Discussion 1. Is there a difference between discrimination and prejudice? 2. Suppose that the labor market is competitive and that employers (white workers, customers) prefer not to hire (work with, interact with) blacks. Does this generate wage discrimination, segregation, or both? 3. Would your answers to question 2 change if the labor market were not competitive? 4. Explain how each of the following models generates wage differentials between blacks and whites. a. Social distance (transactions costs) b. Statistical discrimination c. The theory of self-confirming expectations 5. What policies would be effective in reducing black-white wage differentials under each of the models?

282 Chapter 10 chapter 11

Race Discrimination in the Labor Market

I         can be a powerful force for eliminating discrimination. Firms that discriminate against black workers will be less profitable than firms that are willing to hire workers without respect to race. The same forces work in customer markets. Firms that ignore or under- serve the market among African Americans forgo profits that their nondiscriminating competitors can earn. Yet we also saw that the competitive forces that defeat discrimination can break down if information is imperfect, if consumers have discriminatory tastes, or if the labor market is otherwise imperfect. In this chapter we explore the extent of discrimination in the U.S. labor market. In later chapters we will discuss discrimination in education, customer markets (particularly housing), and the justice system. We begin with the labor market because our theories focused on this market and because it is one of the markets about which economists disagree the most. Some believe the labor market closely resembles the theoretical labor market that eliminates discrimination. Others believe that the types of labor market imperfection discussed in models of persistent discrimination are prevalent. Before discussing the evidence with respect to discrimination, we begin with the trends in the relative wages of blacks and whites because this provides some evidence regarding the factors that account for earnings differentials.

1. Trends in Black-White Earnings Differentials Figures 11.1 and 11.2 show the trends in the ratio of black median earnings to white median earnings for men and women from 1967 to 2004. For all male workers, the ratio of black to white median earnings stood at 58 percent in 1967. By 2003, it had climbed to 81 percent. For those working full-time, year-round, it went from 65 percent in 1967 to 75 percent in 2004, down from its peak of 81 percent in 1999.1

1. Some caution is required. Starting with the 2002 data, the Current Population Surveys on which these data are based have permitted respondents to identify themselves as belonging to multiple

283 Ratio of Median Earnings

0.80 Full-Time, Year-Round Workers 0.75

0.70

0.65

0.60 All Men

1970 1980 1990 2000 Figure 11.1 Ratio of Median Earnings, Black Men / White Men, 1967–2004

Ratio of Median Earnings 1.1

1.0 All Women

0.9 Full-Time, Year-Round Workers 0.8

0.7

1970 1980 1990 2000 Figure 11.2 Ratio of Median Earnings, Black Women / White Women, 1967–2004

Looking at the figures, it is apparent that the changes have not been steady. There was a dramatic increase in the relative earnings of black men from 1967 until sometime in the mid- to late 1970s. After that, their relative earnings stayed constant or even fell until the late 1980s, when the earnings gap began to decline again. For women, the pattern is somewhat different. For all women workers, we see a dramatic increase in the relative earnings of black women from 1967, when the ratio stood at 65 percent, to 1976, when the median black female worker had higher earnings than the median white female worker. Since then the relative earnings of black

races. Before 2002, respondents could identify themselves only as black or white, but not as both. The 2002–2004 figures shown here are for those identifying themselves as only black or only white.

284 Chapter 11 female workers have declined somewhat; they stood at 98 percent in 2004. The story for full-time, year-round workers is similar although less dramatic. In 1967, the ratio was 75 percent. It peaked at 96 percent and, based on the most recent data, now stands at 88 percent. The fact that in the 1990s the wages of black women failed to reverse their decline relative to the wages of white women can be attributed to the very rapid increase in the earnings and labor force participation of white women. As we will see in chapter 14, in the two decades beginning in 1981, relative to the wages of white men, the wages of white women working full-time, year-round increased dramatically, from 58 percent of the median earnings of white men to 76 percent. When looking at these data, it is also important to consider nonparticipation by black and white women. As discussed in detail later, black women who do not have jobs are drawn disproportionately from the low end of the education distribution, while white women are drawn fairly evenly from across the education spectrum. There- fore, the black-white earnings ratio among women understates the differential that would exist if all women participated in the labor force. To some extent, the recent relative decline in the earnings of black women relative to white women probably reflects an increase in the labor force participation of highly skilled white women. For a broader perspective on relative earnings we must turn to the decennial census. Beginning with the 1940 census, we have data on income. Prior to 1940, the decennial census collected information on occupation. James Smith has constructed estimates of relative earnings for the earlier period by combining estimates of income in different occupations with the information about the occupations in which people were engaged.2 Figure 11.3 gives his estimates of the ratio of black to white incomes for men aged 20–64. We can see that there was modest growth in the ratio from 1890 to 1920, followed by a period of stagnation from 1920 to 1940 and then a period of rapid growth from 1940 to 1980. From figure 11.1, we know that the growth from 1970 to 1980 occurred in the first part of the decade. We can achieve some initial insight into why relative incomes increased by examining income changes. Figure 11.4 shows relative incomes by five-year birth cohort for 1940 to 1980 based on the data in Smith’s work. There are a number of important points that can be learned from this figure. The first is that there is a fairly consistent pattern: at any given time, younger cohorts had higher relative wages than older cohorts.3 If we observed only one year of data, we might conclude that blacks faced more discrimination as they aged. However, by looking at cohorts as they aged, we see that, in general, income ratios were constant or increased as a cohort aged. Cohorts in which the relative incomes of blacks were relatively high (or not as low) entered the labor market and replaced cohorts with low

2. James P. Smith, “Race and Human Capital,” American Economic Review 74 (September 1984): 685–98. 3. One exception is due to limitations in the data. In 1950, the 1906–1910 and 1911–1915 cohorts could not be separated. The income ratio for these two cohorts in that year is actually the income ratio for the 1906–1915 group.

Race Discrimination in the Labor Market 285 Income Ratio 0.60

0.55

0.50

0.45

18901900 1910 1920 1930 1940 1950 1960 1970 1980 Figure 11.3 Income Ratio, Black Men / White Men Aged 20–64, 1890–1980 Source: James P. Smith, “Race and Human Capital,” American Economic Review 74 (September 1984): 685–98.

Income Ratio

0.58

0.56

0.54

Birth Year 0.52 1906–10 1911–15 0.50 1916–20 1921–25 1926–30 0.48

1940 1950 1960 1970 1980 Figure 11.4 Income Ratio, Black Men / White Men, by Year of Birth, 1906–1925 Source: James P. Smith, “Race and Human Capital,” American Economic Review 74 (September 1984): 685–98.

income ratios. This will push us toward an explanation that says the skill level of blacks increased relative to those of whites. It will be natural to ask whether the increasing quantity and quality of education for blacks relative to whites can account for the increase in the income ratio. However, we also see that for two periods, 1940–1950 and 1960–1970, there is clear evidence that the relative incomes of blacks grew even within a cohort. For example, the cohort born between 1916 and 1920 was 20 to 24 years old in 1940. By 1950, its income ratio had risen from 50 percent to 53 percent. In contrast, it experienced

286 Chapter 11 almost no increase in its relative income from 1950 to 1960. This pattern suggests that we should look for events in the 1940s and 1960s that can explain improvement in the relative economic performance of black Americans. The migration of blacks from the rural south to the industrial north is a plausible explanation for the change in the 1940s. The civil rights movement and legislation are likely candidates for the change in the 1960s.

2. Explaining the Decline in the Wage Gap The pattern of changes in black-white income differentials requires that we explain why the relative wages of blacks tended to rise from one cohort to the next and why they rose within cohorts in the 1940s, 1960s, and 1970s. For the changes between cohorts, we focus on shifts that change the relative skills that blacks bring to the labor market, and thus on the following: • Relative education. Education levels have increased even more rapidly among blacks than among whites. • Quality of education. Blacks are receiving not only more education but also better education. The elimination of legal school segregation in the south and the reduction of de facto school segregation in the north gave blacks access to better schools. For changes within cohorts, we focus on the following: • Migration. There was substantial black migration from the low-wage rural south to the higher-wage cities of the north. • Reduced discrimination. Both the legal and social environments have changed so that firms are less likely to discriminate against blacks in hiring, wages, or promotion. We add one more explanation to this list: • Declining labor force participation of low-skilled blacks. Fewer black men are working, and those who have dropped out of the labor force are disproportionately workers with low potential earnings. Removing them from the calculation of median earnings artificially inflates the median. Almost all researchers agree that each of these factors has played some role in reducing the earnings gap. However, there is considerable disagreement about their relative importance.

2.1. Education Quantity. The education gap between both black and white men and black and white women narrowed dramatically after 1940. As can be seen in figure 11.5, in 1940, almost one-fourth of white men aged 25 and over had at least completed four years of high school, compared with only about 7 percent of black men. By 2004, the high school gap had been almost completely eliminated. During this period, a gap in the proportion of men with at least four years of college did open up, but the

Race Discrimination in the Labor Market 287 Percentage with High School/College

60 High School, White 40 College, White High School, 20 Black

College, Black 0

1940 1950 1960 1970 1980 1990 2000 Figure 11.5 Percentage of Men 25+ Years Old with Four Years High School / College or More, by Race, 1940–2004

difference in education levels is still much lower than it was sixty years ago. The situation for women is similar (see figure 11.6). James Smith and Finis Welch report that in 1940, black men aged 16 to 64 had an average of 3.7 fewer years of schooling than their white peers. By 1950, this figure had fallen to 3.2 years. In the following decades, it fell to 2.6 years (1960), 2.0 years (1970), and 1.5 years (1980).4 The increase in relative education for black men is thus on the order of one-half year per decade. Generously, this might (given the low return to education among blacks in that period) account for 3 percentage points of the increase in the relative wage of blacks per decade. In contrast, Smith and Welch estimate that the black/white wage ratio increased from 43.4 percent to 72.6 percent between 1940 and 1980. In fact, it is something of a mystery why the relative wages of blacks did not increase more between 1950 and 1960 given both the continued migration from the south and the increases in relative education over this period.

2.2. Education Quality. David Card and Alan Krueger have documented large increases in measures of school quality for blacks relative to whites.5 Class sizes, which in 1915 were almost two-thirds larger for blacks than for whites, had almost equalized by

4. Smith, James P. and Finis Welch, “Black Economic Progress After Myrdal,” Journal of Eco- nomic Literature 27 (June 1989): 519–64. Note that Smith and Welch calculate wage differentials directly rather than basing their calculations on the occupation-income relation. Their estimate of the differential is similar to Smith’s for 1940, but they estimate a much faster narrowing of the differential. 5. David Card and Alan B. Krueger, “School Quality and Black-White Relative Earnings: A Di- rect Assessment,” Quarterly Journal of Economics 107 (February 1992): 151–200.

288 Chapter 11 Percentage with High School/College

60 High School, White 40

High School, College, White Black 20

College, Black 0

1940 1950 1960 1970 1980 1990 2000 Figure 11.6 Percentage of Women 25+ Years Old with Four Years High School / College or More, by Race, 1940–2004

1964. By 1950, the length of the school year was almost the same for the two groups, and the relative wages of teachers converged over the same period. Card and Krueger estimate that 15 to 20 percent of the relative wage gains made by black men between 1960 and 1980 were due to improved school quality, but their estimates are quite imprecise.

2.3. Migration. Migration was unquestionably an important reason for the narrowing of the gap between 1940 and 1960. The black-white wage differential was smaller in the north than in the south, and wages overall were higher in the north. Therefore, blacks moving to the north earned substantially more than they would have earned if they had remained in the south. Roughly one-fourth of southern blacks moved north between 1940 and 1960. However, the northern flow stopped by the early 1960s and began to reverse itself.6 This suggests that there may have been a role for legislation in explaining the narrowing of the gap after 1960. If migration narrowed the gap substantially between 1940 and 1960 but ceased to play a role thereafter, some other factor must have had an increased role in furthering the convergence of black and white earnings.

2.4. Declining Labor Force Participation of Low-Skilled Blacks. The black/white earnings ratios in figure 11.1 are based on men with at least some earnings over the year. As shown in figure 11.7, in the 1970s there was a dramatic increase in the proportion of black men without earnings. Black men without earnings were primarily people

6. John J. Donohue III and James Heckman, “Continuous versus Episodic Change: The Impact of Civil Rights Policy on the Economic Status of Blacks,” Journal of Economic Literature 29 (Decem- ber 1991): 1603–43. Those familiar with Donohue and Heckman will note that this section follows the logic, albeit not the detail, of their presentation.

Race Discrimination in the Labor Market 289 Percentage with Earnings 80 White

70 Black

1970 1975 1980 1985 1990 1995 2000 Figure 11.7 Percentage of Males with Earnings, by Race, 1970–2004

who would have had very low earnings if they had been working. By removing these individuals from the average for blacks, we overestimate median income for blacks relative to whites. There are a variety of ways that we can attempt to correct for the difference in employment rates between black and white men. One approach is to assume that all men who do not have earnings would earn less than the median. If we restrict ourselves to men who have passed the age at which they are still in school but are younger than retirement age, almost all those not working will either be low-skilled or have some disability that prevents them from working. Therefore, at least for prime-age men, this seems a reasonable assumption. Using this approach, Charles Brown calculates that, for the years between 1953 and 1978, about two-thirds of the convergence in the median wage and salary incomes of black and white men can be attributed to declining employment among black men.7 As he notes, this approach is likely to overestimate the importance of declining employment, because at least some of those who do not work would have earned more than the median. Even with this qualification, Brown’s calculation is misleading. Before 1989, the Current Population Survey counted only civilian workers, and it continues to sample only persons in private living quarters, not those who are in institutions. This means that the survey misses the growth in both the proportion of black men who are in the armed forces and the proportion who are incarcerated. To address these issues, we must rely on the census, which is conducted only every ten years. Table 11.1 gives census information for 1940 to 1990. We see that, combining employees, the self-employed, and those in the armed forces, employment among

7. Charles Brown, “Black-Whites Earnings Ratios since the Civil Rights Act of 1964: The Im- portance of Labor Market Dropouts,” Quarterly Journal of Economics 95 (February 1984): 31–44.

290 Chapter 11 Table 11.1 Percentage Employment and Black/White Weekly Earnings Ratio for Men Aged 25–55, 1940–1990

1940 1950 1960 1970 1980 1990 White Men Employed 66.2 68.4 74.2 77.7 78.0 79.1 Self-Employed 18.7 17.4 14.8 11.6 8.3 8.0 In the Armed Forces 0.6 1.9 2.5 2.2 1.7 1.7 Unemployed 9.6 5.7 4.1 3.8 5.9 4.9 In School 0.2 1.2 0.4 0.5 0.8 0.8 Out of the Labor Force 3.3 4.2 3.0 3.5 4.8 4.9 Institutionalized 1.5 1.2 0.9 0.7 0.5 0.6

Black Men Employed 63.9 68.1 74.0 75.9 68.4 67.8 Self-Employed 17.0 10.0 5.9 3.5 1.6 1.9 In the Armed Forces 0.2 1.5 2.6 2.2 2.5 2.6 Unemployed 11.3 8.7 7.9 5.9 10.8 10.2 In School 0.1 1.1 0.3 0.5 1.2 1.1 Out of the Labor Force 4.2 7.5 6.9 8.9 12.7 12.7 Institutionalized 3.3 3.1 3.0 3.1 2.8 3.8

Black/White Earnings Ratios Observed 48.4 62.6 61.8 69 75.7 75.3 Missing below Median 43.3 60 57.4 64.9 64.4 62.9 Missing Equal to Similar 48.2 62.5 61.6 68.4 74.7 73.7 Missing Equal to Lowest Similar 46.9 61.4 58.8 65.8 66.5 66.3

Source: Amitabh Chandra, “Labor-Market Dropouts and the Racial Wage Gap: 1940–1990,” American Economic Review 90 (May 2000): 333–38. prime-age white men was relatively constant over this period. Just under 86 percent of white men were working in 1940. This percentage rose to 91.5 in both 1960 and 1970 before falling somewhat in 1980 and 1990. In contrast, among blacks total employment was relatively constant between 1940 and 1970, then fell dramatically between 1970 and 1980 before stabilizing between 1980 and 1990. The broad pattern is thus similar to that observed in figure 11.7 when the two overlap. The bottom panel of table 11.1 shows estimates of the black-white weekly earnings ratio under different assumptions. Ignoring individuals without earnings (“Ob- served”), we see, as we have previously, significant increases in the earnings ratio between 1940 and 1950 and between 1960 and 1980, with essentially no progress between 1950 and 1960 and between 1980 and 1990. The three last rows use different approaches for “correcting” the data. The first assumes that all workers without earnings would earn less than the median. The second

Race Discrimination in the Labor Market 291 assumes that they would have earnings equal to the average of men of their race with their level of schooling and age. The third assumes that their earnings would equal those of the lowest earner in their race, age, and schooling category. For the most part, all four sets of estimates are similar: relative to white men, black men made significant progress between 1940 and 1950. They made no progress and may even have lost a little ground between 1950 and 1960 before again making significant progress during the 1960s. The real difference in interpretation comes in the 1970s. If we make no correction or assume that workers without earnings would have earned the average earned by similar workers, black men continued to make considerable progress in the 1970s. If we assume that they would have had low earnings relative to all workers or relative to similar workers, the 1970s were a period of stagnation. The truth probably lies between the two conclusions.

2.5. Political, Legal, and Other Factors. The most important piece of civil rights legislation directly affecting the labor market was Title VII of the 1964 Civil Rights Act, which outlawed employment discrimination on the basis of race as well as ethnicity, religion, and sex. The law also created the Equal Employment Opportunity Commission, which is responsible for enforcing Title VII. The Civil Rights Act was strengthened in 1981, when victims became entitled to sue for both compensation and punitive dam- ages in the case of intentional violations of Title VII. Title II of the Civil Rights Act prohibited discrimination in public accommodations (restaurants, hotels, etc.), while Title III gave the attorney general the power to intervene on behalf of those denied access to public facilities. This may have affected the labor market indirectly by making certain types of employment feasible for blacks. The Civil Rights Act was followed in 1965 by Executive Order 11246, which prohibited federal contractors with more than fifty employees and $50,000 in federal contracts from discriminating in any aspect of their business and required them to undertake affirmative action to ensure that they did not discriminate. We saw that there was a dramatic increase in the relative earnings of blacks between 1967 and the mid-1970s. To what extent can these political and legal developments account for this change? James Smith and Finis Welch argue that within education groups, the narrowing of the racial wage gap between 1940 and 1960 was similar to the narrowing between 1960 and 1980. Among those with a college degree or more education, the black-white wage gap narrowed by 21 percent between 1940 and 1960, compared with 24 percent between 1960 and 1980. For people with exactly a high school education, the gap narrowed by 16 percent in the earlier period and 17 percent in the later period, and for those with eighth- to eleventh-grade educations, the gap narrowed by 21 percent in each period. Because the trend between 1940 and 1960 looks a lot like the trend between 1960 and 1980, they conclude: The racial wage gap narrowed as rapidly in the 20 years prior to 1960 (and before affirmative action) as during the 20 years afterward. This suggests that the slowly evolving historical forces we have emphasized . . . —education and migration—

292 Chapter 11 were the primary determinants of the long-term black economic improvement. At best, affirmative action has marginally altered black wage gains around this long-term trend.8 However, as we have seen, there were large differences in progress within each twenty-year period. The rapid decline of the black-white earnings gap in the 1960s and possibly in the 1970s suggests that political and legal changes might have played a role in the convergence. There are two sources of direct evidence that these changes helped to narrow the black-white earnings gap. The first is that initially civil rights compliance was primarily targeted at the south, where we also saw the fastest growth in relative wages. The second comes from the South Carolina textile industry. Textile producers generally sell to the armed forces and were thus subject to the executive order covering government contracts. Before 1965, black women accounted for almost none of the employment in South Carolina textiles, and the proportion of black men was very small, a pattern that had continued through periods of boom and slump and through major wars. Suddenly, beginning in 1965, there was a dramatic shift in the employment pattern of the industry. The timing and the nature of the shift are precisely what would have been anticipated if the executive order was effective.9

3. Evidence on Current Discrimination We have defined discrimination as unequal treatment of equals and have emphasized the importance of discrimination in the case where one group, such as blacks, does consistently worse than another group, such as whites. We were not particularly concerned by the case of statistical discrimination, in which blacks earned the same average wage as whites although they experienced less wage dispersion. Therefore, we will focus on the case in which discrimination takes the form of blacks doing worse than equivalent whites. This immediately raises the question of what we mean by “equivalent.” One approach that we can use is to control for all observable characteristics in a data set and ask whether blacks have less favorable outcomes than whites with the same characteristics. There are two major drawbacks to this approach. The first is that, as we have seen in previous discussions, the measures used in our data sets are very imperfect. Blacks and whites with the same education, age, sex, state of residence, and so on, may be very different with respect to factors that are not included in the data set. For example, they may have grown up in very different families and have attended very different schools. It may be incorrect to attribute the worse outcomes experienced by blacks to discrimination they face as adults if blacks are likely to have had less educated parents and to have attended lower-quality schools, conditional

8. James P. Smith and Finis R. Welch, “Black Economic Progress after Myrdal,” Journal of Eco- nomic Literature 27 (June 1989): 519–64, quote on p. 555. 9. Donohue and Heckman, 1630–40.

Race Discrimination in the Labor Market 293 on measured factors (although the lower school quality may be the result of discrimination they faced as children). In previous chapters we discussed techniques researchers use to get around the fact that the variable in which they are interested might be correlated with missing factors. Many studies relied on actual experiments in which the variable was randomly assigned. Others compared siblings. Such random assignment and quasi-random assignment is impossible when we are concerned with race. The closest approximation to random assignment we can use is an approach known as an audit experiment. In audit experiments, applicants who are as nearly identical as feasible in every respect except race apply for the same jobs or purchase the same good. In some experiments, the researcher sends in written applications that in some way identify the race of the applicants. In other cases, trained actors behave similarly and present identical resumes. We will discuss audit studies later. The second problem is that many of the factors for which we might control may themselves be affected by discrimination. Thus, if we are trying to determine whether similar blacks and whites earn different wages, we might want to look at blacks and whites with the same level of education. However, if otherwise similar blacks and whites achieve different levels of education because blacks expect to face discrimination in the labor market, blacks and whites with the same level of education will not be “similar.” Thus, in many ways measuring discrimination is even more challenging than measuring the effects of policies designed to reduce poverty.

3.1. Getting the Interview. Marianne Bertrand and Sendhil Mullainathan sent job applications to employers in Boston and Chicago.10 They developed a set of resumes and sent four resumes to each employer. They randomly assigned common white names, such as Greg and Meredith, and common African American names, such as Latoya and Tyrone, to the resumes. Because of the random assignment, on average, there was no difference in the qualifications of the “white” and “black” applicants. In fact, the applications did not explicitly reveal the race of the applicant, but employers might infer race from the name. The callback rates were quite different for the two groups. Overall, 10.1 percent of applications with white-sounding names elicited a callback for an interview, compared with only 6.7 percent for those with black-sounding names. Put differently, an applicant with a white-sounding name would expect one callback for every ten applications compared with one callback for every fifteen applications for those with black-sounding names. This ratio was similar for male and female applicants. Another way of looking at the same data is to ask how many employers treated applicants with black- and white-sounding names equally. Fully 82.6 percent of resumes elicited no callbacks to any of the four applicants. Of the remaining 17.4 percent, 4.8 percent resulted in the same number of callbacks for whites and blacks, 3.8 percent in

10. Marianne Bertrand and Sendhil Mullainathan, “Are Emily and Greg More Employable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimination,” American Economic Review 94 (September 2004): 991–1013.

294 Chapter 11 more callbacks of blacks than of whites, and 8.9 percent in more callbacks of whites than of blacks. The fact that an employer called back more whites than blacks or vice versa does not necessarily mean that the employer discriminates again blacks, or against whites in the opposite case. Sometimes blacks and whites are treated differently because of random factors. An employer may have called a white applicant because he liked the school the applicant was said to have attended. Or an employer might have stopped calling after finding ten qualified applicants. The lucky black applicant who was near the top of the pile may have been called, while the unlucky white applicant lower down was not. The 3.8 percent of cases in which more black than white applicants were called is a reasonable estimate of the frequency of such events. If so, 8.9 percent minus 3.8 percent, or 5.1 percent, of employers discriminated. This is out of about 30 percent of employers who called any applicant created by the researchers. Some of the employers who called more blacks than whites may have discriminated against whites. If so, we can still say that discrimination against blacks was 5.1 percentage points more common than discrimination against whites. Although I have described the study as addressing discrimination on the basis of race, it is important to recognize that, in fact, we only know that employers responded differently to applicants with different names. Some other feature of the names may have been important. Also, the study raises a further question, which we will address later: If having a black name is disadvantageous, why do black parents give their children distinctively black-sounding names?

3.2. Getting the Job. Getting a callback is an important step in the job search process, but ultimately we are interested in whether the individual gets the job. If whites are more likely to get an interview but no more likely to get the job, the initial difference in interview rates is relatively unimportant. To address whether whites are more likely to get jobs, studies have used actors with similar characteristics other than race and similar “resumes” to look at discrimination in hiring. James Heckman summarizes the results of three major audit studies using a total of fifteen matched pairs of black and white job applicants.11 All told, these fifteen pairs interviewed for 583 jobs. In the vast majority of cases, neither applicant received a job offer, while in about 15 percent of the cases, both received job offers. The interesting cases are those in which one but not the other was offered the job. In 72 cases, only the white was offered the job, while in 35 cases only the black was offered the job. If we think of these as 107 independent flips of a coin, the probability of the split’s being 72 to 35 is very small. However, it is obvious that, no matter how hard they try, experimenters cannot find pairs of actors who are identical in every respect except race. Sometimes the white

11. James J. Heckman, “Detecting Discrimination,” Journal of Economic Perspectives 12 (Spring 1998): 101–16. The original studies were in Michael Fix and Raymond J. Struyk, Clear and Con- vincing Evidence: A Measurement of Discrimination in America (Washington, DC: Urban Institute, 1993).

Race Discrimination in the Labor Market 295 Table 11.2 Outcomes of Audit Studies for Each of Fifteen Black-White Pairs Seeking Jobs (Percentages)

Pair No. Only White Got Job Only Black Got Job

1 14.3 5.7 2 10.0 15.0 3 6.8 2.3 4 16.7 0.0 5 2.4 0.0 6 26.1 6.5 7 16.7 5.6 8 17.7 6.5 9 18.9 5.4 10 16.7 4.8 11 27.8 0.0 12 0.0 18.9 13 3.0 0.0 14 6.7 13.3 15 0.0 0.0

Source: James J. Heckman, “Detecting Discrimination,” Journal of Economic Perspectives 12 (Spring 1998): 105.

actor will have characteristics that make him more appealing to many employers, and in other cases it will be the black who is more appealing. Despite having been given the same false biography and trained to behave similarly, one will have slightly different mannerisms from the other. If the number of audits per pair is large and the number of pairs is small, we will inevitably find a racial difference.12 To avoid this problem, we can treat each matched pair as a single observation. If the experiment is not biased, half of the time the experimenters will accidentally make the white more attractive and half the time they will make the black more attractive. Assuming no bias in the experiment, finding that the white is more attractive more than half the time would indicate that employers discriminate against blacks. Table 11.2 shows that, out of fifteen matched pairs in three different studies, in eleven cases the white was more likely to get a job offer, in three the black was more likely to get an offer, and in one there were no cases in which one was offered the job and the other was not. If blacks and whites were truly equally likely to get jobs, the probability that eleven out of fourteen pairs would experience more job offers for the white would be less than 5 percent. In fact, we can go further and ask whether in these fifteen cases the difference between the proportion of times only the white got the job

12. To take a simple case, suppose there is only one pair involved in the study. Then for a sufficiently large number of audits, we will be able to detect any difference, no matter how small, in their attractiveness to employers. In reality, we will not know whether this difference is due to race or some other difference between the two individuals.

296 Chapter 11 and the proportion of times that only the black got the job is equal to zero. In other words, is the average difference between the columns in table 11.2 equal to zero? The average difference is, in fact, 6.7 percentage points, and its difference from zero is statistically significant. Of course, it is still possible that either the researchers or the actors inadvertently biased the study by subconsciously making the black applicants less attractive. If either group wanted to demonstrate the existence of discrimination, they might unintentionally have chosen less attractive blacks or acted in a less appealing manner. Although the audit studies using actors can be criticized because the matching of actors may have been imperfect, this criticism does not apply to the study using resumes and different names. On the other hand, using actors addresses the concern that a black advantage in interviews might offset a white advantage in getting interviews. The cumulative evidence suggests that blacks are less likely to be offered a job than are apparently similar whites.

3.3. Further Limitations of Audit Studies. However, there is an additional concern about audit studies: Does having a lower probability of getting a particular job translate into a lower probability of finding employment? Recall that in the Becker model, discrimination on the part of some employers did not translate into wage differentials. Instead the labor market was segregated; racist employers did not hire blacks, but blacks found employment at equal wages at firms that did not discriminate. The audit studies cannot address this question directly. The Becker model implies that the discriminatory tendencies found in the audit studies need not generate wage differentials in the labor market. In his review of the audit studies, Heckman argues that blacks will tend to avoid firms that discriminate. In his argument, this reduces their disadvantage in labor market search. However, in our discussion of theories of discrimination we saw that adjusting search in response to discrimination might not solve the problem. If blacks do not know which firms discriminate, they may respond by applying to lower-paying firms in the expectation that whites will be less likely to apply. As we will discuss in greater detail in chapter 14, David Neumark found more discrimination against women in higher-paying restaurants than in lower-paying restaurants.13 If men are also more likely to apply for jobs in higher-paying restaurants, women will have a strong incentive to avoid the higher-paying restaurants and apply only to restaurants at the lower end of the pay scale. Thus, our assessment of the importance of the labor market audit studies depends on whether we believe that labor markets work well and will cancel out discriminatory tastes or whether they are sufficiently imperfect that prejudice can lead to discriminatory outcomes. This issue cannot be resolved in the abstract. We therefore look for evidence for or against the existence of discrimination at the level of the labor market.

13. David Neumark, “Sex Discrimination in Restaurant Hiring: An Audit Study,” Quarterly Jour- nal of Economics 101 (November 1996): 915–40.

Race Discrimination in the Labor Market 297 3.4. More on the Effect of Names. Roland Fryer and Steven Levitt take a somewhat different approach to studying the effect of names on outcomes.14 They match the birth records of approximately 40 percent of black women born in California in 1973 and 1974 with the birth records of their children. Black women with distinctively black names, on average, had worse adult outcomes as measured by factors such as their education or the per capita income in the census tract where they live. How- ever, these differences can be explained almost entirely by the fact that women with distinctively black names were themselves more likely to come from disadvantaged circumstances. When Fryer and Levitt control for known differences in the birth circumstances of the women in their study, the relation between adult outcomes and name becomes very small and socially unimportant. How can we reconcile this finding with the earlier section that reported that applicants with black-sounding names were less likely than those with white-sounding names to be called for interviews? Unfortunately, there are two basic responses to that question, with opposite implications. The first is that the issues raised earlier about audit studies are important. Blacks are able to avoid firms that discriminate and readily find jobs with firms that do not. As a result, even if they are less likely to be called for interviews at, say, 30 percent of randomly chosen firms, the existence of even this significant minority of discriminating firms is unimportant for their wages and employment levels. The second response points out that in the audit study employers can use black- and white-sounding names to guess the race of the applicant. Once the interview takes place, the employer knows whether the applicant is black or white. Having a white- sounding name does not help the applicant once she reaches the interview stage if the employer will discriminate against her on the basis of her race. Of course, she may im- press some employers who would not have called her for an interview if her name had been distinctively black. In that case, having a white-sounding name will be beneficial. On the other hand, getting an interview with someone who is unlikely to hire her because she is black detracts from time spent more usefully searching for work. Neither of these effects is likely to be large, so it is not surprising that blacks with distinctively black names do only a little worse than blacks with distinctively white names. It does not mean that there is little discrimination against blacks regardless of name. At this point we have reached the end of what can be learned from the audit studies. We turn to statistical analysis of wage differentials.

3.5. Wage Differentials. Ideally we would like to compare the earnings of blacks and whites who were identical in every respect except their race. As we have seen in earlier chapters, in many situations our concern centers on our inability to hold a large number of factors constant. When we examined the effect of teenage motherhood on the mothers, we found that controlling for measured variables reduced the negative impact only slightly. When we compared sisters, we found much smaller negative effects.

14. Roland G. Fryer Jr. and Steven D. Levitt, “The Causes and Consequences of Distinctively Black Names,” Quarterly Journal of Economics 119 (August 2004): 767–805.

298 Chapter 11 Table 11.3 Determinants of Log Hourly Wages for Men Aged 35 to 42

Controlling Controlling Controlling Black Hispanic for Age/10 for Education for AFQT

1 –0.37 –0.20 0.17 (0.04) (0.05) (0.14) — — 2 –0.29 –0.11 0.17 0.10 (0.03) (0.05) (0.12) (0.00) — 3 –0.11 –0.03 0.07 0.26 (0.04) (0.05) (0.13) — (0.01) 4 –0.17 –0.04 0.13 0.06 0.15 (0.03) (0.05) (0.12) (0.01) (0.02)

Source: Kevin Lang and Michael Manove, “Education and Labor Market Discrimination” (Working Paper 12257, National Bureau of Economic Research, Cambridge, MA, 2006). Note: See footnote 15 for a reminder about interpretation of logs.

Even then, we worried that perhaps the sister who became a teen mother differed in other ways from the sister who did not. When we try to measure race discrimination in the labor market, the problem becomes even more difficult, because it is unclear what factors we would like to hold constant if we could measure them. As we will see, the conclusion we reach depends considerably on what factors we believe should be held constant.

3.5.1. Men. Table 11.3 shows the results of comparing wages for black, white non- Hispanic, and Hispanic men aged 35 to 42 based on data from the National Longitu- dinal Survey of Youth. The first row of the table shows (log) hourly earnings differentials among men, controlling only for age. Controlling only for age, black men earned about 31 percent less than white men, while Hispanic men earned about 18 percent less.15 Controlling for age is fairly uncontroversial. Beyond age, almost everything we might consider holding constant could differ because of labor market discrimination. More educated workers tend to earn more than less educated workers, and blacks tend to have less education than whites, but their lower level of education might reflect the fact that they believe they gain less from education. Married men earn more than unmarried men. Blacks are less likely to be married, but, as we have discussed, this might reflect their poorer earnings prospects. The second row shows that controlling for education does somewhat reduce the estimated wage differential for blacks and more or less eliminates it for Hispanics.

15. Recall that log wage differentials are approximately percentage differences, so a coefficient of –.37 implies that the wages of black men are about 37 percent lower than those of white men. The calculations here are more precise.

Race Discrimination in the Labor Market 299 Nevertheless, the black-white earnings gap remains substantial. Black men earn about 26 percent less than white men, even controlling for education as well as age. Derek Neal and William Johnson have argued that we should control only for those factors that cannot respond to market discrimination.16 In particular, they argue that we should control for AFQT (the Armed Forces Qualifying Test),17 a measure of cognitive ability similar to an IQ test. The AFQT was administered to members of the Na- tional Longitudinal Survey of Youth sample in 1980 when they were still relatively young (15–22). The AFQT is likely to capture pre–labor market discrimination—differences in family advantages and in school quality. Neal and Johnson argue that, to the extent that people can choose to invest effort in raising their AFQT, the labor market incentives are as large for blacks as for whites. The third row mimics the authors’ exercise and obtains similar results. Conditional on age and AFQT, there is a small although not trivial difference in the earnings of white, black, and Hispanic males. The estimated wage differential is very small for His- panics and about 10 percent for blacks. Thus, Neal and Johnson conclude that wage differentials are primarily caused by premarket factors, not by discrimination in the labor market. Other researchers have challenged the finding of Neal and Johnson on the grounds that they should have controlled for other factors even if these factors are determined after workers enter the labor market or might be determined by expectations of the labor market. In particular, results may differ if we control for occupation and/or educational attainment. For a given level of AFQT, blacks and Hispanics obtain more education than do whites. The fourth row of table 11.3 shows what happens when we control for both education and AFQT as well as age. The black-white earnings differential rises to about 16 percent and is highly statistically significant. The Hispanic-white differential is still small and insignificant. Should we control for education or just for AFQT? If blacks and whites with the same levels of AFQT and education are otherwise similar, we should control for both. However, if not, controlling for both may give a very misleading picture of discrimination. In particular, there are no controls for school quality and family background that might explain part of the black-white earnings differential. Controlling for measured school quality does not change the results in the fourth row in any important way.18 In addition, the estimates in the table are based only on individuals with earnings. Black men are much more likely to have no earnings than are white men. Prime- age men without earnings are likely to be individuals who would have very low

16. Derek A. Neal and William R. Johnson, “The Role of Premarket Factors in Black-White Wage Differences,” Journal of Political Economy 104 (October 1996): 869–95. 17. The AFQT correlates well with other measures of what is sometimes called general intelligence, or IQ. Like other tests, it undoubtedly reflects a combination of environmental and genetic factors. Neal and Johnson do not claim that it is a measure of innate ability. 18. See Kevin Lang and Michael Manove, “Education and Labor Market Discrimination” (Work- ing Paper 12257, National Bureau of Economic Research, Cambridge, MA, 2006).

300 Chapter 11 earnings if they did work. Correcting for nonemployment somewhat increases the estimated wage gap. The estimates in table 11.3 treat the black-white earnings differential as constant regardless of education or level of cognitive skills. A more careful examination reveals that the earnings differential is lower among more educated workers and may be lower among those with greater cognitive skills. There is essentially no wage differential between black and white men with a college degree and an AFQT score one standard deviation above the mean. On the other hand, there is a substantial earnings gap for high school dropouts.19 Studies of employers are consistent with this finding. Joleen Kirschenman and Kathryn Neckerman found that employers were suspicious of low-skilled blacks. In the view of many employers, poor, low-skilled blacks from the inner city (note the mixture of race, class, and geography) were “unskilled, uneducated, illiterate, dishonest, lacking initiative, unmotivated, involved with drugs and gangs, did not understand work, had no personal charm, were unstable, lacked a work ethic, and had no family life or role models.”20

3.5.2. Hispanics. Although our focus in this chapter is on race discrimination against blacks, it is worth noting that the results in table 11.3 for Hispanics are typical of what has been found generally in the literature. There is general agreement that most or all of the earnings differential between white and non-Hispanic white men can be explained by differences in education and language knowledge. The limited evidence also supports this conclusion for Asian men.21

3.5.3. Women. The situation among women is quite different than among men. Without controls, the black-white wage gap is considerably smaller for women than for men, and it narrows by more than one-fourth if we control for education, which also eliminates the white-Hispanic gap. Controlling for AFQT reverses the wage differential, and, even controlling for both AFQT and education, black women earn more than white women do. Before we declare victory that labor market discrimination has been eliminated for black women, we must consider who works. We can see from table 11.4 that white women who work are somewhat more educated than those who do not work, but the

19. Calculated from Neal and Johnson, table A2. 20. Joleen Kirschenman and Kathryn M. Neckerman, “‘We’d Love to Hire Them, But . . .’: The Meaning of Race for Employers,” in Christopher Jencks and Paul E. Peterson, eds., The Urban Underclass (Washington, DC: Brookings Institution, 1991), 208. See also Phillip Moss and Chris Tilly, Stories Employers Tell: Race, Skill and Hiring in America (New York: Russell Sage, 2001). 21. Stephen Trejo, “Why Do Mexican Americans Earn Low Wages?” Journal of Political Economy 105 (December 1997): 1235–68; Walter S. McManus, “Labor Market Costs of Language Disparity: An Interpretation of Hispanic Earnings Differences,” American Economic Review 75 (September 1985): 818–27; Dan Black, Amelia Haviland, Seth Sanders, and Lowell Taylor, “Why Do Minority Men Earn Less? A Study of Wage Differentials among the Highly Educated,” Review of Economics and Statistics 88 (May 2006): 300–313.

Race Discrimination in the Labor Market 301 Table 11.4 Educational Attainment of Women, by Work Status and Race (Percentages)

Less than College Degree High School High School Some College or More White Women With Wage 5.6 41.1 25.2 28.1 No Wage 13.5 40.3 21.5 24.6

Black Women With Wage 8.2 40.4 34.9 16.4 No Wage 25.6 43.3 22.8 8.6

Hispanic Women With Wage 17.9 41.1 21.6 19.4 No Wage 49.2 40 9.3 1.4

Source: Author’s calculation from the National Longitudinal Survey of Youth, 1979.

difference is not large. Only a little less than 6 percent of white women who work have less than a high school education, compared with almost 14 percent of those who do not work. However, a substantial fraction (25 percent) of those who do not work have a college degree or more, but this is not that different from the fraction of those who do work (28 percent). It is reasonable to guess that the white women who do not work would have lower than average wages if they joined the labor force. However, they would probably not earn much less than the average. In contrast, black women who work are, on average, much more educated than those who do not work. Nonworking black women are over three times more likely to be high school dropouts and about half as likely to have completed college. If these women worked, they would earn wages well below those of the average working black women. The situation is even more extreme among Hispanic women. The proportion of His- panic women with a college degree or more is higher for working Hispanic women than for working black women. But almost none of the nonworking Hispanic women have graduated from college. Derek Neal has estimated the extent of the bias in the estimated black-white wage differential for women from ignoring nonparticipation.22 For data similar to those used here, he calculates that black women earn about 17 percent less than do white women. However, when he adjusts the difference for the nonparticipation of some women, he calculates that the overall wage difference is above 22 percent. In general, Neal’s results and table 11.4 suggest that nonworking black and Hispanic women are

22. Derek Neal, “The Measured Black-White Wage Gap among Women Is Too Small,” Journal of Political Economy 112 (February 2004): S1–S28.

302 Chapter 11 drawn disproportionately from the more disadvantaged segments of the population, while nonworking white women are much more representative of the general population of white women. Some earnings gap undoubtedly remains between black and white women after controlling for AFQT and education.

3.6. Hiring and Employment Differentials. Assessing employment differentials between blacks and whites is even more difficult than assessing wage differentials. Every- thing else equal, most people would prefer to have a higher wage than to have a lower wage. But it is not true that everyone would prefer to have a job than not to have a job. Some people prefer not to work because they are doing something else more valuable to them, such as going to school, taking care of their children, or enjoying retirement. If we find that, among women married to men with high earnings, black women are more likely to be employed than are white women, we are not likely to attribute that to discrimination against the white wives of men with high earnings. It is also important to distinguish between unemployment and nonemployment. In theory, unemployed workers are those who are willing to work at the going wage rate for someone with their skills but are unable to find employment, while those who are out of the labor force do not wish to work at that wage rate. In practice, to be unemployed one must be actively looking for work or waiting to (re)start a job at a fixed date. Some people who are nonemployed are not actively looking for work because they have become discouraged. Others are actively looking for work, but not very seriously. Although many individuals are clearly out of the labor force, others move back and forth between unemployment and being out of the labor force.23 Therefore, for some purposes, it may be more meaningful to distinguish between nonemployment and employment, while for others we may wish to distinguish among employment, unemployment, and being out of the labor force despite the fact that the categories are imperfect. As shown in table 11.5, unemployment and nonemployment are much more common among black men than among white men. In an average month in 2004, 15.4 percent of black men aged 25 to 54 were unemployed, compared with only 3.5 percent of white men in the same age group. Similarly, among those age 25–44, 17.1 percent of black men reported no employment for the year, compared with only 7.6 percent of white men. White men were correspondingly more likely to have worked full-time for the entire year. The situation among women is more complex. Black women are more likely to be unemployed in any given month, but they are also more likely to be employed full- time, and they are substantially more likely to be employed full-time for the entire year. The results for women suggest that we must use caution in ascribing employment differences to employment discrimination.

23. Kim B. Clark and Lawrence H. Summers, “Labor Market Dynamics and Unemployment: A Reconsideration,” Brookings Papers on Economic Activity (1979): 13–60.

Race Discrimination in the Labor Market 303 Table 11.5 Proportion of Labor Force Status of Men and Women Aged 25–54, by Race and Sex, 2004

White Men Black Men White Women Black Women Monthly Average Employed Usually Full-Time 84.0 70.8 57.1 62.5 Employed Usually Part-Time 4.0 4.4 15.4 8.6 Unemployed 3.5 15.4 2.8 5.9 Not in Labor Force 8.5 8.6 24.7 23.0

Employed Year-Round Employed Full-Time, Year-Round 76.2 64.7 48.4 56.2 Employed Part-Time or Part-Year 16.3 18.2 28.5 22.4 Not Employed 7.6 17.1 23.1 21.4

Source: Current Population Surveys, http://www.bls.gov/cps/.

Differences in employment and unemployment rates may reflect other differences between whites and blacks. In our discussion of wage differentials, we saw that skill differentials account for much of the black-white wage differential among men. Figure 11.8 shows the relation between AFQT and nonemployment for a sample of men aged 35 to 43 in 2000. We can see that the relation is quite different for blacks and whites. Although white men with higher AFQT scores are more likely to be employed than are white men with lower AFQT scores, the difference is modest. In contrast, employment increases rapidly (and nonemployment decreases rapidly) with AFQT among black men. The result is that black and white men with AFQT scores one standard deviation above the mean (roughly comparable to an IQ of 115) have similar rates of nonemployment. Black men with lower AFQT scores have a much higher rate of nonemployment than their white counterparts, while the relation is reversed for those with higher AFQT scores. Of course, the difference in employment rates might reflect discrimination. When labor markets are imperfect, if blacks and whites apply for similar jobs and if firms tend to hire whites in preference to blacks, blacks will have longer unemployment durations. They may also avoid competition with whites by applying for lower-wage jobs or simply drop out of the labor force. As discussed earlier, audit studies find that blacks are less likely than whites to be called for interviews or offered jobs, but we also noted that audit studies do not take into account the possibility that blacks are able to avoid applying for jobs where the employer discriminates against them. One way that blacks might avoid discrimination is by applying for jobs where the hiring officer is black. In fact, establishments where blacks make the hiring decisions are more likely to hire blacks. This does not mean that they are less likely to discriminate against blacks or more likely to discriminate in their favor. It is possible that the sorts of jobs where blacks make hiring decisions are also the sorts of jobs for which blacks are more likely to apply and to be the most qualified applicant. For example, we would not

304 Chapter 11 Nonemployment Rate 0.3

0.2

Black 0.1

White

0.0

–2 –1012 Standardized Value of AFQT Figure 11.8 Relation between Nonemployment and AFQT for Men Aged 35–43, by Race, 2000 be surprised if blacks were more likely to be in charge of hiring in stores located in heavily black areas and if these same stores received a higher fraction of black applicants. Michael Stoll, Steven Raphael, and Harry J. Holzer find that at establishments with a black hiring officer, the proportion of black applicants for unskilled and low-skilled jobs is 27 percentage points higher than in establishments with a hiring officer who is not black.24 However, controlling for observable differences reduces this disparity to 12 percentage points. It is possible that the remaining difference is a true effect or the result of additional unmeasured factors. The authors also find that when the hiring officer is black, the probability that a black is actually hired is about 20 percent higher. However, when they control for observed differences among establishments, including the proportion of applicants who are black, this difference falls to 6 percent and is statistically significant only at the 10 percent level.

3.7. A Test from Professional Sports. Suppose fans care only about how well their team is doing and not about the racial composition of the team. Owners also do not care about the race of their players. Instead, they want to win, either because they care directly about winning or because that is all fans care about, and the owners want to maximize their profits. In either case, for a given wage bill, owners would want to hire the best players they can regardless of race. This means that two teams with the same wage bill should have the same probability of winning regardless of the racial composition of the team.

24. Michael A. Stoll, Steven Raphael, and Harry J. Holzer, “Black Job Applicants and the Hiring Officer’s Race,” Industrial and Labor Relations Review 57 (January 2004): 267–87.

Race Discrimination in the Labor Market 305 Table 11.6 Incarceration Rate per 100,000 Population by Age, Sex, and Race/Ethnicity, Midyear 2004

Men Women Age All White Black Hispanic All White Black Hispanic

All 1,348 717 4,919 1,717 123 81 359 143 18–19 1,727 911 5,473 1,957 112 71 262 162 20–24 3,255 1,641 11,054 3,577 264 191 625 304 25–29 3,390 1,666 12,603 3,606 283 203 746 268 30–34 3,060 1,691 10,979 3,438 330 237 905 313 35–39 2,755 1,607 10,036 2,866 346 238 993 331 40–44 2,187 1,314 7,993 2,403 247 162 764 271 45–54 1,162 664 4,546 1,652 101 63 327 136 55+ 247 170 898 473 11 8 29 25

Source: Page M. Harrison and Allen J. Beck, “Prison and Jail Inmates at Midyear 2004,” Bureau of Justice Statistics Bulletin (April 2005): 11. Note: “White” and “Black” exclude Hispanics.

Stefan Szymanski has applied this test to professional soccer in England.25 He finds that, as expected, having better-paid players greatly increases a team’s performance. However, among teams with the same wage bill, those with a higher proportion of blacks do better. In other words, at the same cost, teams could do better by hiring more blacks relative to whites, presumably because, for a given skill level, blacks earn less than whites. Either the fans care about the race of the players and not just about winning or the owners of the team are not maximizing profit because they are prejudiced against blacks.

3.8. Incarceration. We will discuss discrimination in the justice system in chapter 13. Because spending time in prison is likely to make it more difficult to find a good job, it is worth spending some time pointing out the high proportion of black men who spend time in prison or jail. This information is summarized in table 11.6, which shows the number of people in prison or jail by age, sex, and race/ethnicity. It is striking that over one-eighth of black men aged 25 to 29 are in prison or jail. Almost 5 percent of all black men are incarcerated. In contrast, less than 1 percent of white men and less than 2 percent of Hispanic men are in prison or jail. Based on imprisonment rates in 2001, the Bureau of Justice Statistics estimated that 32.2 percent of black men, 17.2 percent of Hispanic men, and 5.9 percent of white men would spend some time in a federal or state prison.26 Many more will spend some time in a local jail.

25. Stefan Szymanski, “A Market Test for Discrimination in the English Professional Soccer Leagues,” Journal of Political Economy 108 (June 2000): 590–603. 26. Thomas P. Bonczar, “Prevalence of Imprisonment in the U.S. Population, 1974–2001” (Spe- cial Report, Bureau of Justice Statistics, August 2003).

306 Chapter 11 Clearly most prisoners are unable to contribute significantly to their families while incarcerated. But the consequences of spending time in prison may be more serious. Men who have spent time in prison have lower wages and lower employment than apparently similar men who have not spent time in prison.27 But apparently similar men who do and do not go to prison are likely to be different along other dimensions that may affect their ability to get and keep a good job. There is fairly strong evidence that people with initially high earnings suffer a substantial earnings loss from imprisonment.28 The evidence that imprisonment affects low-skilled men adversely is less compelling but points in the same direction.29 There are a number of ways in which having been in prison could hurt a worker’s job opportunities. By taking him out of the labor market, it reduces his experience and thus the level of skills he has acquired or worsens his work habits. In addition, prisoners lose social contacts and therefore have fewer friends who can refer them to good jobs. Finally, employers may infer that someone who has been in prison is not likely to be a good worker. An audit study by Devah Pager sheds some direct light on the stigma effect of past imprisonment.30 She used one pair of white male and one pair of black male auditors. Each pair consisted of 23-year-old college students who were matched for similarity of appearance and style of presentation. For each job application, one member of the pair was assigned to “have a criminal record”—eighteen months of time served for felony possession of cocaine with intent to distribute. The order in which they applied for the job was also randomized. In the white pair, the individual “without a criminal record” received a callback 34 percent of the time compared with 17 percent for the other individual. The black “without a criminal record” received a callback only 14 percent of the time, and the other black tester received a callback only 5 percent of the time. There were thus large negative effects of both being black and having a criminal record. In the previous chapter we discussed the possibility that blacks and whites are treated differently because of statistical discrimination. Part of the different treatment of blacks and whites may reflect the greater rate of imprisonment among blacks.

4. Testing for Discrimination: Legal Perspectives The Equal Pay Act of 1963 made it illegal to offer workers different wages for the same job on the basis of race. In fact, this practice was less widespread than might have been expected. Even in the south during the 1930s, firms frequently paid blacks and whites the same wage if they were doing the same job. Robert Higgs reports that in 1936, in

27. For a review see Bruce Western, Jeffrey R. Kling, and David F. Weiman, “The Labor Market Consequences of Incarceration,” Crime and Delinquency 47 (July 2001): 410–27. 28. Joel Waldfogel, “The Effects of Criminal Conviction on Income and the Trust Reposed in the Workmen,” Journal of Human Resources 29 (Winter 1994): 62–81. 29. Western, Kling, and Weiman. 30. Devah Pager, “The Mark of a Criminal Record,” American Journal of Sociology 108 (March 2003): 937–75.

Race Discrimination in the Labor Market 307 southern plants with both black and white laborers, only 28 percent paid more to white laborers than to black ones.31 In a 1935 survey of 109 southern cities, only 9 paid different wages to black and white common laborers. Of course, many firms may have effectively paid different wages for the same work by assigning different job titles to blacks and whites.

4.1. Intent and Adverse Impact. The 1964 Civil Rights Act addressed discrimination in employment and promotion, which prior to the act was commonly practiced. Job advertisements often stated the sex and the race or ethnicity of the person sought. Disparate treatment on the basis of race is now clearly illegal. If blacks are less likely to be hired or promoted within a group of equally qualified candidates, the firm is guilty of discrimination. In considering whether disparate treatment has occurred, intent can be an important element. However, firms may discriminate in a more subtle and possibly unintentional way. They may have a hiring or promotion policy that uses characteristics that are correlated with race or sex. For example, if a firm requires that workers in a certain job be at least six feet tall, the policy will exclude more women than men. Even if the firm hired women and men in proportion to their numbers among the “qualified” applicants, overall, women would be less likely to be hired. Such a policy would be said to have an adverse impact on women. Whether this disparate impact on women constitutes discrimination is a complicated question, because the goal of the policy may not be discriminatory. In a classic case, Griggs et al. v. Duke Power Company, the Supreme Court considered the requirements for admission to a training program for skilled craft workers. Admission was limited to those with high school diplomas and IQ scores above the national median. The court ruled that Duke Power provided no evidence that test scores were job-related and that prior to the establishment of the education requirement, many workers without high school diplomas had performed successfully in these craft jobs. The important point is that the court found that discrimination existed without determining that the policy was intentionally discriminatory. Instead it found that discrimination existed because the required qualifications reduced the number of qualified blacks relative to qualified whites and because Duke Power was unable to establish the business necessity of the requirements.32 In a series of subsequent cases, the Supreme Court initially extended the Griggs decision and then effectively reversed it. In 1989, the court reduced the requirement from “business necessity” to “business justification,” but the 1991 Civil Rights Act

31. Robert Higgs, “Black Progress and the Persistence of Racial Economic Inequalities, 1865– 1940,” in Steven Shulman and William Darity Jr., eds., The Question of Discrimination: Racial In- equality in the U.S. Labor Market (Middletown, CT: Wesleyan University Press, 1989), 9–31. 32. Chief Justice Warren Burger wrote: “The Act requires the elimination of artificial, arbitrary, and unnecessary barriers to employment that operate invidiously to discriminate on the basis of race, and, if, as here, an employment practice that operates to exclude Negroes cannot be shown to be related to job performance, it is prohibited, notwithstanding the employer’s lack of discriminatory intent. . . . The Act does not preclude the use of testing or measuring procedures, but it does proscribe giving them controlling force unless they are demonstrably a reasonable measure of job performance.”

308 Chapter 11 reestablished that a defendant must prove that “the challenged practice is job related for the position in question and consistent with business necessity.” Once the plaintiff has established disparate impact, the burden of proving business necessity falls on the employer. The 1991 Civil Rights Act clarified that the challenged practice is also illegal if an alternative practice would have a less disparate impact and be equally effective for the employer. However, if the employer established the business necessity of the original practice, the burden of proving the existence of an alternate practice would fall on the plaintiff. The 1993 Saint Mary’s Honor Center v. Hicks decision effectively required proof of discriminatory intent.

4.2. Testing for Discrimination. Court cases frequently center around the issue of whether a group is underrepresented relative to its “availability.” It therefore becomes critical to determine the numbers of available workers from different groups. Each side will argue that availability should be determined on the basis of the number of each group within some geographic area or occupation. The question of underrepresentation may hinge critically on how narrowly or broadly the area or occupation is defined. One way to get around the apparently arbitrary choice of occupation or area is to use the number of applicants from each group. However, it can be problematic to use applicants, because if a company has a reputation for discriminating against some group, members of that group are less likely to apply. Finis Welch describes the case of a plant with a nearby population of whites and a more distant population of blacks from which it could attract workers.33 If the court relied on the fraction of blacks in the plant’s work force to determine representative- ness, the extent to which blacks were underrepresented in the plant presumably depended on how broadly the court defined the relevant geographic area from which it should draw workers. This makes it tempting to rely on applications to determine whether blacks were underrepresented at the plant. One important feature of the plant was that it did not advertise its openings but relied on word of mouth to attract applicants. The data suggest that, over the course of about six weeks, hiring any applicant increased white applications by about 7. Hiring whites had no effect on black applicants. Hiring a black applicant, however, increased black applications by about 11.2. At the time of the court case, 26 percent of applicants were black and 23 percent of those hired were black. There was no statistically significant difference between the proportions of applicants and hires who were black. However, if instead of 23 percent, 26 percent of hires had been black, more than 26 percent of applicants would have been black. Welch finds that if instead 45 percent of the hires had been black, 45 percent of the applicants would have been black. Therefore, a small discriminatory benefit for whites among those actually hired translated into a large difference in applications and thus in actual hires.

33. Finis Welch, “Affirmative Action and Discrimination,” in Steven Shulman and William Dar- ity Jr., eds., The Question of Discrimination: Racial Inequality in the U.S. Labor Market (Middletown, CT: Wesleyan University Press, 1989), 153–89.

Race Discrimination in the Labor Market 309 4.3. Statistical Significance and the Small Numbers Defense. There may be few areas in which statisticians have had more impact than in discrimination law. It is relatively rare that a firm has a clearly articulated policy of not hiring blacks. In any individual case, it is unlikely that the plaintiffs can establish that a black applicant was clearly better than the white who was hired and that the reason the black applicant was not hired was his race. Instead, the plaintiffs must establish a pattern in which the firm is revealed to hire fewer blacks than would be expected if it did not take race into account. Expert witnesses have established statistical significance as a basis for determining intent. If the degree of underrepresentation of blacks is unlikely to have occurred by chance, the firm is presumed to have discriminated. The case may focus on experts who disagree about the relevant comparison groups. In the jargon with which we have become familiar, they may argue about which factors they need to control. The expert for the plaintiff will argue that with the appropriate controls, the difference in hiring or promotion probability between blacks and whites is statistically significant. The expert for the defense will argue that a different set of controls is appropriate and that with the right controls, the difference is not statistically significant. Although the use of statistical testimony in court cases is in many ways a positive development, the growing importance of such testimony has two effects that may be less positive. First, when numbers are small, it is hard to find statistical significance. This means that court cases focus on larger firms even if these firms may be less likely to engage in discrimination. In a related study, Jonathan Leonard examined which firms were subject to affirmative action compliance reviews.34 He found that compliance reviews were actually less likely in establishments with the lowest proportions of minorities. One explanation for this result is that when the number of minorities is small, it is very hard to determine whether they are statistically less likely to be promoted. The other problem with the reliance on statistical testimony is that it does not take into account the process by which cases are brought to court. If plaintiff lawyers know that they will not win their case without evidence of statistically significant differential treatment, the only cases that will appear in court are those for which (in at least one expert’s view of the appropriate specification) the treatment was statistically different. This raises two problems. The first is that a firm that wants to discriminate can do so provided that its discrimination can be maintained at a level that does not lead to statistically significant differences. The second is that, if unaware of the statistical significance issue, 5 percent of nondiscriminating firms will have hiring outcomes that differ at the 5 percent significance level from what would be expected under equal treatment. Plaintiffs lawyers will be willing to accept such cases because the statistical significance of the difference will provide strong evidence of discrimination. To see the difficulty, consider the following. It is highly unlikely that any given individual will win the grand prize of a major lottery. The local district attorney observes that Jane Smith has won the lottery. A statistician confirms that, given the number of lottery tickets she has purchased, the odds that Jane Smith would win the lottery were

34. Jonathan S. Leonard, “Affirmative Action as Earnings Redistribution: The Targeting of Com- pliance Reviews,” Journal of Labor Economics 3 (July 1985): 363–84.

310 Chapter 11 less than one in twenty. The district attorney therefore charges Jane Smith with lottery fraud. The example seems absurd. Jane Smith would not be convicted without direct evidence of fraud. However, by its very nature, discrimination is often subtle, and there is often no direct evidence other than the fact that blacks are underrepresented among the workers hired and/or promoted at the firm. If we rely on statistical significance, we will unfairly conclude that some nondiscriminating firms discriminated. If we do not rely on statistical significance, we will have great difficulty establishing that discrimination occurred.

5. Affirmative Action in Employment In 1965, President Lyndon Johnson signed Executive Order 11246, which imposed equal employment opportunity on federal contractors with at least $50,000 in federal contracts or fifty or more employees and required them to take “affirmative action” not to discriminate on the basis of race, religion or ethnicity.35 Initially there was little enforcement of the executive order. Nevertheless, there is good evidence that contractors responded to the executive order by increasing minority employment even in its early years. After 1973, enforcement became stronger, leading to the establishment of the Office of Federal Contract Compliance Programs (OFCCP) in 1978.36 The concept of affirmative action has evolved over time and remains controversial not only in employment but in other areas. However, it should be noted that the term was not new when the executive order was signed. The 1964 Civil Rights Act permitted affirmative action as a response to demonstrated cases of discrimination: “If the court finds that the respondent has intentionally engaged in or is intentionally engaging in an unlawful employment practice charged in the complaint, the court may enjoin the respondent from engaging in such unlawful employment practice, and order such affirmative action as may be appropriate, which may include, but is not limited to, reinstatement or hiring of employees, with or without back pay (payable by the employer, employment agency, or labor organization, as the case may be, responsible for the unlawful employment practice), or any other equitable relief as the court deems appropriate.” Executive Order 11246 differs from the 1964 Civil Rights Act in that affirmative action is not required in response to proof of discrimination. Although it does require that contractors not discriminate (and requires remedies in cases of discrimination), it goes beyond this by requiring contractors to ensure that they do not discriminate. A contractor need not intend to discriminate in order to violate the order.

35. According to Harry Holzer and David Neumark, “Assessing Affirmative Action,” Journal of Economic Literature 38 (September 2000): 483–568, this executive order was merely a reaffirmation of an earlier executive order issued by President John Kennedy. However, it is generally agreed that it was Johnson’s order that established affirmative action; see, for example, Jonathan S. Leonard, “The Impact of Affirmative Action and Equal Employment Law on Black Employment,” Journal of Eco- nomic Perspectives 4 (Fall 1990): 47–63. 36. Leonard, 49–52.

Race Discrimination in the Labor Market 311 The concept of equal employment opportunity is now largely uncontroversial. This is a remarkable change from forty years ago. In contrast, affirmative action is controversial, in part because, when it is not used to correct past discrimination, it appears to violate equal opportunity. Yet because equal employment opportunity cases are often heavily influenced by statistical arguments, the distinction between equal opportunity and affirmative action is often unclear. For example, if blacks are underrepresented in employment in a job category, that is prima facie evidence of discrimination. Firms may try to maintain racial balance in jobs in order to ensure that they do not face charges of violating equal employment opportunity laws. Therefore, firms facing equal employment opportunity enforcement may not act noticeably differently from firms subject to affirmative action considerations. In many cases, it is even difficult to decide whether an action is intended to ensure equal employment opportunity or is a form of affirmative action. For example, is a firm that advertises job openings in both the main city newspaper and a minority newspaper ensuring equal employment opportunity or engaging in affirmative action? If 60 percent of white job seekers and 60 percent of black job seekers read the help wanted advertising in the main paper, the advertising designed to reach additional black job seekers constitutes affirmative action. However, if the firm reached 60 percent of whites through the main paper but could reach 60 percent of blacks only by advertising in both papers, we could reasonably construe its action as ensuring equal employment opportunity. What should we conclude if the firm would reach a lower proportion of blacks than whites if it advertised only in the main paper but would reach a higher proportion of blacks than whites by advertising in both? Or what if, after seeing the ad in the main paper, many blacks check the minority paper for an ad to see if the firm really is interested in hiring blacks? Consequently, it is difficult to distinguish between the effect of requiring equal employment opportunity and that of requiring affirmative action. Harry Holzer and David Neumark report the results of a survey of firms in Atlanta, Boston, Detroit, and Los Angeles.37 Some 55 percent said that they used affirmative action in recruiting, and 42 percent said they used affirmative action in hiring, almost all of which also used it in recruiting. Firms that used affirmative action in recruiting used a much broader range of avenues for finding employees. They were more likely to attract Hispanic and black female applicants. Those that used affirmative action in hiring were more likely to attract black male and black female applicants. Affirmative action in both recruiting and hiring is generally associated with a more rigorous screening process. Firms engaging in affirmative action are more likely to always check the education and criminal records of applicants, check their references, require them to take a drug test or physical, and use a written application. The one exception is that they are less likely to require a personal interview. Affirmative action does seem to have the anticipated effect in that it increases the numbers of minorities, especially

37. Harry J. Holzer and David Neumark, “What Does Affirmative Action Do?” Industrial and Labor Relations Review 53 (January 2000): 240–71.

312 Chapter 11 women, hired and/or their representation among the employees of the firm.38 There is also some evidence that the education gap between women and minorities, on the one hand, and white men, on the other, is larger in firms using affirmative action. Once hired, workers in firms using affirmative action are more likely to be subject to a formal performance evaluation system and receive training, and may be somewhat less likely to have a probationary period. The gap between the performance of white males and other workers is no larger (and possibly somewhat smaller) in firms practicing affirmative action than in those that do not. What should we conclude from these findings? The positive perspective is that all these correlations reflect the causal effects of affirmative action. According to this view, affirmative action requirements push firms to recruit more actively, which allows them to attract more minority and women applicants, to hire qualified minority and women workers, and to make up any deficiencies by having better training and supervision. This view suggests that affirmative action in private sector employment has positive effects on minority and women workers and possibly on all workers. If firms cast their nets more widely and screen more carefully, all workers have more incentive to invest in themselves so that they will be selected for the best jobs. More careful screening and evaluation should also encourage better matching of workers and jobs. A more pessimistic view is that the relation is not causal. Instead, certain firms adopt more rigorous screening and evaluation, perhaps because of the nature of the jobs they have. When screening, evaluation, and training are more important, firms need to search more widely. For these firms, affirmative action is not costly especially if it helps them recruit more applicants, and so they adopt it. According to this view, firms that find affirmative action costly simply do not adopt it because there is little enforcement of affirmative action by the OFCCP. Still, neither view suggests that, as currently enforced, affirmative action in private sector employment is costly.

6. Affirmative Action in Public Employment Some of the fiercest battles over affirmative action concerned public sector employment, particularly as firefighters and police officers. These highly paid and generally secure jobs had traditionally been the bastion of white men. Blacks were often underrepresented among new hires. Layoffs, when they occurred, were frequently based on seniority. Because blacks tended to have less seniority than whites, they were laid off in disproportionate numbers. Blacks turned to the courts for help. Justin McCrary reports that in many cities, police (and fire) department hiring was based on a civil service or similar exam.39 The courts found that general aptitude

38. Note that these two need not be the same. Affirmative action might increase the number of minorities hired, but might also increase their quit rate, so there would be no effect on the proportion of workers in the firm who are minorities. 39. The following is drawn from Justin McCrary, “The Effect of Court-Ordered Hiring Quotas on the Composition and Quality of Police,” American Economic Review (forthcoming).

Race Discrimination in the Labor Market 313 exams violated equal opportunity because of their disparate impact. Blacks were significantly less likely than were whites to achieve the highest scores, and the tests were not closely linked to the requirements of the job. However, linking the content of the exam to the nature of police work did not greatly reduce racial disparities. Blacks were still considerably less likely to do sufficiently well on the exams to be hired. Ultimately, the courts established either interim or long-term quotas in over 90 percent of the cases won by plaintiffs between 1969 and 1980. Because turnover in these occupations is low, many departments remained under the supervision of the courts for a long time. The net result of the quotas was that blacks with lower scores on the exams displaced whites with higher scores. What was the impact on the effectiveness of the police force? McCrary compared the change in crime rates in cities that lost in litigation over police hiring practices with those that did not. He found no measurable difference. For the most part, crime rates were comparable in such cities before and after litigation. There are two explanations for this finding. The first is that black police officers, although weaker in the dimensions measured by the tests, were stronger in some other dimension. For example, they may have understood the African American neighborhoods in their city and used that knowledge to combat crime more effectively. Since a written test could not easily measure such knowledge, the tests missed out on an important component of police ability. The other explanation is that the test score differences between the black police officers who were hired and the white applicants they displaced were not that large. The difference in average scores between whites and blacks was large. But the difference in the scores of the worst-performing whites who would have been hired and the best- performing blacks who were hired because of quotas was not. McCrary estimates that for the 1979 New York City exam, if the highest-scoring candidates were hired, the cutoff would have been 94 (with 110 the maximum), and the average would have been 96.8. With a strict quota of 50 percent for black and Latino hires, the average would have been 95.3. Without some study linking test score to performance as a police officer, we cannot be certain, but this difference does not look large.

7. Concluding Remarks In the labor market, the audit studies show clear evidence of prejudice, but it is less clear to what extent this prejudice translates into wage differentials. Conditional on measured skills and education, black men earn less than white men, but it remains an open question how much of the remaining difference can be explained by unmeasured skills. However, there are two clear results. The first is that if there is discrimination in the labor market, it is concentrated at the lower end of the skill distribution. Black and white male college graduates have similar earnings. It is black high school graduates and, particularly, high school dropouts whose labor market performance is much worse than that of their white equivalents. Second, the difference in the cognitive skills that blacks and whites bring to the labor market is an important source of the difference in their earnings. If we want to

314 Chapter 11 address the differences in earnings between blacks and whites, we must address the black-white test score gap. Opponents of the 1964 Civil Rights Act used to argue that “you can’t legislate morality.” Whether or not they were correct, it has become apparent that “you can legislate social change.” Although the landmark act was only one of the factors that helped narrow the black-white income gap in the second half of the twentieth century, it was an important one. Moreover, it dramatically changed access to education and public accommodations, which in turn helped blacks by giving them better educations and the ability to move to better jobs. But it is clear that the Civil Rights Act, judicial and executive decisions, and the civil rights movement did more than this. They did change morality. Court cases today are rarely about disparate treatment. They are about disparate impact, and one would have to be extremely cynical to believe that the reason is that employers are doing a better job of hiding their discriminatory intent. The debate today is two-fold. First, it is about burden of proof. If a practice hurts blacks, how do we decide whether the practice is nevertheless justified? Should plaintiffs have to show that the practice does not have a business justification, or should the employer have to show that it is a business necessity? Burden of proof will affect how hard firms work to avoid racial imbalance in their hiring. Placing a sufficiently high burden of proof on firms might push them to adopt quotas. Placing a high burden of proof on plaintiffs makes it impossible to establish discrimination without showing discriminatory intent. The second debate is about how we should address the low rate of progress since the mid-1970s and, in particular, what role, if any, affirmative action should play. As we have seen, there is little evidence that affirmative action as currently practiced has hurt private or public sector productivity. However, this does not address the concerns of those who believe that race-based preferences are morally wrong. These are significant and difficult issues. But this should not prevent us from recognizing that they are very different from the issues being debated fifty years ago.

8. Further Reading Bertrand, Marianne, and Sendhil Mullainathan. “Are Emily and Greg More Employ- able than Latisha and Jamal? A Field Experiment on Labor Market Discrimina- tion.” American Economic Review 94 (September 2004): 991–1013. Burstein, Paul. Equal Employment Opportunity: Labor Market Discrimination and Pub- lic Policy. Hawthorne, NY: De Gruyter, 1994. Darity, William, and Patrick Mason. “Discrimination in Employment: Codes of Color, Codes of Gender.” Journal of Economic Perspectives 12 (Spring 1998): 63–90. Donohue, John J., III, and James Heckman. “Continuous versus Episodic Change: The Impact of Civil Rights Policy on the Economic Status of Blacks.” Journal of Economic Literature 29 (December 1991): 1603–43. Fryer, Roland G., Jr., and Steven D. Levitt. “The Causes and Consequences of Distinc- tively Black Names.” Quarterly Journal of Economics 119 (August 2004): 767–806.

Race Discrimination in the Labor Market 315 Heckman, James J. “Detecting Discrimination.” Journal of Economic Perspectives 12 (Spring 1998): 101–16. Holzer, Harry, and David Neumark. “Assessing Affirmative Action.” Journal of Eco- nomic Literature 38 (September 2000): 483–568. Holzer, Harry, and David Neumark. “What Does Affirmative Action Do?” Industrial and Labor Relations Review 53 (January 2000): 240–71. Kirschenman, Joleen, and Kathryn M. Neckerman. “‘We’d Love to Hire Them, But . . .’: The Meaning of Race for Employers.” In Christopher Jencks and Paul E. Peterson, eds., The Urban Underclass (Washington, DC: Brookings Institution, 1991). Leonard, Jonathan S. “Affirmative Action as Earnings Redistribution: The Targeting of Compliance Reviews.” Journal of Labor Economics 3 (July 1985): 363–84. Moss, Phillip, and Chris Tilly. Stories Employers Tell: Race, Skill and Hiring in America. New York: Russell Sage, 2001. Neal, Derek A., and William R. Johnson. “The Role of Premarket Factors in Black- White Wage Differences.” Journal of Political Economy 104 (October 1996): 869–95. Smith, James P. “Race and Human Capital.” American Economic Review 74 (Septem- ber 1984): 685–98. Smith, James P., and Finis R. Welch. “Black Economic Progress after Myrdal.” Journal of Economic Literature 27 (June 1989): 519–64. Welch, Finis. “Affirmative Action and Discrimination.” In Steven Shulman and William Darity Jr., eds., The Question of Discrimination: Racial Inequality in the U.S. Labor Market (Middletown, CT: Wesleyan University Press, 1989), 153–89.

9. Questions for Discussion 1. Describe the trend in the black/white wage ratio since 1940. 2. Why have the earnings of black and white men converged over the past sixty years? What are the roles of migration, convergence in education quantity, declining labor force participation of low-skilled black workers, and government policy? 3. What was the significance of Executive Order 11246? 4. What are the strengths and weaknesses of using statistical evidence to influence or determine the decision in discrimination cases? 5. Discuss the role of burden of proof in disparate impact cases. 6. What have audit studies taught us about the existence of race discrimination? 7. Neal and Johnson find that, controlling for AFQT score, there is no difference between the earnings of black and white men. Is this compelling evidence that there is no discrimination against black men in the labor market? 8. If we control for both AFQT score and education, black men earn less than white men. Is this compelling evidence of discrimination against black men in the labor market? 9. The annual earnings of black women and white women in the labor force are roughly equal. Does this mean that there is no labor market discrimination against black women (relative to white women)?

316 Chapter 11 chapter 12

Race Discrimination and Education

B   U S    overwhelmingly by government and, to a much lesser extent, by nonprofit enterprises, the argument that competition will eliminate discrimination does not apply. We saw earlier that much of the black-white earnings differential can be explained by differences in the skills that blacks and whites bring to the labor market.

1. The Black-White Test Score Gap The fact that test scores explain a large fraction of the black-white wage differential suggests that it would be valuable to understand why this difference exists and what can be done to address it. Unfortunately, the discussion of this question is often conten- tious, in part because racist positions have often been supported by pseudo-scientific research1 and in part because our lack of understanding of the sources of intelligence makes it easy for people to hold strong opinions that are not readily contradicted by evidence. However, a few points are widely accepted. First, there is a significant difference in the IQ scores of blacks and whites that is not easily explained by cultural bias in the tests. Second, there is considerable evidence that intelligence is partially inherited. It is also widely, although not universally, agreed that much of the variation in intelligence among whites is genetic. The American Psychological Association concluded that approximately 75 percent of the variance in adult IQ is genetic.2 Arthur Jensen argued that since so much of the variance in IQ is genetic, the difference in environment between whites and blacks would have to be implausibly large

1. For a critique of this literature, see Stephen Jay Gould, The Mismeasure of Man (New York: Norton, 1981). 2. Ulric Neisser et al., “Intelligence: Knowns and Unknowns,” American Psychologist 51 (Febru- ary 1996): 77–101.

317 in order to explain the measured difference in IQ scores.3 However, there are good reasons for doubting this conclusion. First, kids are right; they really are smarter than their parents. Studies of changes in IQ within countries generally show that IQ scores have been rising by about three points per decade. Some countries have exhibited larger changes over similar time spans.4 Given a rough estimate that the black-white IQ gap is about fifteen points, this means that the difference in IQ between whites and blacks is about the same as between today’s adults and people born about fifty years ago. Although the environment experienced by blacks born in 1960 was probably better than that of the average white born in 1910, the difference is not so huge as to make an environmental explanation for the black-white test score gap unthinkable. Second, it is possible that a single environmental factor could account for one-third of the IQ differential. The consensus among lead experts is that each microgram of lead per deciliter of blood among young children is associated with an IQ loss of one- fourth point. In the 1980s, lead levels among young black children were about 20 micro- grams per deciliter higher than among whites. If the estimates of the IQ effects are correct, differences in exposure to lead would account for a five-point IQ difference. William Dickens and James Flynn argue that people’s cognitive ability influences their environment to some degree.5 To understand the importance of their argument, consider the following example. Suppose that schools give an IQ test to entering students and assign students to different classes based on their IQs. For the purposes of the thought experiment, we will assume that the test is entirely accurate, so it is a perfect measure of cognitive ability at the time the test is taken. Schools can follow one of two policies. In the first case, they assign students to heterogeneous classes. All students are mixed together regardless of IQ. In the second case, students with higher IQs are grouped together and placed in more challenging classes. For the purposes of the thought experiment, let us assume that children with different IQs benefit from different types of education, although the evidence supporting this claim is weak, at least for young children who do not have major learning disabilities. The situation is plotted in figure 12.1. When the children are all mixed together, they retain their original IQs. This is shown in the lower line in the figure. When the children are segregated, both groups increase their IQs over time. However, the children who started out with higher IQs benefit from their special classes and increase their IQs over time by more than the children with lower IQs do. This is shown in the higher line. In either case, if we know a child’s IQ when she is young, we can predict perfectly her IQ when she is older. In the case in which students were assigned to classes randomly, environment could tell us nothing about IQ because all students faced the same environment. But this is also true in the case in which IQ fully determines the envi-

3. Arthur R. Jensen, Educability and Group Differences (New York: Harper and Row, 1973). 4. James R. Flynn, “IQ Gains over Time,” in Robert J. Sternberg, ed., Encyclopedia of Human In- telligence (New York: Macmillan, 1994), 617–23. 5. William T. Dickens and James R. Flynn, “Heritability Estimates versus Large Environmental Effects: The IQ Paradox Resolved,” Psychological Review 108 (April 2001): 346–69.

318 Chapter 12 IQ When Old 140

120 Tracked

100 Mixed

80 100 120 IQ When Young Figure 12.1 Relation between Early and Late IQ (Conceptual) ronment. Once we know the student’s IQ on the original test, we know her IQ when she is older. Even though we designed the thought experiment to ensure a large role for the environment, Jensen would conclude incorrectly from the data that it played no role. Dickens and Flynn also emphasize the potential for multiplier effects. To use their sports analogy, suppose the World Cup comes to the United States. Lee sees a game, gets excited, and starts practicing in the backyard, but without further interaction with other children, his skills will not develop very much, and he will soon lose interest. However, if a number of children get interested in soccer, they may start playing during recess or after school. Each player gets better by playing more and by playing against players who are getting better themselves. If blacks primarily interact with blacks, and if they provide a less cognitively stimulating environment for each other because, on average, they come from more disadvantaged environments, even modest differences in environment can generate a large test score differential for the group. Perhaps the strongest evidence that environment can change the black-white test score gap is the fact that the gap has changed over time. The National Assessment of Educational Progress (NAEP) has been following 9-, 13-, and 17-year-olds since the 1970s.6 The tests measure achievement in reading, mathematics, and science. Figure 12.2 shows the evolution of the test score gap for 9-year-olds. For all three subjects, the test score gap narrowed from the beginning of testing until sometime in the mid-1980s and then held steady or grew before declining again between 1999 and 2004 in reading and mathematics (science knowledge was not tested in 2004).

6. Marianne Perie, Rebecca Moran, and Anthony D. Lutkus, NAEP 2004 Trends in Academic Progress: Three Decades of Student Performance in Reading and Mathematics, NCES 2005-464, U.S. Department of Education, Institute of Education Sciences, National Center for Education Statistics (Washington, DC: Government Printing Office, 2005).

Race Discrimination and Education 319 Differential 60

40 Science

Math 30

Reading 20

1971 1975 19801984 1988 1992 1996 1999 2004 Figure 12.2 White-Black Test Score Differential (NAEP, Age 9), 1971–2004

The story for 13-year-olds is similar (see figure 12.3), except that even in 2004, the test score gaps in math and reading were larger than they were in the mid-1980s. For 17-year-olds, the pattern is somewhat different (see figure 12.4). The gap reached a minimum in 1992. Given the evidence that the test score gap can change, we face the much more difficult task of determining how to change it. Scientific analysis of this question is hindered by the risk of committing what Jensen calls the “sociological fallacy.”7 If differences in test scores are associated with differences in family background, it does not necessarily imply that the test scores are caused by those family background characteristics in ways that can be manipulated by policy. For example, more educated parents raise children with higher test scores. Advocates of the view that environment is important interpret this finding as evidence that increasing parents’ education will benefit their children. Advocates of the view that environment is relatively unimportant interpret this finding as evidence that parents with higher IQs obtain more education and produce offspring with higher IQs. It would be misleading to suggest that there is a consensus about the reasons for the test score gap or what can be done about it. With this caveat in mind, I attempt to summarize what is known. It is clear that the gap exists when students first enter school, though Roland Fryer and Steven Levitt argue that the differential at this stage can be explained by family background factors other than race.8 The gap increases

7. Arthur R. Jensen, “How Much Can We Boost IQ and Scholastic Achievement?” Harvard Ed- ucational Review 39 (February 1969): 1–123. 8. Roland G. Fryer and Steven D. Levitt, “Understanding the Black-White Test Score Gap in the First Two Years of School,” Review of Economics and Statistics 86 (May 2004): 447–64. Fryer and Levitt are unusual in finding that the entire gap can be explained by factors other than race, but the general consensus is that a large fraction of the gap is explained by such factors.

320 Chapter 12 Differential Science Reading 40

Math 35

1971 1975 1980 1984 1988 1992 1996 1999 2004 Figure 12.3 White-Black Test Score Differential (NAEP, Age 13), 1971–2004

Differential 60

Math 50

40 Reading

Science 30

1971 1975 1980 1984 1988 1992 1996 1999 2004 Figure 12.4 White-Black Test Score Differential (NAEP, Age 17), 1971–2004 with education,9 at least through third grade,10 and standard measures of family background and the environment experienced by blacks and whites are insufficient to explain the difference in average test scores between them once students have been in

9. Meredith Phillips, James Crouse, and John Ralph, “Does the Black-White Test Score Gap Widen after Children Enter School?” in Christopher Jencks and Meredith Phillips, eds., The Black- White Test Score Gap (Washington, DC: Brookings Institution Press, 1998), 229–72. 10. Roland Fryer and Steven Levitt, in “The Black-White Test Score Gap through Third Grade,” American Law and Economics Review (forthcoming), find that the gap widens through third grade. There is no evidence that the gap on the NAEP widens between age 9 and age 13 or between age 13 and age 17.

Race Discrimination and Education 321 school.11 In particular, standard measures of school quality, such as class size, teacher experience, and teacher education, are similar for blacks and whites. However, there are some areas of difference in environment that may account for the unexplained difference in scores. First, blacks and whites differ significantly in parenting practices, and it appears that these differences can account for some of the gap.12 As a policy matter, it is difficult to think of changing parenting practices on a wide scale. It may be possible to use preschool programs such as Head Start as an instrument for educating parents as well as students. Indeed, as we saw in our discussion of preschool programs, this is one of the goals of Early Head Start, where the early results of the evaluation study indicate some success. Still, on a broader scale, parents are likely to resist attempts to change their behavior. Second, the way that people respond to blacks and whites may make the environment very different for them. Ronald Ferguson finds that teachers often have lower expectations for blacks, which then become self-fulfilling.13 Claude Steele and Joshua Aronson argue that fear of fulfilling the stereotype that blacks do poorly academically actually lowers performance among blacks.14 Black students do worse when they are asked to record their race on an exam, or when they are told that they are being tested on verbal reasoning ability as opposed to being tested on verbal reasoning but not being told that is the subject of the test. The work of Ferguson and of Steele and Aronson suggests that working with educators to undo the stereotypes can reduce the test score gap. The challenge is that because the stereotypes are self-fulfilling, they are not false. Therefore, undoing the stereotypes requires exposing teachers to experiences that contradict the stereotypes and lead them to develop more positive expectations. Over time, as the test score gap narrows, it is likely that the stereotypes will undo themselves. The issue is whether it is possible to accelerate this process. It is important to note that the stereotype argument does not suggest that black students will do better if they have black teachers (although other views may justify this conclusion). Because the stereotypes are self-fulfilling and thus correct, they are as likely to be held by black teachers as by white teachers, a position supported by some research.15

11. Christopher Jencks and Meredith Phillips, “The Black-White Test Score Gap: An Introduc- tion,” in their The Black-White Test Score Gap (Washington, DC: Brookings Institution Press, 1998), 1–51. 12. Meredith Phillips et al., “Family Background, Parenting Practices, and the Black-White Test Score Gap,” in Christopher Jencks and Meredith Phillips, eds., The Black-White Test Score Gap (Wash- ington, DC: Brookings Institution, 1998), 103–45. 13. Ronald F. Ferguson, “Teachers’ Perceptions and Expectations and the Black-White Test Score Gap,” in Christopher Jencks and Meredith Phillips, eds., The Black-White Test Score Gap (Washing- ton, DC: Brookings Institution, 1998), 273–317. 14. Claude M. Steele and Joshua Aronson, “Stereotype Threat and the Test Performance of Academically Successful African Americans,” in Christopher Jencks and Meredith Phillips, eds., The Black-White Test Score Gap (Washington, DC: Brookings Institution, 1998), 401–27. 15. Ferguson, 299.

322 Chapter 12 George Farkas aptly summarizes the difficulty of assessing the importance of teacher prejudice in school settings. He reviews studies that show important effects on student outcomes of teacher assessment of student behavior and reports that black teachers report more favorable assessments of black students’ behavior than do white teachers. He concludes: If minority and low-income students come to elementary school with less maturity and ability to concentrate and put forth effort on schoolwork, this alone could account for their making less academic progress during the year than White and middle class students. However, we are dealing with teacher judgments of student behavior, and . . . these differ according to the race of the teacher. If this is true, causal inferences become particularly difficult, since there are several competing possibilities that cannot be distinguished by the available evidence. First, it may be that minority students really do behave worse than White students, but only for White and not for minority teachers. Second, it may be that White teachers simply perceive the behavior of minority students to be worse than that of White students. Third, it may be that minority teachers judge the same behavior by minority students more leniently than White teachers do. Further, if students of color do in fact behave worse for White teachers, this could be because White teachers show prejudice toward them, it could be despite White teachers showing no prejudice toward them, or it could be because Black teachers are particularly skillful in getting Black students to behave. Here we see some of the difficulties involved in trying to determine the causes of lower minority achievement, particularly where they involve student effort and possible teacher discrimination.16 In his keynote speech at the 2004 Democratic Convention, Barack Obama decried “the slander that says a black youth with a book is acting white.” He was referring to the pressure that some black students may put on other blacks not to achieve. There is considerable debate about whether, relative to white students, black students put more social pressure on their peers not to achieve. The extreme version of the “acting white” hypothesis is that whites are rewarded socially for academic success, while blacks are punished. This extreme version is not supported by the data. The National Longitudi- nal Study of Adolescent Health asked students both whether they thought they were popular and to list up to five friends. Both black and white students with good grades are more likely to report that they are popular than are students with poor grades.17 However, the relation between self-described popularity and grades is more strongly positive for whites than for blacks.18

16. George Farkas, “Racial Disparities and Discrimination in Education: What Do We Know, How Do We Know It, and What Do We Need to Know?” Teachers College Record 105 (August 2003): 1119–46, quote on p. 1125. 17. Philip J. Cook and Jens Ludwig, “The Burden of ‘Acting White’: Do Black Adolescents Dis- parage Academic Achievement?” in Christopher Jencks and Meredith Phillips, eds., The Black-White Test Score Gap (Washington, DC: Brookings Institution, 1998), 375–400. 18. Roland G. Fryer and Paul Torelli, “An Empirical Analysis of ‘Acting White’” (Working Paper 11334, National Bureau of Economic Research, Cambridge, MA, 2005).

Race Discrimination and Education 323 Roland Fryer and Paul Torelli use the data on friends to draw a more complicated picture.19 With their measure, an individual is more popular the more people who list her among their friends and the more popular the people who do so. Thus, if Latisha and Latoya are each listed twice, but Latisha’s friends are each listed by ten people and no one lists Latoya’s friends, Latisha is more popular than Latoya. Fryer and Torelli limit themselves to friends of the same race.20 Among whites, this measure of popularity is higher for students with better grades. Among blacks, the measure of popularity is also higher for students with better grades, although the rate of increase is lower than for whites, but this positive relation reverses itself at the top of the grade distribution. Black students with straight As are less popular than black students with somewhat lower GPAs and about as popular as students with a B average. To some extent, this difference reflects the fact that black students with better GPAs have more white friends (who are not included in the popularity index for blacks) than do other black students, but this effect is not large enough to make the relation between popularity and grades as strong for blacks as it is for whites. One striking finding in the Fryer and Torelli study is that among blacks, the relation between popularity and grades is much weaker in schools in which fewer than 20 percent of the students are black. However, in schools that are at least 80 percent black, the relation between popularity and grades among blacks is similar to the relation among whites in schools that are at least 80 percent white. Fryer and Torelli interpret this as evidence that there is a penalty for “acting white” in environments in which there is a lot of interracial contact. When group solidarity is not threatened by the presence of a white majority, success is as valued among black students as it is among white students. As always, we must be cautious in interpreting correlations as causal relations. The fact that students with better grades are more popular does not prove that getting good grades makes them popular. It is equally plausible (in my own opinion, more plausible) that the same characteristics that help people do well in school help them to become popular. People who are funny tend to be more popular. And, although those of you reading this book for a college class may doubt this claim, people who are smart tend to be funnier than people who are not. So people who are smart will tend to both get good grades and be popular even if doing well in school does not make one popular. The critical question is whether grades or behaviors that raise grades, such as studying, are more highly rewarded (less severely punished) among whites than among blacks or whether they are simply more positively correlated with popularity among whites. The Fryer and Torelli study cannot resolve this question, but it does at least suggest that peer effects may give blacks, relative to whites, less incentive to undertake the sorts of behaviors that will help them raise their grades, particularly in schools in which blacks are a minority. Third, while blacks and whites attend schools that are similar with respect to some measurable inputs, they do differ in other measurable ways. Blacks are more likely than

19. Fryer and Torelli. 20. This is a somewhat simplified version of the algorithm but captures the essence of their approach.

324 Chapter 12 whites to attend schools with low attendance rates, high dropout rates, high poverty rates, high rates of teacher turnover, and, of course, more black students. It is plausible that blacks fall behind whites in part because blacks attend schools where they have worse peers, who either affect them directly or transform the school environment in such a way that it becomes more difficult for teachers to teach. The difficulty with this argument is that there is little evidence to support a strong peer effect. When we looked at neighborhood effects, we discussed the difficulty of distinguishing peer effects. People with similar characteristics are likely to interact, and people in the same school (or neighborhood) are subject to the same environmental influences. In an attempt to get around these problems, Joshua Angrist and I have looked at the effects of Boston’s METCO (Metropolitan Council for Educational Opportu- nity) program on student performance.21 Under the METCO program, mostly black students from Boston are bused on a voluntary basis to suburban schools that admit them, also on a voluntary basis. Because the METCO students are assigned on a space- available basis, their initial assignment is independent of the performance of other students in the classroom. Although the METCO students, on average, have much lower test scores than the suburban students, there is little evidence that they have much impact on the performance of the suburban students. However, the METCO program involves a relatively modest change in classroom composition. It is quite possible that the effect of large differences in classroom composition would be quite different.

2. Discrimination in Education There is little doubt that blacks historically faced discrimination in education. After reconstruction, southern states established separate school systems for blacks and whites and provided fewer resources per student to those serving blacks than to those serving whites. In its landmark 1896 decision, Plessy v. Ferguson, the Supreme Court upheld the principle of separate but equal, and followed up that decision a few years later in Cumming v. Richmond County Board of Education, which effectively permitted separate and unequal. Until the 1960s, well over 90 percent of southern blacks attended all-black schools.22 In 1954, approximately 0.001 percent of southern blacks were in majority-white schools.23 There is also little doubt that the schools attended by southern blacks were of lower quality than those attended by whites. In 1915, black pupil/teacher ratios were well over 50 percent higher than white pupil/teacher ratios in the south. The white school year was one-third longer, and teacher pay in white schools was more than twice as high.

21. Joshua D. Angrist and Kevin Lang, “Does School Integration Generate Peer Effects? Evi- dence from Boston’s Metco Program,” American Economic Review 94 (December 2004): 1613–34. 22. David Card and Alan B. Krueger, “School Quality and Black-White Relative Earnings: A Direct Assessment,” Quarterly Journal of Economics 107 (February 1992): 151–200. 23. Gary Orfield, “The 1964 Civil Rights Act and American Education” (paper presented to the Conference on the Civil Rights Act of 1964: A Thirty Year Perspective, Washington, DC, November 11–12, 1994).

Race Discrimination and Education 325 By 1954, the date of the famous Brown v. Board of Education decision, these disparities had been substantially reduced. In southern white schools, the pupil/teacher ratio was about one-eighth higher, term length about the same, and teacher salaries about 15 percent higher than in black schools. Moreover, teacher salaries had been converging rapidly. By 1956, the sole important difference in these inputs to education was the somewhat larger class size in southern black schools.24 Despite the convergence of measured inputs to education, in 1954 the Supreme Court struck down legal segregation of schools, concluding that separate is inherently unequal. The following year, in Brown v. Board of Education II, the court rejected immediate desegregation but ordered that desegregation be carried out “with all deliberate speed.” Despite these landmark decisions, desegregation of southern schools progressed slowly. It was not until 1968 that the Supreme Court ruled that voluntary desegregation programs that effectively left schools segregated were inadequate, and the court did not rule that busing could be used to enforce desegregation until 1971. Moreover, the original Brown v. Board of Education decision focused on legal segregation, noting that Boston, which was later to become the center of a violent battle over desegregation, had abolished segregation in 1855. Although the Court recognized that segregation existed outside the south, later decisions were required to address segregation when there was no formal law or school board decision establishing separate schools. Indeed, a case can be made that Brown was most important as a galvanizing force for the civil rights movement. Despite the requirement of “all deliberate speed” in Brown II, in 1964 only 2.3 percent of southern blacks attended majority-white schools.25 Title IV of the Civil Rights Act of 1964 desegregated public schools and strengthened the attorney general’s ability to support desegregation. In addition, Title VI outlawed discrimination in federally assisted programs. Many programs are at least partially funded by the federal government. For example, almost all colleges and universities receive some form of federal funding, if only through federal student loans. Consequently, the scope of the Civil Rights Act is extremely broad, especially in education. The rapid transition of southern schools followed the 1964 Civil Rights Act, which engaged the federal government in the judicial battles against segregation and cut off funding for discriminating school districts. Gary Orfield writes: Had the 1964 Civil Rights Act not been enacted and enforced in the South, the 1954 Brown decision might well have gone down in history as a well-meaning but failed effort to change fundamental social practices in a democratic society. Enact- ment of the law would have been inconceivable without the Brown decision and the civil rights movement it stimulated, but it was the operation of that law, not the courts, that made the Southern schools the nation’s most integrated within a few years. In fact, the Supreme Court did not create a set of principles actually capable

24. Card and Krueger, 166–69. 25. Orfield, 21.

326Chapter 12 of achieving desegregated schools until a half decade after the enactment and enforcement of the Civil Rights Act had begun to transform southern schools.26 Following the passage of the Civil Rights Act, desegregation of southern schools pro- ceeded rapidly. By 1970, one-third of southern black students attended majority-white schools.27 The forces furthering desegregation of schools diminished in the late 1970s and 1980s. Because U.S. cities are segregated by race, full desegregation required busing students from black neighborhoods to white neighborhoods and vice versa. This met resistance not only from conservative whites but also from some white liberals and some black parents who did not want their children bused. In 1974, in a key decision, Milliken v. Bradley, the Supreme Court ruled that the courts could not mandate busing across district lines if there was no evidence of significant wrongdoing by the suburban districts. Despite the fact that the Detroit public school population was overwhelmingly minority, the court ruled that desegregation must be accomplished within the district.

2.1. Resegregation. Some authors have argued that since the mid-1980s, schools have become resegregated.28 Gary Orfield writes: “This resegregation is happening despite the nation’s growing diversity, in particular the rapid expansion in the Latino student population, and is contributing to a growing gap in quality between the schools being attended by white students and those serving a large proportion of minority students.”29 My calculations from the National Center for Education Statistics Common Core reveal that in the 1987–1988 school year, the average black public school student attended a school in which 40 percent of students were white. By the 2001–2002 school year, this number had fallen to 31 percent. On the face of it, this supports the view that the schools are resegregating. However, as Orfield notes, the United States has become much more diverse. I calculate that over the same period, the average white public student went from attending a school that was 84 percent white to attending one that was 79 percent white. Because the public schools are more diverse, both whites and blacks are attending schools with fewer whites. Whites are attending schools with more blacks, more Latinos, more Asians, and possibly even more native Americans. Blacks are

26. Orfield, 3–4. 27. Orfield, 21. Michael A. Boozer, Alan B. Krueger, and Shari Wolkon, in “Race and School Quality since Brown v. Board of Education,” Brookings Papers on Economic Activity, Microeconomics (Washington, DC: Brookings Institution, 1992), 269–326, also conclude that the large increase in desegregation coincides with the 1964 Civil Rights Act. The effects of that act may have been reinforced by the 1965 Elementary and Secondary Education Act, which also withheld federal funds from segregated schools. 28. Gary Orfield and Chungmei Lee, “Brown at 50: King’s Dream or Plessy’s Nightmare?” (Har- vard University Civil Rights Project, Harvard University, Cambridge, MA, January 2004). 29. Gary Orfield, “Executive Summary: Schools More Separate: Consequence of a Decade of Resegregation” (Harvard University Civil Rights Project, Harvard University, Cambridge, MA, 2001).

Race Discrimination and Education 327 attending schools with more blacks, Latinos, and Asians, but this need not imply anything more than that our public schools have become more diverse. Although it is possible that public schools have become more diverse because more white students are attending private schools, this turns out to be incorrect. Roughly 11 percent of white students were in private school in both years. Thus, we have two perspectives. One says that blacks are attending schools with more minority students because of some factor other than the shift in the race/ethnic distribution of public school students. Perhaps school districts are becoming more homogeneous, so within districts desegregation does not promote integrated schools. The other perspective says that the trend is due to shifts in the overall race/ethnic population. How can we address this issue? One way is to conduct a thought experiment. Sup- pose that we took the 2001–2002 school year population and counted each white student as more than one student and each minority student as less than one student in such a way that with this reweighting the proportions of whites, blacks, Hispanics, Asians, and native Americans were the same as in 1987–1988. To do this, it turns out that we would have to treat each white student as about 1.22 students and each black student as about 0.81 student and make similar adjustments for each of the other groups. Having made these adjustments, we would redo our calculations. Note that a school that is all black or all white will continue to be all black or all white. A school that is 50 percent black and 50 percent white will now be about 40 percent black. When we make this calculation, the average black attends a school that is 37 percent white and the average white attends a school that is 85 percent white. Recall that in 1987–1988, these figures were 40 percent and 84 percent. So now we can see that a small amount of resegregation is explained by other factors, but most resegregation is explained by the increase in the public school minority population. One can look at this change optimistically or pessimistically. Because of the growing minority population, far more blacks attend schools that are majority-minority or even predominantly minority. On the other hand, far fewer whites attend schools that are heavily or entirely white. An important issue about which we know relatively little is segregation within schools. Whites and blacks often attend the same schools, but whites are disproportionately in more advanced tracks and blacks in less advanced tracks. However, once we control for past performance, there is no difference in track assignment between blacks and whites.30 The effects of tracking remain controversial and not well studied.

2.2. The Effects of Segregation. By the time the Supreme Court ruled in 1954 that “separate but equal” was inherently unequal, the gap in the resources spent on black and white students had narrowed dramatically. However, schools did not really de- segregate until the late 1960s or early 1970s.31 Thus, school desegregation cannot explain

30. Ronald F. Ferguson, “Can Schools Narrow the Black-White Test Score Gap?” in Christopher Jencks and Meredith Phillips, eds., The Black-White Test Score Gap (Washington, DC: Brookings Institution, 1998), 318–74. 31. Heckman and Donohue, 1627–28.

328 Chapter 12 the wage growth from 1960 through the mid-1970s. However, it is possible that desegregation did reduce the black-white wage differential after that. We have already discussed the evidence regarding resegregation of schools in the United States. We saw that while there has been some trend toward resegregation since the 1980s, much of the change reflects the increased diversity of the American population. Sarah Reber has looked directly at districts that had court-ordered desegregation plans.32 She uses differences in the timing of the court orders to identify the effects of these plans on the actual segregation of the schools. Conceptually her approach is to compare annual changes in segregation in school districts between 1965 and 1975 that were desegregated in, for example, 1965 with the annual changes over the same period for districts that were desegregated later than 1975. She finds that there was an immediate drop in segregation following the court order. Over the following decade, about one-third of that change was eliminated by white flight. There is some direct evidence that school desegregation increased educational attainment among blacks. Jonathan Guryan compares the enrollment rates of black 15-, 16-, and 17-year-olds in large districts that desegregated in the 1960s, 1970s, and 1980s.33 He finds that the dropout rate among blacks did not change between 1970 and 1980 in districts that desegregated in the 1960s or 1980s but dropped by close to 4 percentage points in those that desegregated during the 1970s. This suggests that desegregation reduced the black dropout rate. In the 1990s, courts began to release school districts from their desegregation plans. In 1991, in Board of Education of Oklahoma City v. Dowell, the Supreme Court ruled that if a school district that had eliminated de jure segregation to the extent prac- ticable and was unlikely to return to intentionally segregated schools, it should be released from the court-ordered desegregation plan. Byron Lutz uses an approach similar to those used by Reber and Guryan to study the effects of ending desegregation plans.34 He compares districts that were under court-ordered desegregation plans with those that were released from the plans. He finds a modest increase in segregation following the end of the plans. After four years, ending the desegregation plan reverses its long-run effect (as estimated by Reber) by about one-third. Put differently, the long-run effect of white flight and the response to ending the court-ordered plans is to reduce their effect by 50 percent relative to the initial impact of the plans. Lutz also examines the change in black dropout rates between 1990 and 2000. He finds that in districts where the court-ordered plans were terminated, the dropout rate fell by about 1.6 percentage points less than it did in districts where the court-ordered plans were maintained.

32. Sarah J. Reber, “Court-Ordered Desegregation: Successes and Failures Integrating American Schools since Brown,” Journal of Human Resources 40 (Summer 2005): 559–90. 33. Jonathan Guryan, “Desegregation and Black Dropout Rates,” American Economic Review 94 (September 2004): 919–43. 34. Byron F. Lutz, “Post Brown v. the Board of Education: The Effects of the End of Court- Ordered Desegregation” (Finance and Economics Discussion Series 2005-64, Board of Governors of the Federal Reserve System, Washington, DC, 2005).

Race Discrimination and Education 329 3. Affirmative Action in Education Much of the debate over affirmative action has centered on the role of test scores. In a series of cases, the Supreme Court has found that race can be a factor in college admission but that it must be used in a nuanced way. If a college uses a formula for determining admission, it cannot include race in that formula by, for example, giving blacks an extra five points. Thus, in Grutter v. Bollinger the Supreme Court upheld the race- conscious admissions policy of the University of Michigan Law School on the grounds that the state has a compelling interest in diversity. On the same day, in Gratz v. Bollinger, it struck down the University of Michigan’s undergraduate formula, which gave twenty points to minority students, because that approach was not “narrowly tailored.” Colleges that wish to practice affirmative action have two options. Smaller programs can avoid using a formula. Larger programs will generally have difficulty with this approach. Instead they can attempt to base admission, at least in part, on factors that are race-blind but not race-neutral. In other words, they can give credit in the formula for factors that are more common among black applicants than among white applicants or at least that are more favorable to black applicants than are test scores. For example, in 1996, in Hopwood v. Texas, the court ruled that the University of Texas Law School could not use race as a factor in admissions, a decision that effectively outlawed similar preferences at the undergraduate level at the university. Partially in response to Hopwood, the Texas legislature passed a law granting undergraduate admission to any Texas student who graduated in the top 10 percent of his senior class. If Texas schools were perfectly segregated by race and ethnicity, this would guarantee that 10 percent of each group was entitled to admission. But this is not the case. Whites are considerably more likely than blacks and Hispanics to graduate in the top 10 percent of their class. Still, blacks and Hispanics fare better under the 10 percent rule than they would under a rule that relied on SAT and ACT scores. In addition, in response to Hopwood, the University of Texas adopted a “holistic” approach for the 50 percent of students not admitted under the 10 percent rule. In addition to traditional academic factors, admissions officers weigh such factors as socioeconomic status, school attended, and leadership potential. Critics of Hopwood and Gratz argue that if institutions care deeply about diversity, forcing them to use race-blind mechanisms to achieve that diversity is inefficient. Sup- pose that our objective is to get the best students into the best universities, and suppose also, for simplicity, that the SAT is the best measure of who is a good student. If colleges are free to practice affirmative action, they will choose the black students with the best scores and the white students with the best scores. If they must use a color-blind measure, they will not always get the students with the best scores within each group. Some low-scoring whites will replace higher-scoring whites, and some low-scoring blacks will replace higher-scoring blacks. The more closely the college tries to maintain its original proportion of blacks, the greater this inefficiency will be.35

35. Roland Fryer Jr., Glenn C. Loury, and Tolga Yuret, “Color-Blind Affirmative Action” (Work- ing Paper 10103, National Bureau of Economic Research, Cambridge, MA, 2003), and Jimmy Chan

330 Chapter 12 There is also disagreement about how pervasive affirmative action really is in higher education. Proponents of the view that affirmative action is common argue that there is clearly discrimination against whites at many colleges because the average test score of admitted blacks is lower than the average test score of admitted whites. Therefore, the argument continues, it must be the case that the colleges are requiring whites to have higher test scores in order to gain admission than it requires of blacks. This argument is less compelling than it might appear at first glance.36 To see this, let us first assume that a program relies on a single test to determine which students to admit. Let us further suppose that it admits all students above a single cutoff score common to all race and ethnic groups. Finally, let us assume that, on average, blacks earn lower scores on this test than do whites. In general, then, we would expect that the average test score of admitted blacks will be lower than the average test score of admitted whites. To see this, begin with a simple example. Suppose that ten blacks take the test and receive scores of 1, 2, 3, 4, 5, 6, 7, 8, 9, and 10, respectively. Suppose that ten whites take the test and receive scores of 3, 4, 5, 6, 7, 8, 9, 10, 11, and 12, respectively. If the school system accepts all students with scores of 10 or higher, the average score for accepted white applicants will be 11, while the average score for the one accepted black applicant will be 10. With this example, provided that the cutoff is 10 or lower and greater than 2, the average for blacks will be 1 point lower than the average for whites regardless of the choice of cutoff. For cutoffs above 10, no black applicants are accepted, so no comparison is possible. In the example, despite the absence of any difference between blacks and whites in the criterion used for admission, blacks have lower mean test scores among those admitted to the program. The example presents a case that is not unusual. In the example, the distributions are the same except that the white scores are all two points higher than the black scores. It is a common (although certainly not universal) property of distributions that if we create a second distribution by increasing each element of the first distribution by the same amount, the mean conditional on being above any cutoff value is higher for the second distribution than for the first. In particular, the normal distribution has this property. Many tests are designed to have approximately normal distributions for the groups to which they are administered and have approximately normal distributions for significant subgroups. Let us now consider a different selection process. Suppose that we mistakenly believed that individuals were chosen on the basis of their test scores, but in fact they were chosen on the basis of some other factor that is uncorrelated with test score. Then, because the individuals chosen would be a random sample of the population, the average test score of the blacks who were chosen would be lower than the average test score of the whites who were chosen. and Erik Eyster, “Does Banning Affirmative Action Lower College Student Quality?” American Eco- nomic Review 93 (June 2003): 858–72. 36. William T. Dickens and Thomas J. Kane, “Racial Test Score Differences as Evidence of Re- verse Discrimination: Less than Meets the Eye,” Industrial Relations 38 (July 1999): 331–63.

Race Discrimination and Education 331 If the true selection criterion were weakly correlated with test score, more whites than blacks would be chosen and more individuals with high test scores than with low test scores would be chosen, but the average test score for whites would still tend to be higher than the average for blacks. To the extent that for most admissions criteria, the difference between blacks and whites is smaller than it is for test scores, it is not surprising that blacks have lower test scores conditional on being admitted. This does not mean that no reverse discrimination exists. If a school relied solely on a single admissions test and used a different cutoff for blacks and whites, it would clearly be engaging in reverse discrimination and would not be legally permitted to do so. However, unless an institution bases its admission decisions on a formula that explicitly adjusts for race, it is likely to be difficult to determine whether it engaged in reverse discrimination.

4. Concluding Remarks In a statistical sense, much of the difference between blacks and whites in adult test performance can be explained by differences in cognitive skills and behavioral maturity at school entrance and by the lower quality of the schools attended by blacks. More- over, our strongest evidence that the test score gap widens after children enter school shows that it does so through the third grade. The first finding suggests that Early Head Start and Head Start could play a major role in addressing the test score gap, but only if Head Start is reformed to generate greater cognitive effects (see chapter 6). The second finding suggests that reducing the gap requires finding ways to improve quality in schools with large minority populations (see the section on education reform in chapter 7). The third finding pushes us to focus much of our effort on the early grades. We have seen evidence that blacks are more likely to be punished for acting white in schools in which black students are a minority. This might push us in the direction of resegregating schools, a phenomenon that will occur in any event as the proportion of whites in the school-age population falls. However, the direct evidence is that black students benefited from desegregation and are hurt by resegregation. Whatever pressures black students feel to maintain group loyalty in majority-white schools are more than offset by other benefits. In chapter 10 we saw that one explanation for the difference in earnings between whites and blacks is differences in (nonverbal) language. To the extent that integrated schools are also integrated socially, blacks will have more opportunity to learn the social rules of the dominant group, and white language and social norms will be influenced more by and closer to those in the black community. Whether these benefits are adequate to justify affirmative action in education depends both on how large these effects really are and the value one places on color-blind behavior.

332 Chapter 12 5. Further Reading Angrist, Joshua, and Kevin Lang. “Does School Integration Generate Peer Effects? Ev- idence from Boston’s METCO Program.” American Economic Review 94 (Decem- ber 2004): 1613–34. Card, David, and Alan B. Krueger. “School Quality and Black-White Relative Earn- ings: A Direct Assessment.” Quarterly Journal of Economics 107 (February 1992): 151–200. Dickens, William T., and James R. Flynn. “Heritability Estimates versus Large Envi- ronmental Effects: The IQ Paradox Resolved.” Psychological Review 108 (April 2001): 346–69. Farkas, George. “Racial Disparities and Discrimination in Education: What Do We Know, How Do We Know It, and What Do We Need to Know?” Teachers College Record 105 (August 2003): 1157–88. Jencks, Christopher, and Meredith Phillips. “America’s Next Achievement: Closing the Black-White Test Score Gap.” American Prospect 40 (September–October 1998): 44–53. Jencks, Christopher, and Meredith Phillips, eds. The Black-White Test Score Gap. Wash- ington, DC: Brookings Institution, 1998.

6. Questions for Discussion 1. Discuss the evidence that the IQ–test score gap between blacks and whites is at least partially determined by environment. 2. Describe the principal trends over the past thirty years in the black-white gap on the NAEP. 3. What accounts for the black-white test score gap? 4. What is the evidence on whether “acting white” has negative consequences for black students? 5. Are primary and secondary schools in the United States resegregating? 6. At selective colleges, black students generally have lower SAT scores than do white students. Does this demonstrate that selective colleges practice reverse discrimination? 7. What is the distinction between a color-blind and a color-neutral policy? What are the costs of pursuing color-blind policies that are designed not to be color-neutral?

Race Discrimination and Education 333 chapter 13

Race Discrimination in Customer Markets and the Judicial System

W      1964 C R A dramatically changed civil rights in the United States. Although the 1954 Supreme Court decision in Brown v. Board of Education outlawed “separate but equal” education, relatively few schools were desegregated until Title IV of the 1964 Civil Rights Act outlawed segregation in public education. Similarly, the Montgomery bus boycott, trig- gered by Rosa Parks’s famous refusal to give up her seat to a white, did not precipitate desegregation of public accommodations. Almost a decade after the successful boycott, hotels, restaurants, and water fountains remained segregated. Almost overnight, Title II of the Civil Rights Act integrated these places. The act gave individuals (Title II) and the attorney general (Title III) power to sue if a store, restaurant, hotel, movie theater, or other public accommodation refused services on the grounds of race, ethnicity, or religion. The law also provided for federal funding for legal fees for plaintiffs. It is perhaps most remarkable for the speed with which it succeeded. Before the Civil Rights Act, there were bloody battles over integration. Non- violent protesters who challenged segregated facilities in the south had been beaten, arrested, and sometimes murdered. Shortly after the passage of the act, integration of public accommodations had largely ceased to be an issue. The focus of this chapter is on customer markets, a broader area than public accommodations. The most important customer market is probably housing. Discrimination in housing can promote segregation, and segregation may have important additional consequences by isolating blacks from whites and lengthening commutes for blacks. Such segregation may be reinforced if there is discrimination in credit markets. Dis- crimination in customer markets is most likely when information problems are important or when sellers are likely to care about the race of the customer. There is more risk of discrimination in credit markets where the seller must assess the buyer (e.g., his or her creditworthiness) or where seller and customer interact closely. Although we will not formally analyze models of discrimination in customer markets, the issues of im-

334 perfect information are similar in credit and labor markets. Discrimination based on seller tastes is similar to discrimination by customers. We end the chapter by briefly discussing discrimination in the justice system.

1. Housing Neighborhoods in the United States are highly segregated. If neighborhoods were integrated, the population of most census tracts1 would be between 5 percent and 50 percent black. In fact, only about one-third of census tracts fall in this range. Some 30 percent of blacks live in census tracts that are more than 80 percent black. Put differently, to achieve complete integration, with every census tract having the same proportion of blacks, would requiring moving almost two-thirds of blacks from census tracts with above-average proportions of blacks to census tracts with below-average proportions.2

1.1. Measuring Segregation. There are a many different ways of measuring segregation. In this section we will focus on two common measures, the isolation index and the dissimilarity index.

1.1.1. The Isolation Index. The isolation index3 measures the average proportion of blacks in the neighborhood in which blacks live. That is, for each census tract (or whatever area is being used), we calculate (proportion of total black population that lives in census tract) * (percentage of census tract that is black) and add this up for all census tracts in the metropolitan area. If all blacks live in fully segregated (100 percent black) census tracts, the index will be 1 (or 100 percent). If all census tracts are fully integrated, the isolation index will equal the proportion of blacks in the population. Because the isolation index can never be less than the proportion of blacks in the area studied, it will tend to be higher in metropolitan areas that are more heavily black. Some authors have suggested subtracting the fraction of blacks in the area studied, so the index becomes (average number of blacks in a census tract in which the average black resides) – (percentage of blacks in the metropolitan area). But this simply creates the opposite problem. The isolation index can then never exceed the proportion of nonblacks in the area.

1. The average census tract has about five thousand residents. 2. Information on distribution of population by census tract is from Edward L. Glaeser and Ja- cob L. Vigdor, Racial Segregation in the 2000 Census: Promising News (Washington, DC: Brookings Institution, 2001). 3. Wendell Bell, “A Probability Model for the Measurement of Ecological Segregation,” Social Forces 32 (May 1954): 357–64.

Race in Markets and the Justice System 335 If, in addition, we divide by the proportion of nonblacks in the area, we have (average number of blacks in the census tract in which the average black resides – percentage of blacks in metro area) ——————————————————————————————, (100 – percentage of blacks in the metro area) and then the index ranges from zero to one.4 When it is zero, the metropolitan area has equal proportions of blacks in all census tracts. When it equals one, the census tracts are either all black or all nonblack. To see how to calculate the isolation index, suppose that the area consists of two neighborhoods, each with a population of 100. The black population is 55, of whom 5 live in one neighborhood and 50 live in the other. Then 5/55 of blacks live in a neighborhood that is 5 percent black, and 50/55 live in a neighborhood that is 50 percent black. The proportion of blacks in the area is 27.5 percent. Therefore, we have

5 * —–550+ 50 * —– – 27.5 isolation = ——————————55 55 = .25. 100 – 27.5 Now suppose that we are considering a similar area with two neighborhoods, each with a population of 100. But in this area, there are 110 blacks. A little arithmetic will show that in order to have the same isolation measure as the first area, the new area must have approximately 30 blacks in one neighborhood and 80 blacks in the other. It is not clear why we should view these two places as equally segregated. In fact, it is very difficult to devise segregation measures that allow us to compare areas with different proportions of blacks in the population.

1.1.2. The Dissimilarity Index. The dissimilarity index5 measures the absolute value of the difference between the proportion of the black population located in the census tract and the proportion of the nonblack population located in the census tract: black population in census tract nonblack population in census tract ———————————–— – —————————————— . || black population in metro area nonblack population in metro area ||

We then add this across all census tracts and multiply by 1/2. In our example, 5/55 of blacks and 95/145 of nonblacks live in one census tract, while 50/55 of blacks and 50/145 of nonblacks live in the other. The dissimilarity index is given by

—–5955050– —–– + —– – —–– ||55 145|| || 55 145|| dissimilarity = ——————–————— = .56. 2

4. An additional adjustment is required if the number of blacks in the metropolitan area is smaller than the population of the smallest census tract. See David M. Cutler, Edward L. Glaeser, and Jacob L. Vigdor, “The Rise and Decline of the American Ghetto,” Journal of Political Economy 107 (June 1999): 455–506. 5. Otis Dudley Duncan and Beverly Duncan, “A Methodological Analysis of Segregation In- dexes,” American Sociological Review 20 (April 1955): 210–17.

336 Chapter 13 By convention, an area is viewed as highly segregated if the dissimilarity index is greater than .6, so this area would not be viewed as highly segregated but would be close to the cutoff. It can be shown that the dissimilarity index is the proportion of blacks (or nonblacks) who would have to move to other neighborhoods in order to achieve complete integration. Thus, in our example, if 56 percent of blacks (or about 31 blacks) moved from the heavily black neighborhood to the heavily white neighborhood, we would achieve complete integration. Suppose we compare this area to a similar area where blacks make up 110 out of the population of 200. To obtain the same dissimilarity index requires that approximately 83 of the blacks live in one neighborhood and 27 live in the other neighborhood. Again, it is not obvious whether we would want to think of these two area as equally segregated. Subject to the qualification that it is difficult to compare segregation among metropolitan areas with different proportions of blacks, we use these measures to discuss what has happened to segregation in the United States.

1.2. U.S. Residential Segregation in the Twentieth Century. It is perhaps surprising that there was relatively little residential segregation in the United States at the beginning of the twentieth century. In 1890, the average dissimilarity index for sixty cities was .48, and weighting those cities by their black populations gives an average of .46. By 1940, these figures had grown to .68 and .72 for 109 cities. This process of increased segregation continued through 1970, when the dissimilarity index for 211 cities was .73 and was .79 if weighted by the black population.6 The pattern for the isolation index is similar (see table 13.1). Douglas Massey and Nancy Denton explore the rise of ghettos in the United States.7 They report that in the nineteenth century, in northern cities the black elite lived among whites, and in no city were blacks socially isolated from whites. This was also true in southern cities, where whites lived on large avenues and blacks on the small streets and alleys interspersed among them. While in the south, the black elite did not mix with whites; Jim Crow laws substituted for residential segregation. However, as blacks began to migrate from rural southern areas to urban areas, especially in the north, whites began to force blacks from their neighborhoods. Initially this was enforced by violence. Race riots in which gangs of whites attacked individual blacks were common between 1900 and 1920. After this period, while violence continued, it was more focused on individuals trying to expand the ghetto at its edges. Whites also relied on legal means to “protect their neighborhoods.” They would sign restrictive covenants under which all members of a neighborhood would agree not to sell or rent to blacks. These covenants were enforceable in court until they were

6. Cutler, Glaeser, and Vigdor, 464. The sample of cities changes over time, but the general pattern from period to period is maintained if we limit ourselves to cities covered in both periods. 7. Douglas S. Massey and Nancy A. Denton, American Apartheid (Cambridge, MA: Harvard University Press), 1993.

Race in Markets and the Justice System 337 Table 13.1 Measures of Segregation, Selected Years, 1890–2000

Unweighted Weighted by Black Population Number of Metropolitan Areas Dissimilarity Index 1890 .48 .46 60 1940 .68 .72 109 1970 .73 .79 211 1990 .56 .66 213 2000 .50 .62 291

Isolation Index 1890 .21 .23 60 1940 .37 .46 109 1970 .41 .61 211 1990 .26 .47 213 2000 .20 .39 291

Source: David M. Cutler, Edward L. Glaeser, and Jacob L. Vigdor, “The Rise and Decline of the American Ghetto,” Journal of Political Economy 107 (June 1999): 455–506, table 1. Note: See the text for an explanation of the numbers used.

struck down by the Supreme Court in 1948. Real estate boards would threaten their members with expulsion if they sold or rented to blacks in white neighborhoods. This segregation was enforced or supported by federal agencies. Agencies such as the Home Owners’ Loan Corporation, Federal Housing Authority, and Veterans Administration favored loans in white areas and avoided giving loans in areas that might “turn black.” Public housing was segregated until court decisions (including Gautreaux, which led to the program discussed in chapter 7) forced the federal government to change its policies. The Fair Housing Act of 1968 made it illegal for landlords (in most settings) and realtors to discriminate on the basis of race. Nevertheless, as we will discuss later when we consider audit studies, subtler forms of discrimination may still persist. Realtors may tell black prospective clients that a property is already sold or rented or may direct their attention to a more “suitable” property. The evidence suggests that fair housing laws, along with changing attitudes, reduced segregation. By 1990, the average black lived in a metropolitan area with a dissimilarity index of .66, still highly segregated but noticeably down from its peak of .79 in 1970. By 2000, it had fallen to .62, its lowest level since 1920.

1.3. Explaining Segregation. As table 13.1 shows, although there has been a noticeable decrease in segregation over the past thirty years, the average black lives in a metropolitan area that would generally be considered highly segregated. Why is housing in the United States so segregated? There are at least three possibilities. First, despite fair housing laws, active racism could make it more difficult for blacks to find housing in predominantly white neighborhoods. Owners and real estate agents might avoid renting

338 Chapter 13 or selling property in such neighborhoods to blacks. Second, whites might prefer to live with whites and therefore be willing to pay a premium to live in predominantly white neighborhoods. If blacks do not care whether they live with whites or blacks, whites will bid up the prices in white neighborhoods and blacks will choose to live in black neighborhoods where housing is less expensive. Finally, blacks may prefer to live in black neighborhoods and may be willing to pay a premium to live in such neighborhoods. If whites do not care whether they live with whites or blacks, blacks will bid up the prices in black neighborhoods, and whites will choose to live in white neighborhoods where housing is less expensive. Note that the explanations are not mutually exclusive. Whites might prefer to live with whites, and blacks might prefer to live with blacks and still face discrimination when they look for housing in primarily white neighborhoods. Historically, there was considerable evidence that blacks paid more than whites for equivalent housing.8 Combined with well-documented overt exclusion of blacks from certain neighborhoods, the case is strong that for much of the twentieth century segregation reflected active racism. The current evidence suggests a stronger role for white preferences, because whites now pay more for equivalent housing.9 Survey data suggest that most Americans prefer neighborhoods that are at least partially integrated. Only 20 percent of blacks in four urban areas said that they preferred to live in an all-black neighborhood, while half said that they preferred a neighborhood with equal numbers of whites and blacks.10 We have less precise information on the preferences of whites, but approximately half indicate a willingness to move into a neighborhood that is 40 percent black.11 We need to treat these expressed preferences with caution. To some extent, individuals’ responses reflect their perceptions of neighborhood differences that are correlated with race. Some blacks prefer more heavily white neighborhoods because of characteristics, such as the quality of schools and safety, that are associated with the racial composition of these neighborhoods. The important point is that if blacks and whites have different preferences regarding the composition of the neighborhoods in which they live, these differing preferences can produce segregated neighborhoods even when members of both races prefer integrated neighborhoods. This can be consistent with higher housing costs in white neighborhoods provided that some blacks prefer all-white neighborhoods to all-black neighborhoods. We will show this by way of a somewhat complicated example. Let us suppose that the preferences of blacks and whites are quite close. All blacks prefer neighborhoods that are 40 percent black, while all whites prefer neighborhoods that are 20 percent black. In addition, when choosing among neighborhoods, all individuals seek a neighborhood that is as close to their preferred percentage as possible, and when choosing between two neighborhoods that are equally close, they choose the

8. John F. Kain, “A Pioneer’s Perspective on the Spatial Mismatch Hypothesis,” Urban Studies 41 (January 2004): 7–32, and Massey and Denton, chapter 2. 9. Cutler, Glaeser, and Vigdor, 486–87. 10. Maria Krysan and Reynolds Farley, “The Residential Preferences of Blacks: Do They Explain Persistent Segregation?” Social Forces 80 (March 2002): 937–80. 11. Krysan and Farley, 960.

Race in Markets and the Justice System 339 one with a higher proportion of their own race. Thus, if everything else is equal, whites prefer a neighborhood that is 20 percent black to one that is 10 percent black, which they, in turn, prefer to one that is 30 percent black. Of course, everything else may not be equal. In particular, we would expect housing prices to differ among neighborhoods even if they are identical except for racial composition. In our example, whites will be willing to pay more for a house in a neighborhood that is 20 percent black than they will for one in a neighborhood that is 10 percent or 30 percent black. Moreover, we will assume that the differential price whites are willing to pay grows more than proportionally as the proportion of blacks deviates from 20 percent. Thus, the reduction in price they need to be indifferent between neighborhoods with 30 percent and 40 percent blacks is larger than the reduction they need to go from 20 percent to 30 percent black. The same is true for blacks except that, in their case, we are talking about deviations from 40 percent black. Blacks and whites are similar with respect to how much they are willing to pay to live in a neighborhood with their preferred racial mix. To complete the example, let us assume that the metropolitan area is 30 percent black and that there are three neighborhoods, each of which holds one-third of the area’s housing. If all neighborhoods are 30 percent black and house prices are the same in all neighborhoods, no one wishes to move (at least not because of the racial composition of the neighborhood). So it is possible that all neighborhoods will be integrated. But what happens if a few blacks change neighborhoods with a few whites so that one neighborhood is 29 percent black and another is 31 percent black? Because blacks prefer the neighborhood where blacks make up 31 percent of the population and whites prefer the neighborhood where blacks are 29 percent of the population, more whites and blacks can be better off if whites in the 31 percent black neighborhood trade houses with blacks in the 29 percent black neighborhood. So blacks will leave the neighborhood with a low fraction of blacks, and whites will move in. It might seem that this process would stop when one neighborhood was 20 percent black, as preferred by whites, and one was 40 percent black, as preferred by blacks, but it will not. The blacks in the 20 percent black neighborhood will still prefer the more heavily black neighborhood, while the whites in that neighborhood will prefer to move to the less heavily black neighborhood. So they will keep trading houses until one neighborhood is all white and the other is 60 percent black. It might appear that the process would stop at this point, but it does not. The whites who are in the neighborhood that is 60 percent black would strongly prefer to be in the neighborhood that is 30 percent black. The blacks also prefer the neighborhood that is 30 percent black to the one that is 60 percent black, but not as strongly as the whites do. Therefore, whites in the 60 percent black neighborhood can pay blacks in the 30 percent black neighborhood to trade houses with them or, more realistically, outbid blacks for those houses when they become available. This process continues until two of the neighborhoods are all white.12

12. Readers familiar with game theory may recognize that there are other equilibria when there are many neighborhoods. These equilibria must have the form that integrated neighborhoods all have

340 Chapter 13 Because there are not enough blacks to fill the third neighborhood, some whites will live in the heavily black neighborhood. In order for them to be indifferent between living in a neighborhood that is all white and one that is 90 percent black, housing prices must be lower in the primarily black neighborhood. If the price of housing in the mostly black neighborhood were the same as or higher than in the all-white neighborhood, both blacks and whites would want to move to those neighborhoods. Note that the segregation result in the example does not depend on the metropolitan area being 30 percent black. If instead it were 35 percent black, one neighborhood would be all black, one would be all white, and the other would be mostly white. In this case, housing prices would be highest in the mostly white neighborhood because whites and blacks both prefer this neighborhood to the fully segregated neighborhood, while the all-black neighborhood would have the lowest-priced housing because both groups also prefer all-white neighborhoods to all-black neighborhoods. The view that segregation reflects the greater willingness of whites to pay to live in primarily white neighborhoods is consistent with evidence that equivalent housing is now more expensive in white neighborhoods than in black ones.13

1.4. Evidence from Audit Studies. In 2000, the Urban Institute conducted a large audit study of discrimination in housing rentals and sales.14 Although the study also looked at discrimination against other groups, the majority of the sites looked at discrimination against blacks, and we will focus on the results from those sites. The study relied on over two thousand audits by paired black and white customers in sixteen metropolitan areas. Housing transactions are complex, so it is often not a trivial matter to say whether one customer has been treated better than another. One client may be told an advertised housing unit is not available but shown three similar units, while another may be shown the advertised unit but not told about the others. The researchers used two measures of unequal treatment. In the first, they attempted to place weights on different aspects of treatment and view the client who got more points as having been treated better. The weights were inevitably arbitrary although sensible. In the second measure, the person was recorded as having been treated more favorably only if he was treated as well in all dimensions and better in at least one. The results of the study along with those of a similar audit study conducted in 1989 are summarized in table 13.2. Using either measure, whites were more frequently favored than blacks in housing sales. However, again using either measure, the extent of discrimination has diminished considerably since 1989. The results for rentals are

the same proportion of blacks but some neighborhoods are fully segregated. For example, we could end up with one all-white neighborhood, one all-black neighborhood, and eight neighborhoods that are one-fourth black. However, if a few blacks and whites switched integrated neighborhoods, the same process of unraveling would occur. 13. Cutler, Glaeser, and Vigdor, 486–87. 14. Margery Austin Turner et al., Discrimination in Metropolitan Housing Markets: National Re- sults from Phase I HDS 2000, Final Report (Washington, DC: Urban Institute, 2002).

Race in Markets and the Justice System 341 Table 13.2 Audit Studies of Discrimination in Housing, 1989 and 2000

1989 2000 Weighted Consistent Weighted Consistent Rentals White-Favored 54.6 26.4 49.0 21.6 Black-Favored 41.2 15.3 41.1 19.2 Difference 13.4 11.1 7.9 2.3

Sales White-Favored 56.1 29.0 53.1 17.0 Black-Favored 39.0 16.2 44.8 12.4 Difference 17.1 12.8 8.3 4.6

Source: Margery Austin Turner et al., Discrimination in Metropolitan Housing Markets: National Results from Phase I HDS 2000, Final Report (Washington, DC: Urban Institute, 2002), exhibits 3-5 and 3-16.

less conclusive. Both measures show improvement since 1989 and more favorable treatment of whites. However, in 2000, using the measure of consistently more favorable treatment, the difference in treatment between blacks and whites is statistically insignificant while the improvement between 1989 and 2000 is statistically insignificant when the weighted measure is used. The audit study cannot tell us the consequences of this discrimination for reasons similar to those we discussed when we looked at audit studies of employment discrimination. Because there is a great deal of randomness in the treatment of individuals, a modest difference in the probability of more favorable treatment may have little effect on the quality and price of housing obtained by blacks. Moreover, the audit study measures what happens when blacks and whites seek the same housing from the same real estate agents. To the extent that blacks and whites seek housing in different neighborhoods and use different agents, the audit study may overestimate the extent of discrimination actually faced by blacks. On the other hand, because tastes in housing are very idiosyncratic, Gary Becker’s argument, discussed in chapter 10, that discrimination leads to segregation, not price discrimination, is less compelling when applied to housing than when applied to the labor market. Being excluded from some housing may prevent blacks from choosing the housing they would prefer. In addition, to the degree that we think that neighborhood effects are important, segregation in housing is a concern in its own right.

1.5. Does Segregation Matter? In chapter 7 we noticed that even poor neighborhoods are usually quite diverse, so individuals can find role models with good jobs who own their homes and/or have college degrees. It is less obvious that residents of heavily black neighborhoods have much contact with whites. In 2000, 30 percent of blacks

342 Chapter 13 lived in census tracts that were at least 80 percent black,15 and many of the nonblacks would have been Hispanic whites or Asians. In 1990, over one-ninth of urban census tracts were at least 90 percent black.16 Thus, many blacks live in neighborhoods in which they have little social contact with whites. In principle, such segregation could be good or bad. In the transactions cost model of discrimination discussed in the previous chapter, segregation minimizes the cost of social interaction. There is less need for people who understand the different rules of social interaction used by different groups. From this perspective, blacks living in heavily white neighborhoods must either learn to code-switch between black and white ways of interacting or suffer the cost of not following the norms of the majority-white culture. Blacks living in segregated neighborhoods are spared this cost. On the other hand, in a society in which there are significant differences in average wealth, income, and education between blacks and whites, heavily black neighborhoods may have fewer ameni- ties than can be found in heavily white neighborhoods, and there may be less access to jobs. If it is easier to be successful if one uses typically white modes of interaction, those who acquire these modes as “native speakers” will generally be more successful in life. In chapter 7 we also discussed how hard it is to establish whether neighborhoods really matter. We would not be surprised if blacks who live in primarily white neighborhoods were more successful than those who live in primarily black neighborhoods even if neighborhood had no effect on success. Consider our earlier example in which blacks prefer to live in neighborhoods that are 60 percent white. Blacks with higher incomes are more likely to be able to indulge this taste by purchasing housing in more expensive neighborhoods. Also, blacks who can code-switch are likely to be both relatively successful and more comfortable than are other blacks living in heavily white neighborhoods. One way to get around this problem is to look at the relation between outcomes among blacks and the degree of segregation in the entire metropolitan area in which they live. David Cutler and Edward Glaeser find that, relative to those living in highly segregated metropolitan areas, young blacks (aged 20 to 30) in metropolitan areas with relatively little segregation are more likely to have graduated high school, are less likely to be out of school and not working, have higher earnings, and are less likely to be single mothers. There are at least three reasons to be concerned about this analysis. The first is that there may be other differences among metropolitan areas that are correlated with their degree of segregation. For example, older cities may be more segregated and have less- educated residents. One way to examine this problem is to compare differences for blacks across metropolitan areas with differences for whites. To continue the example, if the explanation were that older cities are both more segregated and have less-educated residents, we would expect whites living in more segregated cities to have less education than do other whites. In general, they do not.

15. Edward L. Glaeser and Jacob L. Vigdor, Racial Segregation in the 2000 Census: Promising News (Washington, DC: Brookings Institution, 2001). 16. Cutler, Glaeser, and Vigdor, 470.

Race in Markets and the Justice System 343 The second concern is that causality may go in the opposite direction. In areas in which blacks are relatively successful, there may be more mixing of neighborhoods. Cutler and Glaeser address this issue by looking at characteristics of the metropolitan area that might affect segregation but that are unlikely to affect the success of blacks except through their impact on segregation. They find that there is less segregation where more government revenue comes from the state rather than from local sources and more segregation in metropolitan areas with more individual communities and, consistent with our discussion of the number of school districts in chapter 7, where there are more rivers. Addressing the reverse causality by allowing these factors to cause segregation does not substantially change their results. The final concern is that successful blacks may migrate to metropolitan areas with less segregation. According to Cutler and Glaeser, both whites and blacks tend to migrate to newer, less segregated metropolitan areas, but there is no difference in the migration patterns of blacks and whites.

1.6. The Spatial Mismatch Hypothesis. In 1968, John Kain argued that housing segregation hurt blacks because it forced them to live farther from employment centers.17 Over time, as fair housing legislation has made housing discrimination illegal and de facto exclusion of blacks from white neighborhoods has declined, middle-class blacks have been able to follow their white counterparts to the suburbs. Jobs and people have left the central city areas together. As the middle class leaves the central city for the suburbs, some jobs will follow them, and these are likely to be, at least in part, service jobs that cater to the middle class as well as jobs that attract the middle class. As jobs leave the central city for the suburbs, some people will relocate to the suburbs, and it is likely to be people who have the most resources and fit in socially. In particular, if African Americans find it more difficult to obtain suitable housing in suburban neighborhoods and if subsidized housing is less readily available in the suburbs, the poor and African Americans will be less likely to follow the jobs to the suburbs. Of course, in principle, it is possible for residents of inner cities to work in the suburbs, but, at least in large cities, the vast majority of commutes either begin and end in the suburbs or begin and end in the central city.18 Jeffrey Zax and John Kain consider the case of a firm that moved from downtown Detroit to Dearborn.19 For some workers this shortened the commute, while for others it lengthened it. Zax and Kain found that white workers whose commute became longer were more likely than were other white workers to move but were no more likely to quit. In contrast, relative to other black workers, black employees whose commute became longer were more likely both to move and to quit. In other words, while some blacks

17. John F. Kain, “Housing Segregation, Negro Employment and Metropolitan Decentraliza- tion,” Quarterly Journal of Economics 82 (May 1968): 175–97. 18. Laurent Gobillon, Harris Selod, and Yves Zenou, “Spatial Mismatch in US Cities: Facts and Theories,” Urban Studies (forthcoming). 19. Jeffrey S. Zax and John F. Kain, “Moving to the Suburbs: Do Relocating Companies Leave Their Black Employees Behind?” Journal of Labor Economics 14 (July 1996): 472–504.

344 Chapter 13 followed the jobs to the suburbs, they were less likely to do so than were whites. Thus, suburbanization of jobs seems to have hurt blacks. William Julius Wilson has argued that the decline of manufacturing, which historically was located in central cities, and the suburbanization of jobs have been particularly problematic for low-skilled blacks.20 With the rise of laws prohibiting housing discrimination on the basis of race, middle-class blacks were able to follow jobs and whites to the suburbs. Those “left behind” in heavily black inner-city neighborhoods were increasingly disadvantaged. According to Wilson, the loss of jobs, particularly low-skilled manufacturing jobs, in the inner city, combined with the departure of many of the better-educated and higher-income residents, was particularly destructive because it reduced the number of positive role models (an issue we discussed in chapter 7) and the number of marriageable men (as discussed in chapter 6). As Douglas Massey and Nancy Denton have emphasized, the increased isolation of poor blacks reflects the fact that housing among the poor remains highly segregated.21 The spatial mismatch hypothesis is unusually difficult to test, in part because it is very difficult to determine causality. People choose where to live in part on the basis of where they work and choose where to work in part on the basis of where they live. Suburbanites are more likely to own cars and to commute by car than are city dwellers, but that reflects, in part, not the fact that living in the suburbs is more attractive to people who own a car but the decision of suburbanites to buy cars. On average, jobs in the central business district pay more than do jobs in the suburbs, but that reflects, in part, the presence of very high-skilled, high-paying jobs in the central business district. On average, blacks live closer to where they work than do whites, but whites’ com- muting times are shorter because they are more likely to drive and travel less congested streets.22

2. Discrimination in Other Markets

2.1. Credit Markets. We noted earlier that residential segregation may be reinforced by lending policies. If banks are reluctant to lend to blacks, they will have more difficulty purchasing homes in general, and possibly in white neighborhoods in particular. A group of researchers at the Federal Reserve Bank of Boston reviewed data from mortgage applications to determine whether race affected the probability of a mortgage application’s being accepted.23 Because they had access to the data used by the banks, they were able to control for most of the “hard data” available to the bank. They found,

20. William J. Wilson, The Truly Disadvantaged (Chicago: University of Chicago Press, 1987). 21. Massey and Denton, chapter 5. The authors argue that segregation significantly worsened the impact of declining job market conditions among inner-city blacks and that outflows of middle-class blacks played a relatively minor role. 22. Bruce A. Weinberg, “Testing the Spatial Mismatch Hypothesis Using Inter-City Variations in Industrial Composition,” Regional Science and Urban Economics 34 (2004): 505–32. 23. Alicia H. Munnell et al., “Mortgage Lending in Boston: Interpreting HMDA Data,” Amer- ican Economic Review 86 (March 1996): 25–53.

Race in Markets and the Justice System 345 even after applying an extensive set of controls, that blacks were about 8 percent less likely to have an application approved than were apparently equivalent whites. Because the Federal Reserve Banks play a critical role in the regulation of financial institutions, the study took on considerable significance and has been subjected to extensive reanalysis. The major criticism is that, despite all their efforts, the researchers were not able to fully capture all the information available to the banks. Including many of the variables potentially left out does not affect the results. The major exception is the bank’s evaluation of whether each credit application “met guidelines.” In- cluding this variable substantially reduces the estimated effect of race, but supporters of the original study argue that the fact that applications from blacks were less likely to “meet guidelines” is a judgment that reflects discrimination rather than an objective measure of the quality of the application that is independent of race. In a thorough review of the debate and a reanalysis of the data, Stephen Ross and John Yinger conclude that the Boston Fed study creates a presumption that discrimination exists in mortgage lending.24 The most plausible explanation under which their finding is incorrect is one in which different mortgage lenders apply different standards for approving loans and blacks tend to apply to lenders with tougher standards. For legal reasons, there are no audit studies of the loan approval process. Audit studies of the preapproval process, however, confirm that blacks receive less encourage- ment to apply and are more likely to be encouraged to apply to other lenders.25

2.2. Customer Markets. Since the passage of the 1964 Civil Rights Act, it has been illegal to refuse service to someone on the basis of race. In situations in which prices are posted and the firm is expected to serve all customers, we may assume that overt discrimination is largely, although not entirely, absent. Exceptions make headlines, as in the case in which Denny’s, a fast-food restaurant, was successfully sued for discrimination against black customers. However, discrimination may be more subtle. Sales assistants may greet white customers more rapidly or treat black customers with suspicion. In a review of discrimination in consumer markets, Peter Siegelman finds only a handful of cases in which complaints of such treatment were upheld by the courts.26 Two of the cases involved instances in which black shoppers were wrongfully detained for shoplifting. In one of the two cases, the corporation had a policy that profiled blacks. Many of the cases that were brought by plaintiffs involved what Siegelman terms “race plus.” In these cases, the customer was initially served but there was a confrontation.

24. Stephen L. Ross and John Yinger, “Does Discrimination in Mortgage Lending Exist? The Boston Fed Study and Its Critics,” in Margery Austin Turner and Felicity Skidmore, eds., Mortgage Lending Discrimination: A Review of Existing Evidence (Washington, DC: Urban Institute, 1999). 25. Robin Smith and Michelle DeLair, “New Evidence from Lender Testing: Discrimination at the Pre-Application Stage,” in Margery Austin Turner and Felicity Skidmore, eds., Mortgage Lending Discrimination: A Review of Existing Evidence (Washington, DC: Urban Institute, 1999). 26. Peter Siegelman, “Racial Discrimination in ‘Everyday’ Commercial Transactions: What Do We Know, What Do We Need to Know, and How Can We Find Out?” in Michael Fix and Margery Austin Turner, eds., A National Report Card on Discrimination in America: The Role of Testing (Wash- ington, DC: Urban Institute, 1998: 69–97).

346 Chapter 13 In one case, lost by the plaintiff on both summary judgment and appeal, the black customers and Hispanic clerk got into an argument over what the customers had ordered. The customers were forcibly ejected from the restaurant. In such cases it is extremely difficult to know whether whites with similar dress and behavior would have been ejected. It is not feasible to conduct audit studies. As a result, cases claiming denial of service are relatively rare and are infrequently upheld. What is clear is that many blacks, especially young black males, experience discrimination. Siegelman estimates that blacks experience discrimination on about 1 percent of shopping trips and about 2.5 percent of restaurant visits. Without an audit study, it is difficult to know whether the discrimination is real or simply felt. Are blacks more likely than whites to be treated rudely, or are they more likely than whites to interpret rude behavior as being the result of prejudice or discrimination? The areas in which discrimination is most likely and more easily detectable are those in which prices are negotiated (for housing, cars, etc.) or in which not all customers are accepted (for insurance, loans, rental housing, and perhaps taxis). We will explore the evidence for discrimination in some of these markets next.

2.3. Taxis. Since we began our discussion of prejudice and discrimination by looking at the decision made by a cab driver, it is perhaps appropriate to begin the present consideration with a discussion of the limited evidence on taxis. An audit study in Washington, D.C., assigned two potential customers, one black and one white, to stand approximately three car lengths apart and to attempt to hail a cab.27 Which potential customer was in front was determined randomly. It took black testers an average of 5.7 minutes to hail a cab, compared with an average of 4.5 minutes for white testers. The difference is small but statistically significant and thus supports the existence of discrimination. By design, the study gave drivers a choice between a white and a black customer. Faced with a choice, drivers might well choose the white customer, but that does not mean that they would choose to continue searching in preference to accepting a black customer when no other choice was immediately available. Because in most real world situations the driver will have to choose between taking a black customer immediately or searching for a fare, the importance of discrimination may be exaggerated by a study in which drivers can choose between two immediate fares. On the other hand, discrimination may be more important in particular circumstances. Black customers may have a hard time getting rides late at night in dangerous neighborhoods. In settings where there are few cabs (and few customers), even a modest difference in the probability of getting a cab may greatly extend the expected waiting time.

2.4. Automobiles. When prices are not posted, firms may discriminate with respect to the price they charge different customers. For example, if sales agents believe that

27. Siegelman, 76–79.

Race in Markets and the Justice System 347 women are poorer bargainers, they may bargain for a higher price when faced with a female customer. The most obvious example of this setting is the market for automobiles. An audit study revealed that when blacks attempt to buy a car, they are offered a higher initial price and face a higher final offer.28 The audit studies suggest that there is significant discrimination in the market for automobiles. In contrast, Pinelopi Gold- berg studied actual purchase prices of cars.29 For most categories of automobiles, she found no evidence of differences in the prices paid by blacks and whites. The sole exception was that minorities paid more for vans and utility vehicles. The difference between the two studies reflects, in part, the fact that in both cases the number of transactions or audits is small. The differences between white men and white women and between white men and black women are similar in the two studies. Goldberg did find a much smaller differential than suggested by the audit study in the case of white and black men. Still, Goldberg’s differential was not trivial. Taken together, the studies suggest that blacks do pay higher prices for similar cars, although the difference may be less than suggested by the audit study.

2.5. Health Care. In a thorough review of the literature, the Institute of Medicine concluded: “Evidence of racial and ethnic disparities in healthcare is, with few exceptions, remarkably consistent across a range of illnesses and healthcare services.”30 To some extent, these differences can be explained by factors other than race, such as family income. However, even in studies that control for other factors, there is generally evidence that blacks and whites receive different medical treatment. The study concludes that differences in treatment for similar conditions are sufficiently large to contribute to higher death rates among blacks.

2.6. The Weakest Link. Though hardly the most important market, the television show The Weakest Link has served as the basis for two interesting studies of discrimination.31 In the television show, contestants answer trivia questions. More correct answers increase the value of the pot that is awarded to the ultimate winner. At the end of each round, the contestants vote off one contestant until only two remain. The two first collaborate to increase the value of the final prize by correctly answering questions and then compete to answer the most questions correctly and win the prize. The optimal strategy for this game is complex. Keeping strong players increases the value of the pot but also increases competition for the prize. Regardless of the optimal

28. Ian Ayres and Peter Siegelman, “Race and Gender Discrimination in Bargaining for a New Car,” American Economic Review 85 (June 1995): 304–21. 29. Pinelopi K. Goldberg, “Dealer Price Discrimination in New Car Purchases: Evidence from the Consumer Expenditure Survey,” Journal of Political Economy 104 (June 1996): 622–54. 30. Institute of Medicine, Unequal Treatment: Confronting Racial and Ethnic Disparities in Health Care (Washington, DC: National Academy Press, 2002). 31. Steven D. Levitt, “Testing Theories of Discrimination: Evidence from ‘Weakest Link,’” Jour- nal of Law and Economics 47 (October 2004): 431–52, and Peter Arciadiacono, Kate Antonovics, and Randy Walsh, “Games and Discrimination: Lessons from The Weakest Link,” Journal of Human Re- sources 40 (Fall 2005): 918–47.

348 Chapter 13 strategy, the show’s host exhorts contestants to vote off the “weakest link” and treats those who do not with scorn. In case of a tie, the person with the strongest performance in the round can choose which of the tied vote-getters to remove from the game, so players must also be concerned about vengeance if they vote against a strong player. The big advantage of studying The Weakest Link is that the researcher has almost all of the information available to contestants. He knows the sex, education, and occupation of the contestant as well as his or her performance so far in the competition. Contestants may also use information on appearance and accent that are not readily summarized by the researcher. Both studies come to the same principal conclusion. There is no evidence that, after accounting for black contestants’ performance on earlier rounds, contestants are more likely to vote off black contestants. One study emphasizes a finding that women are more likely to vote off men, and the other finds evidence that contestants are inclined to vote off Hispanics in early rounds and less likely to vote them off in later rounds. The second study also finds evidence of discrimination against older contestants. Of course, participants in The Weakest Link are not representative of Americans as a whole. Moreover, contestants may be reluctant to take an action on television that might be viewed as discriminatory. Overall, however, the most striking finding is the absence of discrimination against blacks.

3. Discrimination in the Justice System There is little debate over whether there is discrimination, as we defined it in the previous chapter, in the justice system. Black and Hispanic drivers are much more likely than white drivers to be stopped and have their cars searched for drugs. According to at least one study, among those whose cars are searched, the proportion of black and white drivers found to have drugs is roughly equal,32 although this is not a universal finding.33 A little algebra shows that this means that innocent black drivers are much more likely than innocent white drivers to have their cars searched.34 Many blacks respond to this tendency by talking about being guilty of “driving while black.” They experience discrimination in that they are treated differently from innocent whites. People who defend the police argue that the similar fractions, among those stopped, of blacks and whites found carrying drugs prove that the police are not prejudiced but are acting in an appropriate manner. In assessing this matter, there are three important points. The first is to remember our distinction between prejudice and discrimination. Just as prejudice need not lead to

32. John Knowles, Nicola Persico, and Petra Todd, “Racial Bias in Motor Vehicle Searches: The- ory and Evidence,” Journal of Political Economy 109 (February 2001): 203–29. 33. Shamena Anwar and Hanming Fang, “An Alternative Test of Racial Profiling in Motor Vehi- cle Searches: Theory and Evidence” (Working Paper 11264, National Bureau of Economic Research, Cambridge, MA, 2005). 34. Depending on the discrepancy in stop rates, drug carriers who are black may or may not be more likely to be stopped, but, given the discrepancy in the stop rates, it is probable that, even among this group, blacks are more likely to be stopped.

Race in Markets and the Justice System 349 discrimination, the absence of prejudice need not imply the absence of discrimination. It is likely that profiling leads to a much higher proportion of innocent blacks being stopped than the proportion of innocent whites being stopped. Second, suppose that blacks are stopped more frequently because a higher proportion of blacks carry drugs. In general, then, we would expect blacks who are stopped to be more likely to be carrying drugs. The argument is essentially the same as the one that we made in chapter 12 to explain why we would expect whites accepted into elite schools to have, on average, higher test scores than those of accepted blacks. Suppose there are ten identifiable groups with equal numbers of both races. Among whites, five groups never carry drugs, while the remaining five carry drugs with a probability of .05, .10, .15, .20, and .25. Among blacks, three groups never carry drugs, and the remaining seven carry drugs with a probability of .05, .10, .15, .20, .25, .30, and .35. For any probability cutoff the police use to decide whether to stop and search someone, blacks will have a higher probability of actually carrying drugs. If searches occur when the probability is at least .2, then 22.5 percent of the time whites who are searched will be carrying drugs, while 27.5 percent of the time blacks who are searched will be carrying drugs. Obviously, it is possible to construct examples in which searched blacks are less likely to carry drugs, but this will not be true for most distributions. In addition, continuing our assumption that blacks are stopped more frequently because they are more likely to carry drugs, if police sometimes mistake the group to which an individual belongs, the person they search will be less likely to carry drugs if he is white. Put differently, if blacks are more likely to carry drugs, searches conducted by mistake are more likely to produce drugs among blacks than among whites. There- fore, even if it is correct that, on average, searches of whites and blacks are equally likely to produce drugs, this fact suggests that police use a lower cutoff for searching blacks than for searching whites. The third important point is that even when profiling is not based on prejudice, it may not be efficient. Police officers face incentives that encourage them to maximize the number of drug arrests they make. Society is presumably less interested in making arrests than in deterring drug use. Eliminating profiling would reduce the probability of being stopped for blacks and increase it for whites. This, in turn, should increase drug use among blacks and decrease it among whites. Whether drug use would go up or down depends on how much drug use responds to policing in each of the two groups. Thus, eliminating profiling could either increase or decrease overall drug use. In addition to being more likely to be stopped by police, blacks face different treatment than whites after they enter the justice system. Differential treatment does not necessarily imply harsher treatment. One study found that, relative to similar whites, black juveniles who were arrested were more likely to be released (the most lenient treatment) or sent to court (the harshest) but less likely to be required to enroll in a program with a social worker.35

35. Duran Bell and Kevin Lang, “The Intake Dispositions of Juvenile Offenders,” Journal of Re- search in Crime and Delinquency 22 (November 1985): 309–28.

350 Chapter 13 Once blacks enter the court system, they tend to be treated more severely than whites, but it is difficult to eliminate the possibility that the difference is due to factors that are specific to individual cases and correlated with race. One clever study seeks to get around this problem by looking at bail setting.36 Ian Ayres and Joel Waldfogel point out that blacks, on average, face higher bail than apparently comparable whites. Of course, as in other studies, the difference might simply reflect other factors that are not readily available in standard data sets, so apparently similar blacks and whites might present quite different risks of flight. Ayres and Waldfogel point out, however, that there is a very competitive market among bail bondsmen. The cost to the bondsmen of providing a bond is a very small fixed cost plus the probability that the bond will be for- feited times the value of the bond. Because the fixed cost is very small, the price charged to the borrower divided by the amount of the bond is approximately the flight risk. For example, if the flight risk is 50 percent and the bond is $10,000, the price will be $20,000. By dividing the $10,000 bond by its $20,000 price, we find that the bondsmen believe that the flight risk is 50 percent. Because the price charged to blacks, relative to the bond, is lower than for whites, we may assume that, given the amounts of the bonds being set, blacks have a lower probability of flight. Put somewhat differently, for a bond of a given size, the price charged to blacks is no higher and possibly lower than that charged to whites, which implies that the flight risk among blacks is no higher and possibly lower than that among whites. Given this information, we would expect the bonds set for blacks and whites to be similar. In fact, blacks face higher bonds, which, in turn, means that bail is set to generate a lower flight risk among blacks than among whites.

4. Concluding Remarks From the studies we have reviewed, there is strong evidence of historical discrimination in customer markets, including housing. There is also ample evidence of current racial prejudice in such markets. It is less clear whether this prejudice still has important discriminatory effects. As we have discussed previously, blacks may be able to avoid prejudiced realtors and lenders. Blacks may respond to the lower likelihood of being approved for a mortgage by applying to major banks with low mortgage rates but with reputations for not discriminating. On the other hand, if blacks do not have information about which firms discriminate and which do not, they may take actions that worsen the effects of prejudice: they may turn to high-cost lenders or conclude that it is not worth entering the credit market at all. The evidence from the automobile market suggests that the truth lies between the two extremes. Relative to audit studies, differences in transacted prices show evidence of discrimination but somewhat smaller price differentials than were found in the audit

36. Ian Ayres and Joel Waldfogel, “A Market Test for Race Discrimination in Bail Setting,” Stan- ford Law Review 46 (1994): 987–1048.

Race in Markets and the Justice System 351 studies. We do not know whether this result is specific to the automobile market or generalizes to other markets. There is also evidence that patterns of segregation established in the middle of the twentieth century continue to influence present-day residential patterns. Older and more segregated cities continue to be relatively segregated today. We have seen some evidence that segregation hurts blacks, which implies that, at least to some extent, past discrimination continues to have negative effects on blacks. Finally, in chapter 11 we discussed the labor market cost of having been incarcerated, and we discussed the very high rates of incarceration among black men. The evidence that blacks are more likely to be subjected to drug searches, and are likely to have to post higher bail given their flight risk, suggests that discrimination in the justice system contributes to the poorer labor market outcomes of blacks, especially low-skilled men.

5. Further Reading Ayres, Ian, and Peter Siegelman. “Race and Gender Discrimination in Bargaining for a New Car.” American Economic Review 85 (June 1995): 304–21. Ayres, Ian, and Joel Waldfogel. “A Market Test for Race Discrimination in Bail Set- ting.” Stanford Law Review 46 (1994): 987–1048. Balsa, Ana I., and Thomas G. McGuire. “Prejudice, Clinical Uncertainty and Stereo- typing as Sources of Health Disparities.” Journal of Health Economics 22 (January 2003): 89–116. Ladd, Helen F. “Evidence of Discrimination in Mortgage Lending.” Journal of Eco- nomic Perspectives 12 (Spring 1998): 41–62. Munnell Alicia H., et al., “Mortgage Lending in Boston: Interpreting HMDA Data.” American Economic Review 86 (March 1996): 25–53. Ross, Stephen L., and John Yinger. “Does Discrimination in Mortgage Lending Exist? The Boston Fed Study and Its Critics.” In Margery Austin Turner and Felicity Skid- more, eds., Mortgage Lending Discrimination: A Review of Existing Evidence (Wash- ington, DC: Urban Institute, 1999). Schelling, Thomas C. “Sorting and Mixing: Race and Sex.” In his Micromotives and Macrobehavior (New York: Norton, 1978), chapter 4. Siegelman, Peter. “Racial Discrimination in ‘Everyday’ Commercial Transactions: What Do We Know, What Do We Need to Know, and How Can We Find Out?” In Michael Fix and Margery Austin Turner, eds., A National Report Card on Dis- crimination in America: The Role of Testing (Washington, DC: Urban Institute, 1998), 69–97. Yinger, John. “Evidence on Discrimination in Consumer Markets.” Journal of Economic Perspectives 12 (Spring 1998): 23–40.

6. Questions for Discussion 1. Describe the evidence on race discrimination in consumer markets, including credit, housing, automobiles.

352 Chapter 13 2. Supposing that both blacks and whites prefer to live in integrated neighborhoods, explain how differences in the blacks’ and whites’ desired racial mix in integrated neighborhoods could nevertheless result in all neighborhoods being segregated. 3. Does neighborhood segregation have adverse effects on blacks? 4. What is the spatial mismatch hypothesis?

Race in Markets and the Justice System 353 chapter 14

Sex Discrimination

F  - ,   -1950   1980, the ratio of female earnings to male earnings among full-time, year-round workers hov- ered around 60 percent. In the past twenty years, this ratio has increased significantly. In this chapter we will be concerned with understanding both the reasons for the gap and why it diminished. In particular, we will consider the role of discrimination and other factors.

1. Theory Women might earn less than men because they are less skilled, because they choose more pleasant jobs, or because they face discrimination. These explanations are not mutually exclusive. Women might be paid less, in part, because their skill level is lower than men’s but might also be paid less than men with the same skill level because of discrimination. Still, it is useful to discuss the explanations separately. Women and men also tend to be employed in different jobs. Women are more likely than men to be psychologists and less likely to be economists. Again, this could be because women prefer different occupations or because discrimination pushes them into psychology and away from economics. Wage differentials and occupational segregation are also related, but we will address them in turn.

1.1. Human Capital Theory: Wages. One reason that women earn less than men is that, even among those working full-time, women tend to have less experience. Women traditionally have been more likely to withdraw from the labor market after marriage or after the birth of their first child and to reenter only after divorce, after all their children have started school, or after their children have left home. This means that women are likely to have less work experience than men of the same age. The con-

354 sensus estimate is that about half of the traditional 40 percent earnings differential could be explained by the difference in experience.1 The loss of experience has a direct effect by making women less skilled workers, but it goes beyond this. As Jacob Mincer and Solomon Polachek have emphasized, periods of labor force withdrawal are also associated with depreciation of skills.2 Moreover, workers often spend a significant amount of time looking for jobs to which they are particularly well matched. When women withdraw from the labor market, they are not necessarily able to return to the job in which they were previously employed. They may therefore lose much of the benefit of their previous search for a job to which they were particularly suited. Moreover, the prospect of labor market withdrawal will also affect women’s behavior. Women who expect to withdraw from the labor market will not seek jobs or experiences that involve investing in themselves, and they will spend less time looking for jobs to which they are better suited.3 In chapter 6 we saw that one advantage of marriage is that the husband and wife can specialize in different types of production. One can put more effort into producing in the market (working) and the other into home production. If they are equally good at home production, the one who can earn more in the market will put more time into market production while the other puts more time into home production. Even very small differences in wages could lead to large differences in the amount of time spent in home production and working. So, according to human capital theory, this difference builds upon itself. Women expect to spend less time in the market and therefore invest in themselves less than men do. Therefore, they expect that if they marry, their husbands will be likely to earn more than they do and thus that their husbands will focus more on work while they focus on home production. This makes investing in human capital even less worthwhile. One obvious objection to this explanation is that if the human capital explanation were correct, we would expect women to receive less education than men do. In fact, although historically women did receive less education than men, the differences were not large. And the pattern has changed. As can be seen in figure 14.1, women under the age of about 50 are more likely than men of the same age to have at least some education beyond high school, and women younger than their early 40s are more likely to have graduated college than are men of the same age. Thus the direct evidence does not suggest that women invest less in human capital. There are two related and important qualifications. The first is that graduating college

1. Claudia Goldin and Solomon Polachek, “Residual Differences by Sex: Perspectives on the Gender Gap in Earnings,” American Economic Review 77 (May 1987): 143–51. 2. Jacob Mincer and Solomon Polachek, “Family Investment in Human Capital: Earnings of Women,” Journal of Political Economy 82 (March–April 1974): S76–S108, and “Women’s Earnings Reexamined,” Journal of Human Resources 13 (Winter 1978): 118–34. 3. Solomon W. Polachek, “Occupational Self-Selection: A Human Capital Approach to Sex Dif- ferences in Occupational Structure,” Review of Economics and Statistics 63 (February 1981): 60–69.

Sex Discrimination 355 Percentage 70 HS or Less, Male HS or Less, Female BA or More, Male BA or More, Female 60

10 22 27 32 37 42 47 52 57 62 67 72 Midpoint of Age Interval Figure 14.1 Education Distribution by Sex and Age (2001)

is a very gross measure of human capital investment. Women and men tend to major in different subjects. In the mid-1970s, over 40 percent of women college freshmen intended to major in education, fine arts, or the humanities, compared with less than 15 percent of men.4 The second qualification is that college is also an important investment in the marriage market. In the 1950s and 1960s, many women jokingly (and sometimes not jokingly) referred to going to college as “getting their Mrs.” Claudia Goldin has estimated that women college graduates of the 1950s had husbands whose incomes were, on average, 40 percent higher than the husbands of women who only graduated high school. The reason for much of the difference was that the college- educated women married more educated men.5

1.2. Occupational Segregation. Even casual observers can see that women are more likely than men to be in certain occupations and less likely to be in others. For example, women are more likely to be elementary schoolteachers, nurses, flight attendants, and housecleaners. Men are more likely to be cardiologists, pilots, and janitors. Even controlling for measured characteristics such as education and experience, people working in jobs with higher proportions of women earn less, on average, than otherwise comparable workers employed in occupations with lower proportions of women. Our understanding of the sex difference in earnings depends critically on why women tend to be found in certain occupations and why these occupations tend to pay lower wages. For example, if women, on average, care more about job safety than men

4. Solomon W. Polachek, “Sex Differences in College Major,” Industrial and Labor Relations Re- view 31 (July 1978): 498–508. 5. Claudia Goldin, “Career and Family: College Women Look to the Past,” in Ronald Ehrenberg and Francine Blau, eds., Gender and Family Issues in the Workplace (New York: Russell Sage, 1997), 20–58.

356 Chapter 14 do, men will tend to work in more dangerous occupations and get paid more for the greater risk they are taking. Differences in earnings would simply reflect differences in tastes between men and women. Women’s jobs would tend to pay less but be more pleasant. Men who cared a great deal about having pleasant job characteristics would be more likely to be found in “women’s jobs” and receive low pay. Women who were more concerned with earnings would be found in “men’s jobs” and receive high pay. Neither the occupational segregation nor the wage differential would reflect discrimination. Women and men might tend to choose different jobs for human capital as well as taste reasons. As discussed earlier, when women withdraw temporarily from the labor market, they may lose some of their skills. If they expect to withdraw from the labor market while their children are young, they should avoid occupations with high rates of skill depreciation and heavy investment in skills. Paula England points out that if women who expect to leave the labor market choose “female occupations” in order to avoid the depreciation of their skills while out of the labor market, women who expect to remain in the labor market should choose occupations that look much more like those chosen by men.6 Although we can never know who expected to remain in the labor force and who expected to withdraw, on average, those who remain should tend to be those who expected to remain, and those who withdraw should be those more likely to have expected to withdraw. Yet England finds no difference in the occupations of those who eventually withdraw and those who do not.

1.3. Discrimination. Most, perhaps all, of the theories of race discrimination we have discussed can be translated into theories of sex discrimination. Employers, workers, or consumers may prefer not to associate with women, at least in some occupations. Employers, who are primarily male, may have difficulty understanding the social rules used by women. This may lead to wage differences either through costs of social interaction or through the secondary effects of statistical discrimination. And there may be self-fulfilling stereotypes. In addition to these theories, there are theories that focus primarily on sex differentials and are not applicable to race discrimination. Like the human capital model, these focus on the effects of women’s greater likelihood of withdrawing from the labor market, but unlike the human capital model, they are based on employers’ responding in a discriminatory fashion. In these models, the fact that many women withdraw from the labor market makes even women who intend to remain in the labor market unable to obtain employment in certain occupations. Barbara Bergmann has argued that the concentration of women in certain occupations reflects exclusion rather than free choice.7 Historically, women had great difficulty gaining entry into medical school or graduate school in many fields. Pilots were

6. Paula England, “The Failure of Human Capital Theory to Explain Occupational Sex Segrega- tion,” Journal of Human Resources 17 (Summer 1982): 358–70. 7. Barbara R. Bergmann, “Occupational Segregation, Wages and Profits When Employers Dis- criminate by Race or Sex,” Eastern Economic Journal 1 (April–July 1974): 103–10.

Sex Discrimination 357 recruited from the air force, and women, excluded from combat training in the military, were less experienced pilots. Firms would not hire women as janitors because they assumed that women would not know how to make the minor repairs frequently required of janitors. Bergmann argues that overcrowding in heavily female occupations lowers wages in these jobs. Why would firms discriminate against qualified women? One explanation is that firms do not know whether a worker intends to withdraw from the market. Only the worker has this information. The firm may lose money on those who quit. If the job is attractive to a woman regardless of whether she expects to leave the labor market, the firm may refuse to hire women in order to avoid hiring women who expect to leave the labor market. Peter Kuhn shows that this can lead to sex discrimination. The following example captures the flavor of his argument.8 Consider a simple world in which workers are in the labor market for at most two periods, which constitute a lifetime. Suppose that there is a set of jobs that require no training. Workers in these jobs produce (after subtracting the firm’s other costs) $10,000 per period. Therefore, firms with these jobs compete for workers until their wages are bid up to $10,000. There is another set of jobs in which workers initially produce nothing but after spending the first period training produce $40,000, but only at the firm at which they have been trained. We will ignore discounting.9 Let us consider payment in the jobs requiring training. After the workers have been trained, they are worth $40,000 to the firm but can earn only $10,000 elsewhere. We assume that they split the difference, so trained workers earn $25,000 if they remain with the firm that trained them. Because they earn more at the firm that trained them than at any other firm, trained workers do not quit. Because they are worth more than they are paid, trained workers are never laid off. So the firm knows that a worker who comes to be trained will remain with the firm and that it will make a profit of $15,000 on that worker in the second period. Firms that train workers will compete for new workers (in the first period) and will bid their first-period wage up to $15,000 (recall that first-period workers in these jobs produce nothing). So workers who intend to be in the labor force for both periods can choose between a job that will pay $15,000 in the first period and $25,000 in the second period and one that will pay $10,000 in both periods. Because the jobs are otherwise identical, it is not a hard choice. Everyone who expects to be in the labor market for two periods will want the job with training. The problem is that workers who expect to withdraw from the labor market after the first period will also want the job because it pays $15,000 for that period instead of $10,000. So not only all men (all of whom expect to remain in the job market) but also all women (some of whom expect to withdraw) apply for the jobs with training. The

8. Peter Kuhn, “Demographic Groups and Personnel Policy,” Labour Economics 1 (June 1993): 49–70. 9. Discounting addresses the fact that people prefer to have $10,000 now rather than in the future because they can invest the $10,000 now and have more than $10,000 in the future.

358 Chapter 14 firm will not make a $15,000 profit in the second period on women who withdraw. Therefore, it will not be willing to pay them $15,000 in the first period because they might withdraw. If firms knew which women would withdraw and which would not, this would not be a problem. They would offer the ones who would remain the same terms that they offered men and would offer nothing to the ones who would withdraw. But they do not know which women will withdraw. Because it will lose money on any women who withdraw, it will refuse to hire women. Although this captures the intuition behind Kuhn’s model, unfortunately it cannot tell the entire story. A firm that offered women slightly less than $10,000 in the first period would be attractive to women intending to remain in the market but not to those intending to withdraw. Knowing that the wage negotiated after training will be $25,000, those who intend to remain after training will find it worthwhile to earn slightly less than $10,000 in the first period. Those who intend to leave would prefer to work elsewhere and receive $10,000. If other firms were simply excluding women, offering training and a first-period wage of $9,000 would be immensely profitable. Finding the solution to the problem posed by this example involves somewhat messy algebra for very little insight. It is easier to complicate the example somewhat. Usually we assume that people take whatever job pays most, not because we believe it, but because it simplifies the problem and helps our intuition. In this case, it is easier to assume that people care about the nature of the job. Some will take whichever job pays more over the period they will be in the labor force. Some will take the job with training unless it pays a lot less than the job without training, and some will take the job without training unless it pays a lot less than the job with training. This is true for both men and women and is true for women regardless of whether they plan to withdraw from the labor force. Men who take the job with training will receive $40,000 ($15,000 in the first period and $25,000 in the second) over their lifetimes. If they take the job without training, they will receive $20,000 ($10,000 in each period). That is a significant difference, and only men who really prefer the job with no training will take it. Most men will choose the job with training. What about women? If firms were willing to offer $15,000 to women working in the job with training, some of the women planning to leave the market would take the job. Firms would lose money on these women and would not make money on those who stayed. So firms will not offer $15,000. They need to make enough profit on the women who stay to make up for the money they lose on those who leave. The exact first-period wage that firms will end up paying women depends on tastes for the jobs and the proportion of women who leave the market. It will certainly be less than the $15,000 first-period wage that men receive. If most women leave the market, it is likely to be less than the $10,000 that workers receive in the job without training. Note that because the first-period wage is lower for women than for men and the second-period wage is the same, the job with training pays women less than it pays men, even if both remain in the labor market for both periods. Some women who would choose the job with training if they were paid the same as men instead choose the job without training.

Sex Discrimination 359 The theory does explain why both women who remain in the labor market and those who do not earn less than men. It also explains why even women who intend to remain in the labor market are segregated occupationally. However, it does not fully replicate England’s finding that there is no difference in the occupational structure between those remaining and those exiting the labor market.

1.4. Discrimination, Specialization, and Human Capital. One of the difficulties in distinguishing between the discrimination and human capital models is that discrimination affects human capital investment decisions. Women who anticipate that they will not be able to obtain a good job should also think that they are more likely than their future husbands to specialize in home production and therefore should invest less in themselves.

2. Is There Discrimination against Women in the Labor Market? As with blacks, if we had asked this question forty years ago, the answer would have been self-evident. Many advertised jobs were clearly classified as men’s jobs or women’s jobs, and the men’s jobs usually paid more. Newspapers typically had separate columns for “Help Wanted, Male” and “Help Wanted, Female.” There is ample evidence of historical discrimination against married women and women with children. Claudia Goldin describes the “marriage bars” that many employers had that excluded married women from being hired, required that women who married be fired, and/or precluded the employment of pregnant women and women with small children.10 This is not to say that all of the wage differential between men and women was due to discrimination. As discussed earlier, much of the differential could be explained by differences in experience, and there are undoubtedly other characteristics, such as differences in physical strength, that accounted for some of the difference. Data from 1895 show that when they worked in the same factory as men and were paid the same piece rate, women still earned 23 percent less than men. This means that women produced 23 percent less output. During this period, women’s earnings in time-rated jobs in manufacturing were about 40 percent below those of men. If we attribute 23 percentage points of this differential to productivity, there is still a 17 percent differential that is not easily explained by anything other than discrimination.11

2.1. Audit Studies. More recent research has relied on the audit method we discussed earlier.

2.1.1. Restaurants. David Neumark, Roy J. Bank, and Kyle D. Van Nort sent men and women applicants to seek jobs at low-, medium- and high-priced restaurants in

10. Claudia Goldin, Understanding the Gender Gap (New York and Oxford: Oxford University Press, 1990). 11. Goldin and Polachek, 147.

360 Chapter 14 Philadelphia.12 They chose to divide their sample in this way because previous research suggested that high-priced restaurants favor men and low-priced restaurants favor women. Although the “job seekers” delivered their applications in person, they generally did not see the manager when they left the application. Thus, whether the applicant got the interview was based almost entirely on the resume. Because resumes were rotated among applicants over the period, there was no average difference in the resumes. Therefore, whether the applicant got an interview is almost entirely free of concerns about whether the male or female auditors were better job applicants. The overall results were similar for men and women. In twenty-three cases neither applicant got an interview; in twelve both got an interview. In the thirty cases in which only one was interviewed, nineteen restaurants interviewed the man and eleven the woman. If the thirty restaurants that hired only one applicant had flipped a coin, 20 percent of the time there would have been a discrepancy this large. Similarly, in the twenty-five cases in which only one applicant was offered a job, the man received the offer in sixteen cases and the woman in nine. However, if we examine the results in table 14.1, they reveal a much more interesting pattern that confirms the suggestion that more expensive restaurants favor men and less expensive ones favor women. When only one auditor got the interview at a high-priced restaurant, in eleven of fourteen cases the man was interviewed. In contrast, at the low-priced restaurants, in six out of eight cases it was the woman who was interviewed.13 The results are even more striking for job offers. At the high-priced restaurants, when only one auditor was offered a job, in ten of eleven cases it was the man. This is unlikely to have happened by chance. At low-priced restaurants we observe the opposite pattern, with a woman auditor getting the job in all six cases when only one of the pair received an offer. This, too, is unlikely to have happened by chance. As we have discussed, in audit studies we worry that the difference between men and women reflects accidental differences between the auditors we have chosen. Perhaps the man is more attractive or the woman more forceful. One of the strengths of the study by Neumark, Bank, and Van Nort is that the same auditors were used at the high- and low- priced restaurants. Presumably, if the men were chosen at high-priced restaurants because they were more attractive, they would also have been more attractive when they applied to the low-priced restaurants and would have been chosen there.

12. David Neumark, Roy J. Bank, and Kyle D. Van Nort, “Sex Discrimination in Restaurant Hiring: An Audit Study,” Quarterly Journal of Economics 111 (August 1996): 915–41. 13. There are complex issues about the best way to test statistical significance here. Neumark, Bank, and Van Nort report significance tests that are accurate for large samples. Using tests that are exact for small samples, we can reject at the .05 significance level the hypothesis that, if they interviewed only one auditor, the high-priced restaurants were equally likely to interview the man or the woman in favor of the hypothesis that they were more likely to interview the man. We can also reject at the .1 level the hypothesis that the probability of interviewing the man instead of the woman was the same at high- and low-priced restaurants in favor of the hypothesis that the probability of interviewing the man was higher at high-priced restaurants.

Sex Discrimination 361 Table 14.1 Outcomes of Men and Women in Restaurant Hiring Audits

Number of Pairs Neither Both Man Only Woman Only High-Priced Restaurants Offers 23 11 1 10 1 Interviews 23 6 3 11 3

Medium-Priced Restaurants Offers 21 9 4 6 2 Interviews 21 6 7 6 2

Low-Priced Restaurants Offers 21 13 2 0 6 Interviews 21 11 2 2 6

Source: David Neumark, Roy J. Bank, and Kyle D. Van Nort, “Sex Discrimination in Restaurant Hiring: An Audit Study,” Quarterly Journal of Economics 111 (August 1996): 915–41, table 1.

It is highly unlikely that the difference between the high- and low-priced restaurants would have occurred simply by chance. To explain the difference in terms of auditor characteristics, we must believe that the differences between the men and women auditors just happened to favor men at high-priced restaurants and to favor women at low-priced restaurants. It might appear that the benefit of having a better shot at employment at a low- priced restaurant would offset the disadvantage of a lower chance at a high-priced restaurant, so the study does not really show evidence of discrimination. However, wages and tips are higher at the high-priced restaurants. Therefore, if discrimination caused women to be more likely to end up with jobs in low-priced restaurants and less likely to work in high-priced restaurants, it would be reflected in their earnings. When we discussed audit studies that examined race discrimination, we recognized that the audit studies do not necessarily show what happens to workers in the market. Perhaps women looking for jobs in high-priced restaurants can find them readily even if some high-priced restaurants discriminate against them. Still, this study strengthens the case for the view that occupational segregation reflects, at least in part, employer rather than worker decisions.

2.1.2. Orchestras. Perhaps the most compelling evidence of sex discrimination comes from Claudia Goldin and Cecilia Rouse’s study of orchestra auditions.14 Historically, orchestras held their auditions in a way that allowed the judges to view the musicians.

14. Claudia Goldin and Cecilia Rouse, “Orchestrating Impartiality: The Impact of ‘Blind’ Audi- tions on Female Musicians,” American Economic Review 90 (September 2000): 715–41.

362 Chapter 14 Table 14.2 Outcomes of Blind and Not-Blind Auditions of Male and Female Musicians

Blind Not-Blind Proportion Number of Proportion Number of Advancing Person-Rounds Advancing Person-Rounds Preliminaries without Semifinals Women 28.6 112 19.3 93 Men 20.2 247 22.5 187

Preliminaries with Semifinals Women 20.0 20 13.3 15 Men 8.3 12 0.0 8

Semifinals Women 38.5 65 56.8 44 Men 36.8 68 29.5 44

Finals Women 23.5 17 8.7 23 Men 0.0 12 13.3 15

“Hired” Women 2.7 445 1.7 599 Men 2.6 816 2.7 1102

Source: Claudia Goldin and Cecilia Rouse, “Orchestrating Impartiality: The Impact of ‘Blind’ Audi- tions on Female Musicians,” American Economic Review 90 (September 2000): 715–41, table 5. Note: Person-rounds sums the total number of rounds participated in by all members of the sample.

However, starting in the late 1960s, many orchestras began to make auditions blind. Blind auditions hide the musician behind a screen and often use carpeting to muffle footsteps that might reveal the musician’s sex. Orchestras differed with respect to whether they adopted blind auditions. One orchestra in the Goldin and Rouse sample never used blind auditions. Others used them for the preliminary and/or the semifinal round but not for the final round, while yet others eventually adopted blind audition for all rounds. Goldin and Rouse examine the success of musicians who competed in both blind auditions and auditions that were not blind. Table 14.2 summarizes their results. In preliminary rounds, women were more likely to be successful when the audition was blind. In contrast, the outcomes were similar for men regardless of whether the audition was blind. Surprisingly, women were less likely to advance from the semifinal if it was blind. But the pattern of women’s doing better in blind auditions reemerged for the final round. Overall, women were more likely to be offered a job based on an

Sex Discrimination 363 audition in which all rounds were blind than on an audition in which at least some rounds were not blind. For men, there was no such effect. In the past fifty years, there has been a dramatic increase in the proportion of orchestra musicians who are women. Goldin and Rouse conclude that the introduction of blind auditions accounted for about one-third of the increase in the proportion of females among new hires in orchestras. Part of the increase has been due to more women applying for orchestra positions. This increase may have been, in part, a secondary effect of the increased chance of getting a position, so blind auditions may have contributed even more to the increase. Both the human capital model and some models of discrimination explain worse labor force outcomes for women by their greater tendency to withdraw from the labor market. It is natural to wonder whether this explains why women are disadvantaged when auditions are not blind. Goldin and Rouse report that there was no difference in the rate at which women and men took a leave of absence from an orchestra and no difference in tenure or length of leave. Thus, the labor force attachment story does not explain discrimination against these women musicians.

2.2. Beauty. One of the subtle ways in which sex discrimination may reveal itself is through the importance of beauty. Many people believe that women are judged on the basis of their looks, while men are not. In particular, they believe that unattractive women suffer in the labor market. Daniel Hamermesh and Jeff Biddle examine the relation between beauty and earnings.15 They argue that there is considerable agreement, at least within societies, on standards of beauty. Moreover, an individual’s beauty relative to others in his or her cohort does not change much during adulthood. They used surveys in which respondents were categorized as good-looking (above average for their age), average for their age, or quite plain (below average for their age). In all surveys, the majority of respondents were rated as average, but women were somewhat more likely than men to be rated above average or below average. Beauty was associated with earnings. Overall, an individual who is rated above average in looks earns about 12 percent more than one rated below average. Surprisingly, women are not rewarded for looks more highly than are men. If anything, the evidence points in the opposite direction.

3. Discrimination, Marriage, and Children According to Jane Waldfogel, the United States has one of the least generous family leave policies and one of the most expensive child care systems (relative to women’s earnings) among industrialized countries.16 Waldfogel argues that this creates a large gap between the earnings of men and those of women with children.

15. Daniel S. Hamermesh and Jeff E. Biddle, “Beauty and the Labor Market,” American Eco- nomic Review 84 (December 1994): 1174–94. 16. Jane Waldfogel, “Understanding the ‘Family Gap’ in Pay for Women with Children,” Jour- nal of Economic Perspectives 12 (Winter 1998): 137–56.

364 Chapter 14 Both men and women who are married earn more than seemingly comparable men and women who are not married. Waldfogel estimates that the marriage premium is about 12 percent for men and about 4 percent for women. There are at least two explanations for this marriage premium. The first is that married people, especially men, feel more responsibility to their families and therefore work harder. The second explanation is that high-earning individuals are more attractive as marriage partners. Again, this is especially true of men. Individuals with low earnings are more likely to remain single or, if married, are less likely to work, especially in the case of women. One striking difference between men and women is that children are associated with higher earnings for men but lower earnings for women. Waldfogel finds that wages are higher for men with children than for men without children, at least in the case of those with two or more children. For women, however, having children is associated with a wage penalty of about 10 percent, at least for those with two or more children. The difference in hourly earnings between single men and single women is small and perhaps nonexistent, but between married men and married women it is small if they have no children but large if they have two or more children. Waldfogel suggests that the child penalty for married women comes from the effect of labor force withdrawal and argues that more generous family leave policies would help women overcome this disadvantage. Of course, there are other interpretations of the findings. Single men with low earnings may have a harder time finding a spouse than single women with low earnings. Married couples may have more children when the man’s income is high and the woman’s income is low. Still, Waldfogel’s assessment is consistent with evidence that, at least historically, relatively few women could successfully juggle a career and motherhood. Among women who graduated from college around 1910, only 50 percent married and had children. Among women who graduated from college around 1972, by age 40 only 13 to 22 percent (depending on how having a career is measured) had both a career and at least one child.17 The Family Medical Leave Act (FMLA) of 1993 was designed to address this issue. Although modest by international standards, it was extremely controversial. President Bill Clinton signed it after it had been vetoed by President George H. W. Bush. In contrast with provisions of the policies in many other countries, the twelve weeks of leave guaranteed by the FMLA is unpaid (although the employer must continue to pay its share of any health insurance premiums). Moreover, the coverage is modest. It is limited to workers employed for 1,250 hours per year (with an exception for full-time teachers) in firms with at least fifty workers. Individuals can use the leave for maternity, adop- tion, newborn care, care of a sick family member, or their own medical care. Proponents of the FMLA argue that it will benefit firms because it will reduce employee turnover, and turnover is costly. Most economists believe that if guaranteeing family leave would help firms, no regulation would be necessary; firms would offer it on their own because doing so would be profitable. However, this need not be the case.

17. Claudia Goldin, “College Women across the Twentieth Century,” in Francine D. Blau and Ronald Ehrenberg, eds., Gender and Family Issues in the Workplace (New York: Russell Sage, 1997).

Sex Discrimination 365 Firms probably prefer to hire workers who are unlikely to take family leave. Even if all workers are willing to take a pay cut in order to have family leave if they need it, workers who think it is more likely that they will need leave will be willing to take a larger pay cut. In the absence of the FMLA, a firm that tried to increase its profit by offering leave and lowering wages would attract workers with a high probability of taking leave. As a result, most firms might choose not to offer family leave if not required to do so by the FMLA.

4. Sexual Orientation It is even more difficult to define sexual orientation than to define race. Those who want to claim that the number of gays and lesbians is high can use the “one thought” rule: anyone who is occasionally attracted to a person of the same sex is homosexual. At the other extreme, we may limit the definition to people who engage exclusively in same-sex sex. In between, we may rely on whether the individual has recently had same- sex sex or had primarily same-sex sex. Or perhaps it is best to rely on self-identification: a gay man or lesbian woman is anyone who so defines himself or herself. For purposes of studying discrimination, we probably want to focus on a definition that requires people to act on their sexuality. It is difficult to imagine that employers are able to discriminate against people who think about having same-sex sex. It is also probably better to focus on recent sexual activity, because current employers are less likely to know about behavior in the distant past. However, if people make human capital investment decisions based on their identity when they are young, their past behavior may be related to their current earnings through those investment decisions, a point to which we will return shortly. Whether there is discrimination on the basis of sexual orientation is interesting in its own right but also because of what it tells us about sex discrimination. The human capital model discussed earlier argues that women tend to specialize in home production because they earn less than their husbands. Therefore, women who are making human capital investment decisions expect to spend less time in the labor market and choose to invest less in their human capital. This reinforces their tendency to earn less than their husbands do. Similarly, under the human capital model, women choose occupations with less specialized human capital because they expect to withdraw from the labor market. In contrast, under models of discrimination, employers exclude women from certain occupations because they expect them to quit. On average, gay men and lesbian women in long-term relationships have a 50 percent chance of earning more than their partner. This is lower than the probability for a man in a heterosexual relationship and higher than the probability for a woman in such a relationship. Therefore, relative to heterosexual women, the human capital model predicts that lesbian women will tend to specialize less in home production and that, relative to heterosexual men, gay men will tend to specialize less in market production. Thus, we would expect gay men to earn less in the market and lesbian women to earn more in the market than their heterosexual counterparts do.

366 Chapter 14 In addition, gay men and lesbian women should be less likely to expect to have traditional families. Thus, the human capital model implies that when they make their prework investment decisions (e.g., about how much education to obtain and what courses to take), gay men should be less job market–oriented and lesbian women more job market–oriented than their heterosexual counterparts. Dan Black, Hoda Makar, Seth Sanders, and Lowell Taylor examine the relation between sexual orientation and earnings among full-time workers.18 Note that using full-time workers makes it somewhat difficult to detect the effects of specialization in market versus home production. Nevertheless, there is variation in the intensity of work effort and hours worked even among those working full-time.

4.1. Gay Men. The results of Black and his coauthors are largely unaffected by the definition they use to determine who is a gay man. Whether being a gay man is measured as having had more male than female sex partners, having had any male sex partners in the past year (or the past five years), or having had more male than female sex partners in the past year (or the past five years), gay men earn about 14 percent less than heterosexual men. These results are certainly consistent with the human capital model outlined earlier, although they are also consistent with discrimination on the part of employers. However, gay men also tend to have more education than heterosexual men, which is not consistent with the human capital model. Throughout this book we have emphasized that it is impossible to control for all the other factors that may affect outcomes. We have often been concerned with whether a comparison group was appropriate. Black and his coauthors generate an additional result for men that underscores this warning. Men who have not had sex within the past year earn 15 percent less, and men who have not had sex within the past five years earn 22 percent less than apparently similar heterosexual men. Note that the no-sex penalty is even larger than the same-sex sex penalty for men. It is almost impossible to explain this differential by discrimination. American men do not usually tell employers that they have not had sex for the past five years. It is also difficult (although not impossible) to believe that men who do not have sex invest less in human capital. It is much more plausible that the differential reflects one or both of two factors. First, women are more likely to have sex with successful men. So low earnings may cause lack of sex rather than the other way around. Second, some of the features that make a man less attractive as a sex partner may also make him less attractive as a worker. The fact that the estimated penalty for not having sex is at least as large as the penalty for being a gay man should make us cautious about putting too much weight on the conclusion that gay men earn less than heterosexual men because of their sexual preference. We believe that we are missing data on important factors that influence the relative wages of men who do not have sex. These missing factors might also explain some of the differential based on sexual preference.

18. Dan A. Black, Hoda R. Makar, Seth G. Sanders, and Lowell J. Taylor, “The Earnings Effects of Sexual Orientation,” Industrial and Labor Relations Review 56 (April 2003): 449–69.

Sex Discrimination 367 4.2. Lesbian Women. The results for lesbian women are somewhat more sensitive to how sexual orientation is defined. When a lesbian woman is defined as one who has had more female than male sex partners, there is no clear difference in the earnings of lesbian women and other women. When the determining factor is whether she has had a female sex partner in the past year or the past five years, lesbian women earn about 25 percent more than heterosexual women, and when it is restricted to those who have had only female sex partners over the past five years, the differential rises to 40 percent. There is no difference in the earnings of heterosexual women and those who have not had sex either in the past year or previous five years. The fact that lesbian women earn more than heterosexual women is consistent with the prediction of the human capital model. But if the human capital model is correct, we would expect lesbian women to be employed in higher-paying occupations than those in which heterosexual women are employed. In fact, lesbian women do tend to work in higher-paying occupations, but the effect of being a lesbian is small. It accounts for only about 5 percentage points of the 40 percent lesbian-heterosexual differential among women. It is also striking that when a lesbian woman is defined as one who has had a majority of female sex partners over her lifetime, the effect on wages of being a lesbian woman is much smaller. The human capital investment model suggests that the long- term measure might have more effect than a short-term measure. These results are striking in a number of ways. First, they suggest that there is not widespread labor market discrimination on the basis of sexual orientation, although there could be discrimination against gay men and not against lesbian women. Second, they suggest that some of the earnings differential between men and women is likely to reflect market specialization and investment decisions.

5. Trends in the Female/Male Wage Ratio Figure 14.2 shows the ratio of median female earnings to median male earnings among full-time, year-round workers from 1960 to 2004. Because the comparison is restricted to workers who worked both full-time and for the entire year, little, if any, of the difference can be attributed to differences in labor supply. We can see that from at least 1960 until the early 1980s, women earned about 60 percent of what men earned. Then, starting in the 1980s, there was a steady and rapid rise in the relative earnings of women. By 2001, women’s earnings were almost 80 percent of men’s. Among blacks there was a similar pattern, although at a higher level. Starting in 1967 (the earliest year for which data disaggregated by race are available), the ratio of median female to male earnings was about 67 percent. It reached 88 percent in the mid-1990s and was 89 percent in 2004.

5.1. Changes in Human Capital. To what extent can the human capital model explain the pattern in figure 14.2? Figure 14.3 shows dropout rates and college gradua-

368 Chapter 14 Ratio of Median Earnings 0.9

0.8 Blacks

0.7

All 0.6 Whites

0.5 1960 1970 1980 1990 2000 Figure 14.2 Ratio of Median Earnings for Full-Time, Year-Round Workers, Women/Men, 1960–2004

Percentage 60 Dropouts, Men Dropouts, Women

College, Men College, Women 0 1962 1970 1980 1990 2000 2004 Figure 14.3 Dropout and College Graduation Rates for Men and Women over Age 25, 1962–2004 Source: U.S. Census Bureau, “Years of School Completed by People 25 Years and Over, by Age and Sex: Selected Years 1940 to 2004,” http://www.census.gov/population/socdemo/ education/tabA-1.pdf. tion rates for the population over age 25. As can be seen in the figure, the education gap between men and women did widen between the early 1960s and the early 1980s before narrowing again. But over the entire period there was little change in the relative education levels of men and women. So education cannot account for the narrowing of the earnings gap.

Sex Discrimination 369 As early as 1984, James Smith and Michael Ward pointed out that the relative constancy of relative wages over the previous period was due to two offsetting factors.19 On the one hand, women were becoming increasingly skilled relative to men. On the other hand, the skill levels of working women were not rising relative to the skill levels of working men. The rapid increase in women’s labor force participation increased the number of inexperienced women workers in the labor market even though women were becoming less likely to withdraw from the labor market. The authors predicted that over the following twenty years, the relative earnings of women would rise rapidly. Their prediction has proved accurate. In a later study, Smith and Ward found that there was almost no increase in the labor market experience of working women between 1950 and 1980.20 The modest half-year increase in experience was more than offset by a one-year decline in relative education over this period. However, by 1980, the increase in women’s labor force participation had begun to play out. Women constituted almost half of the labor force. There was little room for increased labor force participation of new entrants to bring down the average level of experience. Based on this assessment, Smith and Ward predicted that by the year 2000, women’s earnings would be at least 74 percent of men’s and probably higher. Because their prediction was based solely on improvements in the relative education and experience of working women, the accuracy of their prediction leaves little room for the effect of changes in social attitudes and enforcement of civil rights except for the effect of these changes on women’s own human capital investment decisions. Francine Blau and Lawrence Kahn examined this question more directly.21 They found that the gap in pay between women and men fell by 15.2 percent between 1979 and 1988. Just under 1 percentage point of this change could be explained by the increased relative education of women workers, 5.3 percentage points by their increased experience, and 1.8 percentage points by the decline in unionization, which lowered wages for men. Of the remainder, 4.6 percentage points could be explained by the shift of women toward higher-paying occupations, while 2.8 percentage points could not be explained by measured variables.

5.2. The Pattern by Age and Cohort. Table 14.3 shows the female/male ratio of median (personal) money income by birth year cohort and year.22 It also allows us to

19. James P. Smith and Michael Ward, “Women’s Wages and Work in the Twentieth Century” (Report R-3119-NICHD, Rand Corporation, Santa Monica, CA, 1984). 20. James P. Smith and Michael Ward, “Women in the Labor Market and in the Family,” Jour- nal of Economic Perspectives 3 (Winter 1989): 9–23. 21. Francine D. Blau and Lawrence M. Kahn, “Swimming Upstream: Trends in the Gender Wage Differential in 1980s,” Journal of Labor Economics 15 (January 1997): 1–42. 22. Unfortunately, data on median earnings are available only for a much more recent period. Fortunately, however, for the median full-time, year-round worker, almost all income comes from earnings, except in the oldest age group. Still, women tend to have a somewhat lower proportion of income from earnings (at least in recent years), so the numbers in table 14.3 are slightly higher than they would be if we had information on median earnings.

370 Chapter 14 Table 14.3 Female/Male Ratio of Median Money Income for Full-Time, Year-Round Workers, by Birth Year Cohort, for Selected Years, 1955–2000

Birth Cohort 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000

1891–1900 64.5 1896–1905 64.5 1901–1910 62.2 62.4 1906–1915 58.0 60.3 1911–1920 63.0 57.8 57.6 1916–1925 57.6 56.3 56.7 1921–1930 66.1 56.6 53.9 59.0 1926–1935 65.1 53.9 54.3 62.6 1931–1940 62.1 54.9 56.9 62.3 1936–1945 64.9 56.2 61.4 64.8 1941–1950 65.8 62.3 63.7 1946–1955 68.8 69.0 68.5 1951–1960 75.0 72.3 1956–1965 79.6 73.3 1961–1970 82.3 1966–1975 81.7

Source: June O’Neill and Solomon Polachek, “Why the Gender Gap in Wages Narrowed in the 1980s,” Journal of Labor Economics 11 (January 1993): 205–28, table 2, http://www.census.gov/hhes/income/ dinctabs.html.

look at the pattern within each age group over time by going down the diagonals and across age groups in a given year by going down the columns. Note that older cohorts are at the top of the table, so the highest diagonal gives the ratio for 55- to 64-year-olds and the one below for 45- to 54-year-olds, followed by 35- to 44-year-olds and 25- to 34-year-olds. The same pattern holds if we read down a column. Reading down the diagonals, we can see that within each age group the income ratio initially fell and then increased. The ratio was lowest in 1965 for 25- to 34-year- olds, in 1970 for 35- to 44-year-olds, in 1975 for 45- to 54-year-olds, and in 1980 for 55- to 64-year-olds. Between 1980 and 1990, relative earnings within an age group increased rapidly, with generally modest additional increases between 1990 and 2000. The relation between the earnings ratio and age has changed over time. There has been a fairly consistent pattern: the ratio falls as we move from 25- to 34-year-olds to 35- to 44-year-olds. Historically, the ratio was relatively constant over the 35–54 age range, but after 1980, the earnings ratio was noticeably lower for the 45–54 age group. Similarly, in the earlier period, the ratio was somewhat higher for 55- to 64-year-olds than for 45- to 54-year-olds. More recently, the pattern appears to have been reversing itself. What explains these patterns? It is here that looking at cohorts is helpful. We can see that for people born in the earlier time periods, there was little change in the ratio as the cohort aged from 45–54 to 55–64. For example, for the 1901–1910 cohort the

Sex Discrimination 371 change was 0.2 percentage points (from 62.2 percent to 62.4 percent). However, starting with the 1921–1930 cohort, there was a change in the pattern. As it aged from 45–54 to 55–64 (from 1975 to 1985), this cohort experienced an increase in the ratio of 5.1 percentage points. This change was 7.9 percentage points for the 1926–1935 cohort and 5.4 percentage points for the 1931–1940 cohort. Similarly, in the earlier and later cohorts that we can follow from ages 25–34 to ages 35–44, we see a declining or stable ratio as they aged. However, for the cohorts that made this transition between 1975 and 1995, we see an increase, with the largest increase coming for the cohort that made the transition between 1980 and 1990. Finally, with one exception, the ratio declined for every cohort as it aged from 25–34 to 35–44. The single exception, the 1946–1955 cohort, experienced this age change from 1980 to 1990. The table makes a compelling case that something important happened in the 1980s that went beyond mere increases in human capital. As already noted, over this period women were shifting into higher-paying occupations for a fixed level of human capital, but there was a significant residual increase that is not explained by either increased human capital or occupational shifts.

5.3. The Role of Policy. Sex discrimination is covered by many of the laws we discussed in the previous chapter. Title VII of the 1964 Civil Rights Act prohibits employment discrimination against women. From the perspective of the twenty-first century, it seems natural that the law would cover sex as well as race, religion, and ethnicity. In fact, women were covered by Title VII only because of a strategic error on the part of southern opponents of the Civil Rights Act.23 Southern Democrats introduced the amendment to prohibit sex discrimination in employment because they believed it would encourage Republican moderates to oppose the bill. Even though many liberals opposed the amendment because they believed it would ensure the bill’s defeat, enough voted for the amendment that it passed. To the dismay of the act’s opponents, the Civil Rights Act passed despite the inclusion of women in the protection provided by Title VII. Consistent with the greater support for ensuring equal opportunity employment on the basis of race, President Lyndon Johnson did not require federal contractors to take affirmative action to avoid discrimination on the basis of sex until 1967, two years after the initial executive order addressing affirmative action and race. It was not until the Equal Employment Act of 1972 that the regulations enforcing this were completed. The mid- to late 1970s were an active period in addressing sex discrimination. The Office of Federal Contract Compliance Programs was active in enforcing affirmative action until its activity came to an abrupt halt in the 1980s under the Reagan administration. Similarly, the Equal Employment Opportunity Commission was active in

23. David B. Filvaroff and Raymond E. Wolfinger, “The Origin and Enactment of the Civil Rights Act of 1964,” in Bernard Grofman, ed., Legacies of the 1964 Civil Rights Act (Charlottesville, VA: University Press of Virginia, 2000).

372 Chapter 14 pursuing cases between 1975 and 1982 but not for the remainder of the 1980s.24 In 1978, the Pregnancy Discrimination Act amended the Civil Rights Act to protect pregnant women. In 1986, in Meritor Savings Bank v. Vinson, the Supreme Court ruled that sexual harassment is a form of sex discrimination and thus prohibited under Title VII. In 1988, the court further clarified that abusive behavior based on sex is harassment even if the abuse is not sexual in nature (Hall v. Gus Construction Company). It is impossible to separate the effects of the changes in individual laws and policies from those of general changes in social attitudes. Still, the rapid rise in the relative wages of women between 1980 and 1990 is consistent with the lagged effects of a flurry of enforcement activity in the late 1970s and the effects of legal decisions in the 1980s. There is no compelling estimate of how much of the improvement in the relative earnings performance of women can be attributed to declining discrimination, but a reasonable guess is that its direct effect was modest, probably accounting for 20 percent or less of the narrowing of the gap over the past twenty-five years. On the other hand, it is impossible to know how much diminished discrimination changed women’s expectations about their labor market prospects and thus encouraged them to invest more heavily in marketable skills. This effect could have been large or small.

6. Comparable Worth The apparent failure of women’s wages to catch up with men’s in the 1960s and 1970s encouraged advocates of women’s equality to search for new policies. Some analysts concluded that equal pay for equal work was inadequate. Women and men are found in different proportions in different occupations. Women’s jobs have lower wages than men’s jobs. Therefore, they concluded, equalizing pay within occupations would not do enough to eliminate the gender gap. The failure of men’s and women’s wages to converge despite the civil rights initiatives of the 1960s confirmed this perspective. Con- sequently, some advocates argued that what was needed was not equal pay for equal work but “equal pay for work of comparable worth,” or “comparable worth” for short. It is important to note that the fact that women’s jobs pay less than men’s job need not reflect discrimination. In order to attract workers, jobs that involve more un- pleasant working conditions must pay higher wages. The higher wage is a compensating differential that offsets or compensates for the less desirable job characteristics. For example, dangerous jobs should pay more than safe jobs. If men are less averse to risk than women are, they will be more likely to work in dangerous jobs and will receive a compensating differential for doing so. The critical question then is whether the wage differentials among jobs are compensating differentials or whether women’s jobs pay less because of discrimination. In order to address this question, policy makers must be prepared to make comparisons across very different jobs: should trash collectors or secretaries have higher wages? As a

24. David L. Rose, “Twenty-Five Years Later: Where Do We Stand on Equal Employment Op- portunity Laws Enforcement?” Vanderbilt Law Review 42 (May 1989): 1121–81.

Sex Discrimination 373 result, there is no real way to implement comparable worth without significantly altering the wage-setting mechanism. Comparable worth has been implemented, for the most part, either in government jobs or in countries such as Australia in which government is heavily involved in wage setting.

6.1. Theory. Opponents of comparable worth often argue that it is likely to hurt rather than help women. In theory, comparable worth adjustments work very much as a minimum wage does. By raising wages above those set by the market in some jobs, the comparable worth adjustment reduces demand for workers in those jobs and increases the labor supply. The reduction in demand shifts some workers to jobs that are not covered by comparable worth adjustments. The increase in labor supply lowers wages in those uncovered jobs. Thus, comparable worth will lower demand for workers in women’s jobs. However, it is important to distinguish between demand for women workers and demand for workers in women’s jobs. By creating excess supply in the market for women’s jobs, comparable worth makes it possible for employers to choose the workers they prefer from among all those applying. If employers prefer to hire women in women’s jobs, men who work in such jobs may be the ones displaced by the increased wage. Thus, although under the standard analysis, employment should decrease both in covered jobs and in women’s jobs as a whole, employment of women need not decrease. The standard analysis just described depends critically on labor markets’ being competitive, with lots of similar firms competing for similar workers. The competitive model is unlikely to apply to government employment for three reasons: • In many places government is a large employer, particularly within individual occupations. • The public sector is often unionized. Unions may negotiate over employment as well as wages. • Political factors play a role in determining both hiring levels and the types of workers who are hired.

6.2. The Cases of San Jose and Washington State. One of the earliest and perhaps the most thoroughly studied cases in which comparable worth was implemented was in the city of San Jose. In July 1981, the American Federation of State, County, and Municipal Employees (AFSCME) went on strike and won a comparable worth agreement from the city. The agreement raised wages over a three-year period for 20 percent of the city’s workers, including 54 percent of the workers represented by AFSCME. In a study of the effects of the implementation of comparable worth in San Jose, Shulamit Kahn found that targeted jobs had faster wage growth in San Jose than in comparable California cities.25 Yet employment grew faster in San Jose than in other large California cities and more than in other Silicon Valley municipalities. Kahn found

25. Shulamit Kahn, “The Economic Implications of Public-Sector Comparable Worth: The Case of San Jose, California,” Industrial Relations 31 (Spring 1992): 270–91.

374 Chapter 14 that, contrary to the theory discussed earlier, employment grew faster in comparable worth–targeted jobs than in other jobs. Of course, perhaps employment would have grown faster in targeted jobs anyway. Kahn examined this possibility by looking at other Silicon Valley municipalities. She found that they did not have faster growth in the jobs that were targeted in San Jose. Thus, it appears that, if anything, comparable worth increased employment in targeted jobs in San Jose. Moreover, almost all of this growth in targeted jobs was among women. Kahn concludes that two factors may have combined to create this surprising result. First, wages and salaries in targeted jobs were somewhat low relative to other communities. The increased wages may have allowed San Jose to attract more workers to these jobs. Second, San Jose had an affirmative action program that rewarded department heads for hiring women regardless of the job for which they were hired. By increasing the supply of women to targeted jobs, the comparable worth program facilitated implementation of the affirmative action goals. June O’Neill, Michael Brien, and James Cunningham examined the implementation of comparable worth in Washington State.26 They found that comparable worth did raise wages in targeted jobs. They also found that employment grew more slowly in those jobs than in jobs that were not targeted for comparable worth. Unfortunately, their study did not compare Washington to other states, so we do not know whether the types of jobs targeted for comparable worth were experiencing slower employment growth for reasons unrelated to the comparable worth policy.

7. Concluding Remarks As difficult as it is to determine the extent and nature of race discrimination, it is even more difficult in the case of sex discrimination. In part, this reflects the fact that we are much more inclined to believe that men and women have very different tastes than we are to believe this about blacks and whites. In part, it reflects the fact that differences in biology affect what men and women can and cannot do. And in part, it reflects the fact that our models of discrimination often imply that women will respond to discrimination in ways that are difficult to distinguish from the effects of differences in taste. Thus, it is clear that a significant fraction of the historical 40 percent earnings differential between men and women can be “explained” by the fact that women had less experience. What is not easily determined is how much of that difference in experience reflected women’s desire to spend more time out of the labor market and how much reflected women’s experience of discrimination or their beliefs that they would face discrimination. If women have less experience because they tend to prefer to spend more time outside of the labor market, there is probably very little wage discrimination in the labor market.

26. June O’Neill, Michael Brien, and James Cunningham, “Effects of Comparable Worth Policy: Evidence from Washington State,” American Economic Review 79 (May 1989): 305–9.

Sex Discrimination 375 There is some evidence that supports the view that there is little wage discrimination against women in the labor market. Lesbian women have wages that are similar to men’s. Moreover, they are employed, on average, in higher-paying occupations than are other women. Although it is possible that lesbian women face less discrimination than do heterosexual women, it seems unlikely. The fact that the wage gap is much larger between married men and women with children than between childless married men and women or single men and women is also suggestive of the view that the wage gap reflects choices men and women have made to specialize in different types of production. However, these results are only suggestive. Marital status and parental status are not chosen randomly. There is also some evidence of wage discrimination (including audit studies and patterns of occupational segregation). The historical evidence is strong, and whatever impact legislation and the women’s movement have had (and the evidence supports the view that they have had some), they are unlikely to have completely eliminated discrimination. But there is a subtler issue related to discrimination. The way the U.S. economy is organized seems to force women to choose between having a successful career and having children. Claudia Goldin has demonstrated the rarity of combining the two, and Jane Waldfogel’s evidence at least suggests that women pay a significant wage penalty for having children. Noneconomists are likely to view this as discriminatory. Economists are not. After all, “life is full of choices” and economics is all about how people make those choices. Economists do not say that people face discrimination because their budget does not allow them to purchase everything they would like. We do not say that someone who likes five-star restaurants and living in the wilderness faces discrimination just because it is difficult to indulge both tastes. In the case of the wage penalty that some say women pay, I think the economists are wrong, for two reasons. First, recall that when discussing race, we defined discrimination as any situation in which blacks and whites with identical observed characteristics have systematically different outcomes. So if men who want careers and children can have both but similar women cannot, this constitutes discrimination. Of course, if women who want to stay at home and take care of their children can, but similar men with the same preference cannot, that, too, is discrimination. The second reason is that many institutions are arbitrary in the sense that they evolved for historical reasons. In most places, students no longer use the long summer school vacation to work on the family farm. Entrepreneurs have developed summer camps to accommodate students on summer vacation. If we switched to a system with a three-week March vacation, without doubt there would be entrepreneurs who would respond by offering programs for school-aged children in March. But a single school that decided to move three weeks of summer vacation to March would face tremendous obstacles, because such programs would not be available. If institutions are arbitrary, at least to some degree, we can ask whether the United States has developed a set of institutions that are more or less discriminatory than those

376 Chapter 14 that have arisen in other advanced economies. A reasonable case can be made that, relative to labor market institutions in other advanced economies, those in the United States are unusually unsupportive of families with two working parents.

8. Further Reading Black, Dan A., Hoda R. Makar, Seth G. Sanders, and Lowell J. Taylor. “The Earnings Effects of Sexual Orientation.” Industrial and Labor Relations Review 56 (April 2003): 449–69. Blau, Francine D., and Lawrence M. Kahn. “Swimming Upstream: Trends in the Gen- der Wage Differential in 1980s.” Journal of Labor Economics 15 (January 1997): 1–42. England, Paula. “The Failure of Human Capital Theory to Explain Occupational Sex Segregation.” Journal of Human Resources 17 (Summer 1982): 358–70. Goldin, Claudia. “Career and Family: College Women Look to the Past.” In Ronald Ehrenberg and Francine Blau, eds., Gender and Family Issues in the Workplace. New York: Russell Sage, 1997, 20–58. Goldin, Claudia. “College Women across the Twentieth Century.” In Francine D. Blau and Ronald Ehrenberg, eds., Gender and Family Issues in the Workplace. New York: Russell Sage, 1997. Goldin, Claudia. Understanding the Gender Gap. New York and Oxford: Oxford Uni- versity Press, 1990. Goldin, Claudia, and Cecilia Rouse. “Orchestrating Impartiality: The Impact of ‘Blind’ Auditions on Female Musicians.” American Economic Review 90 (September 2000): 715–41. Hamermesh, Daniel S., and Jeff E. Biddle. “Beauty and the Labor Market.” American Economic Review 84 (December 1994): 1174–94. Kahn, Shulamit. “The Economic Implications of Public-Sector Comparable Worth: The Case of San Jose, California.” Industrial Relations 31 (Spring 1992): 270–91. Mincer, Jacob, and Solomon Polachek. “Family Investment in Human Capital: Earn- ings of Women.” Journal of Political Economy 82 (March–April 1974): S76–S108. Mincer, Jacob, and Solomon Polachek. “Women’s Earnings Reexamined.” Journal of Human Resources 13 (Winter 1978): 118–34. Neumark, David, Roy J. Bank, and Kyle D. Van Nort. “Sex Discrimination in Restau- rant Hiring: An Audit Study.” Quarterly Journal of Economics 111 (August 1996): 915–41. Polachek, Solomon W. “Occupational Self-Selection: A Human Capital Approach to Sex Differences in Occupational Structure.” Review of Economics and Statistics 63 (February 1981): 60–69. Smith, James P., and Michael Ward. “Women in the Labor Market and in the Family.” Journal of Economic Perspectives 3 (Winter 1989): 9–23. Waldfogel, Jane. “Understanding the ‘Family Gap’ in Pay for Women with Children.” Journal of Economic Perspectives 12 (Winter 1998): 137–56.

Sex Discrimination 377 9. Questions for Discussion 1. Describe the mechanisms through which greater labor force withdrawal on the part of women can generate wage differentials. Which of these would generally be viewed as causing discrimination against women? 2. Is occupational segregation explained by women’s preferences? 3. Is there discrimination against women in the labor market? 4. Describe the trend in the female/male earnings ratio since 1960. What accounts for this trend? 5. What do wage differentials between heterosexual and homosexual men and between heterosexual and homosexual women tell us about the reasons for the female/male wage differential? 6. Describe how comparable worth should affect the wages and employment of women. Does actual experience with comparable worth in the public sector support these predictions?

378 Chapter 14 chapter 15

Conclusion: An Agenda to Decrease Poverty and Discrimination?

T        the complexity of evaluating which policies are effective in addressing poverty and discrimination. We must always make decisions based on incomplete information and imperfect research. However, the research also points us strongly in some directions and away from others. In this chapter I outline what I believe are the principal conclusions we should derive from the large body of research on poverty and discrimination. To repeat what I wrote in the introduction, this is the least important chapter. The previous chapters have laid the groundwork both for supporting the conclusions I have reached and for ensuring that you are aware of the weaknesses of the research underlying them. More significant, as discussed in the introduction, knowledge of the facts is insufficient to resolve all issues of policy. Values also play a role, and my policy conclusions reflect my values. Most policies involve trade-offs. Even if we agree what those trade-offs are, we may not agree about which ones to choose.

1. The Value and Limits of Research Social research is often difficult because we believe that it is immoral to experiment on people. This problem is not unique to poverty and discrimination research. Medical researchers face the same issue. At what point are the expected benefits from two different treatments sufficiently close that it is acceptable to compare them rather than simply using the preferred treatment? In some ways, the problem is simpler for many social policies. Popular programs are often oversubscribed. Not all applicants for training programs, Head Start, or some charter schools can be accommodated. If there are no compelling social reasons for preferring some applicants over others, we can use a lottery to allocate places in the programs and test their effectiveness. When there are strong reasons for giving some people priority over others, we can often still evaluate the program if the priorities are clearly defined. If we wish to give priority to the poorest applicants, we can see if there

379 is a sharp jump in the outcome as we pass from applicants who were just poor enough to gain entry to those who just missed the cutoff. If the cutoff changes from year to year, the evaluation is even easier. The bottom line is that if we are going to ration entry into certain programs, we should ration it in a way that facilitates research and then use that research to address the effectiveness of the programs. Some people will argue that research merely competes with programs for resources and that we often know whether programs work without resorting to expensive evaluations. One prominent politician responded to demands that a controversial program be evaluated by saying that the program was so self-evidently good that the only thing that could be learned from an evaluation was whether the evaluation research was any good. Much that seems obvious turns out to be right, but some of it turns out to be false. For example, in this book we have seen that the effect on the adult outcomes of teenage motherhood are much more modest than most people believe. Other people maintain that we are much better off relying on experienced practitioners than on faceless researchers who have no hands-on experience. “You have no right to tell me what to do,” they might say. “You have never taught preschool (run a training program, started a company).” It is tempting to respond by reminding people that experienced practitioners recommended bleeding and leeches, practices that would not have stood up to research, but I expect that many people will view that example as outdated. Instead let me point out that football and baseball managing have been transformed by research. Bill Belichick, the New England Patriots coach, is an economics graduate from Wesleyan. A reporter told him that Berkeley economist David Romer had written a paper saying that National Football League teams punt too much1 and asked him whether he thought it was true. Belichick could have responded by saying that Romer had never played or coached professional football and did not know what he was talking about. Instead, Belichick said, “I read it. I don’t know much of the math involved, but I think I understand the conclusions and he has some valid points.” Belichick also went on to outline some of the limits of the statistical analysis, including the fact that Romer did not account for the psychological letdown from failure.2 Still, Belichick seemed to be influenced by the research. Two weeks before the Super Bowl, right at the beginning of the game, the Patriots found themselves facing a fourth down in their own territory. Based on the wisdom of experienced practitioners, the Pa- triots should definitely have punted. Instead Belichick decided not to punt. The Patriots made the first down and went on to score a touchdown that moved them further on the path to the Super Bowl.3

1. David Romer, “It’s Fourth Down and What Does the Bellman Equation Say? A Dynamic- Programming Analysis of Football Strategy” (working paper, University of California–Berkeley, 2003). 2. Michael Gee, “Football: Kicking Around New Math,” Boston Herald, September 13, 2002, 112. 3. David Leonhardt, “Ideas and Trends: Super Bowl Economics: Incremental Analysis, with Two Yards to Go,” New York Times, February 1, 2004, Late Edition—Final, section 4, column 1, Week in Review Desk, p. 12.

380 Chapter 15 Earnshaw Cook’s Percentage Baseball overturned many strategies used by managers who relied on their experience and instincts.4 Neither Cook nor Romer was a serious candidate to be a professional athlete, and I doubt that either would claim to have the breadth of knowledge necessary to manage a professional team. But their research has been influential in an area where we might have expected total resistance. Common sense and common practice are sometimes wrong. Good research can correct errors. At the same time, not all research is equally valid. We must be aware of the limitations of each study. Belichick was an intelligent consumer of Romer’s research.

2. The Value and Limits of a Strong Labor Market A rising tide does not lift all boats, but it can lift many. Anyone who thinks that a strong economy does nothing for the poor missed the expansion of the Clinton years. This does not mean that all we have to do is find ways to increase GDP per capita, and we can forget about poverty policy. We know that GDP per capita has increased dramatically over the past thirty years without an accompanying decrease in the poverty rate. We saw that one major reason is that wages did not rise at the lower end of the wage distribution. If we want to address poverty, we must focus not just on policies that increase overall earnings but on policies that increase them at the lower end. Moreover, economic growth raises the level of income needed to avoid poverty. Although the official poverty rate is determined by an absolute income, the cost of participating in society goes up when other people are consuming more. The minimum expenditure needed to participate in society is probably somewhere between one-fourth and one-half of the median income, adjusted for household size. Recall that Victor Fuchs suggested the 50 percent value. The estimates based on the Gallup surveys discussed in chapter 2 suggested that a family of four needed an income of about $11,400 to avoid absolute poverty and that this goes up by about 27 cents for every dollar increase in the incomes of married-couple families. That makes the income currently needed for a family of four to avoid poverty about $29,000, which is about 42 percent of the median income for a household with four people. If all incomes go up at the same rate, many of those below this cutoff will continue to be poor, in reality if not by the current official measure. If $29,000 seems too high to you, remember that the upper limit of the Earned Income Tax Credit (EITC) for a married couple with two children is about $35,000. Addressing poverty in the United States fundamentally means addressing the high level of income inequality in this country. Americans place a high value on work. They are reluctant to help those who can work but do not. For those who can work, the re- ward should be an income that places them in a category that allows them to escape poverty in the sense of participating in society. We should have a set of policies that ensures that workers will have the financial means to fully participate in society.

4. Alan Schwarz, in The Numbers Game: Baseball’s Lifelong Fascination with Statistics (New York: St. Martin’s Press, 2004), describes the delayed influence of Earnshaw Cook, Percentage Baseball (Cam- bridge, MA: MIT Press, 1966).

Conclusion: An Agenda? 381 1. We have seen that the minimum wage is a blunt tool for addressing poverty. Nevertheless, the minimum wage is very low by historical standards and can be raised with only a modest impact on employment. As of this writing, the minimum wage would have to rise to $6.45 in order to reach the same level relative to average hourly earnings that it reached in September 1997. A more reasonable goal is to raise the minimum wage to a level comparable to that of the 1970s before the dramatic increase in inequality that began then. If we set the minimum wage at 45 percent of average hourly earnings (still low relative to the 1950s and 1960s), it would be about $7.50. A married couple with two children working (jointly) sixty hours per week for fifty weeks each year would earn $22,500 and receive an EITC of $2,562. In order for this family to receive $29,000, we would have to set the minimum wage at about 55 percent of average hourly earnings. This is less than the peak value of 57 percent in January 1950,5 but it is still high by historical standards. I am reluctant to advocate increasing the minimum wage further without more evidence of its impact on employment at higher levels than we have experienced recently. 2. The high cost of medical insurance is a major problem for employers and low-wage workers. Often employers either do not offer health insurance to low-wage workers or avoid hiring them. Current law establishes low and sharp income cutoffs for Medicaid eligibility, although states have the flexibility to be more generous. Workers are better off keeping their earnings below the cutoff unless they can earn substantially more than the cutoff. There are legitimate concerns that when government pays for the cost of health care, physicians and patients incur costs that exceed their benefit. But for the most part, government can avoid these problems by offering insurance that covers catastrophic costs and/or has significant copayments. 3. Training programs do work, but it is foolish to think that training is much cheaper than traditional education. Returns to good training programs seem to be in line with the estimated returns to education. Not all students do well in traditional education. Training provides a valuable second chance for adults who dropped out of school or left school with few skills. Job Corps shows that training works for the young as well, but it may be that training for the young must be even more intensive than for adults and that the residential aspect of Job Corps is important for this group. The sharp dichotomy between “work-first” and “education-first” approaches is equally foolish. Some people can work and advance if they only get a start. Others are so clearly lacking in skills that they are unlikely ever to move beyond a minimum-wage job without some training.

5. Charles Brown, “Minimum Wages, Employment, and the Distribution of Income,” in Orley Ashenfelter and David Card, eds., Handbook of Labor Economics, Volume 3B (New York and Amster- dam: Elsevier, North-Holland, 1999), 2102–63.

382 Chapter 15 3. Family and Early Childhood Programs The impact of childhood family structure on adult outcomes has been greatly exaggerated, as have the effects of teenage motherhood on the mothers. Still, there is little doubt that growing up in a disadvantaged environment is, in fact, a disadvantage. Our greatest challenge may be to find ways to address this disadvantage. As we have seen, there are promising results from some interventions that begin in pregnancy. These are relatively small programs, but they have been tested on an experimental basis, and there have been benefits to educating parents not only in obvious areas, such as nutrition, but also in general parenting skills, such as how to deal with a crying baby. It is always difficult to know with small programs whether we are just measuring the terrific skills of one very dynamic leader, but there is enough evidence supporting parenting programs that we ought to be trying and testing them on a larger scale. Beyond that, parent education should be combined with quality preschool. In the Brookline Early Education Project (BEEP) program discussed earlier, all participants received high-quality preschool. The varied treatment was the extent of parent education. The BEEP study suggests that parent education programs have a powerful effect in the presence of high-quality preschool. In contrast, in the Abecedarian Program all participants received the “social work treatment” in the form of professional assistance regarding how to take care of their children. But not everyone received the high-quality, full-day preschool program. And again, among those people who are already receiving the parent education, we see a large positive affect from preschool. Similarly, the High/Scope Perry Preschool Program was actually a dual program. It was a part-day preschool program, but there were also frequent home visits by teachers. I have cautioned that the results of Abecedarian, High/Scope Perry, and BEEP programs and a few other well-known early intervention programs should be treated with skepticism. Journals are more likely to publish projects that show positive results, and granting agencies are more likely to support them. We do not know how many other experimental programs failed to produce gains for their participants. And recent work by Michael Anderson has found that some of the programs reporting positive results benefited only girls, not boys.6 However, the preliminary results from the Early Head Start evaluation are very promising, supporting the case for more expanded early intervention. We are finding large cognitive effects, in the short term at least, for the children who participate in Early Head Start. Moreover, better and more universal early child care would make it easier for low- income women to work. If, as a society, we take the position that these women should work, we need to facilitate employment by providing child care. Jonah Gelbach examined the relation among kindergarten enrollment, employment, and welfare receipt

6. Michael Anderson, “Uncovering Gender Differences in the Effects of Early Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects” (unpublished paper, MIT, Cambridge MA, 2005).

Conclusion: An Agenda? 383 among single mothers whose youngest child was five years old.7 The age at which children are eligible to enter kindergarten differs by locality, but a child born in the first half of the year is much more likely to be eligible for kindergarten in the year that he turns five than is a child born in the last quarter of the year. Using the sharp difference in kindergarten eligibility, Gelbach found that when kindergarten was available, the single mothers whose youngest child was five years old had about 125 hours more work force participation when the child was in kindergarten and their welfare receipt was about 10 percent lower. This diminishes but does not eliminate the cost to the public of early childhood education. Combined with the promising evidence that early childhood programs have long-run effects on children and the evidence that poor children are already substantially behind their more advantaged peers when they enter kindergarten, this makes a compelling case for expanding these programs. At the very least, we should have universal public kindergarten in the United States. It is more problematic to expand public education to the prekindergarten level. The reason is that many middle-class parents currently pay for prekindergarten child care and education. If we provide free public education for 3- and 4-year-olds, the cost will be high and much of the cost will subsidize middle-class households. An alternative proposal is to give low-income households vouchers for child care, but we have seen that using vouchers on a large scale can be problematic. The most realistic approach is probably to phase in public education for 3- and 4-year-olds by allowing public school districts to charge for this education on a sliding scale and to eliminate the fees over a long period. Such an approach would replace Head Start with a program in which teachers would be universally qualified and would better integrate preschool and early elementary programs. Child care and early childhood education address the needs of low-income working mothers as long as their child is healthy. We have seen that American labor law is unusually unfriendly to working parents. The Family Medical Leave Act (FMLA) is a first step toward ensuring that parents have some ability to miss work in order to care for a sick child, but by providing unpaid leave and the right to return to the original job, it may be of more value to married middle-class women than to low-wage single mothers. Given the evidence that women with children earn much less than other women, helping middle-class women with children is worthwhile, but it does not address poverty. Part of the difficulty with the FMLA is that it addresses only employees who have been with a firm for at least twelve months and have worked at least 1,250 hours. Turnover is often high in low-wage jobs, so low-wage women may not have worked with a firm long enough to be eligible. Moreover, women who are concerned about FMLA eligibility may be subject to “job lock” because if they change jobs, they will not be able to take time off if their child is ill.

7. Jonah B. Gelbach, “Public Schooling for Young Children and Maternal Labor Supply,” Amer- ican Economic Review 92 (March 2002): 307–22.

384 Chapter 15 4. Education There is little doubt that one of the reasons we need more and better training programs is that some of our schools are failing, particularly in high-poverty areas. But the current education reform fads are unlikely to solve the problem. Putting too much weight on standardized tests just means that schools put too many of their resources into preparing students for the tests. There is ample evidence that schools can improve performance on tests, a good deal less that doing so ends up making graduates more attractive to employers. In theory, testing may help administrators and officials detect underperforming schools. In practice, the high-stakes tests rarely tell them what they do not already know (and, when they do, it is likely that the new information is wrong). Evaluating schools is expensive. The accreditation process for a large public high school costs tens of thousands of dollars. All schools should be required to go through an accreditation process. Principals and teachers should be held account- able for those elements under their control. Leadership matters in schools, as in other organizations. When schools are truly failing, we must be prepared to pay the salary that excellent leaders can command. Moreover, we are beginning to accumulate evidence that African American students fall behind in the early years of elementary school. One of the leading explanations for the positive effect of Head Start on the educational outcomes of whites but not African Americans is that the benefits for blacks are dissipated by low-quality elementary school education. This suggests that we should be focusing a lot of our efforts on what is going wrong in the early elementary years. But under the No Child Left Be- hind Act, testing begins in third grade. Schools need to be tracking outcomes much earlier to figure out why the achievement gap opens up in this period. There is little evidence that charter schools outperform regular public schools. If anything, the lack of experience of charter school leadership tends to make charter schools perform poorly in their first few years. The theory that in the long run charter schools will solve the problem of inner-city schools by creating more competition is inconsistent with the New Zealand experience. There is nevertheless a limited role for charter schools. To reiterate: leadership matters. The relative freedom of charter schools can attract innovative leaders (just as did the relative freedom of Community School District 4 in New York). But charter schools should be carefully selected to ensure that their leaders have the experience and skills to make them effective, and then only if there is a clear mechanism for evaluating the innovations and sharing information about effective strategies with other schools.

5. Addressing the Needs of High-Poverty Neighborhoods The Moving to Opportunity studies suggest that neighborhoods have less important peer effects than many of us would have expected, especially on the basis of the results of the Gautreaux Program. Still, high-poverty neighborhoods are problems. They are neighborhoods that most of us avoid if we can, and they are unsafe and unhealthy for those who cannot or do not avoid them.

Conclusion: An Agenda? 385 Most significant, these neighborhoods reflect the high degree of inequality and the racial divide in the United States. Redlining, the practice of not providing loans for businesses and mortgages in certain neighborhoods, arose because high-poverty and heavily black neighborhoods in the United States were perceived, probably correctly, as high-risk neighborhoods for lenders. We now require financial institutions to lend money in communities where they collect deposits, making redlining illegal. However, this attacks a symptom of economic segregation, not the cause. The high degree of economic segregation in the United States reflects the high level of income inequality. Job creation programs have lost credibility in the United States because programs like those provided under the Comprehensive Employment and Training Act seemed to be providing ineffective training. This overlooks the value of training programs as tagging devices. More important, it appears that, except for those who rise to leadership positions, drug-dealing gangs do not pay well. Jobs with relatively modest pay might provide a viable alternative to gangs for young men and women in high-poverty neighborhoods. Diminishing the power of gangs as employers would also lessen their importance as a source of protection.

6. Race Discrimination and Inequality The sections of this book on poverty and discrimination were largely separate, but the two are interrelated. We discussed the role of self-fulfilling expectations in generating labor market discrimination. To a significant extent, employers and other people hold very negative stereotypes of low-income, low-education African Americans. It is this group that suffers the greatest discrimination in the labor market. When people do not know (or cannot readily infer) the socioeconomic background of a black with whom they are interacting, the negative stereotype continues to influence their interaction. This exacerbates the tendency toward segregation and the social isolation of low- income blacks. The language of business is middle-class white American. I say this as a fact of life, not a moral statement. It could just as easily be Japanese, Spanish, or Ebonics. And indeed, in some ethnic enclaves it is difficult to do business if one does not know some language other than English. However, in the current situation in the United States, it is difficult to be successful in the labor market if one cannot converse in standard American and does not know the social rules of discourse used by middle-class white Americans. Middle-class African Americans are usually at ease with these rules. Often they are also at ease with a separate set of rules governing interaction among African Americans and can code-switch as necessary. In effect, they are bilingual. African Americans educated in segregated schools may not learn how (primarily white) employers expect them to act. They are unilingual, lacking the necessary knowledge of the dominant language. In interactions in both employment and other situations, they may display behaviors that give signals different from those they intend to send. In turn, they may misinter- pret the person with whom they are interacting and respond in an inappropriate way, thereby confirming the negative stereotype.

386 Chapter 15 Table 15.1 Effective Federal Tax Rates, 2001 and 2004

2001 2004

Lowest Quintile 5.4 5.2 Second Quintile 11.6 11.1 Middle Quintile 15.2 14.6 Fourth Quintile 19.3 18.5 Highest Quintile 26.8 23.8 Top 10 Percent 28.6 24.9 Top 5 Percent 30.1 25.6 Top 1 Percent 33.0 26.7

Source: Congressional Budget Office, Effective Federal Tax Rates under Current Law, 2001 to 2014 (Washington, DC: Congressional Budget Office, August 2004), table 2.

Consequently, low-skilled blacks find it particularly difficult to enter the economic mainstream. Inequality promotes economic and racial segregation. And racial and economic segregation promote inequality.

7. Addressing Inequality As we have seen, inequality has increased dramatically in the United States over the past thirty years. There are a number of ways that we can work to reduce inequality. The first is to raise the skill level of those at the low end of the skill distribution. Im- proving education and expanding long-term high-quality training programs can be important here. Earlier we discussed various strategies for reducing earnings inequality. We can reduce income inequality by moving toward a more progressive tax system. The Congressional Budget Office regularly calculates how much households in different parts of the income distribution pay in taxes as a percentage of their total income. Income includes both taxable and nontaxable sources of income and government monetary and in-kind transfers. Taxes include both income and payroll taxes (primarily social security and Medicaid taxes), excise taxes (such as cigarette, alcohol, and gas taxes), and corporate taxes. Table 15.1 shows the effective average tax rates for 2001 and 2004. Two points are obvious. Tax rates for all income groups fell over this period, but they fell much more rapidly for households with higher incomes. The bottom 20 percent of the income distribution saw its effective tax rate fall by 0.2 percentage point, for a tax cut of less than 4 percent. The top 1 percent of the income distribution saw its tax rate fall by 6.3 percentage points, for a tax cut of 19 percent. We could reduce the average tax rate on the lowest quintile of the income distribution to 0 percent by increasing the average tax rate on the top 1 percent to about 28.1

Conclusion: An Agenda? 387 percent, a tax increase of about 5.2 percent. Recall that the top group has an average income of roughly $1.5 million, while the bottom quintile has an average income of around $16,000. Raising the tax on the top quintile (with an average income of over $200,000) by 10 percent to 26.4 percent (still below its 2001 level) would allow us to reduce the effective tax rate on the bottom two quintiles to 0 percent. This might be achieved by further increasing the generosity of the EITC. There is plenty of room to increase the progressivity of the federal tax system without setting tax rates on upper incomes that are high by historical standards.

8. Health Care Our attempt to offset the increases in inequality of the past thirty years should not be limited to supporting the lower end of the skill distribution in the labor market or to the tax system. Although using the tax system to address earnings inequality is important, we must be cautious, because taxing income does generate deadweight loss. As we have seen, providing in-kind transfers can, in some cases, entail less deadweight loss than cash transfers would. These transfers also play an important tagging role, allowing us to target the poor more effectively and thus lowering the cost of poverty policy. Medical insurance is the most important in-kind benefit provided by government. As discussed earlier, the rising cost of medical insurance has wreaked havoc on the low- wage labor market. Companies have either found ways to avoid giving health benefits to low-wage workers or have avoided hiring them. This increases the need for public provision of medical insurance. The problem with the way that Medicaid is currently offered is that there are sharp income cutoffs below which households (or the children in a household) are eligible for Medicaid and above which they are not. This creates a very strong disincentive to work. The Medicaid subsidy should be phased out over some income range rather than eliminated at some arbitrary cutoff. Moreover, Medicaid is unusually high-quality medical insurance. Few recipients would choose health insurance of this quality if they received a cash payment instead. We argued earlier that in-kind transfers could reduce deadweight loss because people choose not to take the benefit, while cash transfers must be taxed away. As currently configured, this argument does not apply to Medicaid, because it is equal to or supe- rior to most private insurance. We should examine whether there are cost-effective ways to ensure that the poor have health insurance without mimicking the private insurance model. Perhaps Medicaid could be provided though a system similar to that of the Veterans Administration.

9. Concluding Remarks An individual is poor if he or she lacks sufficient financial resources to obtain adequate food, shelter, clothing, and medical care and to participate in society. Roughly half a million Americans are homeless. No one should belittle this number or minimize the hardship involved, but the number of homeless is small relative to the number of offi-

388 Chapter 15 cially poor. Roughly one-fourth of the officially poor experience hunger during the course of a year, and four in ten experience food insecurity. These numbers are not small, but it is still true that most of the officially poor are neither hungry nor homeless. The expansion of Medicaid means that poor children will be covered by health insurance and that a significant proportion of their families will be as well. We cannot yet say that poverty in the United States is solely about participating in society, but we are moving in that direction. I have argued earlier that many people who do not meet the official definition of poverty do struggle to meet the standard of full participation in society. Many of the near-poor (by the official definition) also lack health insurance coverage because they do not qualify for public programs. When we use this broader definition of poverty, it is clear that, while the deepest poverty is absolute, for most of the poor the twin issues are participation in society and inadequate access to medical care. As long as the United States is willing to have a highly unequal distribution of income, it will struggle with poverty, because the cost of participating in society will rise as the country becomes richer. To some degree, reducing inequality entails difficult trade-offs, because inequality can create strong incentives to work hard. Americans are probably the hardest-working people in the developed world. It is not difficult to ascribe this high level of effort to the huge potential rewards in a highly unequal society. But we must remember that not all effort is productive. In a winner-take-all society, the important thing is to be the winner, not to be good. We can make ourselves the best by making ourselves better or by making others worse. The corporate scandals of Enron and other corporations remind us that large rewards create strong incentives to reap those rewards, but that is not always beneficial to society. In highly unequal societies, people are forced to spend large sums of money to protect their property rather than on actual consumption. The well-to-do in many developing countries fear kidnapping and must avoid certain types of consumption and spend their money for protection in order to avoid being kidnapped. The high rate of incarceration in the United States and the high level of inequality are related. The situation is made worse by the correlation between race and income. The race riots of the 1960s reminded Americans of the dangers of what the Kerner Commission famously described as “moving toward two societies, one black, one white, separate and unequal.”8 In 1968, the date of the Kerner Commission report, the United States was in a period of declining inequality. Since then, that trend has stopped and reversed itself. The actual causes of increased inequality are complex. Undoubtedly, this inequality reflects some combination of changing technology, increased globalization, and changing institutions, including the decline of unions and the minimum wage. Some of these forces are beyond our control, but it is not beyond our control to offset them. If we wish to reduce poverty and discrimination, we must do so.

8. United States National Advisory Commission on Civil Disorders, Report of the National Ad- visory Commission on Civil Disorders (Washington, DC: U.S. Government Printing Office, 1968).

Conclusion: An Agenda? 389 This page intentionally left blank author index

Page numbers followed by f indicate figures, those followed by n indicate notes, and those followed by t indicate tables.

Abma, Joyce C., 156n Berger, Mark S., 133–34 Brown, Charles, 115n, 290, 382n Adams, Scott, 120n, 121 Bergmann, Barbara R., 357–58 Buddin, Richard, 234n Aigner, Dennis J., 275n Berlin, Gordon L., 135n, 135t, Burdekin, Richard C., 272n Akerlof, George A., 133, 155 248n Burke, Vee, 247n Allen, Joseph P., 187n Berman, Eli, 112n Burt, Martha, 47 Anderson, Elijah, 198, 205 Bertrand, Marianne, 294 Burtless, Gary, 52–53, 137 Anderson, Michael, 383 Besharov, Douglas, 174n Andrews, Margaret, 48n Bettinger, Eric P., 231, 232t, 233, Cain, Glen G., 6, 7n, 275n Angrist, Joshua D., 16n, 149, 325 234 Cameron, Stephen V., 219n Antonovics, Kate, 348n Betts, Julian R., 214n Campbell, Frances A., 176n Anwar, Shamena, 349n Bhattacharya, Jayanta, 51, 183 Cancian, Maria, 143 Arciadiacono, Peter, 348n Biddle, Jeff E., 364 Cao, Zhun, 235n Aronson, Joshua, 322 Bishop, John H., 219 Card, David, 116, 117n, 288–89, Ashenfelter, Orley C., 116n, 126 Bitler, Marianne P., 174–75, 325n, 326n Autor, David H., 64, 111, 118n 258n, 259n Carlson, Steven, 48n Ayres, Ian, 348n, 351 Black, Dan A., 133–34, 273n, Case, Karl E., 224–25 301n, 367 Casper, Lynne M., 151n Bachrach, Christine A., 144n, Black, Sandra E., 224 Cave, George, 131n 151n, 155n Blank, Rebecca M., 53–54, 88, Chan, Jimmy, 330–31n Bachu, Amara, 153n 98, 157n, 244, 245n, 247n, Chandra, Amitabh, 291t Bane, Mary Jo, 45, 55, 55t, 248n Clark, Kim B., 303n 197–98 Blau, Francine, 147, 370 Coate, Stephen, 277n Bank, Roy J., 360–61, 362t Blinder, Alan S., 88, 98 Cohen, Philip N., 151n Barnow, Burt S., 131n Bloom, Howard S., 127f, 127n, Commission on Behavioral and Barsky, Robert, 27n 128t Social Sciences and Education, Bartik, Timothy, 207n, 208n Bonczar, Thomas P., 306n 174n, 183n Beck, Allen J., 306t Boozer, Michael A., 327n Cook, Earnshaw, 381 Becker, Gary S., 212n, 269, 297, Borjas, George J., 114, 251 Cook, Philip J., 323n 342 Bos, Hans, 131n Cookson, Peter W., Jr., 226n Becker, Howard S., 281n Bound, John, 112n Corcoran, Mary, 173 Bell, Duran, 350n Bradbury, Katharine L., 224–25 Costrell, Robert M., 214n Bell, Wendell, 335n Brien, Michael, 375 Crane, John, 148–49 Berg, Ivar E., 213n Brooks-Gunn, Jeanne, 173n Crouse, James, 321n

391 Cullen, Julie Berry, 227–28 Fryer, Roland G., 298, 320, 321n, Holmstrom, Bengt, 218n Culter, David, 88n 323n, 324, 330n Holzer, Harry J., 305, 311n, 312 Cunningham, James, 375 Fuchs, Victor R., 12n, 35, 381 Hotz, V. Joseph, 69n, 75–78, 76t, Currie, Janet, 174–75, 178–80, 134, 161n, 165, 168n 179t, 180t, 183, 192 Gallup, George H., 36n Howell, William G., 229–30 Cutler, David M., 192, 336n, Garces, Eliana, 180t Hoxby, Caroline M., 223–24, 233 337n, 338t, 339n, 341n, 343, Garner, Thesia, 51n Hoynes, Hilary W., 258n 344 Gee, Michael, 380n Hoyt, William E., 229n Gelbach, Jonah B., 258n, 383–84 Dalaker, Joseph, 52t Gennetian, Lisa, 160n, 161, 173n Idson, Todd L., 272n Davis, James A., 199 George, Henry, 2 Imbens, Guido W., 72n, 134 DeLair, Michelle, 346n Germanis, Peter, 174n Institute of Medicine, 348n DeNavas-Walt, Carmen, 43n Geronimus, Arline T., 162–65, Internal Revenue Service, 69n Denton, Nancy A., 337, 345 164t Dickens, William T., 273n, 318, Glaeser, Edward L., 335n, 336n, Jacob, Brian A., 15–16, 17, 204, 319, 331n 337n, 338t, 339n, 341n, 343, 209, 227–28, 239 Dills, Angela K., 238 344 Jargowsky, Paul A., 45, 197–98, Dinardo, John, 118n Glazer, Jacob, 235n 199t, 209n Donohue, John J., III, 289n, Glazerman, Steven, 129n Jastrzab, JoAnn, 132n 293n, 328n Gobillon, Laurent, 344n Jekielek, Susan, 186 Doolittle, Fred, 131n Goldberg, Pinelopi K., 348 Jencks, Christopher, 47, 48, 322n Doyle, Patricia, 51n Goldin, Claudia, 355n, 356, 360, Jensen, Arthur R., 317–18, 319, Doyle, Roy P., 13n 362–64, 363t, 365n, 376 320 Duggan, Mark G., 64 Goorman, Jennifer Appleton, Johnson, Amy, 186n Duncan, Beverly, 336n 188n Johnson, David, 51n Duncan, Greg J., 53, 54, 173n, Gottschalk, Peter, 245n Johnson, William R., 300, 301n 174 Gould, Stephen Jay, 317n Juhn, Chinhui, 102 Duncan, Otis Dudley, 336n Greenberg, David H., 122n Griliches, Zvi, 112n Kahn, Lawrence M., 272n, 370 Eaton, Susan E., 200–201 Grissmer, David W., 221 Kahn, Shulamit, 117, 374–75 Eccles, Jacquelynne, 188n Groeneveld, Lyle P., 7n Kain, John F., 339n, 344 Edin, Kathryn, 68, 148 Grogger, Jeffrey, 249, 250n, 253n, Kane, Thomas J., 238, 331n Ellwood, David, 55, 55t, 148–49 256n, 257n, 259n Karoly, Lynn A., 182, 253n, 256n, England, Paula, 357, 360 Grossman, Jean Baldwin, 185n, 257n, 259n Erickson, Frederick, 274n 186n Katz, Lawrence F., 88n, 118n, Eyster, Erik, 330–31n Gruber, Jonathan, 192 137–38, 139t, 203n Guryan, Jonathan, 329 Katz, Michael L., 155 Falk, Gene, 247n Kearney, Melissa S., 118n Fang, Hanming, 349n Hahn, Andrew, 188n, 189t Keeley, Michael C., 7n Farkas, George, 323 Haider, Steven, 183 Kemper, Peter, 131n Farley, Reynolds, 339n Hair, Elizabeth, 186 Kerr, Steven, 218n Feenstra, Robert, 113n Halle, Tamara, 175n Kirschenman, Joleen, 301 Ferguson, Ronald F., 322, 328n Ham, John C., 191n Klein, Stephen P., 220 Filvaroff, David B., 372n Hamermesh, Daniel S., 364 Klerman, Jacob A., 134, 253n, Fisher, Gordon M., 37n, 38 Hannan, Michael T., 7n 256n, 257n, 259n Fisher, Peter S., 207n Hanson, Gordon, 113n Kling, Jeffrey R., 203n, 307n Fiske, Edward B., 235n Hanushek, Eric A., 211n Knowles, John, 349n Fix, Michael, 295n Harrison, Page M., 306t Knox, Cynthia, 159n, 160t Flynn, James R., 318, 319 Haskins, Ron, 247n, 250n Knox, Virginia, 160n Fortin, Nicole M., 114n, 118n Hauser, Robert M., 18n Knutson, David, 16n Fraker, Thomas M., 40 Haveman, Robert, 88 Korenman, Sanders, 162–65, 164t Frank, Richard, 235n Haviland, Amelia, 301n Kotlowitz, Alex, 204 Frank Porter Graham Child Heckman, James J., 219n, 289n, Kowaleski-Jones, Lori, 174 Development Center, 176n 293n, 295, 296t, 297, 328n Kremer, Michael, 101n Freeman, Richard B., 89–90 Higgs, Robert, 307–8 Krueger, Alan B., 12n, 16n, 116, Friedlander, Daniel, 122n Hogan, Lyn A., 134n 117n, 230n, 288–89, 325n, Friedman, Milton, 4–5 Hollister, Robinson G., Jr., 131n 326n, 327n

392 Author Index Krysan, Maria, 339n Mills, Robert J., 43n Reback, Randall, 229 Ku, Leighton, 174n Mincer, Jacob, 355 Reber, Sarah J., 329 Kuhn, Peter, 358 Moffitt, Robert A., 40, 157, 158, Redd, Zakia, 184–85, 188n 255n, 256n Reed, Deborah, 143 Ladd, Helen F., 235n Moore, K., 161n Ricardo, David, 1 Lalonde, Robert J., 125 Moore, Kristin A., 186 Robins, Philip K., 7n, 122n Landes, Elizabeth M., 115n Moran, Rebecca, 319n Rockoff, Jonah E., 233 Lang, Kevin, 117, 169t, 170t, Morris, Pamela A., 173n Roderick, Melissa, 14–15, 16, 17 172n, 213n, 273n, 299t, 300n, Moss, Phillip, 301n Rodriguez-Planas, Nuria, 188n 325, 350n Mullahy, John, 191n Romer, David, 380, 381 Lee, Chungmei, 327n Mullainathan, Sendhil, 294 Rose, David L., 373n Lefgren, Lars, 15–16, 17 Mullin, Charles, 69n, 75–78, 76t Rosen, Asa, 273n Lein, Laura, 68 Munnell, Alicia H., 345n Rosenbaum, James E., 201, 202t Lemieux, Thomas, 114n, 118n Murnane, Richard J., 111 Ross, Stephen L., 346 Leonard, Jonathan S., 310, 311n Murphy, Kevin M., 102 Rouse, Cecilia E., 230, 362–64, Leonhardt, David, 380n Murray, Charles A., 86, 156, 157, 363t Levine, Kara, 63–65, 64t, 69 250n Rubin, Donald B., 72n Levitt, Steven D., 205–6, 227–28, 239, 298, 320, 321n, 348n Nagaoka, Jenny, 14–15, 16, 17 Sacerdote, Bruce, 72n Levy, Frank, 111 Neal, Derek A., 300, 301n, 302 Saez, Emmanuel, 109n Liebman, Jeffrey, 69n, 203n Neckerman, Kathryn M., 301 Sampson, Amy, 183 Link, Bruce, 46n Neisser, Ulric, 317n Sanders, Seth G., 161n, 165, Lollock, Lisa, 113n Neumark, David, 117–18, 120n, 168n, 301n, 367 Loprest, Pamela, 256n, 257n 121, 297, 311n, 312, 360–61, Schirm, Allen, 188n Loury, Glenn C., 277n, 330n 362t Schmidt, Lucie, 254n Love, John M., 177n Nightingale, Demetra Smith, Schoeni, Robert, 245n Ludwig, Jens, 203n, 323n 122n Scholz, John Karl, 63–65, 64t, 69, Lundberg, Shelly J., 276–77 Noble, Holcomb B., 5n 75–78, 76t Lutkus, Anthony, 319n Nord, Mark, 48n Schwabish, Jonathan, 88 Lutterbeck, Deborah, 40n Schwarz, Alan, 381n Luttmer, Erzo R. P., 158 Offner, Paul, 247n, 250n Schweinhart, Lawrence, 181n, 193 Lutz, Byron F., 329 O’Hara, Brian J., 247n Selod, Harris, 344n O’Neill, June, 371t, 375 Sevak, Purvi, 254n Makar, Hoda R., 367 Oreopoulos, Phillip, 204 Sheppard, Nathaniel, 203n Malthus, Thomas, 1 Orfield, Gary, 325n, 326–27 Shonkoff, Jack P., 173n Mane, Ferran, 219 Shore-Sheppard, Lara D., 191n Manove, Michael, 273n, 299t, Pager, Devah, 307 Short, Kathleen, 51n 300n Palfrey, Judith S., 177n Shroff, Sonali M., 226n Manser, Marilyn, 40n Parijs, Philippe van, 12 Shultz, Jeffrey, 274n Marshall, Alfred, 1–2 Parker, Jonathan A., 27n Siegelman, Peter, 346, 347, 348n Martin, Joyce A., 144n Perie, Marianne, 319n Simmons, Tavia, 151n Massey, Douglas S., 337, 345 Persico, Nicola, 349n Slesnick, Daniel T., 80n Maxfield, Myles, 188n, 189t Peters, Alan H., 207n Smeeding, Timothy, 52–53 Mayer, Christopher J., 224–25 Peterson, Paul E., 229–30 Smith, Annetta C., 47n Mayer, Susan E., 16n, 173 Phelps, Edmund S., 275n Smith, Denise I., 47n Maynard, Rebecca A., 131n Phillips, Deborah A., 173n Smith, James P., 285, 286t, 288, McConnell, Sheena, 129n Phillips, Meredith, 321n, 322n 292–93, 370 McCrary, Justin, 313, 314 Pikerty, Thomas, 109n Smith, Jeffrey, 133–34 McElroy, Susan Williams, 161n, Polachek, Solomon W., 355, Smith, Robert S., 116n 165, 168n 356n, 360n, 371t Smith, Robin, 346n McGuire, Thomas, 235n Poterba, James M., 12n Solon, Gary, 27n, 172n McManus, Walter S., 301n Powers, Elizabeth T., 88n Spence, Michael, 213n Meyers, Alan, 183 Proctor, Bernadette D., 43n, 52t Spiegelman, Robert G., 7n Michalopoulos, Charles, 249 Staiger, Douglas O., 238 Michelsen, Erik, 186n, 188n Ralph, John, 321n Startz, Richard, 276–77 Milgrom, Paul, 218n Ramey, Craig T., 176n Steele, Claude M., 322 Miller, Cynthia, 132n Raphael, Steven, 305 Stevens, Ann Huff, 55

Author Index 393 Stoll, Michael A., 305 Van der Veen, Robert, 12n Williams, Walter, 5n Struyk, Raymond J., 295n Van Nort, Kyle D., 360–61, Wilson, William Julius, 148, Summers, Lawrence H., 303n 362t 345 Szymanski, Stefan, 306 Venkatesh, Sudhir Alladi, 205–6 Wissoker, Douglas A., 7n Ventura, Stephanie J., 144n, 151n, Wolfe, Barbara L., 191n Taylor, Lowell J., 301n, 367 155n Wolfinger, Raymond E., 372n Texas Education Agency, 237n Vigdor, Jacob L., 335n, 336n, Wolkon, Shari, 327n Thomas, Duncan, 178–80, 179t, 337n, 338t, 339n, 341n, Wood, Robert G., 149 180t 343n Tiebout, Charles M., 221, 225 Yaeger, K. E., 7n Tierney, Joseph P., 185n Waldfogel, Jane, 364, 365, 376 Yellen, Janet L., 155 Tilly, Chris, 301n Waldfogel, Joel, 307n, 351 Yelowitz, Aaron, 191n Tobin, James, 5 Walsh, Randy, 348n Yinger, John, 346 Todd, Petra, 349n Ward, Michael, 370 Yuret, Tolga, 330n Torelli, Paul, 323n, 324 Wascher, William, 117–18 Toussaint, Cyril, 131n Watts, Harold W., 6 Zaff, Jonathan F., 186n Trejo, Stephen, 301n Weaver, R. Kent, 258n Zagorsky, Jay, 169t, 170t, 172n Tuma, Nancy B., 7n Weiman, David F., 307n Zargari, Ahmad, 122n Tumlin, Karen C., 251n Weinberg, Bruce A., 345n Zaslow, Martha J., 259n Turner, Margery Austin, 341n, Weiss, Andrew, 213n Zax, Jeffrey S., 344 342t Weitzman, Michael, 183 Zenou, Yves, 344n Welch, Finis, 288, 292–93, Zhu, Pei, 230n U.S. Census Bureau, 22n 309 Zimmer, Ron, 234n U.S. Department of Health and West, Richard W., 7n Zimmerman, David J., 172 Human Services, 65n, 66n Western, Bruce, 307n Zimmermann, Wendy, 251n

394 Author Index subject index

Page numbers followed by f indicate figures and those followed by t indicate tables.

Abecedarian Project, 175–76, 383 American Psychological Association, 317 abortions, 155, 156, 165, 166 American Youth Policy Forum, 188 absolute poverty, 35–36 Armed Forces Qualifying Test (AFQT): controlling academic achievement programs, 184–85, 188 for in wage comparisons, 300, 301; relation to adolescents. See teen mothers; youth nonemployment, 304, 305f; scores of children in AFDC. See Aid to Families with Dependent Children single-parent families, 168, 169t AFDC-UP (Aid to Families with Dependent Ashenfelter dip, 126, 127–28, 252 Children—Unemployed Parent), 158–59, 243, Asian men, wage differentials of, 301 252 assortative mating, 101–2, 150. See also marriages affirmative action: arguments for, 274; Civil Rights audit studies: of discrimination by taxi drivers, Act provision on, 311; in college admissions, 347; of discrimination in automobile sales, 330–32; debates on, 312, 315; distinction from 348; of employment discrimination, 294–97, equal employment opportunity, 312; effects of, 296t; of housing discrimination, 341–42, 342t; 279–80, 312–13; enforcement of, 311, 313, 372; limitations of, 297; of mortgage preapproval Executive Order 11246, 292, 293, 311; by federal process, 346; of sex discrimination, 360–64, contractors, 292, 311, 372; in private sector, 362t 311–13; in public sector, 313–14; by sex, 372; use automobile sales, discrimination in, 347–48, 351–52 by employers, 312 AFQT. See Armed Forces Qualifying Test bail setting, discrimination in, 351 African Americans. See blacks banks. See mortgages Aid to Dependent Children, 63, 66 barrios, 198. See also neighborhoods, high-poverty Aid to Families with Dependent Children (AFDC): baseball teams, 381 availability to two-parent families, 158–59, 243, basketball teams, 271–72 252; benefit levels of, 99, 100f; disincentives for BBBS. See Big Brothers Big Sisters working, 66–68, 244, 248; effects on poverty, 99; beauty, relation to earnings, 364 history of, 66, 252; marital status of recipients, BEEP. See Brookline Early Education Project 159–60, 160t, 161; replacement by TANF, 246; Belichick, Bill, 380, 381 spending on, 63, 64t; trade-off between hours of Big Brothers Big Sisters (BBBS), 185–86 work and income, 67f, 67–68; variations in benefits birth rates: to married women, 145, 146, 146f; to across states, 157; welfare-to-work programs, single mothers, 145, 146f, 251; to teen mothers, 122–23, 134–35. See also welfare programs 161, 162f Aid to Families with Dependent Children— black men: educational attainment of, 148–49, Unemployed Parent (AFDC-UP), 158–59, 243, 287–88, 288f; employment rates of, 148–49, 290, 252 291t, 303, 304t; incarceration rates of, 306t,

395 black men (continued ) rates, 43, 51; racial categories, 265; workers 306–7; marriage rates of, 148–49; without earning less than minimum wage, 116 earnings, 289–92, 290f, 300–301 census tracts, 335 blacks: civil rights movement, 287, 326, 334; Center for Employment Training (CET), 132 educational attainment of, 300, 301, 329; CET. See Center for Employment Training education quality, 287, 288–89, 385; effects of CETA. See Comprehensive Employment and Training names, 294–95, 298; effects of segregated Act neighborhoods, 342–44; Head Start participants, charter schools: administration of, 225, 231; effects 178, 179f, 180, 180f, 181; high-poverty on regular public schools, 234; in New Zealand, neighborhoods of, 198; hiring officers, 304–5; 236, 385; objectives of, 231; performance of, labor force participation rates of, 287, 289–92; 233–34, 385; student performance in, 231–33, language and communication, 273–74, 343, 386; 232t, 240; variation among, 233–34 lead levels in blood of children, 318; marriage rates Chicago: Gautreaux Program, 201–2, 202t; public of, 148–49; middle-class, 344, 386; migration to housing projects, 203, 204 northern cities, 287, 289, 337; police officers, Chicago Public Schools (CPS): charter schools, 233; 314; poverty rates of, 43, 44t, 85, 85f; relative retention in grade studies in, 14–16; school choice wage growth of, 286–87, 288; skill levels of program, 227–28 workers, 286, 287; social pressure and academic child care: cost of, 364; federal block grants for, 250; achievement, 323–24, 332; stereotypes of, 301; increasing access to, 383–84; programs, 176. See teachers, 322, 323; test score gap, 317–25; welfare also early childhood education recipients, 158. See also discrimination; racial child-only families, 258 profiling; segregation; wage differentials, black- children: civic engagement programs for, 186, 188; white cognitive development of, 318–19; effects of black women: educational attainment of, 288, 289f, family structure, 168–72, 169t, 170t; effects of 301–2, 302t; employment rates of, 285, 303, parental death, 171–72; effects of parenting 304t; single motherhood rates of, 157; teen practices, 173; effects of welfare reform, 173, 258, mothers, 162, 163t; unmarried, 148 259; effects on parents’ earnings, 364–66, 376; Board of Education of Oklahoma City v. Dowell, 329 food insecurity of, 49–50; health insurance for, Boston: METCO program, 200–201, 325; school 190–92; immunizations of, 175, 178, 179; desegregation in region, 200–201, 325 influences of public housing, 204; in-kind transfers Brookline Early Education Project (BEEP), 176–77, targeted to, 79; lead levels in blood of, 318; 383 mentoring programs for, 185–86, 188; Brown v. Board of Education, 326 neighborhood effects on, 197, 201–2; nutrition Brown v. Board of Education II, 326 programs for, 174–75; outcomes of poor, 173; budget lines: for AFDC recipients, 67f, 67–68; effects poverty rates of, 43, 44t, 84f, 84–85; programs for of earned income tax credit, 70–72, 71f school-age, 182–90; sibling comparisons, 171, Bush, George H. W., 365 179–80; transmission of poverty to, 172–74; busing. See desegregation, school welfare recipients, 258, 259. See also birth rates; Byrne, Jane, 203 education child support enforcement, 250 cab drivers. See taxi drivers civic engagement programs, 186, 188 Cabrini-Green housing project, 203 Civil Rights Act of 1964: affirmative action allowed California: charter schools, 234; Greater Avenues to by, 311; employment discrimination prohibited by, Independence program, 134; San Jose comparable 308; provisions of, 292, 308, 326, 334, 346; worth agreement, 374–75; welfare-to-work school desegregation provision of, 326; sex experiments, 75–78 discrimination covered by, 372; social impact of, cash transfer programs: advantages of, 78; comparison 315; Title II, 292, 334; Title III, 292, 334; Title to in-kind transfers, 78–81; effects on poverty IV, 326; Title VI, 326; Title VII, 292, 372 rates, 87, 98–99, 100f; efficiency of, 80; Civil Rights Act of 1991, 308–9 supplemental security income, 64t, 64–65, 68, civil rights legislation, 292, 293 253–54; taxes on income from, 80. See also welfare civil rights movement, 287, 326, 334 programs class. See economic status causality, distinction from correlation, 10, 27 Clinton, Bill, 102, 245, 365. See also welfare reform CDCs. See community development corporations cognitive ability: effects of early childhood education, Census Bureau, U.S.: homelessness estimates by, 176, 177, 178, 182; effects of home visit 46–47; POSSLQ couples data, 151, 152f; poverty programs, 175; environmental influences on,

396 Subject Index 317–19; as factor in black-white wage differentials, desegregation, of public accommodations, 334. See 300, 301, 314–15, 317; genetic influences on, also segregation 317–18; influence on environment, 318–19. desegregation, school: across districts, 200–201, See also IQ scores; testing 325; in Boston, 200–201, 325; Brown v. Board cognitive ability tests. See Armed Forces Qualifying of Education, 326; effects on black students, Test 329; effects on wage differentials, 328–29; cohabitation: increase in, 150–52, 152f; measurement enforcement of, 326–27, 329; resegregation issues, 152; rates for women, 152, 153f following, 327–28, 329; resistance to busing communication: nonverbal, 273; problems among plans, 327; school choice programs, 225; in south, social groups, 273–74; between races, 273, 343, 326–27 386 developing countries, poverty in, 31–32 community development corporations (CDCs), 208 differences-in-differences approach, 75 community development policies, 206–8, 386 disability insurance: eligibility rules of, 64, 253; Community Reinvestment Act (CRA) of 1977, spending on, 64, 64t 206–7 disabled individuals, proportion of poor, 90 Community School District 4 (East Harlem), 226–27 discrimination: definitions of, 265–68; distinction community service programs, 186 from prejudice, 265–66; distinction from comparable worth, 373–75 segregation, 268–69; in justice system, 349–51; comprehensive community initiatives, 208 relation to poverty, 386–87; self-confirming Comprehensive Employment and Training Act expectations, 277–80; statistical, 274–75; theories (CETA) of 1973, 121–22, 129 of, 265, 269–80, 281; transaction costs models, concentrated poverty. See neighborhoods, high- 273–74, 343; on The Weakest Link, 348–49. See poverty also sex discrimination confidence intervals, 22–23 discrimination, in customer markets: audit studies of, Consumer Price Index, 37, 39 346, 347, 348; in automobile sales, 347–48, consumer sovereignty, 78, 80 351–52; behavior changed by, 351; in credit contraception, 155–56 markets, 206, 334, 345–46, 351; in health care, correlation, distinction from causality, 10, 27 348; imperfect information in, 334–35; laws CPS. See Chicago Public Schools; Current Population prohibiting, 346; legal cases on, 346–47; by taxi Survey drivers, 347 CRA. See Community Reinvestment Act discrimination, in education: history of, 325–26; credentialism, 213 reverse, 330, 332; segregated schools, 325–26. credit markets, discrimination in, 206, 334, 345–46, See also desegregation, school 351 discrimination, in labor market: audit studies of, crime: drug dealing by gangs, 205–6; effects of Head 294–97, 296t; behavior changed by, 297, 304, Start, 180t; effects of High/Scope Perry Preschool, 309; by customers, 271–72; debates on, 315; 182; effects of Job Corps, 130; effects of Moving definitions of, 267–68; difficulty of measuring, to Opportunity program, 203; effects of police 293–94, 299; disparate impact cases, 308, 314, affirmative action programs, 314; effects of 315; disparate treatment cases, 308; effects of preschool on adult involvement in, 180t, 181, 182; names, 294–95, 298; by employees, 270–71; by in high-poverty neighborhoods, 197, 200, 209. See employers, 267–68, 269–70, 297; employment also imprisonment; justice system differentials, 303–5; evidence on current, Cumming v. Richmond County Board of Education, 293–307; in hiring, 295–97, 296t; in imperfect 325 labor markets, 272–73, 304; intent and, 308, Current Population Survey (CPS): Annual 309, 310–11; in interview callbacks, 294–95, Demographic Supplement, 42–43, 61; description 298; job matching and productivity, 275–77, of, 61; food insecurity supplement, 48–50, 61; 276t; laws prohibiting, 292, 293, 308; legal POSSLQ couples data, 151, 152, 152f; poverty cases on, 308–9; legal tests for, 307–11; in perfect rate calculations, 42–43; sample, 290 labor markets, 269–72, 297; reduction in, 287; customer markets. See discrimination, in customer statistical significance and small numbers defense, markets; housing 310–11. See also occupational segregation; sex customers, prejudice of, 271–72 discrimination; wage differentials dissimilarity index, 336–37, 338t Denny’s, 346 distributions: mean and standard deviation, 20, 21; Denver Income Maintenance Experiment. See normal, 20–21, 331 SIME/DIME drugs, illegal, 182, 205–6, 349–50

Subject Index 397 early childhood education, 176–82, 332, 383–84 standards movement, 18. See also charter schools; Early Head Start, 177, 178, 322, 332, 383 school choice programs; testing; vouchers earned income tax credit (EITC): comparison to Eisenstadt v. Baird, 155 employer subsidies, 136; effects on family income, EITC. See earned income tax credit 65, 69; eligibility for, 69–70; evidence of effects elderly: poverty rates of, 43, 44t, 45, 84, 84f; of, 74–78, 76t; expenditures, 64t, 65; false claims proportion of poor, 90; social security income of, for, 69; as incentive for working, 246; income 45 effect of, 72–78; influence on labor supply, employees, prejudice of, 270–71 70–78; mechanics of, 69; phase-out range of, 70, employers: affirmative action programs of, 312; black 80; rates of, 70, 74–75, 388; substitution effect hiring officers, 304–5; discrimination by, 267–68, of, 72–78 269–70; health insurance provided by, 191; earnings. See incomes; wages migration to suburbs, 344–45; prejudice of, economic growth: failure to reduce poverty, 89–90, 269–70, 297; screening of workers by, 136–37; 102–3, 381; relation to poverty, 1–2, 87–90, 102, stereotypes of black workers, 301 381; relation to welfare caseloads, 252–53, 254f, employer subsidies: effects on employment, 136, 256, 260 137f, 138, 139f, 207–8; effects on equilibrium economic status: intergenerational correlation of, wages, 136, 137f; in empowerment zones and 172–74; middle-class blacks, 344, 386; mobility, enterprise communities, 207; stigmatization of 172 workers, 137; Targeted Jobs Tax Credit, 137–40; education: academic achievement programs, 184–85, Welfare-to-Work Tax Credit, 137 188; accreditation of schools, 385; affirmative employment: effects of employer subsidies, 136, 137f, action in, 330–32; competition among 138, 139f, 207–8; effects of minimum wage laws, communities, 221–25; decentralized systems, 117–18; gangs as source of, 205–6; in poor 221–23, 235–36; diversity of schools, 238–39, neighborhoods, 344–45, 386; spatial mismatch 327–28, 332; early childhood, 176–77, 332, hypothesis, 344–45. See also labor force 383–84; preschool, 177–82, 383, 384; quality of participation schools in high-poverty areas, 211; relation employment discrimination. See discrimination, in between student performance and school district labor market competition, 223–24; retention in grade issue, Empowerment Zones and Enterprise Communities 13–18; returns per pupil of, 123; school lunch and Act of 1993, 207 breakfast programs, 79, 183–84; school quality, enterprise communities, 207 221–25, 239–40; school quality and housing enterprise zones, state, 207–8 prices, 222, 224–25, 237, 238; school quality for Equal Employment Act of 1972, 372 blacks, 287, 288–89, 385; segregated schools, equal employment opportunity, 311, 312 325–26; of skilled workers, 109–10; social benefits Equal Employment Opportunity Commission, 292, of, 182; social pressure and academic achievement 372–73 of black students, 323–24, 332; spending on, Equal Pay Act of 1963, 307 123, 224; teachers’ prejudice, 322–23; tracking, ethnic groups: ghettos of, 197–98. See also Hispanics 57–58, 318, 328. See also desegregation, school; ethnographic studies, 9, 198, 200–201 discrimination, in education; testing; training Executive Order 11246, 292, 293, 311 educational attainment: assortative mating by, 101; expectations: self-confirming, 277–80, 322; of of black men, 148–49, 287–88, 288f; of black teachers, 322 women, 288, 289f, 301–2, 302t; effects of experiments: job training programs, 126–29; natural, desegregation, 329; effects of high-stakes testing, 7–8; with negative income tax, 6–7, 8; quasi-, 7–8; 215–17, 219; as factor in black-white wage randomized, 7; welfare reform, 7, 75–78, 159–61, differentials, 300, 301; of Hispanic women, 302, 173, 245; welfare-to-work programs, 75–78 302t; human capital model, 212; of immigrants, exports. See international trade 113–14; marriage and, 147, 356; of men, 355–56, extreme poverty, 45 356f, 368–69, 369f; relation to earnings, 212–13, 214–15, 368–70, 369f; relation to school quality, Fair Housing Act of 1968, 338 239–40; sorting model, 212–13; of teen mothers, Fair Labor Standards Act (FLSA), 115 162, 163t, 164, 165, 167; of white men, 287–88, families: home visit programs for, 175, 181–82; 288f; of white women, 288, 289f, 301–2, 302t; of median incomes of, 35–37, 36f; poverty rates of, women, 147, 288, 289f, 355–56, 356f, 368–69, 85, 86f; poverty threshold for, 37–39, 41, 381; 369f; of workers, 113–14 welfare dependence in, 244–45 education reform: effectiveness of, 240; No Child Left family backgrounds: as factor in test score gap, Behind Act, 18, 237, 385; programs, 211; 320–22; influence on cognitive ability, 317–19

398 Subject Index family leave policies, 364, 365–66, 384 governments. See federal government; local Family Medical Leave Act (FMLA) of 1993, 365–66, governments; states 384 Gratz v. Bollinger, 330 family structure: changes in, 143; child-only, 258; Greater Avenues to Independence (GAINS) program, effects on children, 168–72, 169t, 170t; poverty 134 rates by, 43–44, 44t; two-parent families, 158–59, Great Society programs, 86, 252 168–72, 243, 252. See also female-headed families; Griggs et al. v. Duke Power Company, 308 marriages; single-parent families gross domestic product (GDP): international trade as Family Support Act of 1988, 158, 243 proportion of, 111–12; relation to poverty rates, federal government: effectiveness of poverty programs, 88, 89f, 381. See also economic growth 63–65; empowerment zones and enterprise Grutter v. Bollinger, 330 communities, 207; housing segregation encouraged by, 338; job training programs, 121–24, 124t, 125; Hall v. Gus Construction Company, 373 poverty measure of, 37–39, 41, 42–43, 58, 83–86, Head Start: Early, 177, 178, 322, 332, 383; 84f; spending on poverty programs, 63–65, 64t; evaluations of, 178–80; goals of, 178; history of, welfare-to-work programs, 122–23. See also specific 178; outcomes of, 178–81, 179t, 180t, 332, 385; departments, agencies, and laws replacement for, 384 Federal Reserve Bank of Boston, 345–46 health care: discrimination in, 348; immunizations, female-headed families: effects of welfare reform, 175, 178, 179; National Health Service (U.K.), 80 255–57, 257f; poverty rates of, 44, 85, 86f, 100; health insurance: for children, 190–92; increasing proportion of, 89, 100, 101f, 106, 143. See also access to, 382, 388; provided by employers, 191; single mothers; single-parent families State Children’s Health Insurance Program, 191. firefighters, affirmative action in hiring, 313–14 See also Medicaid; Medicare Florida, Family Transition Program, 249 heterogeneity. See individual heterogeneity FLSA. See Fair Labor Standards Act high-poverty neighborhoods. See neighborhoods, FMLA. See Family Medical Leave Act high-poverty food, costs of, 37, 38. See also nutrition programs High/Scope Perry Preschool Project, 181–82, 193, food insecurity: of immigrants, 251; incidence of, 50, 383 389; measures of, 48–50 high-stakes testing. See testing Food Stamp Employment and Training Program, Hispanics: educational attainment of women, 302, 123 302t; high-poverty neighborhoods of, 198; food stamp program: political support for, 78–79; incarceration rates of, 306, 306t; poverty rates of, spending on, 65 43, 44t, 85, 85f; teen mothers, 162, 163t; wage food stamp recipients: income sources of, 68; value of differentials of, 299, 299t, 301 food stamps to, 39–40 homelessness: definitions of, 46, 47; estimates of, football teams, 380 46–48, 388–89 foster children, 258 home visit programs, 175, 181–82 Friedman, Milton, 4–5 homosexuals. See gays; lesbians Hopwood v. Texas, 330 GAINS. See Greater Avenues to Independence household composition: cohabiting couples, 150–52; program poverty rates by, 43–44; relation to poverty rates, Gallup polls, 36, 38, 381 143. See also family structure gangs, 205–6 housing: affordable, 208; regional cost differences, 52. Gautreaux case, 201, 338 See also public housing projects Gautreaux Program, 201–2, 202t Housing and Urban Development, Department of, gays: discrimination against, 366–67; earnings of, 201, 202 366–67 housing discrimination: audit studies of, 341–42, GDP. See gross domestic product 342t; in lending, 206, 345–46; redlining practices, gender. See men; occupational segregation; sex 206, 386; segregation caused by, 334, 338–39 discrimination; women housing prices: effects of racism, 339, 341; effects of George, Henry, 2 school choice programs, 229; relation to property ghettos, 197–98, 337. See also neighborhoods, high- tax levels, 222, 223, 224–25; relation to school poverty quality, 222, 224–25, 237, 238 Government Accounting Office, 123 housing programs: federal spending on, 64t, 65; government contractors: equal employment and Gautreaux, 201–2, 202t; Moving to Opportunity, affirmative action requirements of, 292, 311; 202–3; segregation influenced by, 338. See also minimum wage laws and, 120 housing subsidies; public housing projects

Subject Index 399 housing segregation: discrimination as cause of, income tax: increasing progressivity of, 387–88; 334, 338–39; effects of, 342–44; effects of fair negative, 4–7, 8, 12; rates, 387t, 387–88. See also housing laws, 338; explanations of persistence, earned income tax credit 338–41; in high-poverty neighborhoods, 345; individual heterogeneity: as explanation of long-term influence of federal programs on, 338; measures poverty, 56–58, 57t; as explanation of persistent of, 335–37, 338t; persistence of, 338–41, 352; welfare use, 244 in public housing, 338; relation to discrimination, inequality. See income inequality; wage inequality 268–69; restrictive covenants, 337–38; role of inflation: Consumer Price Index, 37, 39; relation to preferences in, 339–41; spatial mismatch poverty rates, 88 hypothesis, 344–45; in twentieth century, in-kind transfers: advantages of, 78–80; comparison 337–38 to cash transfers, 78–81; efficiency of, 80–81; housing subsidies: Gautreaux Program, 201–2; levels federal spending on, 64t, 65; means-testing for, 80; of, 81; Moving to Opportunity program, 202–3; policy shift toward, 63, 65, 68, 81; political policy shift toward, 204; political support for, 79; support for, 78–79; stigma associated with, 80; section 8 rental certificates, 202, 204; tagging tagging function of, 79–80, 133; valuing, 39–40. function of, 79; vouchers, 201, 204 See also food stamps; health insurance human capital model: application to gays and inner cities: depopulation of, 209; job creation in, lesbians, 366–67, 368; explanation of female-male 386; migration of jobs from, 344–45; poverty rates wage differentials, 354–56, 366, 368–70, 369f; in, 44t, 45. See also neighborhoods, high-poverty investments by women, 360; relation between insurance. See health insurance; unemployment education and earnings, 212; of testing, 214, insurance 217–18 intelligence. See cognitive ability; IQ scores hunger: incidence of, 50, 389; measures of, 48–50. intergenerational transmission of poverty, 172–74 See also food insecurity Internal Revenue Service (IRS), 69 international trade, role in wage inequality, 111–13, illegitimate births. See single mothers 114–15 immigrants: educational attainment of, 113–14; IQ scores: explanations of black-white differences, eligibility for welfare benefits, 251; food insecurity 317–19; relation between early and late, 318, 319f of, 251; labor supply increased by, 113–14; sex iron law of wages, 1 ratios of, 149 IRS. See Internal Revenue Service immunizations, 175, 178, 179 isolation index, 335–36, 337, 338t imports. See international trade imprisonment: consequences of, 307; incarceration jails. See imprisonment rates, 306t, 306–7; stigma associated with, 307 Job Corps, 129–30, 131, 382 income effect, of earned income tax credit, 72–78 Job Opportunities and Basic Skills (JOBS) program, income inequality: across families, 89; among women, 122–23, 134 102; causes of, 389; 50-10 male earnings ratio, job placement programs, 134 92–96, 94f, 95f, 106–7; as incentive for working, jobs. See employment 389; increase in as explanation for persistence of JOBS. See Job Opportunities and Basic Skills poverty, 89, 92–96, 109; reducing, 381–82, JOBSTART demonstration project, 131–32 387–88, 389; relation to poverty rates, 92–96, 94f, Job Training Partnership Act (JTPA) of 1982, 122, 97f, 106–7. See also wage inequality 125–29, 133 incomes: effects of job training programs, 127–29, job training programs: for AFDC recipients, 122–23, 127f, 128f, 133–34; effects of welfare reform, 125; Ashenfelter dip in earnings, 126, 127–28; 256–57, 257f; effects of welfare-to-work programs, costs of, 123, 124t, 129, 130, 131; effectiveness of, 135, 135f; family influences on, 172–73; of former 132; effects on earnings, 127–29, 127f, 128f, welfare recipients, 256; median female, 97–98, 133–34; effects on poverty, 123–25; effects on 98f, 147; nonmonetary components of, 39–40; unemployment insurance system, 133–34; perceived needs, 36, 36f; poverty measured relative evaluations of, 125–33, 134; experiments, 126–29; to median family, 35–37, 36f; relation to federal, 121–24, 124t, 125; history of, 121–22; educational attainment, 212–13, 214–15, 368–70, identification function of, 133; need for, 382; 369f; relation to poverty rates, 90–92, 92f, 93f, social benefits of, 130; for youth, 129–33. See also 95–96, 96f, 97–98, 98f, 105–7; relation to welfare welfare-to-work programs caseloads, 253, 254f; relative, 36–37; of teen Johnson administration: affirmative action by sex, mothers, 162, 163t, 164, 165. See also cash transfer 372; Executive Order 11246, 292, 293, 311; Great programs; in-kind transfers; male earnings; wage Society programs, 86, 252; negative income tax differentials; wages proposal, 5

400 Subject Index JTPA. See Job Training Partnership Act manufacturing jobs, loss of, 345 justice system: bail setting, 351; discrimination in, marriages: assortative mating, 101–2, 150; of college- 349–51; incarceration rates, 306t, 306–7; educated women, 356; decline in, 146–56; treatment of black juveniles, 350–51. See also police economic explanations of decline in, 146–50; effects of welfare reform, 159–61, 160t, 243, Kerner Commission, 389 258–59; poverty programs seen as discouraging, 156; promotion of, 250, 258; relation to earnings, labor force participation: of blacks, 287, 289–92, 365; sex ratios and, 149; shotgun weddings, 291t; effects of poverty policy, 87; effects of 153–56, 154f; social explanations of decline in, welfare-to-work programs, 75–78, 76t; Medicaid 150–56; specialization in, 149–50, 355; welfare as disincentive for working, 80, 191, 382, 388; eligibility and, 158–59, 243 welfare as disincentive for working, 5, 66–68, 244, married women: birth rates to, 145, 146, 146f; births 248; of whites, 290–91, 291t. See also to, 144; earnings of, 101–2; employment rates of, employment; labor supply 255, 255f; labor force participation of, 101–2 labor force participation, of women: blacks, 285, 303, Marshall, Alfred, 1–2 304t; effects of welfare reform, 255–56, 255f; Massachusetts: Brookline Early Education Project, former welfare recipients, 256, 257; increase in, 176–77, 383; minimum wage law, 115–16; 370; relation to poverty rates, 101–2; of teen Proposition 21/2, 224–25; tenth-grade math test mothers, 163t, 164; whites, 285, 303, 304t. See scores, 238, 239f. See also Boston also women workers Mathematica Policy Research, 129–30 labor market: demand for skilled vs. unskilled labor, McKinney Homeless Act of 1987, 47 109, 110f; influence on poverty rates, 89–90; mean, 20, 21 unions, 114–15. See also discrimination, in labor Medicaid: children covered by, 190–92; crowding out market; wages of private insurance, 191–92; disincentives for labor market policies: employer-based subsidies, working, 80, 191, 382, 388; effects on labor force 136–40, 207–8; empowerment zones and participation, 87; eligibility of immigrants, 251; enterprise communities, 207–8; increasing eligibility rules of, 80, 190–91, 388; expansion of, earnings of low-wage workers, 103; living wage 191, 192; outcomes for children, 192; spending laws, 120–21. See also job training programs; on, 64t, 65; value of, 40, 42 minimum wage laws; welfare-to-work programs medical expenses, adjusting earnings for, 51, 52t, 53 labor supply: effects of earned income tax credit, medical insurance. See health insurance 70–78; effects of negative income tax, 6–7; income Medicare, value of, 42 effect on, 72; increases due to immigration, men: educational attainment of, 355–56, 356f, 113–14; of low-skilled workers, 113–14; relation 368–69, 369f; effects of family structure, 169–70, to wages, 5–6; of skilled vs. unskilled workers, 170t; effects of preschool education, 182; gays, 109–10, 110f; substitution effect on, 72 366–67; no-sex penalty on earnings, 367. See also language, communication problems, 273–74, 343, black men; male earnings; occupational 386 segregation; wage differentials; white men Latinos. See Hispanics mentoring programs, for children, 185–86, 188 law enforcement. See police merit goods, 78 lead, levels in children’s blood, 318 Meritor Savings Bank v. Vinson, 373 lending. See mortgages METCO (Metropolitan Council for Educational lesbians: discrimination against, 366–68, 376; Opportunity) program, 200–201, 325 earnings of, 366, 367, 368, 376; occupations of, MFIP. See Minnesota Family Investment Program 376 Michigan, charter schools in, 231–33, 234 living wage laws, 120–21 middle-class blacks, 344, 386 local governments, education funding, 221. See also Milliken v. Bradley, 327 property taxes Milwaukee Parental Choice Program, 230 low-skilled workers. See unskilled workers minimum wage laws: debates on, 12; effects on employment, 117–18; effects on poverty, 116–17, male earnings: 50-10 ratio, 92–96, 94f, 95f, 106–7; 119, 119f; effects on wage inequality, 115, 118, median, 95, 95f; relation to poverty rate, 90–92, 118f; effects on wages, 116; effects on young 92f, 93f, 95–96, 96f, 105–7; unexplained, 95, 96f. workers, 116, 117; Fair Labor Standards Act, 115; See also wage differentials history of, 115; increasing minimum, 382; of Malthus, Thomas, 1 states, 115–16, 117–18 Manpower Training and Development Act (MTDA) Minnesota Family Investment Program (MFIP), of 1962, 121 159–61, 243

Subject Index 401 Minnesota Open Enrollment Option, 228–29 New York City: Community School District 4, minors. See children; teen mothers; youth 226–27; police department, 314; school choice in, mixed slums, 198 225, 226–27; school voucher program, 230–31 mortgages: audit studies of preapproval process, 346; New Zealand: charter schools in, 236, 385; discriminatory lending practices, 206, 345–46, decentralization of education system in, 235–36 351; redlining practices, 206, 386 Nixon administration, Family Assistance Plan, 5 mothers: antidiscrimination laws, 373; in labor force, NLSY79. See National Longitudinal Survey of Youth, 68, 173, 259; marital status of, 144; nutrition 1979 programs for, 174–75; proportion on welfare, 247. No Child Left Behind (NCLB) Act of 2001, 18, 237, See also female-headed families; parenting practices; 385 single mothers; teen mothers nonmarital fertility. See single mothers Moving to Opportunity (MTO), 202–3 normal distribution, 20–21, 331 Moynihan, Daniel Patrick, 7 nutrition programs: effects of, 174–75; experiments, MTDA. See Manpower Training and Development 176; political support for, 79; school lunch and Act breakfast programs, 79, 183–84; Special MTO. See Moving to Opportunity Supplemental Nutrition Program for Women, multitask principal agent problem, 218 Infants, and Children (WIC), 174–75; spending multivariate analysis, 104–7 on, 64t, 65

NAEP. See National Assessment of Educational Obama, Barack, 323 Progress observational studies, 5–6, 7–8 names, effects of, 294–95, 298 occupational segregation: causes of, 356–60; National Academy of Sciences, 42, 51, 52–53, 58, comparable worth issue, 373–75; discrimination as 173 explanation of, 357–58; relation to wage National and Community Service Act of 1992, 132 differentials, 356–57 National Assessment of Educational Progress (NAEP): Office of Federal Contract Compliance Programs black-white score gaps, 319–20, 320f, 321f; scores (OFCCP), 311, 313, 372 in Texas, 220–21 orchestras, sex discrimination in hiring, 362–64, 363t National Bureau of Economic Research, 61 Orshansky, Mollie, 37, 38, 39 National Evaluation of Welfare to Work Strategies, out-of-wedlock births. See single mothers 134–35 National Job Corps Study, 129–30, 133 parenting practices: effects on children, 173; as factor National Longitudinal Survey of Youth, 1979 in test score gap, 322; programs to improve, 175, (NLSY79), 62, 300 177, 322, 383 National School Breakfast Program, 183–84 parents: child support enforcement, 250; deaths of, National School Lunch Program, 183 171–72; two-parent families, 158–59, 168–72, natural experiments, 7–8 243, 252. See also marriages; mothers; single- NCLB. See No Child Left Behind Act of 2001 parent families near-poor households: food insecurity of, 50; lack of Pareto principle, 13 health insurance coverage, 389; medical expenses Parks, Rosa, 334 of, 51 Personal Responsibility and Work Opportunity negative income tax, 4–7, 8, 12 Reconciliation Act (PRWORA) of 1996, 3, neighborhoods: effects on children, 197, 201–2; 245–51. See also welfare reform ethnographic studies of, 198, 200–201; influences Plessy v. Ferguson, 325 on individuals, 198–202, 202t, 205; integrated, police: affirmative action in hiring, 313–14; missing 339. See also housing segregation; suburban persons cases, 57; racial profiling by, 266–67, neighborhoods 349–50 neighborhoods, high-poverty: characteristics of policy analysts, disagreement among: limits of residents, 198, 199t; community development statistical arguments, 10–12; values differences, policies, 206–8; crime in, 197, 200, 209; 12–13 economies of, 205–6; gangs in, 205–6; as ghettos, poor neighborhoods. See neighborhoods, high-poverty 197–98; perceptions of, 45; policies to address POSSLQ couples, 151, 152f needs of, 385–86; quality of schools in, 211; racial poverty: definitions of, 31–37, 58, 389; in developing segregation of, 345; trends in, 209. See also public countries, 31–32; duration of, 53–58, 55t; housing projects explanations of long-term, 56–58, 57t; New England Patriots, 380 intergenerational transmission of, 172–74; New Jersey, minimum wage law, 117–18 participation in society and, 34–35, 36–37, 381,

402 Subject Index 389; perceptions of, 32–36; relation to pregnant women: abortions, 155, 156, 165, 166; discrimination, 386–87; returning to, 55, 56 nutrition programs for, 174–75. See also mothers poverty measures: adjusting for medical costs, 51, 52t, prejudice: of customers, 271–72; definition of, 266; 53; adjusting for regional housing cost differences, discrimination based on, 269–73, 297; distinction 52, 52t; alternative, 51–53, 86–87, 87f; data from discrimination, 265–66; of employees, sources, 42–43, 61–62; extreme poverty, 45; for 270–71; of employers, 269–70, 297; in imperfect family of four, 381; including nonmonetary labor markets, 272–73; segregation resulting from, income, 39–40, 86; official federal government, 269–71, 297; of teachers, 322–23 37–39, 41, 42–43, 58, 83–86, 84f; problems with preschool education, 177–82, 383, 384. See also Head official, 38, 39, 41, 52–53; regional differences in Start cost of living and, 41, 42; relative vs. absolute, prisons. See imprisonment 35–37; suggested revisions to official, 42, 51, private schools, 229, 328. See also vouchers 52–53, 58; time periods used, 40–41, 53–55; of program evaluations: adjusting for nonparticipation, World Bank, 31, 35 127, 141–42; of Big Brothers Big Sisters, 185–86; poverty policy: aimed at children, 173–77, 182–90, concerns with, 193–94; of early childhood 383–84; community development programs, education, 176–77, 178–80; of Head Start, 206–8, 386; early education programs, 176–82, 178–80; of job training programs, 125–33, 134; of 332, 383–84; effects on labor force participation, welfare-to-work programs, 134–35, 173 87; federal programs, 63–65, 64t; Great Society property taxes: education funding, 221, 222; limits programs, 86, 252; health insurance, 190–92, 382, on, 224–25; relation to housing prices, 222, 223, 388; health programs, 175; measuring effects of, 224–25 86; in nineteenth century, 1–2; nutrition PRWORA. See Personal Responsibility and Work programs, 64t, 65, 79, 174–75, 176, 183–84; Opportunity Reconciliation Act parenting education programs, 175, 177, 322, public accommodations, desegregation of, 334, 346 383; recommendations for, 381–89; seen as public housing projects: influence on outcomes for discouraging marriage, 156; shift from cash to in- children, 204; moving residents out of, 201–3; kind transfers, 63, 65, 68, 81. See also cash transfer negative image of, 203–4; policy shift away from, programs; earned income tax credit; housing 204; problems in, 203; segregation of, 338 programs; in-kind transfers; job training programs; public schools. See education labor market policies; welfare programs poverty rates: by age group, 43, 44t, 84f, 84–85; with Quantum Opportunities Program (QOP), 187–90, alternative measures, 51, 52t, 86–87, 87f; 189t, 194 calculation of, 42–43; by community type, 44t, quasi-experiments, 7–8 45; comparison of measures, 51–53, 52t; effects of poverty programs, 86; factors explaining variations races: census categories, 265; definitions of, 265; in, 97, 104–7, 105t, 107t; of families, 85, 86f; by language and communication differences, 273–74, family structure, 43–45, 44t, 85, 86f; influences 343, 386. See also blacks; whites on, 87–102; by race, 43, 44t, 85, 85f; relation to racial discrimination. See discrimination GDP, 88, 89f, 381; relation to household racial profiling: by police, 266–67, 349–50; in retail composition, 143; relation to income inequality, stores, 346; by taxi drivers, 266 92–96, 94f, 97f, 106–7; relation to incomes, racism, in housing market, 338–39. See also prejudice 90–92, 92f, 93f, 95–96, 96f, 97–98, 98f, 105–7; Rand Corporation, 220, 221, 234 relation to inflation, 88; relation to median female randomness, 19–20 earnings, 97–98, 98f, 106, 107; relation to median relative poverty, 35–37 male earnings, 90–92, 92f, 93f, 95–96, 96f, research: ethnographic studies, 9, 198, 200–201; 105–7; relation to minimum wage, 119, 119f; observational studies, 5–6, 7–8; value of, 379–81. relation to unemployment rates, 88, 98, 99f, 106; See also audit studies; experiments relation to welfare benefit levels, 99, 100f, 106; residential segregation. See housing segregation relation to women’s labor force participation, restaurants: racial discrimination in, 346–47; sex 101–2; of single-parent families, 44; of teen discrimination in hiring, 360–62, 362t mothers, 144, 162, 163t, 164, 167; trends using restrictive covenants, 337–38 alternative measures, 86–87, 87f; trends using retail stores, discrimination in, 346–47. See also official measure, 83–86, 84f; unexplained, 95, 96, discrimination, in customer markets 97f retention in grade, 13–16 poverty threshold, development of, 37–39 Ricardo, David, 1 Pregnancy Discrimination Act, 373 Robert Taylor Homes, 204 pregnancy prevention programs, 186–87, 250 Roe v. Wade, 155

Subject Index 403 Saint Mary’s Honor Center v. Hicks, 309 skilled workers: demand for, 109, 110f; education of, San Jose, California, comparable worth agreement, 109–10; supply of, 109–10, 110f. See also job 374–75 training programs Scandinavia, births to unmarried mothers in, 151 slums, 198. See also neighborhoods, high-poverty SCHIP. See State Children’s Health Insurance small numbers defense, 310–11 Program social distance model, 273–74 school choice programs: alternatives, 225; charter Social Security: disability insurance, 64, 64t, 253; schools, 225, 231–34, 240, 385; interdistrict, poverty rates lowered by, 45; supplemental security 228–29; intradistrict, 225–28; opponents of, 227, income, 64t, 64–65, 68, 253–54 228; statewide, 228–29; vouchers, 225, 229–31, socioeconomic status. See economic status 234–36 sociological fallacy, 320 schools. See desegregation, school; education sorting model: relation between education and Seattle Income Maintenance Experiment. See earnings, 212–13, 214–15; role of testing, 214–17, SIME/DIME 216t segregation: court cases supporting, 325; distinction South Carolina textile industry, 293 from discrimination, 268–69; in employment, spatial mismatch hypothesis, 344–45 269–71, 297; resegregation of schools, 327–28, Special Supplemental Nutrition Program for Women, 329; of schools, 325–26. See also desegregation, Infants, and Children (WIC), 174–75 school; housing segregation; occupational sports: discrimination in, 271–72, 305–6; use of segregation research, 380–81; wage differentials by race, 306 self-confirming expectations, 277–80 SSI. See supplemental security income senior citizens. See elderly standard deviation, 20, 21–22 sex discrimination: audit studies of, 360–64, 362t; standard error of estimate, 21–22 in employment, 357–64; evidence of, 360–64; standardized tests. See testing history of, 360; influence of beauty, 364; State Children’s Health Insurance Program (SCHIP), institutional, 376–77; laws prohibiting, 372–73; 191 in orchestras, 362–64, 363t; in restaurants, state dependence: as explanation of long-term 360–62, 362t; sexual harassment as, 373; theories poverty, 56, 57–58; as explanation of welfare of, 357–60. See also occupational segregation; dependence, 244 wage differentials, female-male states: enterprise zones of, 207–8; minimum wage sex ratios, of immigrant groups, 149 laws of, 115–16, 117–18; school choice programs sexual activity: effects of increased access to of, 228–29; welfare benefit levels in, 157–58; contraception and abortion, 155–56; no-sex welfare reform experiments of, 7, 75–78, 159–61, penalty on earnings, 367 173, 245 sexual harassment, 373 statistical analysis: causal interpretations of, 10; sexual orientation: discrimination based on, 366–68; confidence intervals, 22–23; controlling for other relation to earnings, 366–68, 376 factors, 27–28; correlations found in, 10; guide to, shotgun weddings, 153–56 19–28; limitations of, 10–12; mean, 20, 21; SIME/DIME (Seattle Income Maintenance multivariate analysis, 104–7; normal distribution, Experiment / Denver Income Maintenance 20–21; relations among variables, 25–27, 26f, 27t; Experiment), 6–7 standard deviation, 20; standard error of estimate, single mothers: birth rates to, 145, 146f, 251; 21–22; t-statistic, 24–25 causes of increase in proportion of births to, statistical discrimination, 274–75 145–46, 153–54; collection of data on, 144; statistical significance, 23–24, 310–11 disability insurance payments to, 254; effects of stereotypes, 279–80, 281, 322–23. See also prejudice welfare benefit levels, 156–58; effects of welfare street gangs, 205–6 features, 158–61; effects of welfare reform, 161, students. See children; education; youth 255; employment rates of, 255, 255f; marriage subsidies. See employer subsidies; housing subsidies rates of, 153, 154f, 161; in other countries, 151; substitution effect, of earned income tax credit, poverty programs seen as encouraging, 156; 72–78 proportion of births to, 144–45, 145f; stigma suburban neighborhoods: busing to schools in, attached to, 155; teenagers, 144, 161, 162f; 200–201, 325; effects on children, 199; jobs in, welfare reform and, 250–51. See also female- 344–45; middle-class blacks in, 344; migration to, headed families 344; studies of public housing residents moving to, single-parent families: after death of parent, 171–72; 201–2, 202t effects on children, 168–72, 169t, 170t; poverty supplemental security income (SSI): as income source rates of, 44. See also female-headed families for welfare recipients, 68; poverty reduced by,

404 Subject Index 64–65; shifting welfare recipients to, 253–54; Armed Forces Qualifying Test; National spending on, 64t, 253–54 Assessment of Educational Progress Supreme Court, U.S.: abortion and contraception Texas: education reforms in, 219; high-stakes testing cases, 155; affirmative action cases, 330; in, 219–21, 240; school performance evaluations, employment discrimination cases, 308–9; 237 restrictive covenants struck down by, 337–38; Texas Assessment of Academic Skills (TAAS), 220, school desegregation cases, 326, 327, 329; sex 237, 238 discrimination cases, 373; support of segregation, Texas Assessment of Knowledge and Skills (TAKS), 325 220, 237 Texas Board of Education, 237 TAAS. See Texas Assessment of Academic Skills theory, importance of, 8 tagging function: of in-kind transfers, 79–80; of job Tiebout model, 221–24, 225, 228 training programs, 133 Title VII, 292, 372. See also Civil Rights Act of TAKS. See Texas Assessment of Knowledge and 1964 Skills TJTC. See Targeted Jobs Tax Credit TANF. See Transitional Assistance to Needy Families Tobin, James, 5 Targeted Jobs Tax Credit (TJTC), 137–40 TOP. See Teen Outreach Program tax credits: employment subsidies, 137–40, 207–8. tracking, educational, 57–58, 318, 328 See also earned income tax credit trade. See international trade taxes. See income tax; property taxes training: retaining workers, 358–59; signals of, taxi drivers: discrimination by, 347; racial profiling 277–80, 278t; of skilled workers, 109–10; wage by, 266 differences after, 358–59; of welfare recipients, teachers: black, 322, 323; cheating on standardized 247–48. See also job training programs tests, 239; expectations of students, 322; prejudice transaction costs models of discrimination, 273–74, of, 322–23; white, 323 343 technological change, role in wage inequality, Transitional Assistance to Needy Families (TANF): 110–11, 114, 115 block grants, 246, 250; caseload reductions, 247, teenagers. See youth 248, 252–54, 260; establishment of, 246; teen mothers: birth rates to, 161, 162f; comparisons reauthorization of, 248; spending on, 63, 64t, 246; to nonteen mothers, 162–65, 163t, 164t; time limits of, 248–50; welfare-to-work programs, comparisons to sisters, 164t, 165; comparisons to 123, 133; work requirements, 246–48. See also women who miscarried or had abortions, 165–68, welfare reform 167t; earnings of, 162, 163t, 164, 165; educational t-statistic, 24–25 attainment of, 162, 163t, 164, 165, 167; employment of, 163t, 164; marital status of, 167; unemployment, distinction from nonemployment, outcomes of, 161–68, 163t, 164t; poverty rates of, 303 144, 162, 163t, 164, 167; pregnancy prevention unemployment insurance, effects of job training programs, 186–87, 250; proportion of births to, programs on receipt of, 133–34 144; races of, 162, 163t; unmarried, 144, 161, unemployment rates: racial differences in, 303–4, 162f; welfare recipients, 162, 163t, 164 304t; relation to poverty rates, 88, 98, 99f, 106 Teen Outreach Program (TOP), 187 Unicon Research, 61 television programs: The Weakest Link, 348–49 unions, 114–15, 374 testing: arguments for, 213–14; black-white score gap, United Kingdom, professional soccer, 306 317–25; cheating by educators, 239; concerns United States Department of Agriculture, 37, 48, 50, about, 237–39, 240; in diverse schools, 238–39; 79 effects on student effort and productivity, 215–17, United States Department of Labor, 122 216t; environmental influences on scores, 320–22; United States government. See federal government evidence of effects of, 219–21; high-stakes for University of Michigan, 330 schools, 236–39, 385; high-stakes for students, University of Michigan Law School, 330 213–21; human capital model, 214, 217–18; University of Texas Law School, 330 overreliance on, 385; randomness in score changes, unmarried mothers. See single mothers 238, 239f; reducing racial gap, 322, 332; relation unskilled workers: demand for, 109, 110f; of scores to retention in grade, 13; role of scores in immigrants, 113–14; marriages rates of, 148–49; college admissions, 330–32; SAT (Scholastic stereotypes of, 301; supply of, 109–10, 110f, Aptitude Test), 13, 330; sorting function of, 113–14; women, 97–98. See also job training 214–17, 216t; teaching to tests, 218–19, 237–39, programs 385; in Texas, 219–21, 240; use of, 240. See also Urban Institute, 341

Subject Index 405 values, 12–13 proportion of population, 252–53, 253f; teen variables: random, 19–20; relations among, 25–27. mothers, 162, 163t, 164 See also statistical analysis welfare reform: anticipated effects of, 12–13; variance, 20 arguments for, 243–45; debates on, 12–13, vocational training: of welfare recipients, 247–48. 245–46; effects of, 249–50, 251–59; effects on See also job training programs; training children, 173, 258, 259; effects on employment, vouchers: housing, 201, 204; school, 225, 229–31, 255–56, 255f; effects on marital status, 159–61, 234–36 160t, 243, 258–59; effects on well-being, 254–58; goal of encouraging marriage, 258; influence of wage differentials: Hispanic and non-Hispanic, 299, experiments, 7; provisions of law, 245–51; 299t, 301; in same-sex couples, 366 rationale for, 3; state experiments, 7, 75–78, wage differentials, black-white: by birth cohort, 159–61, 173, 245; trade-offs involved in, 260; 285–87, 286f; effects of school desegregation, welfare-to-work programs, 134. See also 328–29; explanations for decline in, 287–93; Transitional Assistance to Needy Families factors in, 299–301, 314–15, 317; laws welfare-to-work programs: created in welfare reform, prohibiting discrimination, 307–8; of men, 134; definitions of work, 247–48; effects on 283–84, 284f, 285–87, 286f, 291–92, 291t, 299t, children, 173, 259; effects on earnings, 135, 135f; 299–301; political and legal factors in, 292–93; in effects on labor force participation, 75–78, 76t; professional sports, 306; statistical analysis of, effects on welfare participation, 133; employer 298–303; of women, 283–85, 284f, 301–3. See subsidies, 137; evaluations of, 134–35, 173; also discrimination, in labor market federal, 122–23; job placement programs, 134; wage differentials, female-male: by birth cohort, mandatory, 133, 159; for married couples, 159; 370–72, 371t; comparable worth issue, 373–75; state experiments, 75–78; as tagging devices, 133; decline in, 354, 373; historical, 360, 368, 369f; training, 247–48, 382; work first approach, human capital model of, 354–56, 368–70, 369f; 134–35, 382 occupational segregation and, 356–57; theoretical Welfare-to-Work Tax Credit, 137 explanations of, 354–60. See also sex well-being, effects of welfare reform, 254–58 discrimination white men: educational attainment of, 287–88, 288f; wage inequality: demand factors in, 109, 110–13, employment rates of, 290, 291t, 303, 304t; 114–15; determinants of, 108–15; effects of incarceration rates of, 306, 306t; without earnings, minimum wage laws, 115, 118, 118f; increases in, 290, 290f 109; institutional factors in, 114; supply factors in, whites: effects of Head Start, 179f, 180, 180f, 181; 109, 113–14. See also income inequality high-poverty neighborhoods of, 198; language and wages: effects of employer subsidies, 136, 137f; iron communication, 273–74, 343, 386; poverty rates law of, 1; living wage laws, 120–21; relation to of, 43, 44t, 85, 85f; support of welfare, 158; labor supply, 5–6. See also minimum wage laws teachers, 323; test score gap, 317–25. See also wage wage subsidies. See employer subsidies differentials, black-white Washington State, comparable worth implementation white slums, 198 in, 375 white women: educational attainment of, 288, 289f, Weakest Link, The, 348–49 301–2, 302t; employment rates of, 285, 303, welfare programs: average monthly benefits, 65, 99, 304t; single motherhood rates of, 157; teen 100f; benefit levels and poverty rates, 99, 100f, mothers, 163t 106; benefit levels and single motherhood, WIA. See Workforce Investment Act 156–58; dependence on, 244; disincentives for WIC. See Special Supplemental Nutrition Program working, 5, 66–68, 244, 248; effects of features on for Women, Infants, and Children single motherhood, 158–61; generational patterns WIN. See Work Incentive Program in dependence on, 244–45; participation reduced women: abortions, 155, 156, 165, 166; availability of by job training programs, 130; poverty reduced by, contraception, 155–56; black-white earnings 63–64; spending on, 63, 64t; time limits of, differentials, 283–85, 284f, 301–3; cohabitation 248–50; variations in benefits across states, rates of, 152, 153f; educational attainment of, 147, 157–58. See also Aid to Families with Dependent 288, 289f, 355–56, 356f, 368–69, 369f; effects of Children; Transitional Assistance to Needy family structure, 170, 170t; effects of preschool Families education, 182; gains from marriage, 149–50; welfare recipients: characteristics after welfare reform, incarceration rates of, 306t; income inequality 258; employment after leaving welfare, 256, 257; among, 102; labor force participation rates of, immigrants, 251; income sources of, 68; minors, 101–2, 255–56, 285, 370; lesbians, 366–68, 376; 250, 258, 259; proportion of mothers, 247; marriage rates of, 152, 153f; married, 101–2, 144,

406 Subject Index 145, 146; median earnings of, 91, 147. See also Workforce Investment Act (WIA) of 1998, 122, black women; female-headed families; mothers; sex 187–88 discrimination; white women Work Incentive Program (WIN), 122 women workers: child care availability for, 383–84; working poor: effects of earned income tax credit, 65; comparisons to welfare recipients, 68; effects of proportion of poor, 90, 91f children on earnings, 365, 376; effects of working World Bank, 31, 35 on children, 173, 259; experience of, 354–55, 370; WtW programs. See welfare-to-work programs income sources of, 68; increased opportunities for, 147–48; low-skilled, 97–98; married, 101–2; YOG. See Youth Opportunities Grants median earnings of, 97–98, 98f; relation of beauty youth: effects of employment subsidies, 138, 139f; and earnings, 364; skills of, 355, 370; training of, effects of maternal work, 173, 259; effects of 358–59; withdrawals from labor market, 354–55, minimum wage laws, 116, 117; gangs, 205–6; job 357, 358–59, 365, 376. See also labor force training programs for, 129–33; programs aimed at, participation, of women; occupational segregation; 187–90, 208; treatment of black youth in justice sex discrimination; wage differentials, female-male system, 350–51; welfare recipients, 250. See also work. See employment; job training programs; labor teen mothers force participation; welfare-to-work programs Youth Opportunities Grants (YOG), 208 Worker Profiling and Reemployment Services, 133–34 Youth Services Corps, 132–33

Subject Index 407