Globalization and Child Labor

Globalization and child labor

Inauguraldissertation

zur Erlangung der Doktorwürde der Wirtschafts- und Verhaltenswissenschaftlichen Fakultät der Albert-Ludwigs-Universität Freiburg i. Br.

Krisztina Kis-Katos geb. in Gelendzhik (Russland)

Wintersemester 2009/2010 Albert-Ludwigs-Universität Freiburg im Breisgau Wirtschafts- und Verhaltenswissenschaftliche Fakultät Kollegiengebäude II Platz der Alten Synagoge 79085 Freiburg

Dekan: Prof. Dr. Dieter K. Tscheulin Referent: Prof. Dr. G¨unther G. Schulze Koreferent: Prof. Dr. Heinrich W. Ursprung Datum der Bekanntgabe des Pr¨ufungsergebnisses: 6. Mai 2010 Acknowledgments

I would like to express my sincere gratitude to my thesis supervisor Prof. Günther G. Schulze for all the support and encouragement, academic advice, constructive comments, and the fruitful collaboration on some of the topics of this book; our joint work is included in Chapters 2 and 4. My thanks go also to Robert Sparrow for the productive co-authorship on a study which is presented in Chapter 6 of this book. My research originated in a project on “’Social Dumping’, Trade and Em- ployment in the Context of Globalization of Economic Activity”, which was a joint project with Prof. Heinrich W. Ursprung and his team. The project has been funded by the Volkswagen Foundation from 2002 to 2006, which is gratefully acknowledged. I am also thankful to Prof. Ursprung for reading this thesis. I presented the ideas in this book on several occasions, at the ETSG Meetings (Dublin 2005), at the SMYE Conference (Hamburg 2007), at the IZA Summer School on Labor Economics (Buch am Ammersee 2007), at the IZA/World Bank Conference on Employment and Development (Bonn 2007), at the Conference of the European Economic Association (Budapest 2007, Barcelona 2009), at the PEGNet Conference (Berlin 2007), at the ISS Devel- opment Dialogue (The Hague 2008), at the AEL Conference (Frankfurt 2009), at the IZA/World Bank Child Labor Workshop (Bonn 2009), at the Confer- ence of the German Economic Association (Munich 2007, Magdeburg 2009), at the ZEF (Bonn 2007) and at numerous doctoral seminars in Freiburg; I would like to thank all participants for the many helpful comments, suggestions and discussions. Furthermore, I would like to thank all my friends and (former) colleagues who shared some of the ups and downs of graduate life with me: Julia Alexa Barde, Mathias Czaika, Eva Deuchert, Antonio Farfán Vallesp´ın, Juliane Flied- ner, Stefanie Flotho, Katja Keller, Martin Keller, Aderonke Osikominu, Grischa Perino, Bambang Sjahrir Putra, and Suncica Vujić; thank you for the discussions, encouragement, and advice. I would like to thank Mrs. Giseltraut Herbert for all the kind and selfless help I received from her during these years, and Judith Müller for making life easy at the finishing stages. I am especially indebted to Fitria Fitrani and Bambang Sjahrir Putra for help and advice with Indonesian data. Finally, I would like to thank my parents for their love and support, and my family, István, Márton and Anna for making sure that life has so much more to it than only work.

i List of Abbreviations

APE Average Partial Effect BPS Indonesian Statistical Office (Badan Pusat Statistik) BULOG Indonesian Bureau of Logistics (Badan Urusan Logistik) CAFTA-DR Dominican Republic-Central America-United States Free Trade Agreement EU European Union FDI Foreign Direct Investment GDP Gross Domestic Product GEE Generalized Estimating Equations GMM Generalized Method of Moments GRDP Gross Regional Domestic Product hh. household ILO International Labour Organization IV Instrumental Variables KUK Small Business Loan program (Kredit Usaha Kecil) LIML Limited Information Maximum Likelihood LSMS Living Standards Measurement Study MLE Maximum Likelihood Estimation NAALC North American Agreement on Labor Cooperation NAFTA North American Free Trade Agreement NSS National Sample Survey of India NTB West Nusa Tenggara (Nusa Tenggara Barat) NTT East Nusa Tenggara (Nusa Tenggara Timur) OLS OrdinaryLeastSquares PODES Village Potential Census of Indonesia (Potensi Desa) PROGRESA Education, Health, and Nutrition Program of Mexico (Pro- grama de Educación, Salud y Alimentación) sec. secondary SIMPOC Statistical Information and Monitoring Programme on Child Labour SITC Standard International Trade Classification SML Simulated Maximum Likelihood SNA System of National Accounts SUSENAS National Socioeconomic Survey of Indonesia (Survei Sosial Ekonomi Nasional) TRAINS Trade Analysis and Information System UN United Nations UNCTAD United Nations Conference on Trade and Development UNICEF The United Nations Children’s Fund US United States of America WDI World Development Indicators WTO World Trade Organization

ii Contents

Acknowledgments i

List of Abbreviations ii

Contents iii

List of Tables v

List of Figures vii

1 Introduction 1 1.1 Policyrelevance...... 1 1.2 Planofthebook ...... 4

2 Child labor and its regulatory environment 6 2.1 Introduction...... 6 2.2 Deﬁnitions and world-wide prevalence ...... 7 2.3 Regulation of child labor ...... 12 2.4 Conclusion ...... 24

3 Determinants of child labor supply 26 3.1 Introduction...... 26 3.2 Determinants of child labor and schooling ...... 29 3.3 Data and main variables ...... 33 3.4 Estimationstrategy ...... 42 3.5 Results...... 44 3.6 Conclusion ...... 51

4 Demand and supply interactions 52 4.1 Introduction ...... 52 4.2 ChildlaborinIndonesia ...... 56 4.3 Data and empirical strategy ...... 57 4.4 Results ...... 68 4.5 Conclusion ...... 71

5 Trade liberalization and child labor 74 5.1 Introduction...... 74 5.2 Themodel...... 77 5.3 Trade liberalization and child labor ...... 81 5.4 Empiricalevidence ...... 83 5.5 Conclusion ...... 91

iii CONTENTS iv

6 The eﬀects of trade liberalization in Indonesia 94 6.1 Introduction...... 94 6.2 Theoreticalbackground ...... 96 6.3 Trade liberalization and children in Indonesia ...... 100 6.4 Data and empirical approach ...... 108 6.5 Results...... 116 6.6 Conclusion ...... 131

7 Conclusion 133 7.1 Summary ...... 133 7.2 Policy implications ...... 135

A Statistical Appendix 139

B Mathematical Appendix 147 B.1 First order eﬀects on the number of child laborers ...... 147 B.2 Relative changes in aggregate child labor ...... 147 B.3 Child labor supply response under universal subsistence . . . . 148 B.4 Suﬃcient conditions for an increase in aggregate child labor . . 148

Bibliography 149 List of Tables

2.1 Global numbers on diﬀerent forms of child work ...... 11 2.2 Economic activity of children by regions ...... 11 2.3 Economic activity of children and its regulation world-wide .... 13

3.1 Activities of children by gender ...... 34 3.2 Activities of children by age ...... 35 3.3 Descriptivestatistics ...... 36 3.4 Determinants of household income and business ownership..... 38 3.5 Tests on validity of the instruments ...... 39 3.6 Trivariate probit results on work/schooling of girls ...... 45 3.7 Trivariate probit results on work/schooling of boys ...... 46 3.8 Selected APEs on probabilities of specializing in work/idleness . . 48 3.9 Selected APEs on joint probabilities of work and school ...... 49

4.1 Child labor incidence in small scale manufacturing ...... 58 4.2 Geographic distribution of child labor ...... 59 4.3 Number of ﬁrms per village by specialization in child labor . . . . 60 4.4 Variable deﬁnitions and descriptive statistics ...... 63 4.5 Child labor in small scale manufacturing (w. sample selection) . . 69

5.1 Descriptivestatistics ...... 85 5.2 Changes in child labor and changes in openness ...... 88 5.3 Differential effects of openness for country groups ...... 89 5.4 Changes in child labor and changes in openness (Alt. measure) . . 90 5.5 Differential effects of openness for country groups (Alt. measure) . 91

6.1 Evolution of market work of children over time ...... 105 6.2 Evolution of school enrollment of children over time ...... 108 6.3 Descriptivestatistics ...... 111 6.4 Pooled results on child market work and tariff protection . . . . . 117 6.5 Pooled results on child schooling and tariff protection ...... 118 6.6 Child work, schooling and tariff protection in the district panel . . 120 6.7 Difference estimates and exogeneity test (Tariffs by labor) . . . . . 121 6.8 Difference estimates and exogeneity test (Tariffs by GRDP) . . . . 122 6.9 Difference estimates by time period ...... 123 6.10 Child market work and tariff protection, GMM estimates . . . . . 125 6.11 Child schooling and tariff protection, GMM estimates ...... 126 6.12 Child market work and tariff protection by district type, GMM est. 127 6.13 Child schooling and tariff protection by district type, GMM est. . 128 6.14 Tariff reductions and poverty, GMM estimates ...... 129

v list of tables vi

6.15 Market work by age cohorts, GMM estimates ...... 130

A.1 Age distribution of children by school-class attended ...... 139 A.2 Definitions of explanatory variables ...... 140 A.3 Child labor incidence in small scale manufacturing by island . . . . 141 A.4 Determinants of the number of small firms in a village ...... 143 A.5 Countriesinthesample ...... 143 A.6 Definitionsofvariables...... 144 A.7 The originally poorest countries ...... 144 A.8 Child market work and tariff protection, full GMM results . . . . . 145 A.9 Child schooling and tariff protection, full GMM results ...... 146 List of Figures

4.1 Regional incidence and intensity of child work (province level) . . . 58

5.1 Thehoursofchildlaborsupply ...... 80

6.1 TariffreductionsinIndonesia ...... 101 6.2 Tariffreductionsbysectors ...... 101 6.3 Tariff levels and reductions ...... 102 6.4 Work of children, by gender and age group ...... 104 6.5 Sectoral distribution of child work ...... 104 6.6 School enrollment of children, by gender and age group ...... 107 6.7 Evolution of tariff protection ...... 112 6.8 Initial district conditions and change in child work ...... 114

vii chapter 1

Introduction

1.1 Policy relevance

Child labor is a significant problem; according to the International Labor Or- ganization (ILO 2006) about every sixth child (aged 5 to 14) is working world- wide. Although global child labor has been steadily decreasing over the last decades, large regional imbalances prevail, with child labor being endemic in Sub-Saharan Africa and in some South Asian countries. In 2004 around 6% of children (aged 5 to 14) worked under hazardous conditions (ILO 2006); these forms of work are directly harmful to children and deserve special policy attention in their own right. But other types of work can also be detrimental if they conflict with schooling and thus impede human capital accumulation and child development. The resulting underinvestment in human capital can be inefficient not only from an individual but also from a societal point of view. Given the importance of the issue, a number of initiatives exist to establish legally binding minimum safeguards for all children world-wide. Child labor standards have played historically a central role within the legal framework of the ILO, and abolition of child labor ranks among ILO’s main global priorities. The ILO conventions offer a well-established international legal framework for global minimum child labor standards, although they only apply to ratify- ing countries. In addition to that however, the Declaration on Fundamental Principles and Rights at Work (1998) defined “eradication of child labor” as one of the four core labor standards.1 These are expected to be respected by all ILO member countries alike independently of their development status. However, despite their acknowledged importance, noncompliance with these international as well as with national standards on child labor is a major issue. As the ILO has no real enforcement power on its own, in Northern countries the use of trade sanctions has been frequently proposed in order to compel countries in the South to implement and enforce child labor standards. The idea of linking labor standards to trade was one of the most widely discussed policy options in the early 1990s. In the lead-up to the foundation of the World Trade Organization, the inclusion of a “social clause” in the WTO framework

1 The other three core labor standards include freedom of association and the right to collective bargaining, the abolition of forced labor, and non-discrimination in employment and occupation.

1 introduction 2

has been hotly debated, but was almost uniformly rejected by developing countries as a form of disguised protectionism (cf. Kis-Katos and Schulze 2002). Although trade and labor linkages did not become part of the multilateral trading regime and were taken from the WTO agenda,2 the threat of trade sanctions in response to labor rights violations abroad is not off the table. US senator Tom Harkin has started numerous legal initiatives to impose trade sanctions against countries exporting goods produced by child labor, all of which remained yet unsuccessful.3 Moreover, the US has been including labor clauses in its preferential trade agreements since 1993.4 NAFTA has been the first trade agreement to include a social clause in form of a side agreement, the North American Agreement on Labor Cooperation (NAALC), which required all parties to enforce their own domestic labor laws. The Trade and Devel- opment Act of 2000 (Sec. 412) defined eligibility to preferential trade benefits (under the Generalized System of Preferences, the Caribbean Basin Initiative, the Andean Trade Preference Act, and the African Growth and Opportunities Act) based on implementation of country commitments to eliminate the worst forms of child labor. These labor clauses resulted at most in consultations, in- vestigations and reports, but no trade sanctions were ever applied in response to child labor (or any other) violations. However, in July 2010, the US filed a case under a free trade agreement (CAFTA-DR) against Guatemala for labor rights violations; this is the first ever legal case that could result in trade sanctions in response to breaches of labor standards. Demands for linking trade policies to child labor standards are usually based on moral or efficiency arguments, but protectionist motivations also play a role. Supporters of trade sanctions often voice genuine moral concerns about the well-being of children world-wide. Moreover, trade sanctions are also called for in order to solve problems of international economic efficiency. Such arguments boil down to a fear of a “race to the bottom”—a gradual erosion of standards to inefficiently low levels because of international competition, in which case the imposition of common standards offers a globally efficient solution. Whether this argument is substantiated is questionable; as to date, there is no robust empirical evidence that would support the race to the bottom hypothesis.5 In addition to that, such demands also often reflect political

2 Paragraph 8 of the most recent Doha Ministerial Agreement in 2001 merely restates that “we reaffirm our declaration made at the Singapore Ministerial Conference regarding internationally recognized core labour standards. We take note of work under way in the International Labour Organization (ILO) on the social dimension of globalization”. 3 The Child Labor Deterrence Act of 1999 intended to prohibit importation of goods produced abroad with child labor, although it did not pass the Senate. His latest initiative from 2007—“A Bill to amend the Tariff Act of 1930 to eliminate the consumptive demand exception relating to the importation of goods made with forced labor”—concerns effective enforcement of an import ban on goods produced by slave child (or adult) labor. 4 By contrast, no bilateral trade agreement of the EU combines labor provisions with trade sanctions. 5 Although in an early study Rodrik (1996) finds that the presence of child labor raises introduction 3

economy motives, especially on the side of organized labor: working conditions in the South are often used as a pretext for protectionist demands spurred by rising wage differentials between skilled and unskilled labor and the fear of domestic industries relocating abroad (see also Kis-Katos and Schulze 2002). The main concern with trade sanctions is that under many circumstances they might actually harm those whom they are supposed to protect. The removal of children from the export sectors might just result in them ending up in more hazardous occupations. Whether this will be the case depends crucially on changes in relative factor prices induced by trade restrictions; these will affect in turn both family incomes (and hence the necessity of sending a child to work) and the returns to child labor (an hence the incentives to work). Closely related is the question what happens to child labor in the course of increasing international specialization, especially as rising demand for unskilled labor might create new employment opportunities also for children. More open countries—being also less poor—tend to have less child labor (Edmonds and Pavcnik 2006); however, the shorter term effects of trade liberalization on child labor are less clear-cut. Trade liberalization affects child labor through income and substitution effects resulting from changes in relative goods and factor prices; these might point into opposite directions and leave the net effect on child labor a priori undetermined. Even if the income effects—i.e., a decrease in poverty—reduced child labor, incentives to let children work might increase. The overall effect of trade liberalization remains thus an inherently empirical issue. Both for assessing the effects of increasing international specialization and the potential effects of trade restrictions on children’s well-being, a thorough understanding of the driving factors of child labor is needed. For this, household decisions should be investigated with a special focus on the interplay of demand and supply side factors in shaping child labor outcomes. The effects of trade liberalization can also be directly assessed, both by relating reductions in child labor to trade liberalization in a cross-country setting, and based on the trade liberalization experience of specific countries. This may provide entry points for successful policy interventions, for which a sound knowledge of the factors shaping child labor is a fundamental prerequisite. These are the issues addressed in this book. comparative advantage in labor-intensive goods (measured by exports in textiles as a share of total exports), higher incidence of child labor seems to be associated with lower FDI in all empirical studies (e.g., Rodrik 1996, Kucera 2002). introduction 4

1.2 Plan of the book

Child labor is a very heterogeneous concept, some of its forms being much more harmful and pressing from a policy perspective than others. Chapter 2 presents the various legal and statistical definitions pertaining to child labor, as well as recent estimates on the global extent of the problem. Inspired by the insight that “child labor cannot be just legislated out of existence” (ILO 2002a, p.7), Chapter 2 also investigates the reasons for child labor regulation and some of the potentially superior policy alternatives. Evidence from the theoretical and empirical literature is used to motivate normative as well as positive arguments with respect to child labor legislation. The normative arguments rely either on ethical standards or on economic efficiency considerations. The positive analysis investigates the differing interests concerning child labor restrictions which stem from differences in factor endowments and family structure. Chapter 3 elaborates on the differences between various forms of child labor and their determinants by presenting a case study on the trade-offs between different forms of child work with schooling. The analysis is based on the Survey of Living Conditions (1997/98) of two North Indian states, Uttar Pradesh and Bihar. The study emphasizes the interdependence of different forms of work and schooling and the gender aspects of child work. It jointly estimates market work, domestic work, and school attendance of boys and girls in a trivariate framework, allowing also for combinations of these activities. By including domestic work in the choice set, this method is better suited to address gender differentials in the trade-offs between work and school. Simulation techniques allow hereby for estimating the role of various correlates for specializing in, or combining of specific activities. Among other things, this chapter emphasizes the role of economic opportunities and cultural norms as determinants of child labor. Chapter 4 focuses on the interplay of supply and demand factors determining child labor outcomes, and especially on the role of economic activity. The empirical analysis is based on data from the Indonesian Village Potential Census (PODES) which records both the location of small scale manufacturing firms as well as the prevalence of child labor in these firms in a given village. This makes it possible to distinguish between factors that affect the demand for child labor (i.e., firm location), the supply of child labor, or both. This is of crucial importance as economic activity produces not only wealth—decreasing thus child labor—but also work opportunities for children. Credit availability or school presence not only reduce child labor supply, they also act as location factors for firms and hence increase the demand for child labor; which effect prevails is an empirical issue. The results can also inform us about the potential interplay of demand and supply side effects when, in the course of trade liberalization, the demand for unskilled labor is increasing. Chapter 5 investigates the effects of trade liberalization on the households’ introduction 5

child labor supply in a general equilibrium framework. It focuses on child labor that arises out of subsistence needs. While most studies on the effects of trade liberalization address the interplay of income and substitution effects, this chapter focuses on income effects only and comes to counterintuitive results. It shows that in very poor economies trade liberalization can also increase aggregate child labor supply if the favorable income effects are not large enough to make the poorest households to withdraw their children from work, while child labor increases among the less poor. The predictions of the theoretical model are illustrated based on a country panel where reductions in child labor are related to changes in trade openness for various groups of countries. Chapter 6 addresses the causal effect of trade liberalization on work and schooling of 10 to 15 year old children in Indonesia. It uses the large geographic and economic heterogeneity of the Indonesian archipelago to identify the effects of exposure to trade liberalization on child outcomes. The nov- elty of the approach lies in combining regional variation in economic structure with dynamic effects of trade liberalization. While there are no overall robust findings for schooling, trade liberalization seems to have reduced child labor robustly and consistently; this latter finding is consistent with the presence of Stolper-Samuelson effects favoring the poor. Chapter 6 also presents support- ing evidence that the poverty reducing effects of trade liberalization played an important role in reducing child labor. Chapter 7 summarizes the main findings of the book, and presents policy conclusions. chapter 2

Child labor and its regulatory environment

Abstract

This chapter presents general background information on child labor. It clar- ifies the main definitions on different forms of child labor and outlines the international and national regulatory environment. Subsequently, the main determinants of child labor and the resulting policy implications are summarized.

2.1 Introduction

Child labor has many faces, and clearly, not all of its forms are equally harmful. Some illegal activities (like child prostitution or involvement in drug trafficking) are extremely damaging to the life and health of children. But even activities that are not unlawful if performed by adults can expose children to undue strain, health and safety risks and impede their development. Other forms of work might not be directly harmful to children, but if they conflict with formal school enrollment or reduce school achievement, they can have negative consequences in the long run. Besides these, there are certainly also light forms of economic activity that can help to empower children and are largely unobjectionable (cf. Levison 2000). An adolescent who is earning pocket money by carrying out the Sunday morning newspapers, or helping out a few hours per week in the family business is hard to compare to a child crushing stones in a quarry, or working long hours in the fields. Therefore, in order to assess the potential severity of the problem of child work, two dimensions should be considered: 1. what activities do children perform, for how long, and under what conditions, and 2. what is the age of the children concerned. As to the latter, the United Nations Convention on the Rights of the Child, 1989 (Article 1), defines in general “every human being below the age of eighteen years” as a child. Arguably however, for different measures of child labor the use of different age limits might be instrumental. This chapter introduces the various regulatory and statistical concepts con- nected to child labor. The legal concepts are based on international regulatory

6 child labor and its regulatory environment 7

standards that are especially concerned with minimum age requirements and differentiate between the different forms of child work by their severity, while the most widely used statistical concept has a broader scope and captures economic activity of children in general. After addressing briefly the most relevant measurement issues, recent ILO estimates on the extent of these different forms of child labor are also presented. In many countries, child labor is endemic despite of the presence of minimum age standards prohibiting the work of children under specific age limits. The second part of this chapter is concerned with the questions, how reasonable it is to regulate child labor at all, and how likely it is that national child labor laws get enforced. From a normative perspective, ethical concerns can explain prohibitions of many forms of child labor, but under specific conditions a ban on child labor can also be desirable based on economic efficiency considerations only. Very often however, the use of alternative policy instruments would lead to better outcomes than a simple prohibition. From a positive perspective, there are different special interests that might explain child labor laws as well as their lack of enforcement. The relative importance of the various (especially economic) arguments is investigated in the light of the findings of recent theoretical and empirical literature on child labor.

2.2 Deﬁnitions and world-wide prevalence

The various legal concepts of child labor are defined in an internationally accepted framework by two main ILO Conventions on child labor: the ILO Min- imum Age Convention, 1973, No. 138, and the ILO Convention on the Worst Forms of Child Labor, 1999, No. 182. These conventions lay out a widely acknowledged legal basis for addressing child labor: 154 out of the 183 ILO member countries ratified the Minimum Age Convention until October 2009 and even more, 171 countries ratified the less restrictive Convention on the Worst Forms of Child Labor.

2.2.1 Diﬀerent deﬁnitions of child labor

The legal deﬁnition of child labor (as deﬁned by the ILO Minimum Age Con- vention, 1973, No. 138) includes any type of employment or work by children below the age of 14, and employment under hazardous conditions below the age of 18. The legal concept of child labor excludes however children of age 12 or older who are performing light work which is not harmful for their development and does not preclude school participation.1 National authorities are required to specify a threshold for the maximum weakly hours of work in

1 The above age limits of 12 and 14 apply to developing countries only, other countries are required to adhere to minimum age limits of 13 and 15. child labor and its regulatory environment 8

order to qualify it as light work.2 Article 5 of the Minimum Age Convention also allows for countries with insufficiently developed administrative facilities to exempt some sectors from the minimum age regulations, especially “family and small scale holdings producing for local consumption and not regularly employing hired workers”. The concept of the worst forms of child labor (ILO Convention on the Worst Forms of Child Labor, 1999, No. 182) defines forms of child labor whose eradication should receive absolute first priority. According to Article 3 of this Convention, the worst forms of child labor comprise:

(a) “all forms of slavery or practices similar to slavery, such as the sale and trafficking of children, debt bondage and serfdom, as well as forced or compulsory labor, including forced or compulsory recruit- ment of children for use in armed conflict; (b) the use, procuring or offering of a child for prostitution, for the production of pornography or for pornographic performances; (c) the use, procuring or offering of a child for illicit activities, in particular for the production and trafficking of drugs as defined in relevant international treaties; (d) work which, by its nature or the circumstances in which it is carried out, is likely to harm the health, safety or morals of children.”

The circumstances in Paragraphs (a) to (c) of Article 3 are also termed as the unconditionally worst forms of child labor; they describe clandestine activities that should be abolished and legally prosecuted. Paragraph (d) additionally includes various forms of so-called hazardous work, further speciﬁcation of which is left to national authorities. In its recommendation to national legislators, the ILO (Article 3 of the ILO Recommendation on Worst forms of Child Labor, 1999, No. 190) emphasizes that hazardous work conditions should especially include:

(a) “work which exposes children to physical, psychological or sexual abuse; (b) work underground, under water, at dangerous heights or in conﬁned spaces; (c) work with dangerous machinery, equipment and tools, or which involves the manual handling or transport of heavy loads;

2 Earlier minimum age conventions that pertain to specific sectors (industry, agriculture, trimmers and stokers, non-industrial employment, sea, fishermen, underground work) are more specific on hours limitations. For instance, ILO Minimum Age Convention (Non- Industrial Employment), 1937, No. 33 (Art. 3 (1)(c)) sets two hours per day (a total of 14 hours per week), on either school days or holidays, as the maximum for light work from 12 years of age. child labor and its regulatory environment 9

(d) work in an unhealthy environment which may, for example, expose children to hazardous substances, agents or processes, or to temper- atures, noise levels, or vibrations damaging to their health; (e) work under particularly diﬃcult conditions such as work for long hours or during the night or work where the child is unreasonably conﬁned to the premises of the employer.”

Thus, the concept of child labor can also include forms of work that are not classically considered as economic activity, like hazardous domestic work if it is performed under particularly diﬃcult conditions, for instance for very long hours. Economic activity of children is a broader statistical concept which includes all activities of children that are of economic value in the sense that their contribution enters the UN System of National Accounts (SNA). This—often also labeled as market work—includes all activities of children that contribute to household income, be it directly or indirectly: regular and occasional wage work paid in-cash or in-kind, as well as unpaid work performed both outside or within the family, like work in the family business or in the ﬁelds. It only excludes chores performed within the household like cooking, collecting wood and water, or looking after younger siblings; these chores are labeled domestic work. The concept of economic activity can be applied to any age category, and usually refers to children up to the age of 14 or 17.

2.2.2 Data issues

From all the different definitions presented above, economic activity of children is the only category for which there is systematic evidence for a wide range of countries. Data on economic activity of children is based on nationally representative household surveys, the availability of which has improved considerably over the last decades. The international comparability of such estimates will be further enhanced by current initiatives to harmonize survey instruments (cf. ILO 2008). However, major data problems still prevail: household questionnaires inherently suffer from misreporting as well as from large seasonal variation in the estimates. Obviously, misreporting (especially in the case of activities most harmful to children) will lead to underestimation of child workforce participation. A related problem is the often missing data on the economic activity of the youngest, since in many countries the general purpose household questionnaires do not record work of children below a certain age limit.3 This explains why until recently, internationally comparable data publications contained only information on economic activity of children

3 For instance, SUSENAS, the large scale, regionally representative household survey in Indonesia, which has been used in the empirical study in Chapter 6, records only economic activity of household members aged 10 years or older. child labor and its regulatory environment 10

in the age group 10 to 14 (cf. World Bank 2004). Newer international publications include economic activity of children aged 5 (or 7) to 14 years instead (cf. ILO 2006, World Bank 2009). This considerable improvement in data quality comes at the cost of a loss in backward comparability, as there is virtually no systematic information on economic activity of this broader age group before the year of 2000.4 The other data problem that leads to volatile estimates on child work lies in its often seasonal character: children are much more likely to work during peak seasons (like planting or harvest times in agriculture) as well as during school-breaks. In statistical terms, economic activity of children always refers to a specific recall period, most often it records economic activities during the last seven days before the interview. The timing of the interviews can thus make a large difference for the number of children found working, and compro- mise the comparability across countries but also across survey rounds within the same country.5 Nationally representative household surveys might also systematically underestimate child labor because they are missing information on street children by design, but also because child labor might be strongly clustered in specific geographic regions or sectors. In order to mitigate these problems, the “Statistical Information and Monitoring Programme on Child Labour” (SIMPOC) by the ILO targets also street children and requires the oversampling of urban households, and regions with especially high poverty incidence (Hilton 2003).

2.2.3 Global estimates of child labor

For its global and regional estimates on child labor, the ILO complements the information from the nationally representative household surveys with SIM- POC data. SIMPOC questionnaires are speciﬁcally designed to identify child laborers, and especially the worst forms of child labor, but they are not representative at the national level, and result in considerably higher estimates of economic activity than the national household surveys. According to the most recent ILO (2006) estimates, in 2004 around 190.7 million children in the age group of 5 to 14 were economically active world-

4 The ILO Database on Economically Active Population compiled data on economic activity of children aged 10 to 14 on a ten-year basis; these long time series were also published as part of the World Development Indicators (WDI) (cf. World Bank 2004) but they have been discontinued since then. Currently, the WDI present data on economic activity of children aged 7 to 14 instead, with earliest information coming from around 2000. The cross-country panel analysis in Chapter 5 relies on the earlier time series data on work of children aged 10 to 14 (World Bank 2004). 5 The empirical study in Chapter 3 on children in two North Indian provinces is among the few exceptions as it uses World Bank Living Standards Measurement Study (LSMS) data based on a one-year recall period. By this it avoids mismeasurement due to seasonality, but data problems still arise because of the potentially large recall errors. child labor and its regulatory environment 11

Table 2.1: Global numbers on diﬀerent forms of child work

Total numbers Incidence rate (million) (%) 2000 2004 2000 2004

Child population Age 5–14 1199.4 1206.5 100.0 100.0 Age 5–17 1531.4 1566.3 100.0 100.0 Economically active children Age 5–14 211.0 190.7 17.6 15.8 Age 5–17 351.9 317.4 23.0 20.3 Child laborers Age 5–14 186.3 165.8 15.5 13.7 Age 5–17 245.5 217.7 16.0 13.9 Child laborers in hazardous work Age 5–14 111.3 74.4 9.3 6.2 Age 5–17 170.5 126.3 11.1 8.1 Source: ILO (2006). For deﬁnitions see Section 2.2.1.

wide; about 165.8 million were considered to be child laborers, nearly half of whom were also estimated to work under hazardous conditions (cf. Table 2.1). This resulted in a global activity rate of children of 15.8%; for the wider age category of 5 to 17 years, the activity rate amounted to 20.3%. The global estimates also show that child labor is in rapid decline; between 2000 and 2004 there was an about 2 percentage points reduction in the economic activity of children, with an even larger reduction in hazardous child labor (3 percentage points).

Table 2.2: Economic activity of children (5–14 years) by region

Child population Economically Activity rate (mn) active(mn) (%) 2000 2004 2000 2004 2000 2004

Asia and the Paciﬁc 655.1 650.0 127.3 122.3 19.4 18.8 Latin America and the Caribbean 108.1 111.0 17.4 5.7 16.1 5.1 Sub-Saharan Africa 166.8 186.8 48.0 49.3 28.8 26.4 Other regions 269.3 258.8 18.3 13.4 6.8 5.2 World 1199.3 1206.6 211.0 190.7 17.6 15.8 Source: ILO (2006).

These rapid reductions in child labor are very unequally distributed across child labor and its regulatory environment 12

the continents (cf. Table 2.2). In Latin America and the Caribbean, economic activity rates of children dropped by about two thirds between 2000 and 2004. The resulting economic activity rate of about 5.1% is comparable to average economic activity of children in the heterogeneous group of “Other regions”, which includes developed countries, transition economies, and North Africa and the Middle East. At the same time, the region with highest child labor incidence (26.4% in 2004)—Sub-Saharan Africa—experienced only small relative improvements in child labor rates. According to ILO estimates, in Sub-Saharan Africa the total number of working children is still on the rise, which is due to the high rates of population growth. By contrast, the decrease in the number of child workers by around 5 million in Asia and the Paciﬁc went along with decreasing cohort size. The spectacular reductions in child labor in Latin America coincide with and are—at least partly—due to the success of broad policy initiatives like the Mexican PROGRESA/Oportunidades Program which provided cash transfers to the poor conditional on school attendance (cf. Parker, Rubalcava and Teruel 2007). These recent Latin American success stories emphasize the importance of a sensible combination of consequent regulation, and active pro-poor and pro-school policies.

2.3 Regulation of child labor

ILO conventions No. 138 and No. 182 define the framework of international child labor standards.6 These conventions have been ratified by most ILO member countries (cf. Section 2.2) and have been implemented in national labor laws. However, in many countries, despite of minimum age legislation, work of children younger than the minimum working age is extremely wide- spread. Table 2.3 presents current information both on economic activity rates of children (aged 7 to 14 years) and on the general minimum working age for 74 countries around the globe. Almost all of these countries ratified Convention No. 182, while 65 also ratified Convention No. 138.7 As Table 2.3 also shows, in almost all of these countries there is a minimum age legislation in place (Somalia being the only exception). Nevertheless, economic activity below the legally permitted age is far from negligible. This shows clearly that the issue of child labor is primarily not a matter of missing regulation but of missing enforcement. The evident lack of enforcement raises the question why child labor laws are not effectively implemented. Under what circumstances is regulation of child labor, or, equivalently, the effective enforcement of this regulation, desirable? Motivations behind imposing national or international child labor standards fall into three broad

6 This section is partly based on Kis-Katos and Schulze (2005). 7 Out of the 74 countries in Table 2.3, Ghana, Guinea Bissau, Haiti, Honduras, Liberia, and Uzbekistan did not ratify Convention No. 138, while India, Sierra Leone, and Somalia did not ratify either of the two Conventions. child labor and its regulatory environment 13

Table 2.3: Economic activity of children (7–14) and its regulation world-wide

Country (Year) Work Min. Country (Year) Work Min. (%) age (%) age Europe&CentralAsia Mid.East&N.Africa Albania (2000) 36.6 16 Egypt, Arab Rep. (2005) 7.9 15 Bosnia & Herzeg. (2005) 10.6 15 Iraq (2005) 14.7 15 Kazakhstan (2006) 3.6 16 Yemen, Rep. (1999) 13.1 15 Kyrgyz Republic (2004) 5.2 16 Morocco (1999) 13.2 15 Macedonia, FYR (2005) 11.8 15 Sub-Saharan Africa Moldova (2000) 33.5 18 Angola (2001) 30.1 14 Serbia (2005) 6.9 15 Benin (2006) 74.4 14 Turkey (1999) 4.5 15 Burkina Faso (2000) 50.0 15 Ukraine (2005) 17.3 16 Burundi (2000) 37.0 16 Uzbekistan (2005) 5.1 16 Cameroon (2001) 15.9 14 South Asia Central Afr. Rep. (2000) 67.0 14 Bangladesh (2006) 16.2 14 Chad (2004) 60.4 14 India (2005) 4.2 14 Congo, Dem. Rep. (2000) 39.8 15 Nepal (1999) 47.2 14 Congo, Rep. (2005) 30.1 16 Sri Lanka (1998) 17.0 14 Cˆote d’Ivoire (2006) 45.7 14 East Asia & Paciﬁc Ethiopia (2005) 56.0 14 Cambodia (2001) 52.3 15 Gambia, The (2005) 43.5 16 Indonesia (2000) 8.9 15 Ghana (2003) 6.0 15 Mongolia (2005) 12.4 16 Guinea (1994) 48.3 16 Philippines (2001) 13.3 15 Guinea-Bissau (2000) 67.5 14 Thailand (2005) 15.1 15 Lesotho (2000) 30.8 15 Latin America & Carib. Liberia (2007) 37.4 16 Argentina (2004) 15.1 15 Kenya (2000) 37.7 16 Bolivia (2005) 22.0 14 Madagascar (2001) 25.6 15 Brazil (2004) 7.0 16 Malawi (2006) 40.3 14 Chile (2003) 4.1 18 Mali (2006) 49.5 14 Colombia (2005) 4.0 14 Namibia (1999) 15.4 14 Costa Rica (2004) 5.7 15 Rwanda (2000) 33.1 16 Dominican Rep. (2005) 5.8 14 Senegal (2005) 18.5 15 Ecuador (2004) 4.3 15 Sierra Leone (2005) 62.7 15 El Salvador (2003) 12.7 14 Somalia (2006) 43.5 No Guatemala (2004) 16.8 14 South Africa (1999) 27.7 15 Haiti (2005) 33.4 15 Swaziland (2000) 11.2 15 Honduras (2004) 6.8 15 Tanzania (2001) 40.4 14 Nicaragua (2005) 10.1 14 Togo (2006) 39.6 15 Panama (2003) 5.1 14 Uganda (2006) 38.2 14 Paraguay (2005) 15.3 12 Zambia (2005) 47.9 15 Peru (2000) 24.1 14 Zimbabwe (1999) 14.3 13 Trinidad & Tobago (2000) 3.9 16 Venezuela (2005) 5.4 14 Notes: Economic activity rates of children (aged 7–14) are taken from the World De- velopment Indicators Online Database (World Bank 2009). The year for which the data apply is indicated in parentheses after the country’s name. Information on minimum age legislation is taken from USDOL (2008). Min. age denotes the general minimum working age; national laws make exceptions for work in agriculture, apprenticeships or other forms of light work, while further restrictions apply for hazardous/night/etc. employment. ’No’ stands for no minimum age legislation. child labor and its regulatory environment 14

categories. Child labor bans might be based on purely ethical considerations, they might originate in arguments of economic eﬃciency, but at the same time, imposing them also might serve special national or foreign interests.

2.3.1 Ethical considerations

The worst forms of child labor clearly violate universal human rights and thus must be eradicated through an effectively enforced ban. Among these forms are trafficking (1.2 million), forced and bonded labor (5.7 million), armed conflict (0.3 million), prostitution and pornography (1.8 million) and illicit activities (0.6 million). The numbers in parentheses refer to the children exposed to these conditions in 2000 according to estimates by ILO (2002b). Eradicating these forms of child labor cannot be argued to backfire on the families; a ban on hazardous child labor does not eliminate essential parts of family incomes, because other, non-hazardous, forms of child labor are still available. Imposing a ban on certain forms of child labor is necessarily a paternalistic act as it deprives children, their legal custodians (typically the parents) and the prospective employers of the freedom to enter into contractual labor relationships. Although it is easy to justify paternalism in the case of child protection also from a liberal perspective (e.g., Satz 2003), the extent of such acceptable interference depends on social norms, which typically differ across cultures and change in the course of development (Bhagwati 1995). For instance, in some African countries girls are married or children start apprenticeships at the age of 10; thus effectively marking the transition from a child towards a young adolescent. It is not easy to draw a sharp line between acceptable and unacceptable forms of child labor for each culture, but neither is it justified to impose a line established by Western norms.8 This is complicated as generally accepted norms on child labor themselves may change if child labor regulations are imposed. Preferences and norms may become endogenous through peer group effects: child labor may be acceptable if most people send their children to work but may be stigmatized after an effective ban if only a minority still sends their children to work (López-Calva 2003). The question arises what the relevant social norm is that justifies the extent of child labor regulation from a moral point of view: the existing norm or the potential one which would result from imposing the regulation. Obviously the answer implies a paternalistic consideration on which of the multiple equilibria is the desirable one.

8 That is not to say that such a line should not be drawn. The worst forms of child labor as deﬁned by the ILO Convention No. 182 violate human rights and are not culture-dependent. child labor and its regulatory environment 15

2.3.2 Eﬃciency considerations

Efficiency considerations refer to those cases in which child labor is socially acceptable, but not desired and not efficient. The motivation to disagree with child labor is thus not predominantly an ethical one—although everyone would agree that child labor is not desirable—but some variant of efficiency argument. These efficiency arguments involve multiple equilibria and market imperfections; they can be grouped according to the reasons for child labor: poverty, credit market imperfections, land and labor market imperfections, inefficient intrafamily bargaining resulting from incomplete altruism, and coordination failures leading to development traps of suboptimal education levels.

Poverty

The most obvious reason for child labor is that families need the additional income because they are too poor to survive or live decently otherwise. The link between poverty and child labor is empirically well established (starting with Rosenzweig 1981, cf. also Edmonds 2007). Country studies on child labor find a strong inverse relationship between child labor force participation rates and per capita GDP (e.g., Krueger 1996, Edmonds and Pavcnik 2005a). Studies based on household data often fail to find the same clear-cut results, but this is partly due to problems with empirical methodology: endogeneity of income to child labor, or inappropriate specifications (cf. Bhalotra and Tzannatos 2003). Obviously, the best policy would be to eliminate the underlying cause (rather than addressing only the symptom) through an integrated pro-poor policy approach. In the absence of such policies a ban on child labor will not be effective unless it improves the underlying income situation; it will only drive families deeper into misery or children in illegal and worse forms of child labor.9 Conditions under which a ban on child labor is efficiency enhancing are described by Basu and Van (1998). They model a situation of multiple equilibria in the labor market, in which a ban on child labor reduces effective labor supply to such an extent that the resulting increase in the wage rate more than compensates families for the loss in income from child labor. In such a situation a ban on child labor is the instrument of choice, however it is not clear if and when such a situation exists. In particular, their model has significant drawbacks: First, it is a closed economy model; if factor prices are largely determined through international trade on the world market, the room for fac-

9 The most commonly cited anecdotal evidence concerns the around 50.000 children in the Bangladeshi garment industry who got ﬁred in 1993 as a response to threats of trade sanctions from the US. Many of them were thought to have ended up in worse forms of child labor such as stone crushing or sexual work (UNICEF 1997). The ILO and UNICEF stepped in with various rehabilitation programs, but these interventions came only later (starting around 1996 and fully operational only in 2000), and many children and their families suﬀered in the meanwhile because of the prohibitions (UNICEF/ILO 2004). child labor and its regulatory environment 16

tor price movements is severely limited (e.g., Dixit 2000). The same is true if cross-border movements of workers will be triggered by a shortage of labor and thus incipiently rising wages, brought about by a child labor ban. The second criticism refers to the static nature of the model: Even if families are too poor to survive from adult income only, they might still send their children to school if their education is an investment that pays oﬀ in the future. It presupposes however, that they can ﬁnance this investment, and also are willing to do so. Thus, appropriate access to credit as well as parental trust and altruism play an important role.

Credit market imperfections

In this context, credit markets serve three functions: They allow parents to borrow for investment in human capital of their children, they serve as insurance device in order to smooth consumption in bad years and they allow investing in retirement schemes. If credit markets are imperfect, by denying poor families access to credit, family decisions on child labor may be distorted, resulting in an inefficient situation (Baland and Robinson 2000). In an ideal world, in which family members are completely altruistic toward each other, parents make optimal decisions for investment in the human capital of their children by investing into education to a point where the expected marginal return of investment equals the marginal cost of investment, which includes the opportunity costs of forgone child labor income.10 Parents may be too poor to afford educational expenses and forgo child labor income without appropriate compensation. Optimal decisions would still be possible if parents could take out a loan on behalf of their children and invest it in their education. That presupposes however that those parents have access to credit facilities and that they are confident that their children will pay back the loan.11 Thus, the deeper reason for child labor is not that families are poor, but that they are poor and have no appropriate access to credit (cf. Grootaert and Kanbur 1995). The poor typically have not sufficient collateral, which makes borrowing difficult if not impossible. A ban on child labor alone would not help unless such a ban would increase the factor rewards of adults to such an extent that the family could survive without child income. The chances for that however are bleak, as discussed above. Otherwise an effective ban on child labor would reduce family income thus making schooling even less affordable, and at the

10 Families are modeled as unitary decision-making entities maximizing family utility, which is the unweighted sum of individual utilities. (See Becker 1974 and Barro 1974 for pioneering contributions.) Note that this formulation, focusing on education as investment, ignores any consumption motive for education in the widest sense. 11 An additional issue may arise because those ﬁnancing the investment (the parents) are not identical to those reaping the beneﬁts (the children)—thus trust and altruism play an important role. child labor and its regulatory environment 17

same time reduce the opportunity costs of going to school (as children could not earn money in an alternative occupation). The net effect of substitution and income effects is a priori undetermined, but the very poor, who cannot afford to do without the income of their children, would be worse off.12 A ban could also drive child labor into illegality and thus lead to more clandestine and presumably much more unfavorable occupations and work conditions (cf. fn. 9). A much better policy mix would be providing access to credit, cheaper and better education facilities, and introducing pro-poor policies in general. The importance of access to credit for child labor has been emphasized by Dehejia and Gatti (2005) who find a significant negative relationship between access to credit, as measured by private bank credits to GDP, and child labor incidence in a cross country regression, controlling for a number of other variables such as GDP per capita, urbanization, and initial child labor. The relationship is especially pronounced for the subgroup of poor countries. Flug, Spilimbergo and Wachtenheim (1998) find in a cross-country analysis that financial development is positively associated with schooling. Credit constraints play an important role in explaining child labor as a reaction to income shocks; Beegle, Dehejia and Gatti (2006) find that transitory negative income shocks (bad harvests) increase the level of child labor for Tanzanian farmers, and that households with durable assets take out a loan in response to the shocks. Thus their increase in child labor is less. However, access to credit can also have adverse effects: when land and labor markets are imperfect, easing the credit constraints might also lead to more child labor.

Land and labor market imperfections

In rural societies land ownership is very closely related to household income. Land owners are clearly less poor, and if child labor is rooted in poverty, land ownership should be negatively correlated with child labor. Additionally, child labor should be decreasing with land holdings if land serves as collateral, and thus enables families to finance educational investments of their children.13 The empirical evidence however does not uniformly confirm these expectations. Bhalotra and Heady (2003) find child labor to be increasing in land use—after correcting for endogeneity of the latter—in both rural Pakistan and Ghana.14 The most likely explanation for this anomaly lies in a combination of labor and land market imperfections (Bhalotra and Heady 2003, Wydick 1999).

12 Many children combine work with schooling; if they cannot earn, they might not be able to cover school expenses. 13 Under many circumstances however, land holdings might not be collateralizable, be it because of missing legal entitlements (de Soto 2000) or because markets for land may be too thin (Beegle et al. 2006). In such cases credit access will not depend on land holdings. 14 See also Dumas (2007) for similar evidence from Burkina Faso. child labor and its regulatory environment 18

If landowners cannot efficiently hire workers because of moral hazard problems or periodical shortages of labor, they will rather employ their own children. A well functioning land market could solve the problem because landowners would be able to rent out land instead of employing child labor. However, when both land and labor markets are imperfect, the demand for child labor will be increasing with land size. Basu, Das and Dutta (2009) argue that the relationship between child labor and land ownership should take the shape of an inverse U: when land ownership increases above a threshold, the income effect will overturn the substitution effect from increasing labor demand. For a sample of North Indian households they estimate the turning point to be at around 3 acres of (inherited) land. Labor market imperfections can also contribute to higher levels of child labor if the household operates a family business (as documented by Edmonds and Turk 2004 for Vietnam or in Chapter 3 of this book for North India). Under land and labor market imperfections, access to credit can also have adverse effects. When hiring or supervising labor is difficult or costly, credit access will raise the demand for child labor by fostering economic activity within the household.15 This effect might be reinforced by low school quality, which results in relatively low returns to formal school education as compared to learning-by-doing while working in the family business. When credit is used to acquire business capital, it also raises the expected marginal returns to enterprise specific human capital and hence the incentives to let children work (Menon 2005). Under these circumstances, a child labor ban can be effective as it is less likely to drive children into more hazardous occupations. However, the ban will lead to suboptimal decisions from the household’s perspective, and the low visibility of work within the family will make enforcement of such a ban more difficult. In this case, addressing the land and labor market imperfections is clearly superior to legal restrictions on child labor. A word of caution is also implied in case of poverty reduction programs which transfer money to poor households. The beneficial effects of poverty reduction programs on child labor might be diminished (eventually even re- versed) if the money is used to buy productive assets which in itself is a good idea (Basu et al. 2009). The same applies to land redistribution programs targeting the poor. In order to avoid such unintended effects, these policies should be coupled with measures that increase the flexibility of land and labor markets.

15 The role of credit access for child labor supply and demand in Indonesian small scale manufacturing is highlighted in Chapter 4. child labor and its regulatory environment 19

Intrafamily bargaining and incomplete altruism

Suboptimal investment in education may also occur because those financing the investment (the parents), and those reaping the benefits (the children) are not completely altruistic towards each other; in other words they may not maximize joint utility. If parents are altruistic towards their children, but children are not altruistic towards their parents, optimal decisions are still possible if parents have enough resources to finance educational investment of their children. Parents who do not expect their children to share the returns to educational investment that accrue in the future will just reduce their planned bequest by the current value of their investment costs. Thereby, the targeted bequest of the parents will be transferred partly in form of human capital. This is the ’Rotten Kid Theorem’ described by Becker (1974); see Bergstrom (1989) for a formalization. If, however, the planned bequests are too low, so that the optimal educational investment cannot be financed by reducing bequests accordingly, altruistic parents are willing to take up a loan only if they can expect their children to pay it back from the returns to their educational investment.16 If they expect their children to be imperfectly altruistic, child labor may result. In this case education is inefficient also from a societal point of view: marginal productivity of investment is not equalized across households (Baland and Robinson 2000). Planned bequests can be too low because parents are too poor and cannot save sufficiently or because parents are (partly) selfish and want to consume their resources themselves. In other words, if altruistic parents have not enough resources to finance education and their own retirement payments, they will invest in children optimally only if they expect them to support them at old age. If children are expected to leave the parental household early, parents are less likely to benefit from their investment. Thus they will send their kids to work, thereby effectively transferring a part of the potential future income to the present in order to be able to appropriate it. From a family point of view this is suboptimal investment, created by incomplete altruism of the children. Conversely, if parents are—at least partly—selfish, they will invest subop- timally in their children even if they have enough resources to finance education and retirement.17 Note that from observable behavior of the parents we cannot conclude whether the inefficiency stems from incomplete altruism of the parents or from anticipated incomplete altruism of the children. If we could clearly identify missing parental altruism as the source of inefficiency, compulsory schooling would be the appropriate answer. A ban on child labor

16 If they could borrow against the future income of their children, consumption smoothing would be possible and educational investment would still be optimal. Such contracts are of course neither legal nor enforceable. 17 Again they would do that only if they expected their children to be partly selﬁsh as well because otherwise they could expect to receive their investment in current-value terms in the future. child labor and its regulatory environment 20

would not be the optimal solution also in this instance. The optimal solution would be complete intergenerational contracts, which will not be enforceable. A second best alternative would be some sort of social security system that taxed the educated adults and provided retirement benefits for the parents who sent their children to school coupled with incentives for schooling (Hazan and Berdugo 2002). The hypothesis of parental selfishness finds mixed empirical support. While historical studies argue that in the early twentieth century parents sent their children to work because they could not appropriate their future earnings (e.g., Parsons and Goldin 1989 for the US), Bhalotra (2004) finds that parents are altruistic in general, since if they have to rely more on child labor, they cut back their own consumption as well.

Human capital externalities and development traps

If human capital accumulation offers returns in excess of the private returns to education, it is underprovided in the market equilibrium and thus subsidies for educational investment are called for.18 Child labor bans are not the first best policy response to educational externalities because they do not internal- ize the externality that arises from education and not from the absence of child labor. A specific example of such externalities is an “educational development trap”: Firms will invest in more productive human capital-intensive technologies only if sufficient human capital is available; however as long as firms have not invested in skill-intensive technologies, returns to educational investment are too low to make education a profitable investment compared to the alternative of child labor (Dessy and Pallage 2001). This coordination failure can be addressed either by compulsory schooling or by all measures that make schooling more affordable and more attractive (see above). As there is compelling empirical evidence that better educated parents are significantly more likely to send their children to school, the efforts to improve education will have a lasting impact beyond the existing generation even in the absence of an educational underdevelopment trap (Brown, Deardorff and Stern 2003, Emerson and Souza 2003).

2.3.3 Alternatives to regulation?

Eﬃcient child protection should be concerned with the well being of children as the ultimate goal and not with child labor as such. Thus child labor should be regulated if it is physically or psychologically harmful to the children and a prohibition makes children better oﬀ. This however is not clear in all cases. If

18 There is no clear empirical evidence on the existence of human capital externalities (cf. Acemoglu and Angrist 2000). child labor and its regulatory environment 21

children have no alternative to working and a prohibition of child labor drives families deeper into poverty and children in illegal and worse occupations, child labor regulations are counterproductive. Rather unacceptable forms of child labor should be prohibited, and not child labor in general, and the underlying causes of child labor should be addressed. This should be coupled with a policy approach that systematically raises the relative attractiveness of schooling over child labor. If credit constraints are the issue, the provision of specific credits for educational investment linked to school performance (even without collateral) or more generally, adequately designed microcredit schemes for poor families who do not have access to commercial credits may significantly alleviate the incidence of child labor, as shown above. Policies that enhance land and labor market flexibility may also play an important role by making an increase of child labor in response to improved credit access or increased holdings of productive assets less likely. If problems of intrafamily bargaining and incomplete altruism matter, a tax scheme that holds children liable for their education when they become adults seems advisable (Brown et al. 2003). This could take the form of a pension system that taxed the educated adults and provided for the retirement of parents who have put their children through school. These measures should be complemented by strengthening incentives for parents to send their children to school by improving educational quality and lowering the costs of schooling. This includes educational credits, free school lunches and tuition waivers or scholarships for the poor, provision of transport to school, more present and better qualified teachers as well as other direct income subsidies to the families, which are tied to school attendance. This would make child labor relatively less attractive. Countries in Latin-America had already positive experiences with such financial incentive programs starting with the Mexican PROGRESA (cf. Schultz 2004, Parker et al. 2007), that served as a model for many other programs in Latin America, like the Bono de Desarrollo Humano in Ecuador (Edmonds and Schady 2008) or the Famil- ias in Acción in Colombia (Attanasio, Fitzsimons, Gomez, Lopez, Meghir and Mesnard 2006), etc. A sensible policy should also take into account that child labor and school enrollment are not mutually exclusive (cf. Ravallion and Wodon 2000).19 More flexible school schedules that allowed for combining work with schooling would further reduce the opportunity costs of going to school; moreover, child income

19 Ravallion and Wodon (2000) show that in the course of the Food-for-Education Program in Bangladesh, school enrollment increased signiﬁcantly although the income transfers were signiﬁcantly less than the forgone income by child labor. This increase in schooling was accompanied by a much smaller reduction in child labor, thus child leisure must have been reduced as well. child labor and its regulatory environment 22

might even be used to help to finance education.20 It must also be noted that some forms of child labor impart valuable skills on their own. Beegle, Dehejia and Gatti (2005) find that in Vietnam schooling brings higher returns than the work experience as a child only over a longer horizon (starting from the age of 30) while among the young, former child laborers have higher earnings than their counterparts who went to school.21 These findings suggest that access to long-term credits may be essential for reducing child labor. Obviously, a redistribution of income to the poor or development in general, equitably distributed, would also help to reduce the incidence of child labor (Ranjan 2001, Grootaert and Kanbur 1995, for empirical evidence cf. Rogers and Swinnerton 2001).

2.3.4 Political-economic motives

Although most countries have similarly strict child labor laws, the prevalence of child labor differs substantially across countries (Table 2.3). What explains these stark differences? What are the interests favoring and opposing effective enforcement of child labor laws? Historical evidence shows that child labor regulations have been promoted by union leaders, while capitalists and parents of large families opposed such regulations (Engerman 2003). The econometric study of Moehling (1999) shows that the introduction of child labor regulation in the US was preceded by a significant decline in child labor, so that these regulations could be viewed as codifying a trend rather than setting it. Doepke and Zilibotti (2005) model the different interests: Skilled workers have no interest in child labor laws as they make unskilled labor less abundant and reduce the wage premium of skilled labor over unskilled labor. For unskilled labor (the “working class”) the effect of child labor regulation goes both ways: it abolishes children’s contribution to the family income; at the same time it raises the wage rate of unskilled adult labor. The net effect depends on family size. Young families face a “quantity-quality trade off”: they decide to have either a large family with the children working or a small family and send their children to school. For the former the net effect of the child labor regulation will be negative; it will be positive for the latter. The position of working class families on child labor regulation depends thus on the fertility decision they have made when young. This ’lock-in effect’ characterizes the

20 But, even if child labor might enable children to get an education, it also comes at a cost: empirical studies document decreases in learning quality when children are combining school with work (cf. Heady 2003 for Ghana and Gunnarsson, Orazem and S´anchez 2006 for nine Latin American countries). 21 Further evidence on longer term losses from child labor is presented by Ilahi, Orazem and Sedlacek (2005) who ﬁnd that in a sample of Brazilian adults, those who started working before the age of 13 experience considerable losses of lifetime income, both because of lower human capital endowments and because of marriage sorting as they married individuals with low earnings potential. child labor and its regulatory environment 23

political equilibrium which could either be a situation with high fertility and child labor or with small families and children going to school. A change of equilibrium could be brought about by technical progress which raises returns to education, thereby inducing young adults to decrease family size and invest in education. In a related two sector model with a domestic and an export sector, Doepke and Zilibotti (2009) argue that imposing international labor standards in the export sector can undermine the domestic support for a total ban on child labor. The reason for this is that a ban in the export sector replaces children from the export sector where they compete with adult unskilled labor to the (family production based) domestic sector where adult and child labor are complementary. Thus, a ban in the domestic sector has no additional favorable wage effects for the unskilled adults, they will only experience decreases in the income of their working children. Hence, unskilled adults who send their children to school, and thus would have supported a child labor ban if no regulation were in place, will have less reasons to support a total ban after a partial ban has been introduced.22 Maffei, Raabe and Ursprung (2006) explain the different enforcement of child labor laws by the differing influence that the high skilled elite has in the political process. High skilled families lose from child labor regulation because generally increased education erodes the scarcity rents that the ruling clans (including their children) will have in the future. Using data for 103 developing countries they show that the degree of political repression (as a proxy for the power of the ruling—high skilled—elite) exerts significant negative influence on the degree of enforcement of child labor regulations (cf. Weiner 1991 for similar evidence on India). Krueger and Tjornhom Donohue (2005) model a human capital externality; they show that capital owners will be worse off as labor becomes scarcer while highly productive workers will clearly benefit as they do not send their children to work but benefit from higher wages. Low skilled workers will lose child income and gain from an increase in adult income.23 While high skilled labor

22 The result hinges on the assumption that production technology of the domestic sector is linear in unskilled labor (Doepke and Zilibotti 2009); this in turn determines the ratio of skilled to unskilled in the export sector. An initial ban in the export sector moves all children to the domestic sector, decreases the number of unskilled in the export sector and hence raises unskilled wages. Additionally, because of the decrease in future skill premia, less unskilled families will send their children to school than before. Thus, there are fewer unskilled families who are not hurt by a subsequent ban in the domestic sector. 23 The highly skilled gain from child labor regulation in this model although they lose from it in Doepke and Zilibotti’s (2005) and in Maffei et al.’s (2006) analysis. The reason for this difference is that Krueger and Tjornhom Donohue (2005) model labor in efficiency units as a homogeneous production factor that becomes scarcer as child labor is banned. Those with higher skills and/or productivity (i.e. having a higher labor endowment in efficiency units) gain more from the wage increase than low skilled workers. In the two other models, high and low skilled labor are two distinct production factors, the relative wages of which are child labor and its regulatory environment 24

will benefit from child labor regulation, the working class and the general population will benefit from free education due to the externality. All these models focus on the factor price effects stemming from changes in relative factor endowments, either in the present or in the future. In a small open economy, however, such factor price movements are largely determined by the world market through international trade or international factor movements. Yet, if child labor regulation does not alter goods or factor prices, only parents who send their children to work have an interest against regulation. In a large open economy all factors that are imperfect substitutes to child labor will gain from child labor laws through an improvement of the terms of trade. Hence, the more open and the larger the economy, the less opposition there is against child labor laws (cf. Shelburne 2001).

2.4 Conclusion

The first part of this chapter outlined the main definitions on various forms of child labor, distinguishing between the legal concepts as defined by the ILO (“child labor”, the “worst forms of child labor”, the “unconditionally worst forms of child labor”, or “hazardous child labor”), and the statistical concept of economic activity (or market work) by children. While the more nuanced legal concepts are more useful to assess the severity of the child labor problem world-wide, reasons of data availability and statistical consistency point towards relying on more general information on the economic activity of children. Economic activity of children is also not free of measurement problems (mainly due to mismeasurement and seasonality issues), but it is the only category with data that is comparable across countries as well as within countries over time. This is the reason why, when investigating the determinants of child labor or the effects of trade liberalization on child work, this book is always focusing on child work in general. By doing so, we do not learn about the specific causes of the “worst forms of child labor”, although they are of central policy relevance. Instead, the empirical analysis sheds light on the overall trade-off faced by the families when deciding to send their children to work. However, there is some evidence (ILO 2006) that the general trend of decreasing child work is also mirrored in reductions of its worst forms. The second part of this chapter addressed the regulatory environment of child labor and discussed under what circumstances is regulation the appropriate policy response to child labor. Since most countries ratified the two main ILO conventions on child labor, and have also minimum age standards in place, the main issue is clearly not regulation itself, but the effective enforcement of determined by the factor endowment ratio. Child labor regulation decreases the unskilled labor supply only, thereby reducing the wage premium of skilled workers. child labor and its regulatory environment 25

the legal norms. The arguments in favor of an effectively enforced regulation can rely on ethical, economic, and political economy considerations. From an ethical standpoint, certain types of child labor should be eradicated by all means. These include the most exploitative and degrading forms of child labor, i.e. children as sex workers, soldiers, and working under severely hazardous conditions (in mining, stone crushing or working with dangerous substances, etc.). Regulations against such abusive child labor are necessary and should be enforced unconditionally. The appropriate response to other types of economic activity by children depends on the underlying causes of child labor. When parents are altruistic, a ban on child labor in all likelihood will make families worse off, either by sending children to more clandestine and worse forms of child labor or by threatening the survival of families. Rather, the underlying causes of child labor should be addressed by making credits available to the poor, by increasing the flexibility of local labor markets, and by providing incentives for schooling in form of better and more affordable schools, through school meals, scholarships etc. Underdevelopment traps of economy-wide undereducation and skill-extensive technology can be overcome by the aforementioned measures and by compulsory schooling. If altruism is an issue, child labor is best addressed by incentives for schooling coupled with a social security system that taxes especially the educated adults and provides pensions to the old. From a political economy perspective, domestic regulation of child labor will be supported by those unskilled who do not send their children to work as they are less reliant on child labor income but benefit from a decrease in the supply of unskilled labor and a subsequent change in skill premia. However, if we want that child labor laws, once introduced and enforced, also improve the welfare of the children concerned, regulation has to be complemented with other policies that make sure that these children receive an education instead. chapter 3

Determinants of child labor supply: The work-school trade-oﬀ in India

Abstract

Based on a rural sample of North Indian children and adolescents, this chapter addresses the determinants of participation in work and schooling. The empirical model includes market and domestic work as separate alternatives to schooling in a trivariate probit framework, allowing also for combinations of these activities as well as idleness. This differentiation sheds new light on gender differences in the work-school trade-off in India. The results emphasize the importance of economic opportunities that affect the work-schooling trade-offs for both sexes, and make specialization in schooling less likely. By contrast, the gender non-specific activities (girls’ market and boys’ domestic work) are found to be more closely related to cultural norms.

3.1 Introduction

Household decisions on whether to send children to school or to let them work are clearly interdependent. Child labor and schooling are conflicting alternatives, although they are not exclusive and can often be combined to some extent. Work performed by children reduces the time spent in school,1 as well as their educational attainment (Beegle et al. 2006, Heady 2003, Orazem and Gunnarsson 2004). Models that consider work and schooling decisions in a joint framework can help to better understand their interrelationship and the work-school trade-off. But simple approaches might fail to explore the gender aspects of this trade-off to a full extent because of the large heterogeneity of work, and especially the differences between market and domestic work. Studies investigating the work-school trade-off in a joint framework take two approaches with respect to different forms of work, both of which pose

1 Substitution may be less than perfect: Ravallion and Wodon (2000) ﬁnd that hours of child work decrease by less than the increase in their school participation as a response to a food subsidy in Bangladesh.

26 determinants of child labor supply 27

some problems.2 Those that concentrate on market work only leave domestic work out of the choice set altogether (e.g., Maitra and Ray 2002, Pal 2004, Ba- colod and Ranjan 2008). This underestimates the extent of child labor since, world-wide, most children work for their families, often performing domestic chores (Edmonds and Pavcnik 2005a). It also disregards the gender aspects of child labor as it is mainly girls who perform domestic work.3 Most importantly, this procedure implicitly constrains the determinants of domestic work and idleness to be the same, which masks the underlying differences in the economic incentives for working in the household or staying idle.4 The second approach followed by empirical studies is to treat market work and domestic chores jointly as opposed to school attendance (e.g., Ravallion and Wodon 2000, Ersado 2005, Cigno and Rosati 2005, Chamarbagwala and Tchernis 2010). In countries with relatively large gender disparities, such as the South Asian countries, this captures relatively well the work-school trade-off for boys, but it disregards the inherent differences between girls’ domestic and market work by assuming that both are driven by the same determinants. In the presence of systematic differences between the determinants of market and domestic work as well as domestic work and idleness, it is more appropriate to address market work, domestic work, and schooling as separate alternatives. Simultaneous estimation of the choices between market and domestic work and school attendance can also help to investigate the determinants of inactivity, specialization in a specific activity, or combination of multiple activities. Such an approach is especially well suited to capture the gender differences in the determinants of household decisions on work and school. The North Indian case offers valuable insights into the gender aspects of the work-school trade-off since gender differences in work and schooling are especially large for children in this region. School attendance of girls is considerably lower, and it conflicts mostly with their domestic duties whereas most boys are involved in either school or market work. However, there is no total specialization by gender; a still considerable number of girls works for the market, while some boys are also performing domestic chores. This allows me to investigate the gender differences between the trade-off of the gender-specific activities with schooling, but also participation in the gender unspecific activity.

2 The literature on the trade-off between work and schooling of children started with Canagarajah and Coulombe (1997), Cartwright (1998), Grootaert (1998), Nielsen (1998); for the first study on India see Duraisamy (2000). 3 This point is also emphasized by Levison, Moe and Knaul (2001) who compare the trade-off between school and market work or school and all types of work in Mexico, and find that the first procedure underestimates the trade-off for girls to a large extent. 4 For instance, in the often applied multinomial logit framework (addressing the school only, work only, combine school and work, and stay idle alternatives, e.g., Levison et al. 2001, Maitra and Ray 2002, Bacolod and Ranjan 2008), disregarding domestic work might violate the “Independence of Irrelevant Alternatives” assumption, which presupposes that the relative probabilities of any two alternative categories are not influenced by the existence of other alternatives. determinants of child labor supply 28

The present analysis is based on cross-sectional data from the 1997/98 Survey of Living Conditions of two North Indian states, Uttar Pradesh and Bihar (World Bank LSMS). It estimates jointly participation of 10 to 17 year old children in market work, domestic work and schooling, separately for the two sexes. The regressions condition participation in work and schooling on individual characteristics, asset ownership, household composition and educational attainment, as well as village variables on costs of schooling, proxies for cultural norms, and median wages. The participation equations form a trivariate probit model which is estimated by the method of simulated maximum likelihood. Marginal effects on the joint probabilities of different occupational choices are calculated by a parametric bootstrap procedure. By incorporating schooling, market and domestic work as separate outcomes in a multivariate framework, a more detailed picture on the determinants of child labor emerges. The results show systematic differences between the determinants of market and domestic work as well as those of domestic work and idleness for both boys and girls. Such differences would be blurred when lumping market and domestic work together and could not be captured without including domestic work explicitly in the choice set. These insights are the main contribution of this study. A number of factors traditionally linked to child labor, like tastes for education (proxied by household literacy), or the distance to the nearest school, are mainly related to the trade-offs between the gender-specific work activity and schooling. By contrast, the economic incentive effects of small business ownership or higher adult wages in the village do not differ strongly between boys and girls: they increase participation in market and domestic work for both sexes. Higher school costs in the village also increase the probability that both boys and girls engage in market work. Cultural norms influence participation in the less typical work activity to a larger extent. Caste variables and female workforce participation are relatively strongly related to girls’ involvement in market work but less so in domestic work, while these variables are among the few relevant determinants of boys’ domestic work. Due to computational difficulties, multivariate models have not been typically applied in the analysis of child labor. The only exception to this is the work of Kambhampati and Rajan (2008) who estimate a multivariate probit model for Indian girls, and conclude that country-wide differences in the work-schooling trade-off of female children are driven by differences in cultural norms within the patriarchal kinship systems. This study differs from their work in several aspects. It concentrates on children from two North Indian states, and compares the trade-off between market work, domestic work, and school attendance for both sexes. More importantly, it specifies the probabilities of participation in activities in a more direct way and computes marginal effects of the determinants on trivariate probabilities of specific activities and their combinations. As a result, the gender differences in the determinants of determinants of child labor supply 29

specializing in one given activity or combining various activities can also be explicitly addressed. The remainder of this chapter is structured as follows. Section 3.2 shortly summarizes the evidence on child labor and schooling in India; this is followed by a stylized model on children’s occupational decisions. Section 3.3 describes the data, Section 3.4 outlines the estimation model. Results are presented and discussed in Section 3.5, while Section 3.6 concludes.

3.2 Determinants of child labor and schooling

3.2.1 Child labor and schooling in India

Child labor and schooling can be seen as resulting from a human capital investment decision by the parents, who decide both on family consumption and the time use of their children for schooling, labor, and leisure subject to their income, asset ownership, and preferences. If financial markets were perfect, participation in schooling and child labor would largely depend on the expected net returns to education, which also include the opportunity costs of not working as a child.5 But since credit (and insurance) constraints are binding for the rural poor, poverty is a crucial determinant of the work-schooling trade-off (Basu and Van 1998, Baland and Robinson 2000).6 Thus, regional development affects the work-school trade-off both through reducing poverty and increasing current and future labor demand; the net effect on child work and schooling depends on the changes in the net expected returns to education.7 As imperfect labor markets make hiring and supervising additional labor harder, land ownership has also been shown to increase child labor by raising the marginal productivity of children within the family (Bhalotra and Heady 2003, Basu et al. 2009). Idleness results in this conceptual framework from low net returns to human capital (due to low school availability and quality) paired with too low productivity of children (due to missing productive assets) (Cigno and Rosati 2005).

5 Several studies show that schooling responds to the returns to education in India: Schooling increased strongly when technological change in agriculture raised its returns (Foster and Rosenzweig 1996), while the arising labor demand effects tended to decrease schooling of children from landless households (Foster and Rosenzweig 2004). Kochar (2004) also finds that the probability of rural boys completing middle school in India increases with urban wage growth of the higher skilled (but decreases with middle skilled wages). 6 Schooling in rural India decreases when households are hit by adverse income shocks, even more so if these shocks are not anticipated (Jacoby and Skoufias 1997). 7 Kambhampati and Rajan (2006) document that in India state level growth performance went along with increasing market work participation and reducing school enrollment of children which they attribute to labor demand effects. By contrast, Edmonds, Pavcnik and Topalova (2007) interpret the smaller increases in schooling in those rural Indian districts that were more exposed to trade liberalization as a sign that trade liberalization failed to reduce poverty. determinants of child labor supply 30

Gender inequalities in human capital investment are especially large in India. They are most fundamentally reflected in the survival gap resulting from discrimination against girls in the allocation of food and health care (Sen 1992), especially when families are hit by adverse shocks (Behrman 1988, Rose 1999). They have been argued to result from lower market returns to human capital of females (Rosenzweig and Schultz 1982), but they are also related to the patrilocal family structure (Kambhampati and Rajan 2008): As girls leave the family upon marriage, the future benefits from a girl’s education do not accrue to her own parents; this further reduces the incentives to invest in girls’ schooling.8 The large gender difference in work activities, with boys mostly performing market while girls specializing in domestic work, reflects both gender specific productivity differences in different tasks, as well as parental preferences and cultural norms. Girls are thought to be more suited to care for younger siblings; thus the strength of their comparative advantage in household production will also depend on family composition and birth spacing (Edmonds 2006). Ad- ditionally, in the North Indian patriarchic kinship system girls are often shel- tered from outside influences (especially if they are higher cast or Muslim), which also raises the perceived costs of education and market work for girls (Kambhampati and Rajan 2008). Under such circumstances, market work of girls should be more prevalent in families that own productive assets or operate a family business, since this enables girls to perform market related work without leaving home. Norms with respect to the females’ role can be expected to matter for both girls’ labor market and school participation, and might also be behind some of the unexplained spatial variation in the children propensity to work and study (Chamarbagwala and Tchernis 2010).

3.2.2 A stylized model of occupation choice

The joint decisions on child labor and school attendance can be described in a stylized two-period framework (t = 1, 2) where a unitary decision-maker decides about household consumption and time use of children.9 The number of adults and children is normalized to one each, and general household and community characteristics are depicted by the vectors Θ and Ω respectively. For expositional ease, work decisions in the second period are suppressed, and it is assumed that both adults and children perform full time market work in period two.10 The time spent with school in period one is denoted by S,

8 Dr`eze and Kingdon (2001) also ﬁnd that the schooling of Indian girls depends stronger on monetary incentives (school meals) and school quality variables than that of boys. 9 The decision maker is assumed to weight utilities of children and adults equally, and for simplicity, future consumption is not discounted. 10 The present model builds upon the framework of Bhalotra and Heady (2003) and extends it with domestic work while abstracting from decisions on land tenancy and hiring labor. determinants of child labor supply 31

hours of market related work by Lc1 and Lp1, household work by Hc1 and Hp1 where the subscript c stands for children and p for parents. Parents maximize household utility over the two periods by considering the utility of consumption, the disutility of labor, and the utility of schooling:

max U1(C1,Lc1,Hc1,Lp1,Hp1,S;Θ, Ω) + U2(C2;Θ, Ω) (3.1) Ct,Lc,p1,Hc,p1,S

The intertemporal utility maximization problem is subject to budget and time constraints. The household’s productive assets A0 are assumed to be exoge- nously given at the beginning of the first period; monetary assets are denoted by B0 (bequest) and B1 (borrowing). Credit market imperfections are taken into account using the costs of borrowing g(A0,B1;Θ, Ω), which decrease with the collateral of physical assets, and depend on individual and community characteristics. Household income is generated by a typical household production function ft with decreasing returns. Labor inputs of children and adults directly contribute to the first period income f1, while second period income consists of household production f2 and the child’s second period earnings wc2. These depend not only on education, but also on the labor experience in the first period, and on local labor demand captured by Ω. The costs of schooling P (S; Ω) are convex in the amount of schooling S and depend on school availability and quality in the community. Household chores are defined as a necessary activity (amount H¯ ) which is conditional on household characteristics Θ and produces no direct income. However, if children are helping with domestic work, they raise the earning capacity of the adults in the first period. Parents maximize utility (3.1) subject to the following income and time constraints:

C1 = f1(A0,Lp1,Lc1;Θ, Ω) − P (S;Ω)+ B0 + B1 (3.2a)

C2 = f2(A0,Lp2;Θ, Ω) + wc2(S,Lc1; Ω) − g(B1, A0;Θ, Ω) (3.2b)

Hc1 + Hp1 ≥ H¯ (Θ) Hc1 + S + Lc1 ≤ 1 Hp1 + Lp1 ≤ 1 (3.2c)

They divide their own time and that of their children (both normalized to unity) between market work, household chores, school (children only), and leisure. Consumption, income and time use of parents and children are endogenously determined as a function of asset ownership, and household and community characteristics. The shadow price of income in period one and two are denoted by λ1 and λ2, and the shadow price of the constraint on household work H¯ by λ3. The following ﬁrst order conditions are directly related to the determinants of child labor supply 32

time use of children:11

∂U1/∂C1 = λ1, ∂U2/∂C2 = λ2, (∂g/∂B1)λ2 = λ1 (3.3a)

∂U1/∂Lc1 + λ1∂f1/∂Lc1 + λ2∂wc2/∂Lc1 ≤ 0 (3.3b)

∂U1/∂S − λ1∂P/∂S + λ2∂wc2/∂S ≤ 0 (3.3c)

∂U1/∂Hc1 + λ3 ≤ 0 (3.3d)

As a result, the marginal utility of consumption over the two periods cannot be completely equalized; the difference between marginal utilities depends on the marginal costs of borrowing, reflecting credit market imperfections (equation 3.3a). Children perform market work if equation (3.3b) holds with equality, that is if the sum of the value of marginal product of labor performed by children and their returns from learning-by-doing are at least as high as the disutility of child labor. Children attend school if the marginal utility of education and the returns to schooling are not smaller than the marginal costs of schooling, that is if equation (3.3c) is binding. Children help with domestic chores when the marginal disutility of this work is not larger than the shadow price of parental time (3.3d is binding). This in turn is determined by parental preferences and by the value of marginal returns to the alternative use of parental time. This relatively simple framework has numerous implications for the work- school trade-off of children. It acknowledges that children can also accumu- late farm or business-specific knowledge which reduces the relative returns to schooling and shifts the trade-off towards work. Ownership of productive assets A0 plays an ambiguous role. It raises present and future income and also reduces the costs of borrowing which allows for a better equalization of marginal utilities between present and future consumption (see equation 3.3a). However, it also raises the marginal product of child and adult labor, making market work by children and parents more profitable. This incentive effect affects not only market work by children, but also their domestic duties if the marginal value of parental time increases with the assets.12 The relative quality of human capital accumulated by working might also depend on future employment perspectives: Children of farm and business owners who will in- herit the productive assets benefit more from learning-by-doing. Another part of the incentive effects is due to differences in current and prospective labor market opportunities. Labor market outcomes like adult labor force participation, unemployment, or wages might reflect current labor demand and affect both household income and the incentives to send a child to work.

11 For simplicity, only decisions are considered that are interior with respect to leisure, i.e., the time constraints in (3.2c) are assumed to hold with strict inequality. 12 The eﬀect arises if asset and labor markets function imperfectly, i.e., if land cannot be easily sold or bought, and hired labor is imperfect substitute for child work (Bhalotra and Heady 2003). determinants of child labor supply 33

If returns to formal and informal education ∂wc2/∂S and ∂wc2/∂Lc1 are lower for girls, as is generally argued for India (e.g., Kingdon 1998), girls will be less likely to go to school, or to perform market work. When social norms restrict girls’ work outside the home, the marginal disutility of market work becomes also gender specific. Additionally, as Indian girls usually leave the family upon marriage, the benefits from their formal and informal education cannot be appropriated by the family (Kambhampati and Rajan 2004). This further reduces the incentives to invest in the education of girls. These effects all contribute to girls specializing in household chores. The main difference between children who perform domestic work and those who stay idle results mainly from two factors: all else being equal, children are more likely to stay idle if household income is higher, and if economic opportunities are lower, that is when the marginal value of parental time is lower.

3.3 Data and main variables

3.3.1 Data

The analysis is based on data from the “Survey of Living Conditions” of two North Indian states, Uttar Pradesh and Bihar. The survey was carried out between December 1997 and March 1998 as a part of the World Bank Living Standards Measurement Study (LSMS) series. The quantitative part of the survey is comprised of a household questionnaire and a village level data set with community-level characteristics. It contains data from 120 villages in two selected regions of Uttar Pradesh (Eastern and Southern) and two of Bihar (Northern and Central) where 2250 households were interviewed.13

3.3.2 Activities of children and adolescents

The survey presents detailed socio-economic information about the households and their members, and records the main economic activities of each family member for the previous 12 months. As the data set does not contain information on time use for all types of work, the three dependent variables (market work/domestic work/schooling) are deﬁned as binary indicator variables that show whether a child participated in that activity during the previous year.14 Market work includes all directly productive activities of children, performed within or outside of the household (wage labor, unpaid work on the

13 The survey is not representative of the two Northern Indian states in a strict sense, but it is a very useful source to study behavior of poor households, which are overrepresented in the sample. 14 Due to the relatively long recollection period, this data set is better suited to record seasonally recurring activities (like agricultural child labor). determinants of child labor supply 34

family farm, or in the family business, etc.). Children working in home production are thus also counted as market workers.15 Domestic work includes household chores (i.e. cleaning, cooking, or looking after younger siblings) but also fetching water, collecting ﬁrewood, and foraging. Children are classiﬁed as in school if this has been one of their reported main economic activities for the previous year or if they actually attended school within the last week before the survey.

Table 3.1: Activities of children (10–17) by gender (%)

Male Female Total

One occupation only 86.2 86.3 86.3 Market work 15.9 6.8 11.9 Domestic work 2.4 36.9 17.9 In school 67.8 42.6 56.6 Combine 4.6 7.7 6.0 Market and domestic work 0.5 4.3 2.2 Market work and school 2.9 1.0 2.1 Domestic work and school 1.1 1.9 1.5 All types 0.1 0.5 0.3 No occupation 9.3 6.0 7.8

Total 100.0 100.0 100.0 N 1345 1067 2412 Source: World Bank LSMS, Uttar Pradesh & Bihar 1997/98.

The data set records all major activities of children over the recollection period. This also allows for the fact that work and schooling need not be mutually exclusive, and enables us to consider explicitly those children who combine diﬀerent activities. In the sample, 4.1% of boys and 3.4% of girls are reported to be combining work with school attendance, while girls also combine market and domestic work to some extent (cf. Table 3.1). By contrast, there is also a relatively large number of children who are reported as idle.16 The distribution of children’s activities within the sample reveals a clear gender pattern; while 51.5% of girls aged 10–17 are working, only 22.9% of the boys of the same age perform any kind of work (cf. Table 3.1). The main source of this gender gap lies not in market activities (12.7% of girls and 19.4% of boys perform market work), but in domestic chores which are performed by girls but not by boys (43.6% of girls as compared to 4.1% of boys). A signiﬁcantly higher

15 This broad deﬁnition of market work is useful as it captures better the economic contribution of children (cf. Chapter 2). Globally only a relatively small fraction of children works for wages; most children are employed by their own parents and are working on family farms or in family businesses (Edmonds and Pavcnik 2005a). 16 The phenomenon of idle children is quite common to surveys conducted in India, and has been argued to result from both under-reporting of child work, and low productivity of child labor (Cigno and Rosati 2000). determinants of child labor supply 35

proportion of boys (71.9%) than girls (45.9%) in this age group is enrolled in school.17 The relatively wide age range considered in this study (10 to 17 years) includes both children and adolescents until the age of leaving school.18 The use of this age span is justiﬁable because the age of 18 years marks one of the two major breaks in work participation and school attendance of children (the other break is occurring at the age of 15, cf. Table 3.2).19

Table 3.2: Activities of children (10–17) by age (%)

Working In school At work Being Age N only only & school idle

10 525 19.4 65.5 2.9 12.2 11 184 16.3 73.4 2.2 8.2 12 443 27.1 63.0 4.5 5.4 13 245 22.0 66.5 4.5 6.9 14 244 30.3 60.3 4.1 5.3 15 318 45.9 42.1 4.4 7.6 16 283 55.5 33.2 3.9 7.4 17 143 54.6 36.4 3.5 5.6 18 384 66.4 20.3 3.1 10.2 10–17 2412 31.9 56.5 3.8 7.8 Males 1345 18.8 67.8 4.1 9.3 Females 1067 48.1 42.6 3.4 6.0 Source: World Bank LSMS, Uttar Pradesh & Bihar 1997/98.

3.3.3 Explanatory variables

The vector of explanatory variables includes personal characteristics, socio- economic characteristics of the household, and village level controls (cf. Table 3.3 for descriptive statistics and Table A.2 in the Appendix for deﬁnitions). At the household level, information on household income and asset ownership, family business, educational attainment, and household structure is used. Village level controls include measures of labor market outcomes and median

17 The substantial difference between the numbers of girls and boys within the sample (with a female to male ratio of 0.81) confirms the presence of a gender survival gap (cf. Section 3.2.1). The two North Indian states in the sample are both severely affected by this issue: in 2001 the average sex ratio of females to males was 0.898 in Uttar Pradesh and 0.919 in Bihar as compared to the Indian average of 0.933 (Census of India 2001). 18 Although economic activities of younger children would deserve special attention, the household questionnaire records only economic activities of children aged 10 years or older. 19 The data set does include primary as well as secondary age school children. For the age-grade distribution see Table A.1 in the Appendix. determinants of child labor supply 36

Table 3.3: Descriptive statistics

Females Males Total No. obs 1067 1318 2385 Variable Mean St. dev. Mean St. dev. Min. Max.

Market work* 0.13 0.34 0.22 0.41 0 1 Domestic work* 0.44 0.50 0.04 0.20 0 1 Student* 0.46 0.50 0.72 0.45 0 1 Household income 8.83 0.77 8.79 0.77 5.22 11.20 Family business 0.39 0.49 0.37 0.48 0 1 Marginal land* 0.39 0.49 0.37 0.48 0 1 Small land* 0.28 0.45 0.31 0.46 0 1 Large land* 0.09 0.28 0.10 0.31 0 1 Female literacy rate 0.23 0.38 0.19 0.34 0 1 Male literacy rate 0.59 0.44 0.60 0.44 0 1 Share infants 0.33 0.38 0.27 0.33 0 2 Share children 0.33 0.34 0.31 0.33 0 2 Share boys 0.24 0.29 0.56 0.36 0 3 Share girls 0.53 0.34 0.19 0.27 0 3 Share elderly 0.04 0.11 0.05 0.11 0 1 Share females 0.48 0.13 0.48 0.13 0 1 Birth order 1.88 1.03 1.90 0.97 1 7 Married* 0.11 0.32 0.02 0.15 0 1 Lower castes* 0.49 0.50 0.52 0.50 0 1 Scheduled castes* 0.25 0.43 0.26 0.44 0 1 Muslim* 0.11 0.32 0.09 0.28 0 1 Time to school 3.20 2.25 3.30 2.26 0 15 School costs (v.) 0.30 0.19 0.30 0.19 0.02 1.86 Female work low* 0.31 0.46 0.29 0.45 0 1 Female work high* 0.26 0.44 0.29 0.45 0 1 Median wages (v.) 3.15 0.94 3.12 0.99 0.40 6.50 Thresher owners* 0.10 0.30 0.11 0.32 0 2 Tractor owners* 0.03 0.18 0.04 0.19 0 1 Value of transfers 0.29 1.11 0.24 1.00 0 13 Rooms per adult 1.01 0.72 1.01 0.70 0 10.5 Notes: Indicator variables are marked by asterisks.

school costs. All regressions control for the marital status of a young person, and include a full set of age dummies.

Controls for income and incentive eﬀects

As outlined earlier, there is a clear theoretical linkage between household income and child labor. If child leisure is a normal good, children from wealthier households will be less likely to work. Additionally, when schooling investments are suboptimal due to credit constraints, a rise in income will shift the work-school trade-oﬀ in favor of more schooling. To the extent that working adolescents contribute to household income, simple estimates of the income determinants of child labor supply 37

effect can be expected to be biased downwards. The estimates of the income effect are potentially also attenuated by measurement error in income. Instru- ments for income that are less likely to be influenced by the work of children can help to mitigate this downward bias as income measures are typically less reliable than wealth measures. Yearly household income includes income from four major sources: the wage income of all family members (except for children), the income from running a business, the market value of own agricultural production over the past year, and the value of pensions/transfers received from outside of the household. The variable Household income measures the natural logarithm of the yearly household income per adult household member (aged 18 or above). For each occupation, household income is instrumented by two of the following variables: an indicator of tractor (or thresher) ownership within the household, the number of rooms per adult the household lives in, and the value of transfers from outside of the household (in per adult terms). Machine ownership is used alongside with transfers in regressions for boys and with the dwelling size in girls’ regressions (cf. Table 3.4). All instruments are strong predictors of household income (with F-statistics between 22 and 55), and tests of overidentifying restrictions support their exclusion from the univariate IV-probit regressions of the three dependent variables (cf. Table 3.5). These instruments are thus less influenced by the work of children, while affecting child work and schooling measurably only through wealth effects.20 Household income rises with tractor ownership, the value of received transfers, and the size of the dwelling the household lives in (cf. Table 3.4). In households with large machines, children are less likely to work and more likely to go to school but the major part of this correlation can be explained by a wealth effect. Machines can both substitute for child work but also com- plement it in agricultural production; the sign of this effect is unclear, but it is clearly insignificant in both subsamples, as the tractor (or thresher) dummy is practically orthogonal to the residual of each univariate participation equation. The relative size of the household dwelling is a proxy of household wealth in a broader sense, and is less likely to be determined by child labor. Household income also increases with the value of income transfers (most likely from mi- grant household members). Univariate exogeneity tests or the direct test on the included residual support the use of instruments in all but girls’ schooling equations (cf. Table 3.5); thus, income is not instrumented in this latter case. Ownership of large household assets (land), or the presence of a small scale business within the household can be expected to have both income and substitution effects on child labor. Both land holdings and business activities

20 All variables used as instruments might conceivably affect child labor through channels other than the wealth effect. Nevertheless, the income effects of these variables strongly dominate eventual substitution effects. determinants of child labor supply 38

Table 3.4: Determinants of household income and business ownership

Dependent var. Household income (OLS) Fam. business Sample (1) Girls (2) Boys (3) Heads Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.

Family business 0.432 6.89 0.325 6.37 Marginal land 0.096 1.52 0.068 1.07 -0.170 -2.45 Small land 0.278 3.45 0.302 4.59 -0.450 -4.85 Large land 0.528 5.15 0.611 5.26 -0.658 -4.04 Female literacy 0.305 3.52 0.191 1.87 0.259 1.72 Male literacy 0.080 1.39 0.092 1.33 0.160 3.35 Share infants 0.177 2.72 0.025 0.33 0.099 0.17 Share children -0.029 -0.41 0.067 0.77 0.077 1.10 Share boys 0.282 3.06 0.200 2.43 0.013 -1.55 Share girls 0.276 3.41 0.305 3.65 0.046 -0.37 Share elderly -0.289 -1.35 -0.342 -1.50 0.266 1.87 Share females -0.575 -2.95 -0.517 -2.80 -0.473 -1.68 Birth order 0.003 0.15 0.003 0.10 Married 0.006 0.08 0.011 0.06 0.041 -1.87 Lower castes -0.242 -2.43 -0.044 -0.36 0.333 2.49 Scheduled castes -0.329 -3.12 -0.137 -1.08 -0.059 -0.31 Muslim -0.247 -2.14 -0.096 -0.65 0.580 2.10 Time to school 0.010 0.72 -0.007 -0.56 0.007 0.89 School-costs 0.077 0.41 -0.001 -0.01 -0.142 -1.61 Female work low 0.091 1.37 0.050 0.72 0.168 1.48 Female work high 0.264 2.80 0.118 1.38 0.252 2.32 Median wages 0.034 1.09 0.039 1.20 0.034 0.41 Thresher owners 0.178 2.18 0.030 0.10 Value of transfers 0.103 7.34 0.118 9.50 0.150 1.19 Tractor owners 0.666 3.54 0.490 1.92

No. of obs. (N) 1067 1348 2241 No. of clusters 118 200 120 R-squared (Total) 0.325 0.249 0.049 Alternative instruments Tractor owners 0.796 4.27 Rooms per adult 0.195 4.65 Thresher owners 0.201 1.79 Value of transfers 0.112 8.94 Notes: Estimates (1) and (2) show the ﬁrst stage OLS estimation results for girls and boys using robust standard errors clustered at village level. Estimate (3) shows the results of a (clustered) probit regression of the presence of family business on the same covariates, performed over the sample of household heads. All regressions include a constant, income regressions include age dummies. determinants of child labor supply 39

Table 3.5: Tests on validity of the instruments

F-test of the Tests on overid. restrictions Wald-test of instruments Hansen’s J And.-Rubin exog. F -stat. p-value p-value p-value

Boys Market w. 55.0 0.896 0.904 0.147 (N = 1345) Domestic w. 55.0 0.905 0.962 0.006 Student 55.3 0.975 0.974 0.061 Girls Market w. 32.3 0.668 0.672 0.009 (N = 1067) Domestic w. 22.1 0.618 0.641 0.052 Student 22.1 0.769 0.701 0.645 Notes: Test statistics on the overall strength of the instruments, on the overidentifying restrictions, and on exogeneity are reported. The presented F-test tests the joint significance of the instruments in the first stage regression of household income. The tests of overidentifying restrictions (on the exclusion of the instruments from the second stage) are based on linear probability models, estimated by GMM (Hansen’s J-stat.) and LIML (Anderson- Rubin stat.). The Wald test on exogeneity tests the null hypothesis of exogeneity of family income in univariate probit regressions of each category of time use. lead to a higher yearly income everything else being equal (Table 3.4). At the same time, they can also raise the marginal product of child work and hence the incentives for child work within the family. Asset owner households might also differ with respect to their norms concerning work and school of children.21 As land ownership and family business are also included in the first stage regression predicting income, coefficients on land and family business in the second stage regressions capture these incentive and taste effects to a larger extent. The potentially nonlinear effect of land holdings is accounted for by controlling for different categories of land ownership (in acres) per adult (compared to families with no land holdings). Land is defined as marginal if it is below 0.5 acres per adult, small if it is between 0.5 and 2 acres per adult, and large if it is above 2 acres per adult.22 The effect of family business is captured by an indicator of whether any adult family member is self-employed in a small scale business. Such businesses include processing and selling food (e.g., milk products, flour, cigarettes, or alcohol), small scale manufacturing, and personal services (repair, massage, etc.). It might be argued that the decision of taking up a business depends on whether there are young children who are able to help out and hence it might be jointly determined with child work. As it can be seen from a probit regression of the family business in Table 3.4, the presence of young family

21 Kambhampati and Rajan (2004) argue that large land holdings in Northern India indicate a more patriarchal society with an especially large gender gap in actual and perceived returns to education. 22 For an average landowner family with 3.6 adults, these categories roughly reflect the classification of land by the Indian Census (Bhalotra and Heady 2003, p.208), which defines land as marginal if it is below 1 ha, as small if it is between 1 and 3 ha, and as large if it is above 3 ha (1 ha equals 2.7 acres). determinants of child labor supply 40

members does not make self-employment more likely per se. Nevertheless, for any given family structure, tastes for self-employment might coincide with value judgments about the necessity of child labor. In this case the indicator of family business captures not only labor demand, but also preference effects.23 All regressions include controls of local labor market outcomes. This is of particular interest as child labor can be expected to react to the demand for unskilled labor. Local wage and labor force participation rates reflect economic opportunities within the village and proxy labor demand effect. Median wages (vill.) are based on the village questionnaire and represent the median of daily wages for males in different occupations in agriculture.24 They measure directly the opportunity costs of schooling, while for domestic activities they also approximate parents’ shadow price of time. Female workforce participation also reflects local labor demand, but it is also strongly related to social norms with respect to the economic role of females. As the individual LSMS sample is not representative at village or district level, information from the 55th round of the Indian National Sam- ple Survey (NSS 1999/2000) has been additionally used to build two indicator variables: Female work low takes one in villages where female labor force participation is below the median both based on LSMS (village level) and NSS (district level) data, while Female work high indicates above median participation levels based on both data sets. Whether children are more or less likely to work in villages where female workforce participation is higher, is a priori unclear. To the extent that market work by females reflects labor demand effects, it also indicates more opportunities for market (and eventually also domestic) work of children. In the long term, better opportunities in the labor market might favor both market work and schooling of girls as means of human capital accumulation. But the social aspect of female workforce participation is equally important, and especially relevant for girls: In villages where more females work, girls’ labor force participation is also less prohibited by social norms.25

Further controls

For given levels of income and wealth, controls of educational attainment act as proxies for tastes and value judgments concerning education and work within

23 Table 3.4 also shows that the probability of operating a business falls with land ownership, and rises with male and female literacy. Self-employment is negatively associated with the share of females within the family, is more likely among Muslims and less likely among the members of scheduled castes and tribes. 24 Only male wages are included in both girls’ and boys’ regressions because information on male wages is the one which is most consistently available throughout the villages. 25 These eﬀects might be partly counteracted by the rising decision making power of females. If females are more concerned about child work and schooling, their economic power will shift the work-school trade-oﬀ in favor of more schooling. determinants of child labor supply 41

the family. Male literacy rate and Female literacy rate measure adult literacy within the household for both sexes. These variables capture general attitudes towards education and work within the household. The regressions also include various controls of household composition. Since all economic variables are normalized by the number of adult household members, controls of household composition are also measured in relative terms. The variables Share infants, Share children, Share girls and Share boys measure ratios of children to adult family members for the age groups 0 to 5 years, 6 to 9 years, and 10 to 17 years (by gender) respectively. The larger the relative share of infants in a household, the larger the potential need for help in child-care related activities, especially from older girls. A larger relative share of children and adolescents might also raise the need for their economic contribution. In order to capture potential birth order effects, the birth order among the siblings of the same sex is also included. Birth order effects might reflect parental preferences for first or later born as well as the presence of credit constraints: Earlier born children might have to work more while having older siblings might help to postpone employment of the young (Psacharopoulos and Patrinos 1997, Emerson and Souza 2002).26 Female share additionally controls for the sex composition of the adults in the family; Elderly share measures the share of elderly (aged above 66 years). Schooling costs involve two main dimensions: the monetary costs of schooling, given by tuition fees, school supplies, uniforms, etc., and the opportunity costs of time, measurable by school availability. Differences in the direct costs of schooling are proxied by the median of yearly expenses for a primary school student in each village.27 In order to reduce the problem of comparability of school types, only costs of attending primary school (classes 1 to 5) are included. School availability is controlled for by the variable Time to school which measures the time it takes to reach the nearest secondary school for each household. Higher costs of schooling can be expected to reduce school participation, while also reducing the opportunity costs of child work.

26 Edmonds (2006) shows that in Nepal the comparative advantage of older girls in household chores changes with younger siblings’ number, gender, and birth spacing. On the role of the gender aspects of household and sibling composition see also Parish and Willis (1993) and Morduch (2000). 27 This measure might overestimate the true costs of schooling if school choice is endogenous to the households’ willingness to pay for education, or if school costs are positively correlated with unobservable school quality. This might counteract the expected negative effect of school costs on education; in this case the negative effect found in Section 3.5 gives an upper bound estimate of the true effect. determinants of child labor supply 42

3.4 Estimation strategy

The empirical analysis estimates participation in market work, domestic work and school simultaneously. The three latent variables, market work L∗, household chores H∗, and school attendance S∗, depend on a vector of explanatory variables X, three unknown vectors of parameters βL, βH , βS, and the normally distributed error terms ǫL,ǫH ,ǫS. As the three choices are conﬂicting alternatives of children’s time use, and are determined simultaneously by the same decision making process, the same X vector of explanatory variables is included in all three equations.

∗ ′ L = X βL + ǫL ∗ ′ H = X βH + ǫH (3.4) ∗ ′ S = X βS + ǫS

The three equations from (3.4) are then mapped into three binary variables Yj (j = L,H,S) that take one if the child engages in a given activity, and zero otherwise.

′ Yj = 1(X βj + ǫj > 0) j = L,H,S (3.5)

Endogeneity of income is addressed by a two-step limited information procedure (Rivers and Vuong 1988) which decomposes the vector of explanatory variables X into the endogenous income variable x and the vector of exogenous variables Z1. At the ﬁrst stage of the two-step procedure, income is regressed on the set of exogenous explanatory variables Z1, and a set of instruments Z2. At the second stage, the residuals v from the ﬁrst stage are included as an additional regressor in each equation.

′ ′ b x = Z1δ1 + Z2δ2 + v (3.6a) ′ Yj = 1(Z1β1j + αjx + θjv + ej > 0) j = L,H,S (3.6b)

The underlying assumption is that theb error terms in the income and participation equations are jointly normal and hence the error terms in the latter can be decomposed into two error components θjv and ej (Wooldridge 2002, 472-475). The ﬁrst parts of the error components are correlated with v, and θj is directly estimated for each equation, the secondb parts are independent of v and x and jointly normal.28

28 The t-test of θj = 0 can be interpreted as a test of exogeneity of x1 within the given b equation. This procedure estimates the related coeﬃcients α and β1j only up to a scale. This is taken into consideration by estimating the average partial eﬀects (Wooldridge 2002, 475). determinants of child labor supply 43

The joint estimation of the three participation equations (3.6b) involves the evaluation of the loglikelihood over i = 1,...,N observations, based on a joint trivariate probability:

N ′ ′ ′ ln L = ln Φ3(κLiZiγL, κHiZiγH , κSiZiγS, κLiκHiρLH , κLiκSiρLS, κHiκSiρHS) i=1 X ′ ′ where Φ3 is the trivariate normal cumulative density function, Ziγj = Z1β1j + αjx + θjv (j = L,H,S) are the combinations of explanatory variables and coefficients as in (3.6b), ρLH , ρLS , ρHS are the correlation coefficients of the three error terms,b and κL, κH , κS are the corresponding sign variables that equal one if a child engages in a given activity, and minus one otherwise (Greene 2003, 710). The estimation of this function requires the computation of derivatives of third order integrals for which no general solutions exist. However, the problem can be addressed by simulation techniques: The method of simulated maximum likelihood allows the estimation of a trivariate probit model by using the Geweke-Hajivassiliou-Keane smooth recursive estimator (see Greene 2003, pp. 931-933). The estimation assumes that the error terms of the three participation equations ǫL,ǫH ,ǫS are jointly normally distributed with a covariance matrix Σ. The three correlation coefficients between the three sets of error terms ρLH , ρLS, and ρHS summarize the association between unobservable individual-specific factors determining the likelihood of being engaged in different types of occupations and are estimated along with the model. The Geweke-Hajivassiliou-Keane smooth recursive estimator decomposes the original three-dimensionally correlated error terms into a linear combination of uncorrelated one-dimensional standard normal variables. The trivariate distribution is thus transformed into three sequentially conditioned univariate distributions. In order to evaluate the resulting integral, D random draws of these standard normal variables are taken from truncated normal distributions, and a sample average of the simulated probabilities is used to estimate the probability that enters the likelihood function.29 The average partial effects (APE) have been estimated by averaging sample partial effects, computed for each individual.30 As two-step procedures estimate the coefficients only up to a scale, a procedure proposed by Wooldridge (2002, 475) has been used: Partial effects of probit equations have been calculated for each individual by including θjv2, the first-stage OLS residuals mul- tiplied by their estimated coefficient (see equation 3.6b). Thus partial effects b 29 Estimations have been implemented with Stata,b using the mvprobit, mvnp and mdraws routines of Cappellari and Jenkins (2003, 2006). For given sample sizes (1067-1318 observations), relative stability of the simulated γ and ρ parameters was ensured with about D = 300 random draws. 30 Estimation of APE-s on marginal probabilities has been carried out based on the user- defined Stata routine margeff (Bartus 2005), while estimation of APE-s on trivariate probabilities reused parts of this routine. determinants of child labor supply 44

have been averaged across the first stage residuals of the sample.31 Standard errors of the APE-s for the trivariate probabilities have been estimated by a computationally intensive empirical Bayes procedure. 2000 replications of the estimated coefficient vectors (γL, γH , γS, ρLH , ρLS, ρHS ) were redrawn from a multivariate asymptotically normal distribution (characterized by the estimated variance-covariance matrixb Σ),b andb theb standardb b deviation of the partial effects was computed. This serves as an approximation of the standard error of the partial effects. b

3.5 Results

Tables 3.6 and 3.7 present the results from the trivariate probit regressions. All regressions report robust standard errors that are clustered on the village level, allowing for correlation between unobserved characteristics of children within the same village. The estimated correlation coefficients between market work, domestic work, and schooling reflect the strength of the main unexplained trade-offs between the three types of occupation. They confirm the fact that domestic work and school are the two most conflicting alternatives for girls, market work and school for boys. The estimated correlation coefficient between the unexplained part of domestic work and school of girls amounts to -0.90, between market work and school of boys to -0.87. For all other occupations the respective correlations are much smaller and less significant. The estimated average partial effects of the explanatory variables on the marginal probability of each occupation are given in Tables 3.6 and 3.7. Tables 3.8 and 3.9 present selected average partial effects on the joint trivariate probability of a given combination of the three activities where average is taken across all girls or boys in the sample (cf. Section 3.4). They show the average effect of each explanatory variable on the probability that a child specializes in one given activity (market work, domestic work, or school), combines different activities, or stays idle. As expected, the probability that a child works falls and the probability that he or she goes to school rises with rising income. The effects of income on specializing in work are similar for the primary occupations of girls and boys (domestic and market work respectively). An APE of about -0.2 in Tables 3.6 and 3.7 indicates that by increasing yearly per adult income by 1000 Rupees, the probability that a child performs market work in a family with yearly per adult income of 9000 Rupees decreases by 2.2 percentage points. In a family with yearly per adult income of 5000 Rupees, the same effect is 4 percentage points. Household income also raises the probability of girls (but not boys) staying idle. This gender difference might be due to the lower perceived returns

31 Wooldridge (2002, 475) shows that average partial eﬀects calculated this way are consistent. determinants of child labor supply 45

Table 3.6: Trivariate probit results on work/schooling of girls

Dependent var. Market work Domestic work Schooling Coeff. APE t-st. Coeff. APE t-st. Coeff. APE t-st.

Household income -1.363 -0.232 -2.69 -0.691 -0.223 -3.52 0.388 0.107 5.11 Family business 0.913 0.167 3.29 0.278 0.089 2.17 -0.135 -0.037 -1.23 Marginal land 0.008 0.001 0.05 0.183 0.059 1.41 0.131 0.036 1.06 Small land 0.384 0.071 1.64 0.209 0.067 1.51 0.255 0.071 1.64 Large land 1.076 0.247 3.08 0.727 0.230 3.09 0.010 0.003 0.06 Female literacy -0.384 -0.065 -1.35 -0.556 -0.179 -3.22 1.055 0.291 6.17 Male literacy 0.170 0.029 1.02 -0.247 -0.080 -2.25 0.491 0.135 4.22 Share infants 0.365 0.062 1.55 0.443 0.143 3.25 -0.346 -0.095 -2.25 Share children 0.213 0.036 0.98 0.085 0.027 0.53 -0.202 -0.056 -1.18 Share boys 0.037 0.006 0.13 0.223 0.072 1.32 -0.245 -0.067 -1.52 Share girls 0.671 0.114 2.42 0.232 0.075 1.34 -0.330 -0.091 -1.87 Share elderly 0.046 0.008 0.09 -1.444 -0.465 -3.62 1.096 0.302 2.79 Share females 0.357 0.061 0.63 -0.608 -0.196 -1.45 0.337 0.093 0.86 Birth order 0.006 0.001 0.11 -0.107 -0.034 -1.98 0.170 0.047 2.94 Married -0.281 -0.044 -1.75 0.538 0.178 3.07 -0.945 -0.251 -4.55 Lower castes 1.012 0.164 2.45 0.312 0.101 1.53 -0.501 -0.138 -2.29 Scheduled castes 1.236 0.254 3.03 0.131 0.042 0.59 -0.348 -0.096 -1.43 Muslim 0.621 0.126 1.36 0.445 0.144 1.96 -0.473 -0.127 -1.86 Time to school -0.004 -0.001 -0.17 0.086 0.028 3.99 -0.063 -0.017 -2.67 School-costs (v.) 0.778 0.132 2.60 0.081 0.026 0.33 -0.630 -0.174 -2.09 Female work low -0.539 -0.082 -3.33 0.104 0.034 0.92 0.037 0.010 0.29 Female work high 0.422 0.077 2.31 0.029 0.009 0.25 0.176 0.049 1.26 Median wages (v.) 0.209 0.036 2.92 0.104 0.034 2.01 -0.022 -0.006 -0.40 Resid. income eqn˙ 1.198 2.30 0.534 2.60 Age dummies yes yes yes t-st. t-st. t-st.

Corr. coeff. ρ21 = −0.167 -2.46 ρ31 = −0.152 -2.49 ρ32 = −0.900 -43.35 Notes: Results of the trivariate probit model are estimated by SML with 300 pseudorandom draws, clustered at village level. t-statistics refer to the estimated coefficients and are based on robust standard errors. The model also includes age dummies and a constant. Average partial effects (APE) are calculated with respect to the marginal probability of each occupation. Sample size is N = 1067 observations. Wald-test of the model χ2(92) = 2513.85, p = 0.0000. determinants of child labor supply 46

Table 3.7: Trivariate probit results on work/schooling of boys

Dependent var. Market work Domestic work Schooling Coeff. APE t-st. Coeff. APE t-st. Coeff. APE t-st.

Household income -0.539 -0.123 -2.25 -3.233 -0.238 -1.76 0.847 0.207 1.97 Family business 0.288 0.067 2.25 1.116 0.124 1.79 -0.260 -0.064 -1.50 Marginal land -0.105 -0.024 -0.78 0.149 0.012 0.65 0.382 0.092 3.00 Small land -0.007 -0.002 -0.04 1.105 0.125 1.86 0.081 0.020 0.50 Large land 0.103 0.024 0.36 2.175 0.437 1.88 0.208 0.049 0.57 Female literacy -0.397 -0.091 -1.82 -0.523 -0.039 -1.09 0.800 0.196 3.82 Male literacy -0.442 -0.101 -3.81 -0.272 -0.020 -0.98 0.651 0.159 5.86 Share infants -0.017 -0.004 -0.13 0.245 0.018 0.92 -0.128 -0.031 -0.89 Share children 0.287 0.066 1.80 -0.077 -0.006 -0.29 -0.154 -0.038 -0.92 Share boys 0.329 0.075 2.32 0.915 0.068 2.22 -0.425 -0.104 -3.10 Share girls -0.179 -0.041 -1.02 0.927 -0.077 1.50 0.014 0.003 0.07 Share elderly -0.484 -0.111 -1.13 -1.042 -0.146 -1.30 0.630 0.154 1.44 Share females -0.071 -0.016 -0.22 -1.982 -0.039 -2.07 0.214 0.052 0.64 Birth order -0.036 -0.008 -0.71 0.006 0.000 0.08 0.049 0.012 1.08 Married 0.308 0.077 1.06 0.000 0.000 0.00 -0.422 -0.112 -1.64 Lower castes 0.234 0.053 1.31 -0.879 -0.072 -2.96 -0.299 -0.072 -1.54 Scheduled castes 0.215 0.051 1.10 -0.925 -0.066 -2.10 -0.202 -0.051 -0.95 Muslim 0.260 0.063 1.15 -1.129 -0.047 -2.17 -0.510 -0.135 -2.29 Time to school 0.041 0.009 2.42 0.015 0.001 0.45 -0.052 -0.013 -2.93 School-costs (v.) 0.315 0.072 1.92 0.372 0.027 1.08 -0.573 -0.140 -3.07 Female work low -0.089 -0.020 -0.73 -0.278 -0.018 -1.34 0.179 0.043 1.34 Female work high -0.098 -0.022 -0.67 0.196 0.015 1.08 0.149 0.036 1.20 Median wages (v.) 0.076 0.017 1.62 0.183 0.014 1.86 -0.073 -0.018 -1.66 Resid. income eqn˙ 0.437 1.85 3.062 1.66 -0.695 -1.62 Age dummies yes yes yes t-st. t-st. t-st.

Corr. coeff. ρ21 = −0.081 -1.05 ρ31 = −0.874 -37.86 ρ32 = −0.308 -4.35 Notes: Results of the trivariate probit model are estimated by SML with 300 pseudorandom draws, clustered at village level. t-statistics refer to the estimated coefficients and are based on robust standard errors. The model also includes age dummies and a constant. Average partial effects (APE) are calculated with respect to the marginal probability of each occupation. Sample size is N = 1318 observations. Wald-test of the model χ2(93) = 3132.90, p = 0.0000. determinants of child labor supply 47

to schooling for girls. The probability that girls specialize in domestic chores is also decreasing with household income. The results clearly show that incentive effects affect girls and boys similarly: While income increases with land ownership and business activities (cf. Table 3.4), for any given level of income, children are more likely to work and less likely to specialize in school in families that own large land or operate a family business. Both girls and boys from these households are more likely to do market work or to combine market work with other occupations, they are less likely to specialize in school and less likely to stay idle. These effects shift the work-school trade-off in favor of more work; this might be due to the fact that demand for family labor is increasing with asset ownership, but potentially also to the relatively higher returns to learning-by-doing in these families. In a similar vein, agricultural wages in the village capture incentive effects by proxying opportunity costs of non-working, and parental shadow valuation of time spent in the household: with higher wages the probability of going only to school is reduced and the likelihood is higher that both boys and girls are involved in any type of work activity. The work-schooling trade-off is also affected by household composition variables. The dependency ratio of smaller children to adults within the family has a significant influence on girls’ work and schooling: With one additional small child per adult the probability that a girl performs only domestic duties rises by 6% points, while the probability that she only goes to school falls by about 9% points. The share of same-sex teen-aged children in the family raises the probability of market work and reduces the probability of school attendance. A possible explanation for this is sibling rivalry: for any given level of per adult income, the more adolescents are in the family, the more necessary is their economic contribution. Among female siblings, later born girls fare better due to birth order effects, they are less likely to perform domestic chores and more likely to go to school. Surprisingly, birth order effects cannot be detected among male siblings.32 The results also show that although household income is associated with a rising share of females and elderly (cf. Table 3.4), their non-monetary contributions are important for the outcomes of child work and schooling. For any given income level, the share of elderly within the family raises the probability of school attendance for children of both sexes and reduces market work for boys and domestic work for girls. A higher share of females among the adults reduces the probability that children of both sexes specialize in domestic work, as females are more likely to share the burden of household chores. Preferences for education, captured by adult literacy, mostly determine the trade-off between participation in schooling and the gender-specific activity.

32 Naturally, both variables might also reflect inherent differences between households with different fertility and human capital investment strategies (few well-educated or many uneducated children) as predicted by the theory (Becker and Lewis 1973). determinants of child labor supply 48

Table 3.8: Selected APEs on probabilities of specializing in work/idleness

Work, no school Idle Outcome Market only Domestic only Combine M&D Do nothing APE t-st. APE t-st. APE t-st. APE t-st. Girls: Household income -0.034 -2.08 -0.165 -2.37 -0.138 -4.07 0.070 2.72 Family business 0.024 2.20 0.036 0.84 0.108 3.63 -0.046 -2.95 Marginal land -0.010 -1.05 0.016 0.45 -0.010 -0.63 -0.013 -0.79 Small land -0.007 -0.85 0.001 0.01 -0.005 -0.23 -0.020 -1.06 Large land -0.001 0.00 0.080 1.06 0.123 2.23 -0.075 -4.43 Female literacy rate -0.014 -1.08 -0.112 -2.27 -0.063 -2.23 0.020 1.01 Male literacy rate 0.015 1.27 -0.093 -2.17 0.018 1.16 -0.022 -1.40 Lower castes 0.027 2.22 -0.016 -0.40 0.058 2.45 -0.018 -1.30 Scheduled castes 0.037 2.14 -0.046 -0.93 0.076 2.30 -0.021 -1.22 Muslim -0.017 -1.30 0.091 1.91 -0.020 -0.93 -0.003 -0.19 Time to school -0.002 -1.19 0.020 3.68 0.003 1.46 -0.004 -1.85 School costs (v.) 0.021 1.58 0.020 0.33 0.035 1.50 0.022 1.06 Median wages (v.) 0.007 1.44 0.005 0.22 0.024 2.87 -0.019 -2.22 Predicted prob. 0.035 0.309 0.061 0.119 Sample share 0.068 0.369 0.043 0.060

Boys: Household income -0.150 -2.49 -0.044 -2.61 -0.076 -1.21 Family business 0.074 2.39 0.017 1.60 0.005 0.17 Marginal land 0.001 0.05 -0.002 -0.42 -0.041 -2.44 Small land 0.018 0.62 0.022 2.17 -0.011 -0.47 Large land 0.041 0.63 0.041 1.48 -0.069 -2.11 Female literacy rate -0.068 -2.12 -0.006 -0.63 -0.051 -1.95 Male literacy rate -0.045 -2.13 -0.003 -0.63 -0.037 -2.13 Lower castes 0.021 1.16 -0.019 -2.11 0.028 1.57 Scheduled castes 0.010 0.39 -0.014 -1.92 0.020 0.88 Muslim 0.036 1.12 -0.010 -2.62 0.021 0.95 Time to school 0.006 2.02 0.001 0.83 0.004 1.36 School-costs (v.) 0.077 2.28 0.004 0.68 0.010 0.33 Median wages (v.) 0.013 1.10 0.006 1.69 0.009 0.94 Predicted prob. 0.129 0.012 0.155 Sample share 0.159 0.024 0.093 Notes: Estimation results are based on the trivariate probit model. The average partial eﬀects (APE) are calculated with respect to the joint trivariate probability of each outcome. Market only refers to the outcome P (L = 1,H = 0,S = 0), Domestic only to P (L = 0,H = 1,S = 0), Do nothing to P (L = 0,H = 0,S = 0). t-statistics are based on standard errors approximated by parametric bootstrap. Sample size is N = 1067 observations for girls and N = 1345 observations for boys. determinants of child labor supply 49

Table 3.9: Selected APEs on joint probabilities of work and school

Combine school with Outcome School only Market work Domestic work APE t-st. APE t-st. APE t-st. Girls: Household income 0.324 4.95 -0.029 -1.26 -0.017 -0.81 Family business -0.162 -4.41 0.027 1.93 0.004 0.24 Marginal land 0.007 0.21 -0.007 -0.72 0.016 1.04 Small land 0.012 0.28 -0.001 -0.08 0.019 1.10 Large land -0.184 -3.64 0.007 0.42 0.033 1.21 Female literacy 0.180 3.22 -0.004 -0.21 -0.002 -0.09 Male literacy 0.047 1.36 0.031 2.21 0.000 -0.01 Lower castes -0.076 -1.75 0.030 2.01 -0.011 -0.96 Scheduled castes -0.080 -1.69 0.046 1.93 -0.018 -1.24 Muslim -0.041 -1.23 -0.023 -2.00 0.016 0.79 Time to school -0.019 -3.37 -0.003 -1.75 0.004 2.04 School costs (v.) -0.079 -1.69 0.010 0.57 -0.030 -1.90 Median wages (v.) -0.036 -2.05 0.011 1.75 0.006 0.81 Predicted prob. 0.356 0.038 0.079 Sample share 0.426 0.010 0.019

Boys: Household income 0.369 4.19 -0.017 -0.57 -0.046 -1.79 Family business -0.157 -4.11 0.020 1.36 0.017 1.15 Marginal land 0.021 0.62 0.018 1.66 0.002 0.27 Small land -0.087 -2.25 0.006 0.39 0.031 1.83 Large land -0.208 -2.62 0.038 1.13 0.078 1.44 Female literacy 0.135 2.33 -0.005 -0.33 0.000 -0.05 Male literacy 0.088 3.05 -0.002 -0.13 0.002 0.27 Lower castes 0.016 0.49 0.001 0.12 -0.034 -2.38 Scheduled castes 0.018 0.41 -0.001 -0.06 -0.022 -2.11 Muslim -0.032 -0.62 0.007 0.52 -0.015 -2.98 Time to school -0.012 -2.53 0.0002 0.10 0.000 0.09 School costs (v.) -0.125 -2.01 0.025 1.81 0.002 0.25 Median wages (v.) -0.039 -2.00 -0.0002 -0.06 0.007 1.48 Predicted prob. 0.634 0.047 0.016 Sample share 0.678 0.029 0.011 Notes: Estimation results are based on the trivariate probit model. The average partial eﬀects (APE) are calculated with respect to the joint trivariate probability of each outcome. School only refers to the outcome P (L = 0,H = 0,S = 1), Combine school with market work to P (L = 1,H = 0,S = 1), combine school with domestic work to P (L = 0,H = 1,S = 1). t-statistics are based on standard errors approximated by parametric bootstrap. Sample size is N = 1067 observations for girls and N = 1345 observations for boys. determinants of child labor supply 50

Both male and female literacy make children more likely to go to school and less likely to work or to stay idle. The results corroborate the well-known importance of female education, which plays the more decisive role for both girls’ and boys’ work and education. School availability and average school costs are as expected, strongly negatively related to school attendance (and especially exclusive specialization in school) of both sexes. School distance is more closely related to the trade-off between the gender specific activity and schooling; it increases the likelihood that boys perform exclusively market work or girls specialize in domestic work. Cultural norms, proxied by controls for caste and religion, strongly influence the economic role of the sexes, and are especially important in explaining the gender unspecific activities. Muslim boys, as well as boys from lower and scheduled castes, are less likely to perform domestic work; girls from lower and scheduled castes are more likely to work for the market, or to combine domestic chores with market work. Splitting market work into wage work and home production and estimating a model with four distinct categories shows that ceteris paribus girls from scheduled castes are the most likely to work for wages, while girls from lower castes are the most likely to help out in family business.33 The effects of cultural norms are also strongly reflected in the role of adult female workforce participation in the village which significantly raises market related work and reduces exclusive school attendance of girls for any given level of income. With higher female labor force participation, children are also more likely to combine school with market work. The workforce participation of adult females clearly shifts the trade-offs towards girls’ market work as opposed to domestic chores or schooling. The results also support the view that children are more likely to stay idle if they have less economic opportunities to work. Girls tend to be inactive in households that do not have large land holdings or family business, and in villages where wages and/or female workforce participation are low. Boys’ idleness reflects economic opportunities to a lesser extent, with the exception of land holdings, instead depending strongly on the preferences for schooling proxied by educational attainment of parents.34

33 These findings support those of Kambhampati and Rajan (2004) who find similar patterns of caste-based differences among all Indian girls. They argue that this reflects the less patriarchal cultural norms among the lowest castes, which put less restrictions on the work of girls outside the household. 34 Although ability of the children is not measured, idleness can also be expected to crucially depend on individual abilities. As demonstrated by Bacolod and Ranjan (2008) for the Philippines, in a family the least able children are the ones to stay idle, especially among the relatively richer families. determinants of child labor supply 51

3.6 Conclusion

This chapter has presented jointly estimated participation equations in market work, domestic work, and school of North Indian children by the method of simulated maximum likelihood. The regressions controlled for a wide range of individual and household characteristics, like household composition, educational attainment, religion or caste. Additionally, land ownership, participation in business activities, and controls for village or district level labor market outcomes have been included in all regressions. Household income has been instrumented in a two-step procedure. This also allowed for a better separa- tion of the incentive effects arising from the ownership of large land, family business, or female labor force participation and wages. The empirical analysis has shown that the income and incentive effects that affect the trade-off between school and work are of relatively similar magnitude for both Indian boys and girls. But since the major work activities are different for the two sexes, the trade-offs between work and school differ as well: schooling mostly conflicts with household work for girls, and market work for boys. This gender difference could be due to a gender gap in the relative returns to both formal education and learning-by-doing as well as cultural norms with respect to the females’ role. Interestingly, several of the usual explanatory factors, like tastes for education (proxied by household literacy), or the distance to the nearest school, are mainly related to the trade-offs between the gender-specific work activity and schooling. By contrast, cultural norms influence participation in the less typical work activity to a larger extent. Caste variables and female workforce participation are relatively strongly related to girls’ involvement in market work but less so in domestic work, while cultural norms are among the few relevant determinants of boys’ domestic work. For instance, the highly significant role of female workforce participation at the village and district level shifts girls’ activities from domestic towards market work. Studies that concentrate on market work only, are bound to neglect the major part of these trade-offs for girls. The results also emphasize the special role of economic incentives, arising from the ownership of family business or larger land holdings but also from village level wages. The effects of economic incentives are considerably less gender specific: they increase participation in both market and domestic work of children of both sexes, without reducing the probability of school attendance per se, but at the same time making exclusive school attendance or even idleness less likely. chapter 4

Demand and Supply Interactions: Child labor in Indonesian small scale manufacturing

Abstract

This chapter analyzes the geographic incidence of child labor in small manufacturing firms in Indonesia at the village level. It uses a unique data set covering virtually all Indonesian villages and urban neighborhoods and distinguishes between demand and supply side determinants of child labor. By correcting for sample selection it becomes clear that a number of counterintuitive results in a conditional probit model—e.g. child labor that is unaffected by credit access and school proximity—are the result of an interplay between supply and demand side determinants. Credit access and school proximity reduce child labor supply but simultaneously constitute positive location factors for firms thereby increasing the demand for child laborers. Thus, to effectively reduce child labor, growth oriented policies, such as enhancing school and credit facilities, should be complemented by policies specifically geared towards increasing school attendance.1

4.1 Introduction

To this day, child labor remains a big developmental problem, not only because it affects child development negatively, but also because it significantly reduces human capital formation and thus growth performance (cf. Duflo 2001). In order to design appropriate policies to reduce child labor, it is necessary to understand its determinants. This chapter contributes to our understanding by analyzing demand and supply factors of child labor in Indonesia separately. It employs a unique data set that comprises all Indonesian villages and uses the geographic variation in economic conditions to distinguish between determinants of the demand for, and the supply of child labor.

1 This chapter is based on Kis-Katos and Schulze (fth).

52 demand and supply interactions 53

Empirical evidence suggests that child labor is associated with poverty. In cross country analyses GDP per capita turns out as a very powerful determinant of child labor (e.g., Krueger 1996, Cigno, Rosati and Guarcello 2002); and open economies have less child labor due to gains from trade (Edmonds and Pavcnik 2006). At the micro level, evidence on the effect of income or wealth on child labor is less clear, partly because of the endogeneity of income or expenditures (Bhalotra and Tzannatos 2003). Edmonds and Pavcnik (2005b) show for Vietnam that trade liberalization has reduced child labor due to positive income effects. Bhalotra (2003) finds that labor supply of boys (not of girls!) in rural Pakistan depends negatively on the own wage—the income effect thus dominates the substitution effect. The analyses of exogenous income transfers also show that child labor declines with income.2 Duryea, Lam and Levison (2007) show that Brazilian children are more likely to drop out of school and work instead if the male household head becomes unemployed. These findings suggest the existence of a strong negative income effect on child labor. Other studies find a positive relationship between wealth and child labor for agricultural households that own land (Bhalotra and Heady (2003) for Pakistan and Ghana and Kanbargi and Kulkarni (1991) for Karnataka/India), which they explain by imperfect land and labor markets.3 Basu et al. (2009) argue that this relationship has an inverted U shape. If investment in human capital is profitable, the poor could, in principle, borrow against future earnings and send their children to school. Yet, in many developing countries the poor are thought to have insufficient access to credit, which gives rise to child labor (Baland and Robinson 2000, Ranjan 2001). As (insufficient) access to credit is typically not observable,4 it has to be inferred from reactions to income shocks. These studies show that in many developing countries child labor increases (decreases) as a reaction to negative (positive) income shocks suggesting that credit constraints are binding.5 If

2 The Food for Education Program in Bangladesh (Ravallion and Wodon 2000), the Mex- ican PROGRESA (Schultz 2004), or Brazil’s Bolsa Escola program (Bourguignon, Ferreira and Leite 2003) are examples. Cardoso and Souza (2004) analyzes conditional cash transfers in this program. If the program reduces the cost of schooling at the same time, it provides an additional incentive to substitute school attendance for labor (even though both occupations are not mutually exclusive). 3 Parikh and Sadoulet (2005) make a similar argument for family businesses in Brazil. 4 The only exception is Guarcello, Mealli and Rosati (2009) who measure credit restrictions directly through a survey recording credit history and find that credit constrained families in Guatemala are less likely to send their children to school and more likely to increase child labor in response to negative income shocks. 5 Dehejia and Gatti (2005) show that income volatility affects child labor more in countries with low financial development. School attendance in rural India declines in response to unanticipated seasonal fluctuations in household income (Jacoby and Skoufias 1997). Tanza- nian farmers react to transitory income shocks with increased child labor, but this increase is lower for farmers with durable assets, who can be assumed to have better access to credit due to their collateral (Beegle et al. 2006). Edmonds (2005) shows that the introduction of a large pension scheme for black South Africans reduced work of those children living with demand and supply interactions 54

credit availability leads to higher capitalization, returns to child labor may increase and thus schooling may decrease (cf. Wydick 1999, Guatemala). By contrast, Cameron (2001) and Suryahadi, Priyambada and Sumarto (2005) do not find any significant increase in Indonesian child labor in response to the sharp decline in income during the 1997/98 crisis. Cost and accessibility of schooling have been shown to be a major determinant for child labor. Programs that reduce the cost of schooling have proved to raise school attendance and reduce market work (cf. fn. 2). Accessibility of education proves to be a major determinant of school enrollment e.g. in Indonesia (Pradhan 1998) and in rural Tanzania (Kondylis and Manacorda 2006). Extra household work rises with the cost of schooling in rural Pakistan (Hazarika and Bedi 2003). These microeconometric studies focus on the determinants of child labor supply, given by household characteristics such as income, wealth, parental education, family structure, etc. The degree of child labor that we observe, however, is the result of the interplay of demand and supply. For example, Manning (2000) argues (but does not show explicitly) that the high unemployment in Indonesia during the 1997/98 crisis has made it hard for children to find jobs so that child labor did not increase. Thus, negative income shocks may increase the supply of child labor and simultaneously decrease the demand for it. We observe only the net effect. Recently, a number of studies have included demand determinants in their empirical setup. Labor market conditions, influencing both demand for and supply of child labor, have been shown to matter: Duryea and Arends Kuen- ning (2003) show that adolescent employment in urban Brazil increases with the wage rate. Child labor declines with rising adult employment (Manacorda and Rosati 2007, Brazil). Wahba (2006) shows that it decreases in Egypt with rising wages for illiterate adult males. Fafchamps and Wahba (2006) show for Nepal that proximity to urban centers increases school attendance and decreases total child labor, but makes it more likely that children are involved in wage work. Growth of economic activity increases child labor (Kambhampati and Rajan 2006 and Swaminathan 1998 for India, Kruger 2007 for Brazilian coffee production), which points to a dominance of the demand effect over the supply effect. This chapter argues that the distinction between supply side and demand side determinants of child labor is important, not only for analytical reasons, but especially because an effective policy to combat child labor needs to reduce the supply of child labor. The forces driving up the demand for child labor are typically those that increase the demand for adult labor as well and therefore are beneficial to overall development. Yet the distinction between demand and recipients of the pension, but not of those living with future pensioners, again pointing to existing credit constraints. demand and supply interactions 55

supply factors is far from obvious. While availability of credit is regarded as reducing child labor supply, it may also increase child labor demand as it may lead to higher economic activity and higher demand for (child) labor. The net effect is uncertain a priori. Likewise, economic activity may locate where schools provide an educated labor force and may in its wake increase demand for child labor. Conversely schools may locate where economic activity and thus the return to education is highest. It may turn out that school and credit availability are associated with higher child labor; yet this association does not invalidate the remedy of better schools and better access to credit against child labor—it is the result of a dominant demand effect.6 This chapter addresses these issues. It uses a unique data set of all villages and urban neighborhoods in Indonesia (i.e., more than 68 thousand), which includes a rich set of village characteristics, notably child labor incidence in small industries, credit availability, presence of schools, income shocks, a wide range of poverty-related variables, local unemployment, geographic indicators, and the economic structure of villages/urban neighborhoods, etc. The large variation of these village level variables across all of Indonesia allows me to distinguish between the effect of these and other variables on the location decision of small firms—and thereby indirectly on child labor demand—and their effect on the supply of child labor. The empirical results show that child labor in small industries is significantly associated with poverty, negative income shocks, and unemployment. Credit availability increases child labor, but only through a demand side effect: It increases the likelihood of small industries locating in a village, and if so, the number of small firms in that village and thereby the demand for child labor. In other words, the set of villages/urban neighborhoods that have small firms is a biased sample of all Indonesian villages/neighborhoods with some of the determinants of child labor also being responsible for the sample selection. After correcting for the sample selection bias, credit availability significantly reduces child labor. This signifies the supply side effect of credit availability. On the local level, the demand effect outweighs the supply effect. By a similar argument, school availability significantly reduces child labor only after correcting for location decisions of small firms. Again, the likelihood of small firms locating in a village increases with school availability. This chapter proceeds as follows. The next section lays out the regional and sectoral incidence of child labor in small industries throughout Indonesia. Section 4.3 describes the data and sets out the empirical model. Section 4.4 reports the results, Section 4.5 concludes.

6 Even if credit and school availability are negatively associated with child labor, their true eﬀect may be underestimated due to the endogeneity of economic activity. demand and supply interactions 56

4.2 Child labor in Indonesia

Indonesia has experienced a steady decline in labor force participation rates of children aged 10 to 14 from 22.1% in 1960 to 7.1% in 2002 and 5.2% in 2007.7 This reduction was due to rising living standards, falling family sizes and a structural change that reduced the labor force in agriculture and the cottage and small scale industry, the main sectors in which children worked (cf. Manning 2000). The reductions in child labor went along with significant improvements in school enrollment. Assisted by a large primary school construction program (Sekolah Dasar INPRES) that was launched in 1973, primary school enrollment became almost universal already in the mid 1980s (e.g., Lanjouw, Pradhan, Saadah, Sayed and Sparrow 2002).8 The economic crisis in 1997/98 did not lead to a large secondary school drop out or increase in child labor (Cameron 2001, Suryahadi et al. 2005), but halted the improvements in child labor for the next two years. Primary enrollment benefited from a social safety net scholarship program which successfully prevented a decrease in primary enrollment (Sparrow 2007). Despite improvements in school enrollment and child labor over the last decade, local public primary education is still delivered inefficiently in Indonesia (Lewis and Pattinasarany 2010). Child labor and secondary school enrollment are both still closely linked to poverty (Suryahadi et al. 2005). Although Indonesia currently pursues a nine year universal basic education policy (up to the age of 15), around 30% of the children drop out after completing primary school (around the age of 12). The drop-out rates are considerably higher for the poorest quintile of the population (almost 50%) than for the richest quintile (12%) (Paqueo and Sparrow 2006). Schooling and work are not mutually exclusive; around half of the working children still go to school, working considerably fewer hours per week than their only working counterparts (Suryahadi et al. 2005). Indonesian children work predominantly in agriculture, but small scale manufacturing is the second most important sector for child work. At the time of the present analysis, in 2002, 65.3% of the working children (10–15 years old) worked in agriculture, 12.2% in manufacturing and 11.3% in trade.9 The analysis of child labor in small industries (such as leather, wood, ceramic, and metal processing, weaving and food production) is not only interesting because it is the second most important sector for child labor and has not been the focus of attention in the analysis of child labor. It is particularly interesting because—unlike the ubiquitous agricultural sector—small industries are

7 Source: World Bank (2004) and SUSENAS 2002 and 2007. From 1960 to 2002, world- wide labor force participation rates of children declined from 24.9% to 10.6% (World Bank 2004). 8 From 1974 to 1978 more than 61,000 primary schools were built. This has led to large increases in primary school enrollment rates and had significant long-term labor market impacts (cf. Duflo 2001, 2004). 9 These figures are based on data from SUSENAS 2002. demand and supply interactions 57

present only in a subset of villages and urban neighborhoods. This allows us to disentangle demand and supply side inﬂuences on child labor through the endogenous location of these economic activities.

4.3 Data and empirical strategy

4.3.1 Data

The data source is the Village Potential Statistics, (Potensi Desa; PODES) of 2003 which was collected in the fall of 2002. The data set is compiled by Statistics Indonesia (Badan Pusat Statistik, BPS) in the context of the periodic census and covers 69125 villages and urban neighborhoods, all villages/neighborhoods in Indonesia.10 It consists of a rich set of data at the village level; in particular it reports for each village the number of firms in different small scale manufacturing industries (in leather, wood, metal, ceramic, weaving, food, and other industries) and whether children are employed in these sectors. In addition, the data set contains a rich set of village level variables on the availability of different kinds of credit, presence of or distance to schools and markets, various other infrastructure variables, population, unemployment rate, geographic and poverty variables. In one dimension this data set is much more comprehensive than those of existing microeconometric studies which are typically confined to small geographical areas: it comprises data on virtually every village in Indonesia.11 Because Indonesia is so vast and so heterogeneous in structure, poverty incidence, and geography, this generates enough variation to capture the roles that these variables play in determining child labor incidence. This great geographic coverage comes at the cost of not being able to assess the number of child laborers in any village, but only sectoral and geographic child labor incidence. The main dependent variable is an indicator variable that takes one if village heads report that children are working in any of the small scale manufacturing industries in the village. Thus, this study cannot directly assess the intensity of child labor, nor can it capture child labor in other sectors, notably not in agriculture. Nonetheless, the geographic child labor prevalence variable is reasonably correlated with information on the intensity of child labor

10 Due to data encoding errors, the main regressions are using observations on 68344 villages. 11 The Indonesian ‘100 village survey’ that has been used to analyze the eﬀect of the 1998 crisis on child labor and schooling covers only 10 out of 292 districts at that time and oversamples poor and rural districts (cf. Cameron 2001, Suryahadi et al. 2005). Even the rich longitudinal Indonesian Family Life Survey (IFLS) data set represents only 13 out of 30 Indonesian provinces (cf. Thomas, Beegle, Frankenberg, Sikoki, Strauss and Teruel 2004). Studies on child labor in other countries typically cover only a geographic portion of the country. demand and supply interactions 58

Figure 4.1: Regional incidence and intensity of child work (province level)

Market work of children Child work in manufacturing .15

.025 Bali

Bali E.NTG W.NTG .02 SE.Sul. N.Sum. S.Sul.S.Kalim.

.1 Bangka Gorontalo

W.Kalim.C.Sul. .015 C.Kalim.

S.Sum. Bengkulu S.Kalim. E.Jawa LampungC.Jawa

W.Sum. E.Kalim. W.NTG N.Sul.Banten .01 W.Jawa C.Jawa

.05 Yogya Gorontalo Jambi SE.Sul. Riau Jakarta E.Jawa Yogya N.Maluku Bangka W.Jawa .005 W.Sum. Papua W.Kalim. Lampung Aceh S.Sul.Banten N.Sum.S.Sum. Share of children (10−14) working (Susenas) Maluku C.Sul. E.NTGJakarta Riau Jambi Papua N.MalukuMalukuAceh N.Sul.BengkuluC.Kalim. E.Kalim. 0 0 0 .1 .2 .3 .4 0 .1 .2 .3 .4 Share of villages with child work in small firms (PODES) Share of children (10−14) working in manufacturing (Susenas) Share of villages with child work in small firms (PODES)

based on SUSENAS, the Indonesian national household survey. The province level correlation between average labor force participation of children (aged 10 to 14) in manufacturing and geographic child labor prevalence amounts to 0.664, the correlation between overall workforce participation of children and the above child labor measure amounts to 0.524 (cf. Figure 4.1).12

Table 4.1: Child labor incidence in small scale manufacturing

Child labor incidence % Small ﬁrms By industry (u.m.) (c.m.) % of vill. No.(c.m.)

Leather 0.4 17.5 2.3 7.2 Wood 4.4 19.9 22.3 7.9 Metal 0.8 18.1 4.2 9.0 Ceramic 4.1 24.7 16.7 29.1 Weaving 1.8 18.9 9.6 26.1 Food 7.3 24.2 30.3 15.0 Other 2.8 23.0 12.0 25.3 Total 15.1 29.5 51.3 17.3 Notes: The unconditional mean (u.m.) of child labor incidence gives the average prevalence rate of child labor over all villages. Conditional means (c.m.) are calculated for the subsample of those villages where small ﬁrms operate in the given industry (for Total: in any industry).

12 The corresponding district level correlations amount to 0.383 and 0.077. Data refer to 2002. demand and supply interactions 59

Table 4.2: Geographic distribution of child labor

No. of Child labor % Small ﬁrms Province villages (u.m.) (c.m.) % of vill. No.(c.m.)

Sumatera 21113 11.0 31.3 35.2 5.6 N. Aceh D. 5773 9.8 28.8 34.0 5.0 Sumatera Utara 5377 8.0 30.1 26.7 4.2 Sumatera Barat 874 19.0 31.5 60.3 18.7 Riau 1625 9.8 28.8 34.0 3.8 Jambi 1189 12.1 31.0 39.0 4.5 Sumatera Selatan 2707 9.4 27.6 34.2 4.6 Bengkulu 1163 12.1 48.0 25.3 2.4 Lampung 2128 19.6 39.1 50.2 9.3 K. Bangka Belitung 317 14.8 22.7 65.3 11.2

Jawa 24952 19.5 27.3 71.5 30.2 DKI Jakarta 267 15.4 24.6 62.5 13.8 Jawa Barat 5753 21.2 30.2 70.1 24.4 Jawa Tengah 8554 21.3 27.1 78.6 40.9 DI Yogyakarta 436 28.4 30.5 93.3 84.2 Jawa Timur 8463 16.2 24.4 66.4 23.6 Banten 1479 19.2 32.3 59.5 15.6

Bali/Nusa Tenggara 3974 20.6 30.7 66.9 46.4 Bali 686 38.3 45.6 84.1 69.3 Nusa Tengg. Barat 738 34.0 44.3 76.8 63.4 Nusa Tengg. Timur 2550 11.9 20.0 59.3 35.3

Kalimantan 6014 15.4 39.3 39.2 7.7 Kalimantan Barat 1438 9.1 25.3 36.0 5.0 Kalimantan Tengah 1328 11.9 41.6 28.6 3.7 Kalimantan Selatan 1949 23.9 43.4 55.0 15.2 Kalimantan Timur 1299 13.2 44.2 29.8 3.5

Sulawesi 7659 16.2 29.6 54.6 9.7 Sulawesi Utara 1196 14.0 28.7 49.0 7.2 Sulawesi Tengah 1440 11.8 24.4 48.5 4.7 Sulawesi Selatan 3084 17.9 32.7 54.7 12.0 Sulawesi Tenggara 1563 18.7 29.8 62.8 11.9 Gorontalo 376 15.2 25.1 60.4 8.4

Maluku 1577 7.4 23.2 31.7 4.4 Maluku 836 5.6 20.4 27.5 4.0 Maluku Utara 741 9.3 25.1 36.6 5.3

Irian Jaya (Papua) 3507 3.2 35.1 9.1 1.6

Total 68796 15.1 29.5 51.3 17.3 Notes: Provinces are as of 2002. The unconditional mean (u.m.) of child labor gives the average prevalence rate of child labor in all villages. Conditional means (c.m.) are calculated for the subsample of those villages where small businesses operate. demand and supply interactions 60

Table 4.3: Number of ﬁrms per village by specialization patterns in child labor

Average No. of ﬁrms Means test No. obs. in given sector H-alt. p-value

Villages with child labor Food Wood in neither sector 382 (A) 15.7 (B) 6.1 A

Villages with child labor Food Ceramic in neither sector 170 (A) 20.4 (B) 8.7 A

Villages with child labor Wood Ceramic in neither sector 492 (A) 8.4 (B) 12.0 A

Villages with child labor Food Others in neither sector 218 (A) 21.2 (B) 20.3 A

Villages with child labor Wood Others in neither sector 510 (A) 7.2 (B) 19.4 A

Villages with child labor Ceramic Others in neither sector 239 (A) 12.7 (B) 21.0 A

The sectoral pattern of geographic incidence of child labor in manufacturing is given in Table 4.1. Child labor incidence is strongest in the food, ceramic and wood sectors, which have highest prevalence in villages. Conditional incidence is highest in food and ceramic sectors—a quarter of villages that have small firms in that sector have children working there. Child labor incidence by province is given in Table 4.2. Child labor incidence— measured as percentage of villages in which children work in small industries— is highest in Java (20%) and Bali/Nusa Tenggara (20%) and lowest in Irian Jaya (3%) and Maluku (7%). A very different picture emerges for child labor incidence conditioned on the existence of small firms: It is highest in Kali- mantan (39%) and Iran Jaya (33%) and lowest in Maluku (23%) and Java (27%). Regional incidence of child labor by sector is given in Table A.3 in the Appendix. The probability of child labor in manufacturing in a location is clearly increasing in the number of firms in that location. This can be seen in Table 4.3 which relates the specialization patterns of child labor across industries to the average number of firms in those industries. As location factors may be industry-specific, the table only looks at villages where two industries from any industry-pair are located, and compares the average number of firms in a given industry with and without child labor occurrence (given child labor outcomes in the other industry). In 22 (19) out of 24 comparisons, the average number of firms is significantly larger in villages where child labor occurs in that industry compared to villages without child labor in that industry at the 10% (1%) significance level. Since child labor incidence is determined by the number of firms in a given location, factors determining firm location matter for regional child labor incidence as well. This finding refers to the demand side for child labor.

4.3.2 Empirical strategy

Observable market outcome

Observable child labor incidence is an equilibrium outcome of the interplay of the demand for and the supply of child labor. The supply of child workers needs to ﬁnd its demand if child labor is to be observed; an exclusive focus on supply side determinants in empirical analyses thus implicitly assumes that the demand side will accommodate any change in the supply of child laborers. This is a restrictive assumption; in the case of child labor in small industries it is too restrictive as only around half of all villages have small industries at all (cf. Table 4.1). As documented in Table 4.3, the probability of child labor in a given location rises with the number of small ﬁrms; this suggests that the level of demand and supply interactions 62

economic activity is the driving force behind the demand for child labor. Thus the demand for child labor is affected by all variables that affect the location decisions of firms (and their subsequent growth). Among these variables are the proximity to market, the availability of qualified labor and credit, the quality of infrastructure and the geographic location, which influences accessibility and thus transport costs. Lastly, local purchasing power should matter as well. Poverty is a strong determinant of child labor supply as it indicates the need to rely on children’s income for household survival (cf. literature in section 4.1). Thus, poverty related variables should turn out significantly. The existence of negative income shocks (e.g. through diseases or unemployment) and the inability to cushion these effects through borrowing (i.e. insufficient access to credit) should increase child labor supply. Lastly, opportunity costs of child labor, especially the proximity and accessibility of schools, influence the decision to send children to work (cf. section 4.1). A particular problem arises in this context because some of the variables may capture demand and supply side effects at the same time. For instance, access to credit may reduce child labor supply, but it is also a location factor for (small) firms. Schools increase the quality of the labor force and thus capture a location factor, but they also increase the opportunity costs of child labor. Before addressing this issue and a possible sample selection bias created by endogenous firm location, village level supply and demand factors for child labor will be discussed.

The supply factors

In accordance with the literature, the supply of child labor can be expected to depend positively on the degree of poverty in a village. As the village level data set does not include detailed information on income, the regressions include the following proxies:13 The variable Bad housing measures the share of families that live in non-permanent houses. Because fertility is closely related to poverty, the controls include average family size in the village. Lastly, the share of families with electricity is inversely related to poverty. Income shocks may create the need to smooth consumption by sending children to work. Negative shocks are measured by the number of deaths by epidemics that occurred in the previous year as a percentage share of total population. To the extent that epidemics are not randomly located across Indonesia, the epidemic death rate might also be proxying for village poverty. The level of unemployment may increase child labor supply. If the main income earner becomes unemployed, labor supply by children may increase in order to (partly) compensate the loss in income.

13 Deﬁnitions and descriptive statistics of all variables used are given in Table 4.4. demand and supply interactions 63

Table 4.4: Variable deﬁnitions and descriptive statistics

Variable Mean St. dev. Min. Max. Deﬁnition

Child labor in small firms* 0.151 0.358 0 1 Indicates child labor being present in small businesses in the village Credit access* 0.538 0.499 0 1 Indicates general access to credit for households in the village Epidemic death rate (%) 0.029 0.261 0 28.9 Incidence rate of deaths bc. of an epidemic in previous year (in %) Bad housing 0.467 0.336 0 1 Share of families in the village who live in a non-permanent building Family size 4.259 0.953 1.0 24.4 Average family size in the village Unemployment (%) 3.948 6.052 0 100 Percent share of the unemployed to village population Primary school presence* 0.895 0.306 0 1 Indicates the presence of at least one primary school in the village Sec. school distance 0.639 1.455 0 10 Measures the distance to the nearest secondary school (in 10 kms) Families w. electricity 0.563 0.335 0 1 Share of families with electricity Village market* 0.241 0.428 0 1 Indicates the presence of a permanent or semi-permanent market in the village Urban* 0.174 0.379 0 1 Indicates urban villages or neighborhoods Coastal* 0.130 0.336 0 1 Indicates villages located on the coast Lower altitudes* 0.909 0.288 0 1 Indicates villages located not higher than 800m above sea level Distance to market* 1.218 2.184 0 10 Measures the distance to the nearest permanent market (in 10 kms) Small business credit* 0.319 0.466 0 1 Indicates access to the small business credit program (KUK, Kredit Usaha Kecil) in the village Population (.000) 2.992 3.811 0.01 69.77 Number of village inhabitants (in .000) No. small firms 17.34 63.80 0 1994 Number of small scale manufacturing firms in the village Notes: The descriptive statistics refer to the observations used in the regressions, N = 68344. Dummy variables are indicated by an asterisk. demand and supply interactions 64

Credit allows poor households to send their children to school instead of having to rely on their income to support the family. Especially if families experience a negative transitory income shock, they may borrow instead of using child labor as a consumption smoothing device. Villages where households have access to credit may thus be less likely to experience child labor, other things being equal. Yet, even if formal credit programs are available at the village level, the poorest households will not necessarily have access to credit facilities because of a lack of collateral. Even if people have access to credit, credit facilities may be insufficient or people may opt to compensate the loss of income by a combination of credit and child labor. The decision to send children to work rather than to school depends also on the opportunity costs of work. These are higher if schools are nearby and thus easy to reach. Obviously, children could work and go to school, but the likelihood that children work can be expected to be lower if they go to school (cf. Bhalotra and Tzannatos 2003). The likelihood of child labor can thus be expected to go down with the number of ’relevant’ schools in the village. Regressions include the presence of primary schools and separately the distance to the nearest secondary school. To make sure that the distance to secondary school does not act as just a proxy for the distance to the nearest larger town, the regressions also control for the distance to the nearest permanent market place. Lastly, the probability of child labor occurrence should increase with overall population, as there are more potential child laborers. To allow for nonlinear influence, the empirical specification includes a third order polynomial of population.

The demand factors

The demand for child labor depends on the overall demand for labor, which is a function of economic activity, and on the availability of the alternative to child labor—adult labor. The latter is proxied by the level of unemployment. Therefore, the higher the unemployment rate, the lower the demand for child labor, other things being equal. Yet, as noted in the previous section, unemployment raises the supply at the same time—only the impact on the market outcome can be observed. The level of economic activity in the sector of small manufacturing industries is proxied by the number of small firms. Because the number of firms alone does not completely capture the level of economic activity (as firms can have different activity levels) the regressions include location factors for these small firms as well, assuming that the factors that make firms locate in a certain village/neighborhood make them flourish as well. Small firms in these sectors are located only in 51.3% of all villages (Table 4.1). In order to iden- demand and supply interactions 65

tify the relevant location factors, a Heckman selection model is used with a dummy for firms being present as dependent variable at the selection stage and the number of small firms as dependent variable at the outcome stage.14 This guides the subsequent choice of variables for the demand side of child labor. Input availability is a major location factor. The regressions include the availability of skilled labor as proxied by the presence of a primary school and the distance to the nearest secondary school. Both turned out significant in the Heckman regression on the location of small firms: secondary school distance and primary school presence exerted significant negative and positive influences, respectively, on the probability of firm presence and, if present, on the number of firms. Both variables should at the same time reduce the supply of child labor (see above). Availability of credit, particularly when it is specifically designed for small businesses, is a major determinant for the decision to locate in a certain village. Access to credit also enhances the level of economic activity and thus the demand for child labor, other things being equal. In the empirical model two different measures of credit availability are used: broad credit access, that includes all forms of formal credits, and small business credit that caters specifically to the firms under consideration. PODES collects data on various forms of subsidized and unsubsidized formal small scale credits available within a village. Most importantly, 31.9% of the villages report the presence of an unsubsidized investment credit program for small businesses (KUK, Kredit Usaha Kecil). This credit program has been introduced in 1990 in order to provide loans of limited size to small businesses and is targeted at small firms and not households (Lapenu 1999). PODES records information on other loan programs such as farm credits or housing credits and on whether any other form of formal credit is available.15 Based on this, two binary variables are coded: Credit access takes one if there is any form of formal credit available in a village, while small business credit takes one if the village has access to the firms’ investment credits. The broad measure of credit access is a demand factor and a supply factor at the same time as it tends to increase economic activity and opens up the opportunity to send children to school without an unsustainable drop in consumption. In contrast, the small business credit works only on the supply side as it is available only for business purposes. Market size and proximity to the relevant market are major determinants for firm location and firm success. Market proximity is measured by the dis-

14 The regressions use the share of families with electricity as an instrument to identify the ﬁrst stage as it turns out to be only correlated with the presence but not with the numbers of ﬁrms. Regression results are given in Table A.4 in the Appendix. 15 In more than half (53.8%) of the villages households have access to some form of formal credit; in 21.9% of villages there is access to microcredit, but no access to the business credit program (KUK). demand and supply interactions 66

tance to the nearest permanent market. Purchasing power of the local population is proxied by population size and the variables indicating poverty, i.e. the share of population with (non-permanent) bad housing, average family size, and share of families with electricity. The latter variable also indicates the quality of infrastructure and thus should exert a positive influence beyond the income and wealth effects. These variables all are demand factors as well and exert an influence opposite to their effects on child labor supply (see above). Better access to infrastructure and lower transport costs are reflected by the geographic variables, indicating a coastal location or a location in lower altitudes. The variable indicating urban neighborhoods or villages captures in addition to closer proximity to markets a better infrastructure as well as a higher local purchasing power. They are all expected to raise economic activity. To control for unobservable characteristics of various parts of the archipelago such as differences in culture, institutions, and mentality, regressions include province or district fixed effects.

Econometric speciﬁcation

As only the existence, but not the extent of child labor in a village can be observed, the probability of child labor incidence L in village j, region k is estimated by probit models of the form

P r(Ljk =1) = P r(xjkβ + λk + ǫjk > 0) (4.1) where xjk is a vector of controls specified above and λk stand for province or district fixed effects. Because child labor in small manufacturing firms can occur only in locations that have such small firms, the probit model is run only on the subset of villages with small firms. As child labor can be observed only in those villages where small manufacturing firms operate, probit models lead to biased estimates on those variables that affect not only child labor but also industrial location. As argued before, this applies to a number of variables. For example, the opportunity costs of child labor should decrease with the distance to the nearest secondary school, that is, school availability reduces child labor incidence. At the same time, if small firms locate preferably in villages with a population with higher basic skills, the effect of school availability on child labor will be biased downwards and may even change its sign. In other words, for any given number of firms, school availability reduces child labor; but since the number of firms depends positively on school availability and more firms increase the likelihood of child labor, school availability has two countervailing effects on local child labor incidence—the opportunity cost effect and the location effect. The former refers to the supply side of, the latter to the demand side for child labor. In a simple probit estimation, only the net effect can be explained for the restricted demand and supply interactions 67

sample of villages that have small firms. It may have either sign. If the location were random, we would still observe the effect of school availability on the supply of child labor. A possible demand side effect of increased schooling would be limited to overall higher economic activity, if at all, but no longer lead to a sample selection bias with firms locating in villages with more schools.16 A similar argument can be made e.g. for credit availability. This sample selection problem is addressed by estimating a probit model with sample selection correction (cf. Wooldridge 2002, 570), where at the selection stage the binary variable Sjk is used, which indicates whether small firms operate in village j in region k. Firm location is assumed to depend on 17 the same vector of locational factors xjk as before:

P r(Sjk) = P r(xjkγ + zjkφ + λk + νjk > 0) (4.2) The model is estimated by MLE under the assumption of joint normality of ǫjk,νjk. At the outcome stage, the existence of child labor is explained by the same regression as in (4.1). In these models, identification comes from two sources. First, by including province or district fixed effects λk, within province/within district variation in economic conditions is used to explain child labor. As economic and cultural backgrounds vary vastly between Indonesian regions, regional fixed effects can gauge large parts of this unexplained heterogeneity. Second, the models with sample selection use the availability of a credit scheme targeted to small businesses as an instrument zjk. After controlling for overall credit access, the availability of specific business credits that are targeted only at small businesses for business purposes should not capture any supply effect of child labor (cf. previous section). Households do not have direct access to it and thus cannot use it to finance consumption that they would need to forgo if they decided to send their children to school. At the same time, the availability of specific small business credits is an important location factor for small businesses (and increases their number), and thereby raises the demand for child labor by increasing firm activity.18 Thus, after controlling for overall credit availability, small business credits affect child labor only through their effects on the presence of small firms, but not through any channel independent of the location of firms. 16 Strictly speaking, the demand effect at the village level consists of a sample selection effect—villages with more schools tend to have a larger share of small firms—and a level effect: more schools lead to overall higher economic activity. The former can be controlled for, the latter effect will be observed jointly with the supply side effect. 17 The only difference between the two sets of controls is that in equation (4.2) xjk includes only a third order polynomial in population size (instead of the bivariate third order polynomial in population size and firm numbers of equation (4.1).) 18 Small business credits have turned out significant in the Heckman regression on small firm presence and number of firms present; they were completely insignificant in the second stage of the probit with sample selection (based on the presence of small firms in the first stage). demand and supply interactions 68

4.4 Results

Table 4.5 reports two sets of results. Columns 1 and 2 contain the results from the probit model on child labor occurring in a village/urban neighborhood conditional on small manufacturing firms being present in that location. The exclusion of the locations without small firms makes sense, because PODES measures only child labor in that sector. Including locations that cannot have the type of child labor measured in this study would lead to biased estimates. The conditional probit model of columns 1 and 2 thus estimates the determinants of child labor incidence for the set of locations that have small firms.

However, this sample is not representative for all of Indonesia, because the villages with small firms have different characteristics than the ’average’ village; they are more attractive locations for small firms (cf. Table A.4). Determinants of child labor will have a different impact in a random sample than in this biased sample. Thus additionally the results of a probit model with sample selection are also reported. The endogenous variable at the selection stage is a binary variable that is one if small firms are present in a location (columns 5 and 6); the endogenous variable at the outcome stage is a binary variable for the occurrence of child labor in that location (columns 3 and 4). Both models include fixed effects at the province level (odd numbered columns) or at the district level (even numbered columns) to control for unobserved heterogeneity. While district level fixed effects allow to capture unobserved heterogeneity at a more disaggregated level they come at the cost of reduced observations—districts that have only villages/neighborhoods either with or without child labor drop out in both models; in the model with sample selection districts that have only villages with or without small firms are dropped in addition. Overall, results are similar. Table 4.5 reports marginal effects evaluated at the sample mean. In the conditional probit model, access to credit has no significant effect on child labor. This result is at odds with conventional wisdom, but it masks two countervailing effects: Credit access increases significantly the likelihood of firms being present in a location by around 15% points. This shows a significant demand side effect of credit access. At the outcome stage credit access reduces child labor. This effect is significant for the model with province controls—given that there are firms, the availability of credit reduces the probability of child labor occurrence by 7% points. This portrays the typical supply side effect of credit availability on child labor. The effect of credit programs specifically targeted at small businesses increases the probability of child labor in the conditional probit model significantly by 2% points; this effect can be traced back to a significantly higher probability of small firms locating in that village. As argued in the previous section, small business credit serves to identify the probit model with sample selection—it turns out highly significant at demand and supply interactions 69

Table 4.5: Child labor in small scale manufacturing (with sample selection)

Dependent Child labor in small ﬁrms Small ﬁrm presence Model Cond. probit Probit with sample selection (A) Probit (B) Outcome stage (C) Selection stage (1) (2) (3) (4) (5) (6)

Credit access -0.0020 -0.0010 -0.0720* -0.0404 0.1499** 0.1409** (0.0143) (0.0145) (0.0334) (0.0342) (0.0116) (0.0117) Epidemic death rate (%) 0.0966** 0.0839** 0.0932** 0.0873* 0.0028 0.0099 (0.0305) (0.0310) (0.0334) (0.0349) (0.0124) (0.0109) Bad housing 0.0265 0.0388* 0.0507* 0.0618** -0.0598** -0.0662** (0.0195) (0.0182) (0.0225) (0.0239) (0.0202) (0.0199) Family size 0.0039 0.0006 0.0107† 0.0045 -0.0159** -0.0124** (0.0049) (0.0041) (0.0059) (0.0050) (0.0050) (0.0043) Unemployment (%) 0.0033** 0.0028** 0.0033** 0.0030** 0.0002 0.0006 (0.0006) (0.0007) (0.0008) (0.0009) (0.0007) (0.0007) Primary school presence 0.0294† 0.0281† -0.0160 0.0047 0.0873** 0.0761** (0.0159) (0.0140) (0.0254) (0.0223) (0.0150) (0.0107) Sec. school distance 0.0047 0.0040 0.0110* 0.0089 -0.0111** -0.0116** (0.0052) (0.0050) (0.0055) (0.0059) (0.0040) (0.0039) Families w. electricity 0.0509* 0.0604** 0.0355 0.0528† 0.0340 0.0461* (0.0231) (0.0211) (0.0289) (0.0276) (0.0217) (0.0204) Village market 0.0103 0.0085 0.0018 0.0061 0.0253** 0.0165* (0.0072) (0.0067) (0.0091) (0.0079) (0.0075) (0.0067) Urban -0.0035 -0.0019 -0.0143 -0.0133 0.0354** 0.0632** (0.0106) (0.0101) (0.0129) (0.0134) (0.0146) (0.0134) Coastal -0.0080 -0.0063 -0.0218 -0.0129 0.0326** 0.0212 (0.0153) (0.0149) (0.0170) (0.0177) (0.0144) (0.0136) Lower altitudes -0.0183 -0.0209 -0.0568* -0.0471† 0.0900** 0.0899** (0.0242) (0.0236) (0.0259) (0.0283) (0.0241) (0.0194) Distance to market -0.0031 -0.0017 0.0041 0.0032 -0.0129** -0.0139** (0.0034) (0.0035) (0.0045) (0.0048) (0.0028) (0.0029) Small business credit 0.0207† 0.0206† 0.0590** 0.0546** (0.0113) (0.0108) (0.0125) (0.0122) Rho -0.6633* -0.4936† (0.3769) (0.3236)

Further controls Yes Yes Yes Yes Yes Yes Fixed effects Prov. Dist. Prov. Dist. Prov. Dist. No. of Prov./Dist. 30 365 30 360 30 360 N (observations) 35093 35021 35093 34867 68344 67706 Pseudo R2 0.035 0.090 Notes: The table reports marginal effects (evaluated at the sample mean) from probit regressions conditional on the presence of small firms in a village, and from a probit model with sample selection correction. All equations include a constant and province or district fixed effects. Further controls include a third order polynomial in population size at the selection stage and a bivariate third order polynomial in population size, the number of small firms, and all their interactions in child labor regressions. Robust standard errors (clustered at district level) are reported in parentheses. *,**,† denote significance at the 1, 5, and 10% level. demand and supply interactions 70

the selection stage, but is insignificant at the outcome stage. This is not surprising, as the variable captures the additional effect of specific credit facilities for small businesses (beyond the broad measure of credit), which attract small businesses but are not available to households to finance consumption. The conditional probit model suggests that the presence of a primary school increases child labor incidence significantly, while the distance to secondary school has no impact. Again, this runs counter the notion that lower costs of schooling (i.e. more accessible schools) reduce child labor. The results that control for sample selection show that the presence of a primary and the proximity of a secondary school are significantly associated with a higher probability of firms locating in that village/neighborhood and thus indicate favorable location factors, especially a labor force with better skills. At the outcome stage, however, child labor incidence increases with rising distance to the nearest secondary school (model with province controls) while primary school presence has no statistically significant impact. This set of results shows that close-by schools attract firms and would-be child laborers alike, but that this relationship is stronger for firms than for children. Unemployment and epidemic death rate do not influence location decisions, probably due to the transitory nature of the underlying shocks. Therefore, both factors have the same effect in the conditional probit model and at the outcome stage of the model with sample selection: They significantly increase child labor. For the epidemics this was to be expected; in case of unemployment the increased child labor supply effect seems to outweigh the effect of a larger idle adult workforce which could be hired instead of children. Bad housing increases child labor in the conditional probit model with district controls and is insignificant in the model with province controls. Family size, the other proxy for poverty, does not turn out significant in both specifications of the conditional probit model. Yet, again, these results veil the underlying forces: Poverty variables are associated with a lower likelihood of firms being present, they serve as a locational disincentive (as the local purchasing power is lower); at the outcome stage, a larger share of households with bad housing significantly increases the probability of child labor. In other words, the conditional probit model underestimates the supply side effect of bad housing, or more generally of poverty, due to a countervailing demand side effect. The same applies for family size, yet the relationship is not as close. Closer distance to the next permanent market and being an urban or a coastal neighborhood/village make the location more attractive for small industries and thus increase the probability of small firms residing in that location. Yet they have no significant effect on child labor at the outcome stage and are insignificant in the conditional probit model as well. Locations in lower altitudes are more attractive for firms, but have a lower probability of child labor at the outcome stage; in the conditional probit the variable is insignif- demand and supply interactions 71

icant. The share of families with electricity increases child labor incidence in the conditional probit model; in part due to a positive effect on the probability of firms locating in that village as the infrastructure is better and also due to a higher probability of child labor at the outcome stage (only in the model with district controls). One potential explanation for this latter effect could be that infrastructure quality (electrification) increases the intensity of firm activity (beyond its location and numbers) to an extent that the demand effect on child labor dominates the supply side effect of more better-off households. The regressions control for a third order polynomial in population size, which turns out significant in all regressions (not reported): with rising population the probability of child labor increases significantly (in a nonlinear way). This captures a supply side effect: As the number of potential child laborers increases, it becomes more likely that child labor occurs. The polynomial in the number of firms in the village is also highly significant— child labor becomes more likely the larger the number of firms in the village.19 This captures an important demand side effect: The sample selection correction dealt with the sample selection bias in the probit model that arose from analyzing only villages/neighborhoods that had at least one small manufacturing firm. The outcome stage thus includes variables that measure supply and demand factors given that there is a least one firm. An obvious control for the demand level is the number of firms in the village/neighborhood. Thus the other variables at the outcome stage include demand side effects only insofar as they have not been controlled for by the number of firms, but they capture the entire supply side effect.

4.5 Conclusion

This chapter has analyzed the geographic incidence of child labor in small manufacturing firms in Indonesia with the help of a unique data set. Previous empirical studies on child labor have either used large household data sets or cross-country data sets. In contrast, this study was based on a data set covering virtually all Indonesian villages and urban neighborhoods. The geographic variation in these data allowed for a distinction between demand and supply side determinants of child labor in a way that would not be possible in other types of data sets. The conditional probit model on child labor for the set of villages that have small manufacturing firms have provided counterintuitive results. For instance, overall credit access and the distance to secondary schools do not affect child labor in the model; villages that have primary schools actually have a higher probability of child labor. Because the set of villages/urban neighborhoods

19 The interaction effects between population size and firm numbers are also significant. demand and supply interactions 72

that have small industries—a prerequisite for observing child labor in that sector—is a non-representative sample of all Indonesian villages, a probit model with sample selection has been applied. This estimating procedure addressed explicitly the factors that lead to the presence of small firms and thus allow the demand for child labor to materialize. Many of these location factors—e.g., proximity to schools, access to credit, market size as measured by the absence of poverty—are variables that affect the supply of child labor at the same time. The selection stage thus focused on demand factors only. The outcome stage of the probit model with sample selection estimated the effect of the variables on the occurrence of child labor, given that small firms are present and controlling for the number of firms present. Thus at that stage main determinants of the demand for child labor were controlled for and therefore the results are more in line with conventional wisdom of the literature that has focused predominantly on supply side determinants. For instance, the findings show that access to credit attracts small firms and raises their level of economic activity, thereby raising the demand for child labor. At the same time access to credit reduces child labor supply. As a result of these countervailing effects, the net effect in a conditional probit model is insignificant. Yet, this blurs the underlying relationships. As a second example, poverty related variables such as the share of people living in non-permanent houses, exert a negative influence on the probability of small firms being present in a location, and if so, on the number of small firms. Thus poverty reduces the demand for child labor. At the same time it increases the supply of child laborers. An aggregated view on the impact of poverty on child labor would thus underestimate the extent to which poverty makes families ready to send their children to work. A similar line of reasoning applies to school accessibility. This disaggregated perspective not only enhances our understanding of child labor incidence; more importantly, it has significant policy implications. Most factors that increase the demand for child labor are linked to the level of economic activity. As an increase in economic activity is desirable for many reasons, notably poverty alleviation, these variables do not provide an entry point in the fight against child labor. Yet, the finding that unemployment increases child labor incidence does provide such an entry point on the demand side. Unemployment increases the supply of child labor (as household income is reduced by unemployment); at the same time however it increases the pool of available adult laborers, who are a substitute to child laborers. Thus making adult laborers more attractive to employers compared to child laborers could reduce child labor incidence. As most child laborers work in their parents’ businesses our finding points to moral hazard problems in the labor market preventing potential employers to hire adult labor outside the family instead of making their own children work despite the high returns to educational investment. On the supply side the results of this study corroborate earlier findings— demand and supply interactions 73

enhanced credit and school accessibility and poverty-reducing policies decrease child labor supply. As the supply side provides the better entry point for policy interventions, old prescriptions still apply. Yet they will not necessarily lead to a reduction of child labor in the short run. At a regional level these policies may instead lead to an increase in child labor due to increased economic activity. This may be frustrating to local policy makers, yet it does not invalidate their general approach. In the worst case, we may see an inverted U shape relationship between child labor incidence and economic development. However, if policies that make a location more attractive for economic activity, such as provision of credit, schools, and other infrastructure, are complemented by policies that are specifically designed to reduce the supply of child labor, for instance through cash transfers conditional on school attendance or free school meals or free school transport, it may be possible to escape the transitory increase in child labor resulting from the applying the ’right’ policies. The findings in this chapter thus suggest that it may not be enough to pursue growth enhancing policies by providing education and infrastructure in order to fight child labor. chapter 5

Trade Liberalization and Child Labor: The Role of Income Eﬀects

Abstract

This chapter considers the effects of trade liberalization on child labor that arises out of subsistence needs. It argues that favorable income effects are most likely to reduce the need for child labor in the South, even when export goods have a necessity character. However, in very poor economies, aggregate hours of child labor can also increase as a result of more open trade. Although the poorest families are the ones who benefit the most from trade in a Heckscher-Ohlin setting, their income gains might not be high enough to make them withdraw their children from work, while adverse income effects can raise the incidence of child labor among the less poor. Empirical support for the argument is provided by the finding that in a country panel, increases in trade openness are associated with significantly smaller reductions in child labor among the poorest food exporters than among food exporters on average.1

5.1 Introduction

Triggered by the globalization debate over the last decade, awareness about child labor has increased hugely, both among policy makers and in the public. The link between international trade and child labor has been hotly debated in particular. Does globalization increase the incidence of child labor, or conversely, does trade liberalization reduce the numbers of working children? While some of the most inﬂuential theoretical analyses (Basu and Van 1998, Baland and Robinson 2000) address child labor in a closed economy setting only,2 several studies look at the interactions between trade openness and child labor. These studies can be divided into two main categories: Earlier

1 This chapter is based on Kis-Katos (2007). 2 The closed economy focus restricts the applicability of some of the policy conclusions. The favorable impacts of a ban on child labor (Basu and Van 1998, Baland and Robinson

74 trade liberalization and child labor 75

open economy models treat child labor either as a fully exogenous factor of production (Brown, Deardorff and Stern 1996) or as a “public bad” (Maskus and Holman 1996, Maskus 1997). More recent studies address endogenous child labor supply which is based on human capital investment decisions that are made under binding borrowing constraints (Grote, Basu and Weinhold 1998, Ranjan 2001, Jafarey and Lahiri 2002). These studies show that international trade affects child labor via two main effects: 1. income effects arise since goods and factor price changes alter real family income, 2. incentive/substitution effects arise from changes in the present and future wages of a child that change the opportunity costs and returns to education.3 Although the income effects in these models unambiguously reduce child labor, this reduction can be more than offset by an increase in the opportunity costs of a child’s time (because of rising child wages and falling future skill premia). Thus, the effect of trade liberalization on child labor remains mainly an empirical issue. The empirical evidence on the relationship between trade openness and child labor does not show a unified picture. In a cross-country setting, measures of trade openness and economy-wide child labor are negatively correlated. Cross-sectional studies find that more open economies tend to have lower child labor force participation rates, even when other factors, like per capita income, economic structure, or educational quality, are controlled for (Shelburne 2001, Neumayer and de Soysa 2005). By contrast, based on a smaller panel of developing countries, Cigno et al. (2002) do not find a robustly significant association between trade openness and child labor. Edmonds and Pavcnik (2006) argue that the only channel through which the effect of trade openness is transmitted is by raising per capita GDP.4 Studies based on microempirical data, which are generally better suited to investigate the relative strengths of income and substitution effects, show that the overall effect of trade liberalization on child labor differs across countries. Edmonds and Pavcnik (2005b) find that the increase in the price of

2000, Krueger and Tjornhom Donohue 2005, Doepke and Zilibotti 2005) hinge on the presence of wage adjustments that result from a restriction of child labor supply. However, in a small open economy factor prices will be largely determined through international trade at the world market, and hence the room for factor price movements is severely limited. Under these conditions, a direct ban on child labor can turn out harmful for the poor and is bound to encounter enforcement problems (Dixit 2000). 3 By contrast, trade liberalization does not alter child wages in the model of Dinopoulos and Zhao (2007) as children are assumed to be paid nutrition efficiency wages. Within the short term perspective of a sector specific factors model—with skilled labor as the common factor, child and adult unskilled labor as specific factors in the agricultural sector, and capital in the modern sector—the effects of trade liberalization depend on the specialization patterns of the economy. If the agrarian good produced by child labor is exported, a relative increase in its price will increase child labor without changing the efficiency wage paid to the children. 4 To resolve potential endogeneity problems, they construct a measure of trade openness that is based on economic geography only (following Frankel and Rose 2002). However, the applicability of their results is limited as measures of geographic openness, although exogenous to child labor, are unable to capture the effects of trade liberalization itself. trade liberalization and child labor 76

rice in Vietnam caused by trade liberalization reduced significantly child labor through favorable income effects. By contrast, Edmonds et al. (2007) conclude that schooling in India increased less in those rural districts that were stronger exposed to tariff cuts due to their industrial employment structure.5 Kruger (2007) shows that in coffee producing parts of Brazil, child labor in poor and middle income families increases with regional coffee production, and hence incentive or demand effects must dominate the favorable income effects. This study puts forward an additional explanation for why child labor increases as a result of trade liberalization. While previous theoretical studies stressed mainly the trade-off between the favorable income and the unfavorable incentive effects, it is important to realize that within the poorest countries, even the real income effects of trade can raise aggregate child labor. The analysis follows Basu and Van (1998) by addressing child labor that arises out of subsistence needs,6 and adapting their main framework for a small open developing economy. Within this framework, trade liberalization can both raise or reduce the need for child labor in an individual family. The reaction of individual child labor supply depends on the distributional impacts of trade at the family level. The potential decrease in aggregate child labor is due to the poverty reducing effect of international trade, both via an increase in average income and a reduction in inequality.7 However, an increase in trade openness will fail to reduce child labor in aggregate terms if the economy is extremely poor and gains from trade are not large enough. This possibility arises if trade liberalization moves relatively better-off families towards subsistence while the favorable income effects reduce child labor by less at the lower end of the income distribution. Nevertheless, the overall policy message remains positive: Income effects of trade can reduce the need for some of the worst forms of child labor, captured in this setting by longest hours of constrained work. The empirical evidence presented in this chapter supports the potential relevance of this argument. In a country panel that extends over four decades

5 See on this also Chapter 6, which shows that child labor decreased by more in those Indonesian districts that were more strongly exposed to trade liberalization. This was most likely due to the favorable income effects for the poor. 6 Although the complexity of causes of child labor—resulting from an interaction between the work-schooling trade-off (cf. Chapter 3), market imperfections, social norms, missing parental altruism and intrafamily bargaining failures—is now widely acknowledged, poverty is still regarded as one of its most substantial explanatory factors. See for a review of early empirical studies Grootaert and Kanbur (1995). More recent evidence on the determinants of child labor based on microempirical research is summarized by Bhalotra and Tzannatos (2003) and Edmonds (2007). 7 The role of income inequality has been addressed by Swinnerton and Rogers (1999) and Rogers and Swinnerton (2001) who analyze the potential effects of income redistribution on child labor in a somewhat similar framework in a closed economy setting. This chapter incorporates their main insights, but it considers a way of income redistribution that has arguably larger policy relevance: In developing economies international trade can achieve actual redistribution of income while the scope for income redistribution via transfers is very limited. trade liberalization and child labor 77

(1960–2000), increases in trade openness were associated with significantly smaller reductions in child labor within the group of the poorest food exporters as compared to food exporters on average, after controlling for per capita growth, urbanization, as well as regional and time effects. The remainder of this chapter proceeds as follows. Section 5.2 sets up a basic model of a small open economy with endogenously determined child labor. Section 5.3 derives the comparative static effects of a marginal increase in trade openness on child labor outcomes for an individual family and the economy. Section 5.4 offers empirical support to the main theoretical argument, while Section 5.5 concludes.

5.2 The model

5.2.1 Production and prices

Consider a small open economy of Heckscher-Ohlin type that produces two goods, food X1 and a composite manufacturing good X2. The manufacturing good is taken as num´eraire and the relative price is given by p. Production relies on two intersectorally mobile factors, labor L and composite (human and physical) capital K, that are fully owned by the domestic households. All markets are perfectly competitive and hence both factors are fully employed. The resulting zero proﬁt and full employment conditions can be written as:

aL1w + aK1r = p aL2w + aK2r = 1 (5.1) aL1X1 + aL2X2 = L aK1X1 + aK2X2 = K where w stands for wages per efficiency unit of labor, r for the returns to capital, and aLi(w, r), aKi(w, r) [i = 1, 2] are the optimal per unit input coefficients of labor and capital that depend on factor prices. The country is imperfectly integrated into the world economy; barriers to trade such as tariffs and transportation costs drive a wedge between the domestic relative price p and the world market price pw:

p = βpw. (5.2)

By assumption, good 1 (food) is produced more labor intensively. As the home country is relatively more abundant in (unskilled) labor than the rest of the trading world, food is assumed to be the export good of the economy.8 Thus,

8As food is considered a necessity in this model, at least up to a certain level of consumption, it will only be exported at given relative prices p if the labor-abundant economy is productive enough to provide for domestic consumption, and hence no trade pattern reversals occur. This assumption will be made throughout the model. trade liberalization and child labor 78

β < 1. As long as both goods are produced (factor endowments stay within the diversiﬁcation cone), factor prices w(p) and r(p) are determined by the given output prices only.

5.2.2 Factor ownership and household income

Assume that there are N households in the economy that are regarded as single decision making units. Household size is normalized to one parent and one child per household. Household i is endowed with 1 unit of adult labor and with ki ∈ (0, kmax) units of composite capital, both inelastic in supply. Capital distribution in the economy can be described by a density function i g(k). Adult income ya is given as a sum of adult wage and capital income,

i i ya = w + rk . (5.3a)

Children can also contribute to household income as they are assumed to be perfect substitutes to adult labor in production (“substitution axiom” by Basu and Van 1998). Denote the fraction of time that a child of family i spends working by li where 0 ≤ li ≤ 1. Since children are less productive than adults with the eﬃciency parameter γ (0 <γ< 1), if they are working, they earn the i income yc:

i i yc = wγl (5.3b)

5.2.3 Household preferences over consumption and child labor

Assume that all households have identical preferences over consumption of i i food c1 and the manufacturing good c2. In order to emphasize the eﬀects of subsistence needs, a certain amount s of food consumption is considered a necessity. As long as s is not achieved, families consume only food and none of the manufacturing good. Utility of consumption is thus deﬁned in two steps:

i i ′ ι(c1) if c1 ≤ s with ι(s)= u0, ι > 0 U i = (5.4) i i i ( u0 + u(c1 − s, c2) if c1 >s

The ﬁrst line of the utility function U i describes a situation where the family lives at or below the poverty line and spends all of its resources on food. The second line gives the utility function of the household living above the subsistence threshold, which is an increasing, strictly concave and homothetic i function with two arguments, consumption of additional food c1 − s and of the i second good c2. Child labor is assumed to exist only as long as it is needed to achieve subsistence consumption (“luxury axiom” by Basu and Van 1998). trade liberalization and child labor 79

Thus, children are compelled to work only in those families where adults cannot ﬁnance the minimum amount of food on their own.9

5.2.4 Household decisions and child labor outcomes

Consumption decisions arise from households maximizing utility (5.4) subject to the family budget constraint. In families where adult income is higher than i the costs of subsistence, ya > ps, utility is maximized subject to the adult i i i budget constraint ya = pc1 + c2. With homothetic preferences, these families spend constant fractions of adult income on both additional amounts of food i i c1 − s and the composite manufacturing good c2. In families in which adult i income falls short of the subsistence costs, an income gap ps−ya arises. By the “luxury axiom”, children work only to the extent needed to close this income gap:

i i 0 ≤ yc ≤ ps − ya (5.5a)

Child income is given by the child wage (equation 5.3b), while child labor in family i is determined by the income gap and child wages:

i 1 if ya ≤ ps − wγ,

i i l =  ps − ya i (5.5b) c  wγ if ps − wγ

ks ks = max ((ps − w(1 + γ))/r, 0 ) η = g(k)dk (5.6a) Z0 ks ks = max ((ps − w)/r, 0 ) θ = g(k)dk (5.6b) Zks 9 This presumes that for any adult income above the costs of subsistence the marginal disutility of child labor always exceeds the marginal utility of consumption. trade liberalization and child labor 80

k i

(ps−w)/r k s

γ))/ (ps−w(1+ r k s

0 i 1 lc Figure 5.1: The hours of child labor supply

With capital endowment of ks, a family is just able to ﬁnance subsistence consumption when children work full time; with capital endowment of ks, adult income just covers the costs of subsistence and no child labor is necessary.

5.2.5 Aggregate child labor supply and goods and factor market equilibria

Aggregate hours of child labor supply LC are given by the sum of individual households’ child labor supplies: pS − Y L = L + L = ηN + θ Aθ (5.7) C Cη Cθ wγ

LCη denotes the total hours of labor by children who work full time; LCθ is the sum of interior child labor hours where child income YCθ = wγLCθ is 10 just covering the aggregate “income gap”, pSθ − YAθ. Since markets clear perfectly, factor employment equals aggregate factor supply measured in eﬃ- ciency units, i.e. L = N + γLC . The resulting domestic income is given by Y = wL + rK = pX1 + X2. The balanced trade equilibrium of the economy is described by setting the value of export supply equal to import demand at world market prices. In equilibrium, the given world market prices and market frictions determine domestic prices. Together with the production technologies and factor endowments they determine factor rewards w, r, and hence family incomes, consumption and production patterns, as well as LC , the level of child labor in the economy. 10 The aggregate “income gap” is the diﬀerence between aggregate food consumption Sθ = sθN and aggregate adult income YAθ = wθN + rKθ in those families in which child labor is interior. trade liberalization and child labor 81

5.3 Trade liberalization and child labor

In the policy discussion, greater openness to trade is seen both as one of the causes of child labor problems world-wide and as a potential cure for them. Globalization critics often blame international trade for setting new incentives for child labor abuse in the South. By contrast, many economists believe that gains from trade can make a powerful contribution to eradicating child labor. However, while the average income eﬀect always reduces child labor, reductions in inequality in the South can also raise child labor (cf. Rogers and Swinnerton 2001).

5.3.1 Impact of trade on goods and factor prices

Trade liberalization brings domestic relative prices p closer to the world market prices pw. In terms of the present model, β rises (hats denote percentage changes):

p = β > 0, (5.8a) and hence food becomes relatively more expensive, which raises the costs of b b subsistence. At the same time factor prices also adjust, wages rise in terms of both goods, while capital returns fall in absolute value (Stolper and Samuelson 1941):

r< 0 < p< w. (5.8b)

For any given level of child labor, the distributional impacts of these price b b b changes are well-known: Despite the rise in subsistence costs, the poorest ben- eﬁt from a trade opening as wage increases more than compensate them for a rise in the price of food. Relatively richer families experience a relatively smaller real income gain or even a loss of real income as their capital income diminishes. As a result, income inequality in the economy is unambiguously reduced. The changes in the “income gap” trigger a child labor supply response, individual child labor decisions adapt, and hence economy-wide child labor changes.11

5.3.2 The reaction of individual child labor supply

The impact of trade liberalization on child labor depends on the changes in adult and child income relative to the rise in the relative price of food. In 11 There will be also a further production effect that arises due to a change in child labor supply, and that alters the economy’s production possibilities set. This leads to a reallocation of resources between the sectors via the Rybczynski effect (1955). Since under the assumption of imperfect specialization it has no further impacts on family income, this last production effect will be ignored in the present analysis. trade liberalization and child labor 82

families where children work some but not all the time, the reactions of child i labor supply lc can be divided into three parts:

ps w rki li = p − w + r − w. (5.9) c yi yi yi c c c First, there is an adverseb price effectb that reducesb realb familb y income as it makes food relatively more expensive. Second, there is an adult income effect (within the parentheses) that can take both signs depending on factor endowments of the family and the relative magnitudes of factor price changes. The larger the share of wage income in total adult income, the more favorable are the adult income effects. Third, there is a child income effect that unambiguously reduces child labor supply. The overall real income effect is clearly favorable for the poorest families as they experience the real gains from a wage increase but remain mainly unaffected by the falling returns to capital. Child labor supply in these families decreases. Families that have more substantial capital holdings adjust their supply of child labor upward or downward depending on the overall sign of the real income effect.

5.3.3 Change in economy-wide child labor supply

Changes in economy-wide child labor depend on the distribution of the income effects to individual families living at or near to the subsistence threshold. With a price change, aggregate hours of child work (5.7) change for two reasons: as a result of the accumulated changes in individual child labor supply among those families where child labor is interior, and due to changes in the number of children working full time and those working at all. In the present model, marginal changes of the number of child laborers have no first order effects on aggregate child labor (i.e. dLC = dLCθ, see Appendix B.1). Thus, the relative change in aggregate hours of child work is determined as:

pS wθN rK L = θ p − w + θ r − w. (5.10a) C Y Y Y C C C which is the economy-wideb equivalentb of expressionb (5.9).b b Under the assumption of perfect competition, factor price movements depend on the change in goods prices p . The change in child labor supply (5.10a) can thus be reformu- lated in terms of the relative price change (using the magnification effects of Jones (1965),b see Appendix B.2): 1 1 LC = − w(θN + γLCθ − aL1Sθ) p (5.10b) δ YC Child labor supplyb reactions depend on two main factors: Firb st, on the production structure, captured by the difference in unit cost shares between industries, δ(> 0), and by the unit input coefficient aL1, and second, on the importance of trade liberalization and child labor 83

subsistence consumption relative to family endowments. Child labor decreases only if the labor supply of those living at the poverty line θN +γLCθ is greater than the labor embodied in their subsistence consumption aL1Sθ. If no families are living under the poverty line, i.e. η = 0 and LCθ = LC , the income effect of more open trade unambiguously leads to a decrease child labor (cf. Appendix B.3). Intuitively, if there are no families where children would work maximum hours, income gains to the poorest are directly translated into reductions in child labor supply among them. As average income in the economy increases because of the gains from trade, aggregate hours of child work cannot increase, irrespectively of the increases in child labor supply among the less poor. How- ever, if some parts of the population live below the poverty line, i.e. η > 0, there might be situations where the favorable income effects of trade are overturned by the real income losses to the relatively less poor. Such outcomes can arise if despite their real income gains, the poorest still have to send their children to work full-time as they still cannot reach subsistence, while relatively richer families are constrained to rely on more child labor because of their income losses.12 Aggregate hours of child work can increase if: 1. the redistribution of income induced by trade liberalization raises child labor incidence, and 2. the income gains from trade are not large enough to outweigh these effects. A sufficient condition for such an outcome is given in an economy where: 1. large numbers of families live below the poverty line, while there are relatively fewer families with intermediate levels of child labor, i.e., the economy is very poor and the income distribution is strongly skewed to the right; 2. the comparative advantage in the labor intensive good is relatively small and hence preferable factor price adjustments will be less strong (cf. Appendix B.4).

5.4 Empirical evidence

The empirical implication of the above arguments is that for exporters of subsistence goods (such as food), trade liberalization can be expected to reduce the need for child labor, although not everywhere alike. The sufficient conditions for an adverse income effect of trade on child labor predicted by the model are more likely present within very poor economies. Other things being equal, the effect of trade liberalization on child labor participation rates should be less favorable in poorer countries where child labor is needed for subsistence

12 As adult incomes diﬀer only because of capital endowment diﬀerences in the model, the distributional impacts of changing factor prices can be interpreted in terms of household poverty. Would one distinguish between worker and capitalist households instead, child labor would increase in poor capitalist households, decrease in poor worker households, while it would not react within the extremely poor worker households. Increases in aggregate child labor would be most likely if a large part of poor capitalist households sent their children to work and increases in child labor among them were not compensated by reductions in child labor among poor worker households. I owe this point to an anonymous referee. trade liberalization and child labor 84

of a large share of families. Thus, if the theoretical argument is empirically relevant, one should ﬁnd that, after controlling for changes in average income, increases in trade openness reduce child labor in countries that are exporters of subsistence goods, but this eﬀect is smaller (or even of opposite sign) for the poorest among them.13

5.4.1 Data and variables

To address this question, a panel of 91 developing and industrialized countries is used, where most variables are measured each decade between 1960 and 2000.14 Unlike in other empirical studies, the main focus of the regressions is not on the determinants of child labor per se but on the differential impacts of changes in trade openness on changes in child labor over time. The panel includes only countries that reported positive values of child labor in 1960 and only up to the time when child labor participation rates reach zero.15 As a total, the sample consists of 300 observations over four time periods, with an average of 3.3 observations per country.16 Most data is taken from the World Development Indicators (World Bank 2004). Basic summary statistics are presented in Table 5.1 (cf. Table A.6 in the Appendix for variable definitions). The dependent variable ∆Chlab is based on estimates of the ILO; it gives the percentage point change in labor force participation rates of children aged between 10 and 14 years over a decade. As widely argued, child labor force participation rates present only a crude measure of child labor. Nevertheless, they serve as the best available proxy for the prevalence of child labor in a cross-country setting and are widely used in empirical work.17 The main explanatory variable, ∆Open is defined as the percentage point change in the openness indicator over a decade where openness is measured by

13 In the model, poverty, which causes an adverse child labor response, is multidimensional. It requires low productivity, large subsistence poverty with children working full-time, and relatively high poverty among small owners of capital. Although the “group of poorest countries” used in the empirical analysis does not cover all of the above aspects, it can be still expected that such adverse circumstances are more likely present in extremely poor economies. 14 For a list of countries see Table A.5 in the Appendix. 15 Countries that did not exist in their present form in 1960 are thus excluded from the analysis as well as countries where information on trade openness has been entirely missing. 16 The main panel includes 23 countries with 2, 18 countries with 3, and 50 countries with 4 observations. 17 Caution is required when exploiting the time aspect of the data. As in most countries, and especially in developing economies, economic censuses are rare, the ILO relies heavily on projections (both inter- and extrapolations) for its estimates of economic activity rates (ILO 2000). Thus, the reductions in child labor force participation rates will appear considerably smoother in the data than in reality. This study uses changes in child labor over ten-year periods that are less eﬀected by the issue. trade liberalization and child labor 85

Table 5.1: Descriptive statistics

Variable No. obs. Mean St. dev. Min. Max.

∆Chlab 300 -0.030 0.025 -0.152 0.059 Chlab1960 300 0.238 0.174 0.001 0.794 Chlab1970 168 0.220 0.155 0.006 0.631 ∆GDP 300 0.172 0.259 -0.841 1.050 ∆Urban 300 0.056 0.043 -0.015 0.306 ∆Open 300 0.070 0.187 -0.550 0.939 ∆Fraser 168 0.062 0.121 -0.330 0.420 FoodExp 300 0.547 0.499 0 1 I15 (15 poorest) 300 0.180 0.385 0 1 I20 (20 poorest) 300 0.247 0.432 0 1 I25 (25 poorest) 300 0.300 0.459 0 1 I30 (30 poorest) 300 0.360 0.481 0 1 I35 (35 poorest) 300 0.417 0.494 0 1 East Asia & Pac. 300 0.113 0.318 0 1 South Asia 300 0.060 0.238 0 1 Latin America 300 0.293 0.456 0 1 Sub-Sah. Africa 300 0.310 0.463 0 1 North Africa & ME 300 0.080 0.272 0 1

the ratio of exports plus imports to GDP. While it measures changes in actual trade flows, this openness indicator captures trade policy only indirectly. Al- ternatively, some regressions use changes in the Openness Index of the Fraser Institute (Gwartney, Lawson and Gartzke 2005), ∆F raser, as a more direct measure of trade policy changes.18 As a component of its Economic Freedom of the World Index, the Fraser Institute estimates the Freedom to Trade Inter- nationally on a scale between 0 and 10 (which has been rescaled here between 0 and 1). The Index is based on information on tariff barriers (and eventually other regulatory trade barriers), on the difference between the actual and expected size of the trade sector, on black market premia, and on international capital market controls. Between 1960 and 2000 there was a general increase in trade openness based both on actual trade flows and on broader measures of trade policy, as well as a steady decline in child labor participation rates: Within the sample child labor force participation rates declined on average by 3 percentage points per decade while trade openness increased on average by 6 to 7 percentage points. Additional explanatory variables include the extent of child labor at the beginning of the analysis, GDP growth, urbanization, as well as regional and time

18 The correlation coeﬃcient between ∆Open and ∆Fraser amounts to 0.33. The other commonly used measure of trade openness, the Sachs and Warner (1995) openness indicator, could not be used because within the group of the poorest countries there were too few large switches of trade regime which makes a decomposition of the eﬀects of openness change between poor and less poor impossible. trade liberalization and child labor 86

controls. The labor force participation rate of children in 1960, Chlab1960, captures relevant scale effects as countries with originally higher child labor incidence can experience greater reductions in child labor. The variable ∆GDP controls for average income effects by measuring the growth of real per capita GDP over a decade. It is expected to unambiguously reduce child labor as GDP per capita is found to be the most powerful explanatory factor of prevalence of child labor in a cross-section of countries (see e.g., Krueger 1996, Edmonds and Pavcnik 2006). The variable ∆Urban measures the percentage point change in the share of urban population within a country. Although children working in manufacturing receive much of the public attention, child labor is still most prevalent in rural societies (Edmonds and Pavcnik 2005a). Urbanization can be thus expected to reduce child labor. Five regional controls (dummies for East Asia and the Pacific, South Asia, Latin America, North Africa and the Middle East, and Sub-Saharan Africa) have also been added to the regressions; they capture further structural differences between child labor trends across the subcontinents. The time dummies capture general economic trends in the world as reductions in child labor have been significantly larger in the last two decades than in the first two decades. Based on the predictions of the model, three country groups are treated differently: the poorest countries among food (subsistence good) exporters, other food exporters, and the rest. Two indicator variables are used for this purpose. F oodExp indicates countries that were or are large exporters of subsistence goods. The variable takes on the value of one if the exports of food and agricultural raw materials make up for more than a half of merchandising exports of a given country in at least one of the decades during the period 19 of analysis. The indicator variable IGroup stays for the originally poorest countries within the sample. IGroup takes the value of one if the country belongs to the group of countries that were the poorest based on their per capita GDP in 1960, where Group size varies between 15 and 35.20

5.4.2 Estimation strategy and main results

The main estimation equation takes the following form (for country i, decade t):

′ ∆Chlabit = Xitβ + γf(∆Openit)λt + ǫit (5.11)

Elements of the vector Xit include the amount of child labor at the beginning of the ﬁrst period, Chlab1960i, changes in GDP per capita over a

19 The categories food and agricultural raw products comprise the commodities of SITC sections 0 (food and live animals), 1 (beverages and tobacco), 2 (crude materials except fuels), and 4 (animal and vegetable oils and fats). From section 2 divisions 27 and 28 (crude fertilizers, minerals, metalliferous ores and scrap) are excluded. 20 See for the list of the poorest countries by this deﬁnition Table A.7 in the Appendix. trade liberalization and child labor 87

decade, ∆GDPit, the change in urbanization rates ∆Urbanit, and regional controls; λt stands for the set of the time controls. In order to analyze whether there is a diﬀerence in the eﬀects of trade liberalization across country groups, the changes in openness ∆Openit are interacted with the indicator variables 21 F oodExpi and IGroup,i:

γf(∆Openit)=∆Openit (γ1 + F oodExpi [γ2 + γ3IGroup,i]) (5.12)

The estimations have been carried out using a generalized linear model for panel data (GEE) (Liang and Zeger 1986) where the correlation structure of the error terms has been explicitly estimated. The main strength of the estimator is that it does not make the assumption of uncorrelated error terms within a group, and hence it can account for the fact that improvements in child labor conditions in a country are correlated over time. Tables 5.2 and 5.3 show the main regression results. Everything else equal, child labor force participation rates decreased by more in countries with higher child labor prevalence. On average, a 10 percentage point increase in GDP growth over a decade led to a further reduction in child labor by 0.14 percentage points. Child labor decreased also by more in countries with larger increases in urbanization. As compared to countries of Europe, Central Asia and North- America, decreases in child labor have been somewhat smaller in Latin America and South Asia, and considerably smaller in Sub-Saharan Africa. Changes in openness are not related to child labor on average (column 2 of Table 5.2), nor is there such a relationship for food exporters (column 3).22 However, the effects of openness differ between country groups: column 4 of Table 5.2 shows that increases in trade openness are significantly related to increases (or smaller reductions) in child labor for the fifteen poorest exporters of subsistence goods, for which the product of the indicator variables I15 × F oodExp equals one. The results of column (4) are replicated in Table 5.3 for differing Group sizes of the poorest countries, and the total effect for each of the three country types (average, subsistence exporters, poor subsistence exporters) is also reported. The estimated coefficient in row (a) (γ1 of equation 5.12) captures the average impact of an increase in trade openness for all countries. As before, on average, there is no significant relationshipb between changes in child labor and in openness. However, the estimated effect for an average food ex- porter (a)+(b) (γ1 + γ2 of equation 5.12) is negative; as to be expected, trade

21 Additionally to that, in a series of regressions (not reported here but available on request) the changesb in opennessb have been interacted with the amount of child labor in 1960, Chlab1960i. This generally increased significance of the results and the fit of the regressions which indicates that the impacts of openness were stronger in countries with originally higher levels of child labor. 22 As all regressions control for changes in GDP, the openness variable does not capture the total income effect of trade liberalization, only effects that go beyond raising average per capita income. trade liberalization and child labor 88

Table 5.2: Changes in child labor and changes in openness

Dependent variable: ∆Chlab (1) (2) (3) (4)

Chlab1960 -0.107 -0.107 -0.107 -0.110 (8.66) (8.49) (8.37) (8.08) ∆GDP -0.014 -0.014 -0.014 -0.014 (2.75) (2.65) (2.63) (2.67) ∆Urban -0.051 -0.051 -0.051 -0.048 (1.77) (1.78) (1.78) (1.64) ∆Open -0.005 -0.005 -0.005 (0.86) (0.79) (0.83) ∆Open × FoodExp 0.000 -0.004 (0.01) (0.50) ∆Open × FoodExp × I15 0.074 (1.73) East Asia & Pac. -0.004 -0.004 -0.004 -0.003 (0.90) (0.78) (0.78) (0.71) South Asia 0.008 0.008 0.008 0.007 (1.71) (1.60) (1.60) (1.56) Latin America 0.007 0.007 0.007 0.007 (2.28) (2.19) (2.19) (2.25) Sub-Sah. Africa 0.023 0.023 0.023 0.023 (4.00) (3.93) (3.95) (4.05) North Africa & ME -0.001 -0.001 -0.001 -0.001 (0.28) (0.32) (0.32) (0.29)

N (Observations) 300 300 300 300 n (Countries) 91 91 91 91 Wald-test [χ2(k)] 161.8 (11) 154.7 (12) 180.1 (13) 207.4 (14) Notes: The models are estimated using the GEE approach where the correlation structure of the error terms is explicitly estimated. Additional controls include time dummies. The reported absolute values of the t-statistics (in parentheses) are based on semi-robust standard errors. Results of Wald-tests of the models are reported, with the degrees of freedom of the χ2 statistics in parentheses.

liberalization is associated with reductions in child labor for exporters of subsistence goods. Most interestingly, increases in openness are associated with smaller reductions in child labor for the poorest subsistence good exporters. The differential effect for the poorest food exporters as compared to all food exporters (γ3 of equation 5.12) is significantly positive, while the overall effect (a)+(b)+(c) (γ1 +γ2 +γ3) is positive although not statistically significant. The differentialb effect vanishes gradually when the size of the group of the poorest is enlarged. Atb theb sameb time, the fit of the model clearly deteriorates when group size is increased (see Table 5.3) which also indicates that the effect is most relevant for the poorest countries. trade liberalization and child labor 89

Table 5.3: Diﬀerential eﬀects of openness for country groups

Dependent variable: ∆Chlab (1) (2) (3) (4) (5)

(a) ∆Open -0.005 -0.005 -0.005 -0.004 -0.004 (0.83) (0.78) (0.79) (0.70) (0.68) (b) ∆Open × FoodExp -0.004 -0.009 -0.010 -0.010 -0.013 (0.50) (1.04) (1.16) (0.89) (1.05)

(c) ∆Open × FoodExp × IGroup 0.074 0.048 0.045 0.019 0.021 (1.73) (2.06) (2.15) (1.12) (1.32) Estimated (a)+(b): -0.010 -0.014 -0.015 -0.015 -0.017 (1.29) (1.88) (2.07) (1.46) (1.60) Estimated (a)+(b)+(c): 0.065 0.034 0.031 0.004 0.004 (1.47) (1.47) (1.45) (0.30) (0.31) Wald-test [χ2(14)] 207.4 193.1 193.3 183.1 185.1

Group size 15 20 25 30 35 N (Observations) 300 300 300 300 300 n (Countries) 91 91 91 91 91 Notes: The models are estimated using the GEE approach with time variant correlation structure of the error terms. The reported absolute values of the t-statistics (in parentheses) are based on semi-robust standard errors. Additional controls include a constant, Chlab1960, ∆GDP , ∆Urban, time dummies, and ﬁve region dummies (for East Asia and the Paciﬁc, South Asia, Latin America, North Africa and the Middle East, and Sub-Saharan Africa).

5.4.3 Robustness of the results

The results remain broadly the same when trade liberalization is measured not by actual trade flows but by the policy-based Fraser Index ∆F raser which proxies changing trade policies (see Tables 5.4 and 5.5). The use of the Fraser Index of trade openness restricts sample size to 66 countries and makes the sample more biased towards the better-off countries. This might be one of the reasons why the explanatory power of the regressions significantly decreases when using ∆F raser instead of ∆Open. GDP growth and changes in urbanization retain their signs but lose significance. Neverthe- less, there is still a significant difference between the effects of increasing trade openness for poorest and less poor food exporters; while trade liberalization reduces child labor for food exporters on average, it does not reduce it for the poorest exporters of food (see Table 5.5). A differential effect for the poorest countries also remains when the sample is restricted to a smaller set of countries where the mechanisms described by the model are more likely to apply. In a series of further robustness checks, the sample has been restricted in two ways: 1. to 68 non-high income countries trade liberalization and child labor 90

Table 5.4: Changes in child labor and changes in openness (Alt. measure)

Dependent variable: ∆Chlab (1) (2) (3) (4)

Chlab1970 -0.111 -0.111 -0.109 -0.112 (6.91) (6.91) (6.79) (6.96) ∆GDP -0.009 -0.009 -0.009 -0.009 (1.17) (1.17) (1.21) (1.09) ∆Urban -0.062 -0.062 -0.067 -0.069 (1.29) (1.27) (1.37) (1.40) ∆Fraser 0.000 0.010 0.010 (0.01) (0.99) (0.98) ∆Fraser × FoodExp -0.031 -0.038 (2.56) (2.75)

∆Fraser × FoodExp × I15 0.038 (1.59) East Asia & Pac. -0.001 -0.001 -0.002 -0.002 (0.21) (0.21) (0.25) (0.25) South Asia 0.017 0.017 0.017 0.017 (2.07) (2.05) (2.18) (2.17) Latin America 0.011 0.011 0.010 0.011 (1.73) (1.74) (1.76) (1.81) Sub-Sah. Africa 0.026 0.026 0.025 0.026 (3.14) (3.17) (3.19) (3.28) North Africa & ME -0.006 -0.006 -0.006 -0.006 (0.68) (0.68) (0.71) (0.75)

N (Observations) 168 168 168 168 n (Countries) 66 66 66 66 Wald-test [χ2(k)] 122.5 (10) 125.4 (11) 129.3 (12) 149.6 (13) Notes: The models are estimated using the GEE approach with time variant correlation structure of the error terms. Additional controls include time dummies that are not reported. The reported absolute values of the t-statistics (in parentheses) are based on semi-robust standard errors. Results of Wald-tests of the models are reported, with the degrees of freedom of the χ2 statistics in parentheses.

(according to the definitions of the World Bank) where child labor can be supposed to pose a more relevant problem, and 2. to 46 main food exporters. In both cases, a significant differential effect of trade liberalization within the poorest countries is found when changes in trade are scaled by original levels of child labor. As to be expected, the difference is more pronounced within the sample of food exporters where the hypothesized income and price effects of trade in subsistence goods most likely apply. These empirical results do not yield to a strictly causal interpretation. If changes in openness and changes in child labor both merely reflect underlying trade liberalization and child labor 91

Table 5.5: Diﬀerential eﬀects of openness for country groups (Alt. measure)

Dependent variable: ∆Chlab (1) (2) (3) (4) (5)

(a) ∆Fraser 0.010 0.010 0.010 0.011 0.011 (0.98) (0.98) (0.99) (1.01) (1.01) (b) ∆Fraser × FoodExp -0.038 -0.038 -0.036 -0.035 -0.035 (2.75) (2.75) (2.42) (2.06) (2.05)

(c) ∆Fraser × FoodExp × IGroup 0.038 0.038 0.019 0.009 0.009 (1.59) (1.59) (0.81) (0.47) (0.48) Estimated (a)+(b): -0.028 -0.028 -0.026 -0.023 -0.024 (2.54) (2.54) (2.12) (1.80) (1.80) Estimated (a)+(b)+(c): 0.010 0.010 -0.007 -0.015 -0.015 (0.47) (0.47) (0.36) (1.16) (1.16) Wald-test [χ2(13)] 149.6 149.6 136.7 139.0 140.0

Group size 15 20 25 30 35 N (Observations) 168 168 168 168 168 n (Countries) 66 66 66 66 66 Notes:The models are estimated using the GEE approach with time variant correlation structure of the error terms. The reported absolute values of the t-statistics (in parentheses) are based on semi-robust standard errors. Additional controls include a constant, Chlab1970, ∆GDP , ∆Urban, time dummies, and ﬁve region dummies (for East Asia and the Paciﬁc, South Asia, Latin America, North Africa and the Middle East, and Sub-Saharan Africa).

shifts in the general economic environment, openness will not be exogenous to child labor. Unfortunately, there are no good instruments that would predict changes in trade policy.23 However, there are no a priori reasons to expect the general policy environment to inﬂuence both child labor and trade policies diﬀerently within the poorest food exporters only. Thus, the patterns found here can be considered at least indicative.24

5.5 Conclusion

This chapter investigated potential impacts of international trade on child labor that is motivated by subsistence needs. For this purpose, goods and factor price adjustments have been examined under the additional assumption that the Southern exportable is a necessity. Within the Heckscher-Ohlin framework, the

23 Geographic instruments and the concept of geography based trade openness (see Frankel and Rose 2002) are of no help in this case as they capture openness to trade in an inherently static manner and hence do not help to predict policy changes. 24 OLS estimates over country averages of the variables yield qualitatively similar results; the estimated coefficients change in magnitude, but the differential effect of trade liberalization on poor and less poor countries still remains. trade liberalization and child labor 92

poorest worker families benefit the most and the capital-rich families benefit the least or even lose from more open trade. Hence, income effects of trade can be expected to reduce child labor incidence in general. However, reductions in child labor cannot be expected to be uniform across the households. In very poor worker families children might have to work full time despite of the favorable income effects, while in capital-richer families child labor can even increase. Losers of trade liberalization might also include families who live near to the poverty line, and who are adversely affected by the reduction of returns to their human and physical capital holdings and by the increase in the costs of subsistence. On aggregate, child labor can even increase as a result of more open trade, despite the increases in average income. The analysis concentrated deliberately only on income effects of trade, ar- guing that they determine the changes in the actual need for child labor in a family. In reality, the supply of child labor responds also to changing economic incentives, and especially to changes in expected net returns to schooling. Even so, income effects retain their importance as long as parents consider child labor as a “bad”, or more importantly, as long as poverty or missing access to credit put effective constraints on their decisions. From a policy perspective, child labor that arises from poverty constraints has a special role: As long as the child’s earnings are needed for subsistence of the family, a policy interven- tion can only be successful if it raises the current income of the poor (be it by redistribution, credit provision or other means). Thus, this chapter complements the existing theoretical literature on child labor in an open economy setting, without questioning the importance of other explanatory factors for child labor besides poverty constraints. Incentive effects can further weaken the favorable reactions of child labor supply and hence, the overall effect of trade liberalization can turn out less favorable than described by the model. The empirical evidence presented in Section 5.4 shows that the mechanisms of the model can actually play a role in explaining the impact of globalization on child labor: While not significant on average, trade liberalization was associated with smaller reductions in child labor for the poorest food exporters than for all food exporters. The distributional impacts of trade liberalization can serve as one explanation for these differences, even though the relevance of other effects (like changing economic incentives) cannot be excluded. From a policy perspective, the potentially unfavorable reaction of the numbers of working children should be of less concern if the scale of individual child labor matters as well. The overall message of the analysis is thus positive: Al- though it is not sure that poverty-induced child labor decreases on aggregate, the need for the worst forms of child labor (captured here by longest working hours) will be reduced if the favorable income effects accrue to the poorest. Especially, the above arguments should not serve to legitimate the imposition of trade sanctions by Northern countries in order to combat child labor. Trade sanctions are completely counterproductive as they reduce the real income of trade liberalization and child labor 93

the poorest worker households and harm those families that most likely have to rely on child labor in order to survive. chapter 6

The eﬀects of trade liberalization on child work and schooling in Indonesia

Abstract

This chapter examines the effects of trade liberalization on child work and schooling in Indonesia. The estimation strategy identifies geographical differences in the effects of trade policy through district level exposure to reduction in import tariff barriers. For this a balanced panel of 261 districts is used, which is based on four rounds (1993 to 2002) of the Indonesian annual national household survey (SUSENAS). The empirical analysis relates workforce participation and school enrollment of children aged 10 to 15 to geographic variation in relative tariff exposure. The main findings show that increased exposure to trade liberalization is associated with a decrease in child work among the 10 to 15 year old. The effects of tariff reductions are strongest for children from low skill backgrounds and in rural areas. Schooling seems less responsive to tariff changes in the short run, although there are some lagged positive effects due to tariff reductions, but only in urban districts and for boys. Favorable income effects for the poor, induced by trade liberalization, are likely to be the dominating effects underlying these results.1

6.1 Introduction

The effects of trade liberalization on work and schooling of children are widely debated and public and political interest in the issue is high. From a theoretical perspective these effects are a priori unclear (e.g., Ranjan 2001, Jafarey and Lahiri 2002) as trade liberalization acts potentially through several channels, changing relative prices, real income distribution, wages and net returns to education. The arising income and substitution effects can both raise and reduce schooling and workforce participation of children.

1 This chapter is based on Kis-Katos and Sparrow (2009).

94 the effects of trade liberalization in indonesia 95

Empirical evidence on the issue is scarce. Cross-country studies generally find trade liberalization to be associated with lower incidence of child labor on average (Cigno et al. 2002), a relationship that seems most likely to be driven by the effect of trade on income, as more open economies have less child labor because they are richer (Edmonds and Pavcnik 2006). The empirical evidence in Chapter 5 shows differential effects of trade openness, with smaller reductions in child labor for the poorest food exporting countries. However, empirical studies based on micro data and direct evidence from trade reforms are required to understand the heterogeneous effects from trade liberalization and identify the main channels at work. For example, Edmonds and Pavcnik (2005b) find that rice price increases due to a dismantling of export quotas in Vietnam led to an overall decrease in child labor in the 1990s, especially due to the relatively evenly distributed favorable income effects. By contrast, Edmonds et al. (2007) find that in rural India, districts that have been more strongly exposed to trade liberalization have experienced smaller increases in school enrollment on average, which they argue is primarily due to the unfavorable income effects to the poor and the relatively high costs of education in these districts. This study contributes to the microempirical literature by examining the trade liberalization experience of Indonesia in the 1990s, which, given the vast geographic heterogeneity of the archipelago, offers an interesting case study on the effects of trade liberalization on child work and schooling. In preparation to and following its accession to the WTO, Indonesia went through a major reduction in tariff barriers: average import tariff lines decreased from around 19.4 percent in 1993 to 8.8 percent in 2002. During that same period the workforce participation of children aged 10 to 15 years more than halved while school enrollment steadily increased. Due to Indonesia’s size and geographic variation in economic structure, the various districts have been very differently affected by trade liberalization, which offers us a valuable identification strategy. The identification strategy of this study follows that of Topalova (2005) and Edmonds et al. (2007), and combines geographic variation in sector composition of the economy and temporal variation in tariff lines by product category, yielding geographic variation in (changes in) average exposure to trade liberalization over time. This study extends their approach in several ways. First, it defines two alternative measures of geographic exposure to trade liberalization, by weighting tariffs on different products by the shares these sectors take in (i) regional GDP, and (ii) the regional structure of employment. These measures reflect different dimensions of households’ exposure to trade liberalization: the former through the distributional effects of local economic growth, the latter through labor market dynamics. In addition to this, the data allows us to go beyond the fixed effects approach employed in earlier studies and investigate the dynamic effects of trade liberalization. the effects of trade liberalization in indonesia 96

The analysis draws on a variety of data sources. Indonesia’s annual national household survey (SUSENAS) provides information on the main activities of children and their basic socio-economic characteristics. The empirical analysis uses four rounds of this repeated cross section data, spaced at 3–year intervals between 1993 and 2002. As the SUSENAS is representative at the district level, the analysis is applied both at the individual level using pooled repeated cross section data with district fixed effects, and at the district level with pseudo panel data for 261 districts. The data on economic structure of the districts comes from information on regional GDP (GRDP) of the Central Bureau of Statistics in Indonesia (BPS), while district level employment shares and further controls are based on SUSENAS. Additional district level information is derived from PODES, the Village Potential Census. Finally, information on tariff lines comes from the UNCTAD-TRAINS database. The findings show that stronger exposure to trade liberalization has lead to a decrease in child labor among the 10 to 15 year old. The effects are strongest for children from low skill backgrounds and in rural areas. For schooling, no consistently robust results emerge: overall, schooling seems less responsive to tariff changes in the short run, although some lagged positive effects of tariff reductions arise in urban districts and for boys. Favorable income effects for the poor induced by trade liberalization are likely to be the dominating effects underlying these results. The next section of this chapter provides a theoretical framework. The third section elaborates on the context of the tariff reductions in Indonesia, and the developments in child labor and education for the study period. Section 4 presents the data and sets out the identification strategy. The results are then discussed in section 5 while section 6 concludes.

6.2 Theoretical background

Although child labor is determined by an interaction between the necessity and the opportunities to work, credit constraints, returns to school, as well as parental preferences, its close link to poverty is undisputed (Edmonds and Pavcnik 2005a). Hence, reductions in trade barriers are more likely to lead to reductions in child labor if they are going to benefit the poor in the economy. Based on standard Stolper-Samuelson reasoning, trade liberalization has been commonly expected to alleviate poverty in developing countries (e.g., Bhagwati and Srinivasan 2002). However, as increases in unskilled wages also raise the opportunity costs of children not working, the overall effects on child labor are a priori not clear. Even in its simplest version, the Stolper-Samuelson reasoning does not necessarily imply a reduction in child labor due to trade liberalization, as the resulting income and substitution effects point in different directions. In a the effects of trade liberalization in indonesia 97

Heckscher-Ohlin economy with two mobile factors, low and high skilled labor, and two industries producing one export and one import-competing good, reducing import tariffs leads to a decrease in the relative price of the imported good with respect to the numéraire (export good). On the production side, there will be a shift towards the production of exportables with low skill intensity, which in turn raises the demand for unskilled labor and reduces the skill premium in the economy.2 The price changes will also lead to consumption shifts, and the overall effects of trade liberalization are expected to be positive (gains from trade). Households will be affected by changing goods and factor prices through two main channels. First, changes in wages and goods prices alter the real income of the households. Second, shifts in the relative prices of goods and opportunity costs of not working result in substitution effects which lead to a further reallocation of consumption and labor supply. While real incomes of the poor low-skilled households should increase after trade liberalization, the overall reaction of child labor is not clear-cut, since rising real wages of the unskilled increase the incentives to work. More formally, consider a household consisting of one child and one adult where the adult chooses consumption of two goods (c1, c2), and allocates the child’s time between work (l), and schooling (1−l) in order to maximize household utility. The utility of schooling (ν(1 − l)) reflects both disutility of child labor and the discounted present value of the returns to school. The decision is made subject to the household’s budget constraint and the time constraints for the child, assuming that financial markets are typically imperfect such that the household cannot borrow against the child’s future income in order to invest into education, even if the discounted net returns to education would be positive.3 The budget constraint states that the expenditures on consumption goods and schooling (σ(1 − l))4 cannot be higher than the adult’s income (y) plus the income from child labor (γwl).5

max u(c1, c2)+ ν(1 − l) s.t. y + γwl = pc1 + c2 + σ(1 − l), 0 ≤ l ≤ 1 c1,c2,l 2 These price eﬀects might be both mitigated and enhanced in the presence of non- tradable goods (e.g., inputs producing education): If the import-competing sector is more capital intensive than both the exporting and the non-traded sectors (as it might be expected in a developing economy), the relative price of the non-traded good with respect to the num´eraire (exportable) will rise. Overall demand and production shifts will in this case depend on the relative factor intensities of each industry and the gross substitutability of all goods in consumption (Komiya 1967). 3 Credit constraints and imperfect smoothing seem a reasonable assumption for most developing countries, at least for those households that send their children to work, as credit constraints are among the main causes of child labor (e.g., Beegle et al. 2006). For Indonesia the study in Chapter 4 of this book shows that credit availability is closely related to the incidence of child labor in small scale manufacturing. 4The direct costs of schooling (excluding opportunity costs) are denoted by σ, and for simplicity, are assumed to be linear in school time. 5 If the adult is unskilled, adult income equals the unskilled wage y = w, if the adult is skilled, adult income equals the skilled wage. In both cases adult labor is assumed to be inelastic in supply. the effects of trade liberalization in indonesia 98

The relative price of the exportables (c1) is denoted by p, unskilled wages by w. The child is assumed to be a perfect although less productive (γ < 1) substitute for unskilled adult labor (substitution axiom by Basu and Van 1998). Given the optimal allocation of income over the two consumption goods, optimal (uncompensated) child labor supply (l∗) is determined by relative goods and factor prices, school costs and adult income: ∗ l = l(p,wγ,σ,y) (6.1) For an interior solution (where the child combines work and schooling), the optimality condition is given by: ′ wγ vy(p,wγ,σ,y)= ν (1 − l) − σ vy(p,wγ,σ,y) (6.2) which is expressed in terms of an indirect utility function (v(p,wγ,σ,y)). The work-school trade-off depends thus on the relative magnitudes of the real value of the marginal product of child labor (left hand side of equation (6.2)) and the net real marginal returns to education (right hand side of equation (6.2)), with marginal utility of income (vy) denoting the inverse of the price deflator. A child will be doing at least some work, if real child wages are greater than the marginal real net return to education if the child spends all its available time on learning (l = 0).6 After reducing import tariffs, imported and import competing products become relatively less expensive (dp > 0), the overall effect of which can be seen by totally differentiating the uncompensated child labor supply equation (6.1):

∗ ∂l ∂l ∂w ∂l ∂y dl = +γ + dp (6.3) ∂p ∂w ∂p ∂y ∂p + +? ? + − +? |{z} The first term within the|{z} parentheses|{z} captures|{z} the|{z} direct effect of the relative price change: a decrease in import prices makes schooling relatively more expensive; if schooling and the import competing good are gross substitutes this will increase child labor.7 The first part of the second term captures both the income and substitution effects from an increase in child wages: if the substitution effect dominates, rising wages increase child labor supply.8 6 Conversely, a child will be spending at least some time going to school if real child wages are smaller than the marginal net returns to education at no schooling (l = 1). For ease of exposition, abstract from the possibility that a child stays idle, which will be most likely if both real net returns to education and value of marginal product of child work are low. 7 This substitution effect can be mitigated or enhanced by real income changes resulting from the shift in relative goods prices (depending on consumption patterns). 8 Additionally, there might be dynamic effects of falling skill premia, which make invest- ′ ment into education less profitable, reduce ν (1 − l) and thus raise ∂l/∂w further. But as technological upgrading is certainly an issue in the long run, this gives an additional motive for human capital accumulation and makes the longer term relevance of short term falls in skill premia questionable. the effects of trade liberalization in indonesia 99

The second part of the term captures the change in unskilled wages with an increase in import prices and is negative by a Stolper-Samuelson argument. The third term captures the effect through adult income: if the adult has unskilled labor, an increase in the import price should decrease adult income, and hence increase child labor. The overall sign of these effects depends on whether the favorable income effects or the substitution effects are dominating.9 Depar- tures from the Stolper-Samuelson reasoning that result additionally in negative income effects for the poor make an increase in child labor more likely. The expected favorable effects of trade liberalization on child labor depend crucially on the incomes of the poor increasing due to trade. Although the Stolper-Samuelson reasoning presents a very powerful argument in favor of these expectations, under many circumstances trade liberalization might fail to benefit the poor (e.g., Davis and Mishra 2007). If a developing country trades not only with more but also with less skill-abundant countries than itself, reductions in tariffs on goods with the lowest skill intensity may also hurt the poor by reducing the demand for least skilled labor. The expected increases in unskilled wages can also be reduced or even missing if the effects of trade liberalization are accompanied by skill biased technological change. In contrast, reductions in tariffs on goods that are not produced within a country will have no effects on producers and will only benefit consumers of those goods. Favorable income effects are more likely to occur if intersectoral worker mobility is high and markets are competitive, which corresponds to a longer run perspective.10 If workers’ skills are industry specific instead and hence the between industry mobility is low, workers might be harmed in the short run by reductions in protection. In a constrained economic environment, with imperfect smoothing, such short term economic shocks can also have long term consequences for the poor. For instance, decisions on withdrawing a child from school in face of a shock are often irreversible and can have intergenerational effects. Empirical evidence on the effects of globalization on poverty is partly incon- clusive, since, contrary to the Stolper-Samuelson predictions, many empirical studies do not observe reductions either in poverty or in wage inequality in developing countries that reduced tariffs unilaterally (cf., Harrison 2007).11

9 Although the above arguments have been presented on the intensive margin (with both work and school being interior), the effects translate easily to the extensive margin as well: The workforce participation of a child is influenced through the same channels as presented above, and the share of children working and/or enrolled in school is influenced by changes in district poverty and wages/labor market conditions, and hence by the same income and substitution effects. 10 Reductions in protection are more likely to benefit the poor if labor mobility between industries is high and labor market policies are flexible, and if social safety nets are well- functioning (Harrison 2007). 11 There are however studies showing that accounting for geographic (e.g., Chiquiar 2008) the effects of trade liberalization in indonesia 100

For Indonesia, however, the pro-poor effects of trade liberalization are not unlikely: Suryahadi (2003) documents rising unskilled wages over the period of trade liberalization in the 1990s, while Sitalaksmi, Ismalina, Fitrady and Robertson (2009) find improvements in perceived working conditions. Indeed, the results of the present study seem to suggest that tariff reductions have induced positive income effects and reduced poverty, eventually leading to a reduction in rural child labor. Our empirical analysis will focus on the effects of trade liberalization prop- agated through changes in the composition of economic activity, and will abstract from eventual changes in consumption patterns.12

6.3 Trade liberalization and children in Indonesia

6.3.1 Trade liberalization in the 1990s

Trade liberalization in Indonesia took place over more than fifteen years. From the mid-1980s the former import substitution policy has been gradually re- placed by a less restrictive trade regime, tariff lines have been reduced while at the same time a slow tarification of non-tariff barriers took place (Basri and Hill 2004). This laid the ground to the next wave of trade liberalization in the mid-1990s, with rising foreign firm ownership and increasing export and import penetration.13 Tariff reductions were particularly strong in the 1990s, with Indonesian trade liberalization policy in that decade being defined by two major events: the conclusion of the Uruguay round in 1994 and In- donesia’s commitment to multilateral agreements on tariff reductions, and the Asian economic crisis in 1997 and the post-crisis recovery process. After the Uruguay round Indonesia committed itself to reduce all of its bound tariffs to less than 40% within ten years. In May 1995 a large package of tariff reductions was announced which laid down the schedule of major tariff reductions until 2003, and implemented further commitments of Indonesia to the Asia Pacific Economic Cooperation (Fane 1999). While the removal of specific non-tariff barriers was accompanied by a temporary rise in tariffs (especially in the food manufacturing sector), this did not affect the overall declining trend in any major way. or within-industry (e.g., Verhoogen 2008) heterogeneity can help to identify Stolper- Samuelson linkages in developing countries. 12 This implicitly implies that differences in district level trends in the composition of consumption are assumed to be unrelated to the districts’ economic production structure; in which case not controlling for the consumption channel will not confound the estimates. 13 Arguably, cronyism and specific protection of a few industries with ties to the Soeharto- family—especially chemicals, motor vehicles and steel—reduced the effect of overall liberalization. However, the largest part of the cronyism occurred in nontraded sectors and did not further affect protection of the traded sectors (Basri and Hill 2004, p.637). the effects of trade liberalization in indonesia 101

Figure 6.1: Tariﬀ reductions in Indonesia

Average tariff lines Standard deviation of tariff lines 20 20

17.2

14.7 15 15

11.5 10 10 Average tariffs Average tariffs

6.6 5 5 0 0

19891990 1993 19951996 199920002001200220032004 19891990 1993 19951996 199920002001200220032004 Years Years

Figure 6.2: Tariﬀ reductions by sectors

Average tariff lines Standard deviation of tariff lines 20 20

Manufacturing 15 15

Manufacturing 10 10 Agriculture Agriculture Effective applied tariffs Effective applied tariffs 5 5 Mining Mining 0 0

1993 1995 1996 1999 2000 2001 2002 1993 1995 1996 1999 2000 2001 2002 Years Years

Figure 6.1 shows the reduction in tariff lines over time and the variation between industries. On average, nominal tariffs reduced from 17.2 percent in 1993 to 6.6 percent in 2002. In this period the strongest reductions occurred from 1993 to 1995 and during the post crisis period after 1999. Tariff dispersion decreased especially in the post-crisis period when reductions were the effects of trade liberalization in indonesia 102

more universal. While tariffs decreased across the board, there were marked differences in initial levels and in the extent of the decrease (see Figure 6.2). Manufacturing started with relatively high tariff barriers but also showed the strongest reductions. For example, wood and furniture saw tariffs decline from 27.2 to 7.9 percent, textiles form 24.9 to 8.1 percent and other manufacturing from 18.9 to 6.4 percent. The average tariffs for agriculture were already much lower in 1993, at 11.5, and which reduced to 3.0 percent. Figure 6.3 shows that tariff reductions and tariff levels are reasonably positively related; all outliers showing significant increases in tariffs are related to alcoholic beverages and soft drinks that were subject to a major retarification of non-tariff barriers.

Figure 6.3: Tariﬀ levels and reductions

Changes of tariffs Outliers 10 150 0 100 50 −10 Tariff changes 1993−2002 Tariff changes 1993−2002 0 −20 −30 −50 0 10 20 30 40 0 50 100 150 200 Tariff levels Tariff levels

Existing studies on the effects of Indonesian trade liberalization document both increased firm productivity and improvements of working conditions in manufacturing. At the plant-level, Amiti and Konings (2007) find that trade liberalization affected firms’ productivity through two main channels: falling tariffs on imported inputs fostered learning and raised both product quality and variety, while falling output protection increased the competitive pressures. Comparing the two effects they argue that gains from falling input tariffs were considerably higher. Firm productivity has also been strongly affected by FDI flows, as firms with increasing foreign ownership experienced restructuring, employment and wage growth, as well as stronger linkages to export and import markets (Arnold and Smarzynska Javorcik 2005). At the same time, working conditions seem to have improved, especially in manufacturing. Using individual employment data, Sitalaksmi et al. (2009) argue that the increase in export-oriented foreign direct investment went along with rising relative wages in the textile and apparel sector. Additionally, working conditions, proxied by the effects of trade liberalization in indonesia 103

workers’ own assessment of their income, working facilities, medical beneﬁts, safety considerations and transport opportunities, improved over time in the expanding manufacturing industries as compared to agriculture. Based on a microsimulation exercise Hertel, Ivanic, Preckel and Cranﬁeld (2004) argue that full multilateral trade liberalization is expected to decrease household poverty in Indonesia, although self-employed agricultural households would be the most likely losers of trade liberalization in the short-run, which is mainly due to the assumption that self-employed labor is immobile in the short- run. In the longer run some former agricultural workers will be moving into the formal wage labor market and the poverty headcount could be expected to fall for all sectors. However, the mobility of low skilled labor, and hence the speed and ability to exploit the opportunities from trade liberalization, may be underestimated by Hertel et al. (2004). For example, Suryahadi, Suryadarma and Sumarto (2009) show that during the 1990s the agriculture employment share dropped from 50 to 40 percent, while the services share increased from 33 to 42 percent. In addition, they attribute most of the poverty reductions in that decade to growth in urban services. This is further supported by Suryahadi (2003), who documents a fast increase in the employment of skilled labor force as well as a decline in wage inequality (i.e. faster wage growth for the unskilled) during trade liberalization in Indonesia, although he does not establish causality.

6.3.2 Child work

Indonesia experienced a steady decline in child work in the thirty years before the Indonesian economic crisis, but this decline halted with the onset of the crisis (e.g., Suryahadi et al. 2005). Nevertheless, market work among children aged 10 to 15 increased only slightly in response to the economic crisis (e.g., Cameron 2001).14 During the crisis children have been moving out of the formal wage employment sector into other small-scaled activities (Manning 2000), but the labor supply response seems to have been concentrated with older cohorts. The overall decline in child work over the study period is portrayed in Figure 6.4, for boys and girls, and by diﬀerent age groups. Child work is here deﬁned as any work activity that contributes to household income. From 1993 to 2002, the incidence of child work halves for children of junior secondary school age (13 to 15 year old), and is cut by more than 70 percent for children age 10 to 12. This decline is observed for both boys and girls, although boys engage in market work more than girls. In 2002 market work incidence for boys age 13 to 15 years is 14.8 percent, and 2.3 percent for boys at age 10 to

14 Information on working children below the age of 10 is not available at a systematic basis. the effects of trade liberalization in indonesia 104

Figure 6.4: Work of children, by gender and age group

Boys’ market work Girls’ market work .6 .6 0.530

0.418 0.407 .4 .4

0.302 0.283

0.221

.2 0.148 .2 0.100 0.080 0.054 0.023 0.016 Proportion of girls doing market work Proportion of boys doing market work 0 0

1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Years Years

Aged 10−12 Aged 13−15 Aged 10−12 Aged 13−15 Aged 16−18 Aged 16−18

12. Among girls market work incidence is 10.0 and 1.6 percent for the same age groups, respectively.

Figure 6.5: Sectoral distribution of child work

Work of children aged 10−12 Work of children aged 13−15 25 25

Else 20 20

Services Trade 15 15 Manufacturing 10 10

Else Agriculture 5 5 % of children working by sector % of children working by sector

Agriculture 0 0

1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Years Years

Agriculture is the main sector for child work, and developments in this sector are driving the overall trends, as shown in Figure 6.5. In 1993 just over the effects of trade liberalization in indonesia 105

Table 6.1: Evolution of market work of children over time

Share of boys aged 10–15 doing market work By head’s educational attainment By district type Year None Primary Low sec. Higher Rural Urban Total N 1993 0.247 0.159 0.086 0.037 0.197 0.044 0.174 63,009 1994 0.232 0.139 0.072 0.041 0.176 0.043 0.156 63,556 1995 0.236 0.139 0.078 0.037 0.180 0.050 0.161 59,992 1996 0.202 0.124 0.073 0.036 0.153 0.045 0.137 61,234 1997 0.179 0.107 0.054 0.029 0.129 0.028 0.115 58,487 1998 0.197 0.122 0.082 0.033 0.144 0.048 0.130 56,783 1999 0.178 0.111 0.065 0.030 0.130 0.040 0.117 54,907 2000 0.150 0.092 0.052 0.024 0.106 0.026 0.096 51,003 2001 0.173 0.107 0.070 0.031 0.124 0.036 0.112 56,379 2002 0.140 0.080 0.046 0.021 0.090 0.030 0.082 56,243 N 222,837 191,241 67,801 99,714 477,500 104,093 581,593 581,593

Share of girls aged 10–15 doing market work By head’s educational attainment By district type Year None Primary Low sec. Higher Rural Urban Total N 1993 0.177 0.113 0.080 0.069 0.143 0.067 0.131 59,895 1994 0.169 0.103 0.060 0.064 0.128 0.066 0.119 59,582 1995 0.152 0.102 0.072 0.066 0.125 0.064 0.115 57,102 1996 0.137 0.085 0.058 0.053 0.106 0.052 0.098 58,430 1997 0.109 0.067 0.046 0.043 0.082 0.037 0.075 55,427 1998 0.127 0.084 0.060 0.050 0.097 0.060 0.091 53,814 1999 0.120 0.068 0.051 0.045 0.087 0.044 0.080 51,936 2000 0.094 0.061 0.041 0.030 0.070 0.030 0.065 47,832 2001 0.109 0.070 0.053 0.043 0.081 0.047 0.076 52,938 2002 0.092 0.049 0.032 0.029 0.059 0.037 0.056 52,752 N 207,841 180,188 64,162 97,517 447,531 102,177 549,708 549,708 Notes: Participation shares are weighted by survey weights. N refers to the number of observations in the sample, rural districts denote Kabupatens, urban districts denote Kotas.

75 percent of child work in the age group 10 to 12 occurred in agriculture, while two in three child workers aged 13 to 15 worked in agriculture. The dominance of the agricultural sector in child work translates into a 79 and 69 percent share in the overall reduction in child work for the two age groups, respectively. However, the relative changes from 1993 to 2002 are remarkably constant across sectors. The trends in child work vary greatly by location and education attainment of the head of household (Table 6.1). Child work incidence is much higher in rural districts compared to municipalities, but rural areas experienced the largest decline, both in absolute and relative terms. These patterns mirror the trend dominance of the agricultural sector. Child work incidence decreases the effects of trade liberalization in indonesia 106

with the level of education of the head of household. Boys living in households where the head of household has not ﬁnished primary education, are almost six times more likely to work than boys from households where the head of household holds a degree higher than junior secondary school; for girls this ratio is about three. For all the levels of education we see child work incidence decreasing.

6.3.3 Schooling

Indonesia has shown strong improvements in education attainment over past decades, reaching almost universal primary school enrollment already in the mid 1980s (e.g., Jones and Hagul 2001, Lanjouw et al. 2002). Indonesia’s current 9 year basic education policy aims at achieving universal enrollment for children up to the age of 15; that is, up to junior secondary school. But while junior secondary school enrollment has certainly improved, the large drop out of around 30 percent in the transition from primary to junior secondary (around 70 percent) remains a problem. In particular striking are the relatively low transition rates among the poor. Among the poorest 20 percent of the population, almost half of the children that ﬁnish primary school drop out at junior secondary level; this is in stark contrast to the 12 percent drop out rate for the richest quintile (Paqueo and Sparrow 2006). Other problems that are still cause for concern are delayed enrollment and relatively high repetition rates. The main barriers to education concern both demand and supply factors. Paqueo and Sparrow (2006) ﬁnd that enrollment is sensitive to the level of school fees, in particular for secondary education. However, indirect costs form an even larger obstacle to enrollment, in the form of tuition fees, text books and uniforms, and transport costs. On the supply side, quality of education is a major source of concern in Indonesia: in particular, teacher quality and absenteeism, and lack of access to secondary schools in remote and rural areas (World Bank 2006). The economic crisis did not lead to a large school dropout, as was ini- tially feared after a similar experience in the late 1980s, although the increase in enrollment did stagnate in 1999. Households appeared to have employed alternative short term smoothing strategies to protect the education of their children, in particular children in secondary school as this is associated with relatively higher sunk costs and higher future returns (Thomas et al. 2004). A second explanation can be found with the success of a social safety net scholarship program in preventing a decrease in primary enrollment (Sparrow 2007).

Figure 6.6 shows the recent trend in enrollment by age group (irrespective of enrollment level). Enrollment among primary school age children has been the effects of trade liberalization in indonesia 107

Figure 6.6: School enrollment of children, by gender and age group

Boys’ schooling Girls’ schooling

1 0.973 1 0.973 0.953 0.958 0.963 0.970 0.944 0.947 .8 .8 0.790 0.795

0.697 0.676 .6 .6

0.507 0.488 Proportion of girls in school Proportion of boys in school 0.446 0.402 .4 .4

1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 Years Years

Aged 8−9 Aged 10−12 Aged 8−9 Aged 10−12 Aged 13−15 Aged 16−18 Aged 13−15 Aged 16−18

near universal throughout the period 1993 to 2002. There is a strong increase in enrollment among secondary school age cohorts, with a slight decrease in the post crisis years. By 2002, 80 percent of children age 13 to 15 were enrolled in school, and about half of the 16 to 18 age group. A striking feature for Indonesia is that, unlike for child work, we see no gender gap. The diﬀerential trends in enrollment by household characteristics and location are shown in Table 6.2. While school enrollment is higher among children in municipalities, it is the strong increase in the rural districts that has driven the national trend during the 1990s. Enrollment is higher among boys and girls from relatively highly educated households. But similar to rural districts, the group with the lowest initial level of enrollment experience the largest relative and absolute gains from 1993 to 2002. The remainder of this analysis focuses on primary school age children close to the transition point, age 10 to 12, and junior secondary school age children, age 13 to 15. For children younger than 10 enrollment is almost universal and information on work is not available. the effects of trade liberalization in indonesia 108

Table 6.2: Evolution of school enrollment of children over time

Share of boys aged 10–15 enrolled in school By head’s educational attainment By district Year None Primary Low sec. Higher Rural Urban Total N 1993 0.737 0.859 0.944 0.978 0.810 0.943 0.830 63,009 1994 0.751 0.874 0.959 0.977 0.830 0.949 0.848 63,556 1995 0.764 0.876 0.951 0.982 0.834 0.940 0.849 59,992 1996 0.774 0.884 0.950 0.979 0.847 0.943 0.861 61,234 1997 0.787 0.886 0.959 0.982 0.861 0.948 0.874 58,487 1998 0.779 0.879 0.949 0.983 0.854 0.946 0.867 56,783 1999 0.792 0.887 0.958 0.982 0.865 0.943 0.877 54,907 2000 0.793 0.889 0.960 0.979 0.868 0.951 0.879 51,003 2001 0.797 0.890 0.958 0.981 0.871 0.948 0.882 56,379 2002 0.788 0.882 0.950 0.980 0.872 0.949 0.882 56,243 N 222,837 191,241 67,801 99,714 477,500 104,093 581,593 581,593

Share of girls aged 10–15 enrolled in school By head’s educational attainment By district Year None Primary Low sec. Higher Rural Urban Total N 1993 0.734 0.850 0.937 0.941 0.805 0.907 0.821 59,895 1994 0.743 0.868 0.950 0.942 0.824 0.919 0.838 59,582 1995 0.764 0.868 0.948 0.946 0.828 0.923 0.843 57,102 1996 0.770 0.877 0.953 0.955 0.842 0.928 0.856 58,430 1997 0.775 0.884 0.952 0.958 0.853 0.934 0.865 55,427 1998 0.783 0.881 0.948 0.961 0.855 0.932 0.867 53,814 1999 0.791 0.891 0.954 0.962 0.864 0.938 0.876 51,936 2000 0.800 0.894 0.961 0.969 0.872 0.946 0.883 47,832 2001 0.817 0.896 0.966 0.967 0.882 0.937 0.890 52,938 2002 0.801 0.889 0.951 0.969 0.878 0.944 0.887 52,752 N 207,841 180,188 64,162 97,517 447,531 102,177 549,708 549,708 Notes: Participation shares are weighted by survey weights. N refers to the number of observations in the sample, rural districts denote Kabupatens, urban districts denote Kotas.

6.4 Data and empirical approach

6.4.1 Data

Indonesia’s national socio-economic household survey, SUSENAS, provides information on the outcome variables and socio-economic characteristics for individuals and households. The SUSENAS is conducted annually around January- February, typically sampling approximately 200,000 households, and is representative at the district level. The district will be the main unit of analysis in this study, as districts take a key role as the main administrative units in Indonesia, and the regional labor markets are also best deﬁned in district terms. the effects of trade liberalization in indonesia 109

Districts are defined as municipalities (Kota) or predominantly rural areas (Kabupaten). Each district (both the Kota and Kabupaten) can be further divided into urban precincts (Kelurahan) and rural villages (Desa). It is important to emphasize the difference between these two urban/rural indicators, since the empirical analysis will use both variables. A district classified as a rural Kabupaten mainly consists of rural villages, but may also include small towns that are registered as urban precincts in the data. In similar vein, districts classified as urban Kota mainly contain urban precincts and neighborhoods, but may also cover some rural areas at the fringes, which are then registered as villages. The exception are the five districts comprising the capital Jakarta, which are defined completely as urban. The Kota/Kabupaten classification will therefore appear as a fixed effect in the empirical analysis, but the differential effects of tariff reduction for municipalities and rural districts will be also investigated. In addition, the Desa/Kelurahan division will be included as time variant control variable within districts. The outcome variables record whether a child has worked in the last week and whether a child is enrolled in school. As mentioned earlier, market work is defined as activities that directly generate household income, irrespectively of whether it was performed in the formal labor market or within the family. It does not include household chores that constitute domestic work. The SUSE- NAS also provides information on education attainment of other household members, household composition, monthly household expenditure and sector of employment. The sectoral share of GDP per district is derived from the Regional GDP (GRDP) data of the Central Bureau of Statistics in Indonesia (BPS). The district GRDP are available from 1993 onward, and breaks down district GDP by 1 digit sector, of which the tradable sectors are agriculture, manufacturing and mining/quarrying. Information at lower level of aggregation is available (down to 3 digits), but the availability is not consistent over time. Information on tariff lines comes from the UNCTAD-TRAINS database. These reflect the simple average of all applied tariff rates, which tend to be substantially lower than the bound tariffs during the 1990s (WTO 1998, WTO 2003). As data on tariff lines is not available for some years (1994, 1997, and 1998), information from four three-year intervals is used (1993, 1996, 1999, and 2002) both in the pooled cross section and in the district panel. The relevant product categories can be matched to sectoral employment data derived from SUSENAS and the GRDP sectors at the 1 digit level. Not every district can be included in the sample: Districts in Aceh, Maluku and Irian Jaya have not been included in the SUSENAS in some years due to violent conflict situations at the time of the survey. In addition, the 13 districts in East Timor were no longer covered by SUSENAS after the 1999 referendum on independence; these districts are not part of the analysis. Another problem the effects of trade liberalization in indonesia 110

is that over the period 1993 to 2002 some districts have split up over time. To keep time consistency in the district definitions, the study redefines districts to the 1993 parent district definitions. Since the SUSENAS rounds are representative for the district population in each year, a district panel can be constructed by pooling the four annually repeated cross sections. This yields a balanced panel of 261 districts, which reduces to 244 districts when merged with the GRDP data, as GRDP information is not available for all districts. In addition to the pooled data, a district pseudo-panel is also created by computing district level means for each variable, weighted by survey weights. The advantage of pooling the cross-section data is that individual level data can account for individual heterogeneity, both in terms of characteristics and the impact of trade liberalization. For example, the differential impact for high and low skilled labor, urban and rural areas, and gender can be investigated. On the other hand, in the pseudo-panel the observation unit is the district which allows us to investigate dynamic effects at the district level.15 Table 6.3 provides descriptive statistics. Pooling the four years of SUSE- NAS data yields a sample of 458,401 observations for children age 10 to 15. The top panel of the table shows the outcome variables and the individual and household characteristics that will be used in the regressions. The bottom panel shows the descriptive statistics for the different tariff measures after they have been merged to the individual data. The tariff variables reflect a district’s exposure to tariff protection based on either GRDP or employment shares. The table also reports the district specific poverty head count ratio (P0) and poverty severity (P2). The poverty measures are based on per capita expenditure data from SUSENAS and province—urban/rural—specific poverty lines.16

6.4.2 Regional tariﬀ exposure

Following Topalova (2005) and Edmonds et al. (2007), tariff exposure measures are constructed by combining information on geographic variation in sector composition of the economy and temporal variation in tariff lines by product category. This yields a measure indicating how changes in exposure to tariff reductions vary by geographic area over the period 1993 to 2002. This study extends this strategy by considering two alternative measures of economic structure at district level. First, similar to previous studies, tariff changes are related to the employment shares of sectors within districts. This

15 In order to allow for heterogeneity in the district panel, it is constructed not only for the whole sample but also for subsamples, divided by age, gender, and household head’s education. 16 Details on the method for construction of the poverty lines are described in Pradhan, Suryahadi, Sumarto and Pritchett (2001) and Suryahadi, Sumarto and Pritchett (2003). the effects of trade liberalization in indonesia 111

Table 6.3: Descriptive statistics

Variables No. obs. Mean St.dev. Min. Max. Pooled: Child market work 458401 0.123 0.328 0 1 School enrollment 458401 0.859 0.348 0 1 Female 458401 0.486 0.500 0 1 Age 458401 12.45 1.71 10 15 Female head 458401 0.081 0.272 0 1 Household size 458401 5.727 1.815 1 22 Rural 458401 0.668 0.471 0 1 Head’s ed.: primary 458401 0.329 0.470 0 1 Head’s ed.: secondary 458401 0.117 0.321 0 1 Head’s ed.: higher 458401 0.174 0.379 0 1 Tariff weighted by labor shares 458401 5.416 3.086 0.176 14.90 Tariff weighted by GRDP shares 432161 4.441 2.356 0.160 13.85 District panel (aged 10–15 years): Child market work 1044 0.121 0.080 0.011 0.488 School enrollment 1044 0.862 0.076 0.545 0.991 Average age 1044 12.46 0.112 12.11 13.03 Female share 1044 0.487 0.027 0.385 0.598 Rural share 1044 0.646 0.317 0 1 Share of hh-heads w/o education 1044 0.376 0.160 0.028 0.848 Tariff weighted by labor shares 1044 5.264 3.080 0.176 14.90 Tariff weighted by GRDP shares 976 4.278 2.314 0.160 13.85 Total population: Poverty headcount ratio (P0) 1044 0.268 0.171 0 0.871 Squared poverty gap (P2) 1044 0.017 0.019 0 0.155

reflects how households are exposed to trade liberalization through local labor market dynamics. In addition to this, districts differ in relative exposure in terms of sector shares in district GRDP. These two measures may differ strongly, as agriculture typically has relatively high employment but low economic production shares, while the opposite holds for manufacturing. It is a priori not clear which measure will be more effective in capturing district exposure to trade liberalization. This will depend on the extent to which tariff changes are geared towards labor intensive industries.

For each sector (h) the annual national tariﬀ lines Tht for the relevant product categories are weighted by the 1993 sector shares in district (k) GRDP the effects of trade liberalization in indonesia 112

or active labor force (L):

H GRDP GRDPhk,1993 Tkt = × Tht (6.4) 1993 =1 GRDPk, Xh H L Lhk,1993 Tkt = × Tht (6.5) 1993 =1 Lk, Xh The evolution of tariﬀ protection, weighted by the GRDP and employment shares, is shown in Figure 6.7. Exposure is higher when the tariﬀ lines are weighted by employment shares as compared to GRDP. This emphasizes the role of agriculture in terms of employment as compared to economic production.17

Figure 6.7: Evolution of tariﬀ protection

Mean of district tariffs weighted by labor shares Mean of district tariffs weighted by GRDP shares 8.27 8 8

6.66 6 6 4 4

2.52

Tariffs by GRDP 2.20 2 2 Tariffs by labor shares 0 0

1993 1995 1996 1999 2000 2001 2002 1993 1995 1996 1999 2000 2001 2002 Years Years

Overall Agriculture Overall Agriculture Manufacturing Mining Manufacturing Mining

Since regionally representative data on the sectoral composition of households is usually available only at the one or two digit level, the empirical analysis cannot distinguish tariff reductions on locally produced import-competing goods from tariff reductions on goods which are not produced locally. Instead, the focus of this study lies on the interactions between overall trade liberalization and the regional differences in economic structure, which determine the 17During the analyzed time-span, rice prices were regulated, as the national trading com- pany (BULOG) had an import monopoly on rice, while export bans on rice were also effective. Given the governments control of rice import and export, tradable agricultural good production excludes rice production, and the labor and GRDP shares in tradable agriculture are reduced by the share of rice fields in agricultural plantations within each district. This latter information comes from the 1993 village agricultural census (PODES). the effects of trade liberalization in indonesia 113

extent to which a region might be negatively affected by reductions in protection but also the extent to which it might be able to benefit from the efficiency gains associated with more competition in the local economy.

6.4.3 Identiﬁcation

Static analysis: pooled district panel

Identification of the impact of tariff reductions relies on the geographic panel nature of the combined data, and in particular on the variation in tariff exposure over districts and over time. District fixed effects control for regional variation (δk), while time-region fixed effects control for aggregate time trends (λrt), allowing these to differ by the five main geographic areas of the archipelago: the islands of Java, Sumatra, Kalimantan and Sulawesi, and a cluster of smaller islands consisting of Bali and the Nusa Tenggara group.18 A set of time variant household and individual control variables is also included (Xikt): age, gender and education of the household head, household size, and whether a household resides in an urban precinct or rural village (i.e. the Desa/Kelurahan composition of districts). The main specification for the pooled district panel is

′ Pr(yikt = 1) = Pr(α + βTkt + Xiktγ + λrt + δk + ǫikt > 0) (6.6) where yikt reflects either work or schooling for child i in district k at time t. The model is estimated separately for the municipalities and rural districts. The differential impact of trade liberalization is further explored by interacting the tariff exposure measure with the education of the head of household, as proxy for high or low skilled labor.

Potential sources of bias

The main identifying assumption is that time variant shocks ǫikt are orthogonal to Tkt. This would seem a reasonable assumption, given that Tkt consists of the baseline economic structure and national changes in tariff regime. Thus, any temporal or regional variation endogenous to child work activities would be controlled for by time and geographic fixed effects. However, the identifying assumption would be violated if changes in district tariff exposure are endogenous to different local growth trajectories. Within the Indonesian context, regional variation in growth trajectories may be partly determined by initial conditions regarding sectoral composition, in particular agriculture.

18 Although Bali is typically grouped with the economic center Java, here the islands of Bali, NTT and NTB are grouped together because of close similarity of child workforce participation an its movements over time across these islands. the effects of trade liberalization in indonesia 114

Figure 6.8: Initial district conditions and change in child work 1993-2002

(a) Initial child work (b) Initial rural population .1 .1 0 0 −.1 −.1 −.2 −.2 −.3 −.3

Change child work 1993−2002 0 .1 .2 .3 .4 .5 Change child work 1993−2002 0 .2 .4 .6 .8 1 Child work 1993 Rural population share 1993

Change child work 1993−2002 0 .2 .4 .6 .8 Change child work 1993−2002 0 .2 .4 .6 Labour force share agriculture 1993 GRDP share agriculture 1993

A first trend to note is that districts with a higher initial incidence of child labor experience larger decreases in child labor over time. This is reflected by Figure 6.8a, which depicts a strong correlation between child work incidence in districts in 1993 and the decrease in child work from 1993 to 2002. With the bulk of child work located in agriculture, child work could be expected to decrease faster in districts with a relatively large share of the population active in agriculture and living in rural areas in 1993. These patterns are confirmed by Figure 6.8b for the initial rural population share, Figure 6.8c for the initial agricultural labor force share and Figure 6.8d for the GRDP agriculture share. Regional diversity in structural change from the primary to secondary and tertiary sectors and in economic outcomes is a prominent feature of Indonesia’s economic geography. Hill, Resosudarmo and Vidyattama (2008) show evidence of strong regional variation in economic growth and structural change since the 1970s. However, they find only weak positive correlation between economic growth and structural change in districts. A related initial conditions problem, discussed at length by Edmonds et al. (2007), lies with the non-tradable sector. Districts may experience different growth paths, depending on the size of the non-tradable sector. Since the initial sectoral composition of district economies is at the heart of the effects of trade liberalization in indonesia 115

Tkt, such differential trends in child labor could confound the estimated effects of trade liberalization. The scope of these confounding effects will be explored through an initial conditions sensitivity analysis and by exploiting the panel features of the data.

Dynamic analysis: district pseudo-panel

Collapsing the pooled district panel to a district pseudo-panel provides more options to further address the potential source of bias and allow a dynamic analysis, at the cost of losing the individual variation in the data. The district pseudo-panel analogue to (6.6) is ¯ ′ y¯kt = α + βTkt + Xiktγ + λrt + δk +ǫ ¯kt (6.7) wherey ¯kt is the fraction of children in district k that work or are enrolled in school in a given year t. This specification is still prone to bias through time-variant unobservables. However, with the fixed effects removed after a first-difference transformation of (6.7), it provides a first indicative test of exogeneity of tariff exposure. The assumption of strict exogeneity, E{Tkt¯ǫks} = 0 for all s and t, implies that Tkt should add no extra explanatory information in the regression ¯ ′ ∆¯ykt = β∆Tkt + ϕTkt +∆Xiktγ + λrt + ∆¯ǫkt (6.8) which provides the testable hypothesis that ϕ = 0. As suggested by Edmonds et al. (2007), the scope of the bias related to initial conditions can be investigated further by introducing initial sector shares as control variables. Therefore, initial conditions interacted with year dummy variables are added to equation (6.8). Initial conditions are reflected by the 1993 labor and GRDP shares (for specifications with T L and T GDP , respectively) of the agriculture, mining, manufacturing, construction, trade, and transport sectors (with utilities as reference group), in addition to adult literacy rates in districts. If the tariff measures are endogenous to child work or schooling, or if they capture differential trends in child work between districts, child work and schooling would also be correlated with future changes in district tariff exposure. This is tested by regressing changes iny ¯ from 1993 to 1996 on changes in T from 1999 to 2002 (i.e. ∆Tkt+2). Finally, in order to fully exploit the pseudo-panel by taking a dynamic specification, a lagged dependent variable and lagged tariff measure are included. ¯ ′ y¯kt = βTkt + φTkt−1 + θy¯kt−1 + Xiktγ + λrt + δk +ǫ ¯kt (6.9) The lagged dependent variable accounts for state dependence, and potential confounding differential trends in child labor between relatively high and low the effects of trade liberalization in indonesia 116

child labor districts. The lagged effects of tariff changes can identify short and long term effects. The immediate effect of a percentage point change in tariff exposure is reflected by β. The total long term change iny ¯ as a result of a percentage point change in tariff exposure, taking into account lagged effects of tariff changes and its dynamic multiplier effect troughy ¯kt−1, is approximated by (β + φ)/(1 − θ). However, introducing a lagged dependent variable to the model compro- mises consistency of fixed effects estimates, in particular when the time dimension of the panel is limited. A GMM approach is adopted to resolve any bias from the lagged dependent variable and potential endogeneity of tariff exposure, by using an Arellano-Bond (1991) difference GMM estimator, with a two-step Windmeijer (2005) correction.19 System estimation is not suitable as this requires the identifying assumption that the instruments are not correlated with the fixed effects. This is a problematic assumption since a main cause of concern for this analysis lies with the correlation of changes in child labor and tariffs with the initial characteristics of districts. This is also reflected in the Hansen over-identifying restrictions test results, which strongly reject the validity of the instruments in case of system GMM. Tariff exposure and the lagged dependent variable are treated as endogenous, and adult literacy as pre-determined. First differences of these variables are then instrumented with their lagged levels.20

6.5 Results

6.5.1 Static analysis

The results from the static analysis apply specification (6.6) to pooled cross section data. The estimated effects of tariff reductions on work are given in Table 6.4. The table only reports the coefficients for tariff exposure, omitting the other covariates for ease of presentation. These include a child’s age and gender, household size, gender and education of the household head, and a dummy variable indicating whether a households resides in a rural village or urbanized precinct within the district. The basic specification (model A) indicates that a decrease in tariff exposure is associated with a decrease in child work for 10 to 15 year old children, but the size of the effect varies between urban and rural areas and also depends on the nature of the exposure measure. A percentage point decrease in

19 Estimation is implemented by using the xtabond2 command in StataTM(Roodman 2003). 20 The length of the panel (4 rounds) does not allow us to meaningfully address dynamic effects that go beyond one time lag. The number of instruments used in the estimations is 25, N is 488 for tariffs weighted by GRDP shares and 522 for tariffs weighted by labor shares. the effects of trade liberalization in indonesia 117

Table 6.4: Pooled results on child market work and tariﬀ protection

Dependent Child market work (aged 10–15 years) Tariﬀs weighted by Labor shares GRDP shares Districts All Rural Urban All Rural Urban (1) (2) (3) (4) (5) (6) Model A Tariﬀ 0.0148** 0.0146** 0.0118† 0.0066* -0.0018 0.0037 (0.0015) (0.0032) (0.0011) (0.0030) (0.0034) (0.0033) Adj. R2 0.132 0.137 0.052 0.130 0.136 0.052

Model B Tariff × 0.0157** 0.0155** 0.0136* 0.0081** -0.0003 0.0049 Head’s ed.: none (0.0015) (0.0032) (0.0067) (0.0030) (0.0035) (0.0037) Tariff × 0.0133** 0.0135** 0.0131† 0.0062* -0.0022 0.0059 Head’s ed.: primary (0.0014) (0.0031) (0.0068) (0.0030) (0.0036) (0.0038) Tariff × 0.0094** 0.0103** 0.0072 0.0036 -0.0031 0.0013 Head’s ed.: secondary (0.0014) (0.0030) (0.0056) (0.0027) (0.0033) (0.0030) Tariff × 0.0037** 0.0044 0.0095 0.0001 -0.0060* 0.0022 Head’s ed.: higher (0.0014) (0.0030) (0.0069) (0.0022) (0.0026) (0.0036) Adj. R2 0.133 0.137 0.052 0.131 0.136 0.064

In all models: District fixed effects Yes Yes Yes Yes Yes Yes Region×year interact. Yes Yes Yes Yes Yes Yes No. obs. 458,401 375,400 83,001 432,161 349,160 83,001 No. districts 261 209 52 244 192 52 Notes: All models are estimated by OLS, weighted by sampling weights. Further controls include a full set of gender and age indicator interactions, household size, and dummies on heads’ education, female head, and living in a rural neighborhood. For the divided samples, rural districts refer to Kabupatens, urban districts to Kotas. Standard errors (clustered at district level) are in parentheses. **,*,† denote significance at the 1, 5, and 10% level.

labor weighted tariff exposure leads to a 1.5 percentage point decrease in work incidence. This result is mainly driven by the effect in rural districts, where the estimates are larger and more precise than for municipalities (1.5 and 1.2 percentage points, respectively). The estimated effects are smaller for GRDP weighted tariff exposure, which would suggest that tariff changes affect households mainly through labor markets and less through distributional effects of economic growth. Model B investigates differential effects by skill level. The tariff exposure measure is interacted with the level of education of the head of household, defined as (i) not completed primary school, (ii) completed primary school, (iii) completed secondary school and (iv) completed higher education. The benefits of tariff reductions are relatively higher for low skill households. The effects of tariff reductions on school enrollment are presented in Table the effects of trade liberalization in indonesia 118

Table 6.5: Pooled results on child schooling and tariﬀ protection

School enrollment of children (aged 10–15) Tariﬀs weighted by Labor shares GRDP shares Districts All Rural Urban All Rural Urban (1) (2) (3) (4) (5) (6) Model A Tariﬀ -0.0040** 0.0018 -0.0018 -0.0024 0.0020 -0.0041† (0.0013) (0.0026) (0.0043) (0.0019) (0.0026) (0.0022) Adj. R2 0.186 0.195 0.080 0.186 0.195 0.080

Model B Tariff × -0.0048** 0.0011 -0.0013 -0.0043* 0.0020 -0.0049† Head’s ed.: none (0.0014) (0.0027) (0.0061) (0.0020) (0.0027) (0.0026) Tariff × -0.0027* 0.0026 -0.0039 -0.0015 0.0047† -0.0056† Head’s ed.: primary (0.0013) (0.0025) (0.0050) (0.0019) (0.0026) (0.0031) Tariff × 0.0006 0.0046† 0.0020 0.0018 0.0062** -0.0017 Head’s ed.: secondary (0.0012) (0.0024) (0.0055) (0.0018) (0.0026) (0.0025) Tariff × 0.0047** 0.0083** -0.0028 0.0041* 0.0080** -0.0035 Head’s ed.: higher (0.0014) (0.0024) (0.0070) (0.0018) (0.0025) (0.0030) Adj. R2 0.187 0.195 0.080 0.186 0.196 0.080

6.5 for the full sample and by district type. A percentage point decrease in exposure to labor weighted tariff protection is associated with an 0.4 percentage point overall increase in school enrollment, but the effects are not statistically significant for rural or urban districts separately. For GRDP weighted tariffs the overall effect is smaller and not precise, although there is a statistically significant effect for municipalities. Similar to the results for child work, the effects on schooling are larger for low skill households. However, some of the positive coefficients are statistically significant, which implies a decrease in enrollment due to a decrease of tariff exposure, in particular for higher educated rural households. the effects of trade liberalization in indonesia 119

6.5.2 Sensitivity analysis and exogeneity tests

The static results are based on district fixed effects, and could be confounding the effects of trade liberalization and differential growth paths. This section will examine this potential source of bias. First, the pseudo-panel estimation results for both random and fixed effects specifications are presented in Table 6.6. As expected, the correlation between tariff exposure and the outcome variables is partly driven by time invariant characteristics of districts and changes in demographics and human capital. In general, controlling for fixed effects (columns 2 and 6) reduces the tariff coefficients, compared to the random effects specification (columns 1 and 5). Adding covariates yields specification (6.7), and further reduces the effects but improves the fit (columns 3 and 7). The effects of tariff changes on child work remain precise and are consistent with the pooled cross section results, although the coefficients are slightly smaller. But the coefficients for schooling reduce strongly in size, in particular for the GRDP weighted tariffs. Simple inclusion of the lagged tariff variable (columns 4 and 8) indicates that immediate and longer-term effects of trade liberalization might differ, in particular for schooling, and that dynamic effects of tariff changes should be considered. Adding a one-period lag of tariff exposure eliminates the immediate effects on schooling, suggesting that the observed correlations are in fact lagged effects. In fact, the labor share weighted coefficients suggest that the model without lags seems to pick up the net result of an initial negative effect on schooling, which is outweighed in the long term by a positive effect. However, the standard error for the initial effect is too large to confirm this. The GRDP weighted results also show the larger effect with the lagged tariff changes, but the results lack precision. For child work the labor share weighted results are not sensitive to including lagged tariffs, while the GRDP weighted results show an initial large effect which is attenuated over time. The exogeneity of the tariff measures and sensitivity to initial conditions are therefore addressed more specifically in Tables 6.7 and 6.8. Columns (1) to (4) show that the first difference estimates for child work resemble the fixed effects estimates presented earlier. For schooling, on the other hand, the coefficients are small and statistically insignificant (columns 5 to 8). The tests for strict exogeneity of tariffs exposure with respect to child work is given in column (2) of Tables 6.7 and 6.8. For both the labor and GRDP weighted tariff measures the zero hypothesis of strict exogeneity is not rejected. The estimated effects on child work are also robust to including the initial conditions and year interaction terms in case of the labor share weighted tariffs (column (3)). The coefficients increase in size but lose precision when the interaction terms are included. For schooling, the results are not so robust. The hypothesis of strict exogeneity is rejected for the labor share weighted the effects of trade liberalization in indonesia 120 0.0019 -0.0019 cy rates (0.0015) -0.0037* s (0.0021) res † 3 0.395 7 0.380 –15 years) 0011) (0.0017) 0015) (0.0027) 027** -0.0020 040** -0.0012 -0.0004 denote significance at the 1, 5, and district panel † ithin the age group; the set of further controls includes the 0.0030 (0.0023) (0.0028) -0.0056* eholds with a household head with no education, adult litera level) are in parentheses. **,*, Tariffs weighted by labor shares Tariffs weighted by labor sha No Yes Yes Yes No Yes Yes Yes NoNo No No Yes Yes Yes Yes No No No No Yes Yes Yes Yes Tariffs weighted by GRDP shares Tariffs weighted by GRDP share (1) (2) (3) (4) (5) (6) (7) (8) RE FE FE FE RE FE FE FE (0.0013) (0.0015) (0.0017) (0.0027) (0.0011) (0.0012) (0. (0.0022) (0.0022) (0.0023) (0.0042) (0.0016) (0.0016) (0. Table 6.6: Child work, schooling and tariff protection in the L GDP T T year interactions × : The dependent variables are expressed as district shares w 10% level. average age of children, the share of girls, the share of hous and the rural share. Standard errors (clustered at district Tariffs 0.0151** 0.0128** 0.0113** 0.0108** -0.0052** -0.0 Lagged tariffs R-squared (within)ObservationsNo. districts 0.552 0.553 1044 0.606 261 1044 0.548 261 1044 0.423 261 783 0.428 261 1044 0.53 1044 261 1044 261 261 783 261 Tariffs 0.0111** 0.0111** 0.0062** 0.0155** -0.0055** -0.0 Lagged tariffs R-squared (within)ObservationsNo. districts 0.497 0.497 976 0.572 244 976 0.524 244 976 0.400 244 732 0.401 244 0.50 976 244 976 244 976 244 732 244 DependentPanel A Market work (aged 10–15 years)Panel B Schooling (aged 10 Panels A and B: District fixed effects Notes Region Further controls the effects of trade liberalization in indonesia 121 -0.0014 ade, and (0.0015) on, adult res s first differences 10–15 years) 0.0039 denote significance at the † † ffs weighted by labor) -0.0027 (0.0017) ) of agriculture, mining, manufacturing, construction, tr L T re, share of households with a household head with no educati red at district level) are in parentheses. **,*, eracy rate in districts. The set of further controls include -0.0013 0.0046 0.0027** -0.0016 (0.0015) (0.0047) (0.0009) (0.0020) No No Yes No No No Yes No Tariffs weighted by labor shares Tariffs weighted by labor sha Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes (1) (2) (3) (4) (5) (6) (7) (8) (0.0015) (0.0027) (0.0080) (0.0013) (0.0019) (0.0049) year interact. Table 6.7: Difference estimates and test for exogeneity (Tari × +2 year interactions t × : Initial conditions are 1993 labor/GRDP shares (for L T Dependent ∆ Market work (aged 10–15 years)Region ∆ SchoolingNotes (aged Initial conditions Further controls ∆Tariffs 0.0094** 0.0078** 0.0096 -0.0001 0.0033 Tariffs ∆Tariffs R-squaredObservationsNo. districts 0.201 783 261 0.202 783 0.238 261 783 0.13 261 522 0.216 261 0.225 783 261 0.259 783 261 783 0.20 261 522 261 1, 5, and 10% level. transport (with utilities as reference group) and adult lit of all controls fromliteracy Table rates 6.6: and the average rural child share. age, girls’ Standard sha errors (cluste the effects of trade liberalization in indonesia 122 -0.0021 ade, and s (0.0021) on, adult s first differences 10–15 years) denote significance at the † ffs weighted by GRDP) -0.0018 (0.0026) ) of agriculture, mining, manufacturing, construction, tr GDP T / L re, share of households with a household head with no educati red at district level) are in parentheses. **,*, T eracy rate in districts. The set of further controls include 0.0021 0.0069 0.0019 -0.0007 (0.0019) (0.0046) (0.0012) (0.0029) No No Yes No No No Yes No Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Tariffs weighted by GRDP shares Tariffs weighted by GRDP share (1) (2) (3) (4) (5) (6) (7) (8) (0.0023) (0.0038) (0.0069) (0.0016) (0.0027) (0.0078) year interact. × Table 6.8: Difference estimates and test for exogeneity (Tari +2 year interactions t × : Initial conditions are 1993 labor/GRDP shares (for GDP T 1, 5, and 10% level. transport (with utilities as reference group) and adult lit of all controls fromliteracy Table rates 6.6: and the average rural child share. age, girls’ Standard sha errors (cluste Dependent ∆ Market work (aged 10–15 years)Region ∆ SchoolingNotes (aged ∆Tariffs 0.0073** 0.0102** 0.0173* -0.0001 0.0026 0.0080 Tariffs Initial conditions Further controls ∆Tariffs R-squaredObservationsNo. districts 0.185 732 244 0.186 732 0.226 244 732 0.13 228 488 0.206 244 0.208 732 0.238 244 732 244 732 0.19 228 488 244 the effects of trade liberalization in indonesia 123 are of P shares d at district 0025) (0.0027) 0.0000 -0.0001 10–15 years) 1999-2002) 1996 1999–2002 1993–1996 1999–2002 controls from Table 6.6: average child age, girls’ share, sh teracy rates and the rural share. Standard errors (clustere YesYes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes denote significance at the 1, 5, and 10% level. (1) (2) (3) (4) (5) (6) (7) (8) † (0.0027) (0.0028) (0.0035) (0.0041) (0.0022) (0.0017) (0. Table 6.9: Difference estimates by time period (1993–1996 and year interactions × : The set of further controls includes first differences of all level) are in parentheses. **,*, households with a household head with no education, adult li Time period 1993–1996 1999–2002 1993–1996 1999–2002 1993– Tariffs weighted by Labor shares GRDP shares Labor shares GRD Dependent∆Tariffs ∆ Market work (aged 10–15 years)Region Notes 0.0088** 0.0110** 0.0045 ∆ Schooling (aged 0.0127** -0.0018 0.0020 - R-squaredObservations 0.136 261 0.245 261 0.107 244 0.318 244 0.179 0.217 261 0.162 261 0.168 244 244 Further controls the effects of trade liberalization in indonesia 124

tariff measure, although exogeneity is not rejected when the initial condition interaction terms are included. The sign and size of the coefficients are also not robust to specification, although the standard errors are large. Finally, there is no correlation between the two outcome variables and future tariff changes. The coefficients for two year lead changes in tariff exposure (∆T1999−2002) are small and not statistically significant for both schooling and work (∆y1993−1996). For child work, the initial results from the district pseudo panel are fairly robust to specification, suggesting that the negative relationship between tariff reduction and child work is not driven by differential growth trajectories of district economies and the reduction of the agricultural sector. However, the results also indicate that the effects of trade liberalization are not static events but are dynamic in nature. These dynamics are overlooked in a simple fixed effect analysis, which may in fact capture the confounding result of short- and long-term impacts. By contrast, the robustness of the results for schooling is questionable. The strong positive correlation between schooling and tariff reduction does seem to conceal confounding trends. Although not all tests for exogeneity are rejected, the effects on schooling are sensitive to specification and controlling for initial economic structure. The economic crisis in 1997/98 also raises interpretational concerns, as the devaluation of the Rupiah resulted in short-term price spikes which affected especially the poor. Although the effect of the price spike has largely subsided by the 1999 SUSENAS round, and the overall negative effect of the crisis is controlled for by the region-time fixed effects included in every regression, concerns still might remain that the crisis might confound the effects of tariff reductions. This is especially the case if the effects of the crisis were correlated with the economic structure of the districts. In order to investigate these concerns, Table 6.9 reports difference estimates for two separated time periods: 1993–1996 (pre-crisis) and 1999–2002 (post-crisis). The results confirm the robustness of the previous findings on the effects of trade liberalization on child work which are largely unaffected by the split. By contrast, the effects on schooling disappear. This again questions the robustness of the schooling results, although it might also reflect the dynamic nature of the effects on schooling discussed before.

6.5.3 Dynamic analysis

The main GMM estimation results for the dynamic specification are summarized in Tables 6.10 and 6.11.21 The results are presented by age group, gender 21 The full specification and detailed results are reported in the Appendix, Tables A.8 and A.9. Note that the Hansen over-identification test rejects the validity of the instruments the effects of trade liberalization in indonesia 125

Table 6.10: Child market work and tariﬀ protection, GMM estimates

Dependent (yt) Child market work (aged 10–15 years) Tariﬀs weighted by Labor shares GRDP shares

Tt Tt−1 yt−1 Tt Tt−1 yt−1 Age 10–15 0.0086** -0.0026 0.3876** 0.0106* -0.0085† 0.2794 (0.0028) (0.0026) (0.1252) (0.0043) (0.0050) (0.1774) Age 10-12 0.0058** -0.0045 0.3638* 0.0064† -0.0035 0.2465 (0.0023) (0.0022) (0.1474) (0.0034) (0.0041) (0.1885) Age 13-15 0.0118** -0.0006 0.2368** 0.0161* -0.0143* 0.1843 (0.0040) (0.0037) (0.0813) (0.0063) (0.0068) (0.1167) Male 0.0099** -0.0044 0.3482** 0.0110† -0.0077† 0.2545 (0.0033) (0.0031) (0.1071) (0.0058) (0.0041) (0.1288) Female 0.0081* -0.0011 0.3334** 0.0122** -0.0053 0.2995** (0.0032) (0.0028) (0.1124) (0.0049) (0.0033) (0.1053) No education hh-head 0.0165** -0.0026 0.3430** 0.0274** -0.0119* 0.2938** (0.0057) (0.0038) (0.1040) (0.0094) (0.0052) (0.1133) Primary hh-head 0.0107** 0.0011 0.2149* 0.0126* -0.0015 0.2454* (0.0036) (0.0035) (0.0935) (0.0053) (0.0040) (0.0963) Junior Sec. hh-head 0.0100* 0.0050 0.0352 0.0041 -0.0002 0.0577 (0.0041) (0.0043) (0.0848) (0.0054) (0.0044) (0.0978) Senior Sec. hh-head 0.0169* -0.0070 0.1125† 0.0055 0.0001 0.1312† (0.0067) (0.0055) (0.0661) (0.0039) (0.0027) (0.0742) Notes: Coefficients from difference GMM estimates on tariffs and lagged market work are presented in rows for different subsamples of children (by age, gender and household head’s education). All models include region and time interactions, average age, share of females (where applicable), rural share, adult literacy, and the share of unskilled headed families (where applicable). Standard errors (clustered at district level) are in parentheses. N = 522 for tariffs weighted by labor shares, and N = 488 for tariffs weighted by GRDP shares. **,*,† denote significance at the 1, 5, and 10% level.

and household head education level. Decreasing district tariff exposure by one percentage point, leads to a short- term decrease in child labor incidence of the 10–15 year old by 0.86-1.06 percentage points depending on the tariff measure. Recursive substitution over the four periods gives us the overall effect of the decrease in local tariff exposure: when considering labor sector shares, the tariff reductions explain around half (49%) of the average reduction of child labor of 8.98% points. The overall effect is even larger when tariff exposure is weighted by district GRDP sector shares, explaining around 70% of the reductions. These figures clearly show that the local effects of tariff reductions are considerable, but because of the

at 10 percent level for the urban child work estimates. The Hansen test also rejects at 10 percent level for some of the sub-samples, in particular the youngest age group, boys and households with no educated head of household. Hence, these results need to be interpreted with some caution, although the previous analysis shows little evidence of endogeneity of tariﬀ exposure with respect to child work. For school enrollment, which seems more prone to endogeneity, the Hansen test does not reject the exogeneity of the set of instruments. the effects of trade liberalization in indonesia 126

Table 6.11: Child schooling and tariﬀ protection, GMM estimates

Dependent (yt) Schooling (aged 10–15 years) Tariﬀs weighted by Labor shares GRDP shares

Tt Tt−1 yt−1 Tt Tt−1 yt−1 Age 10–15 0.0046* -0.0023 0.3672** 0.0034 -0.0014 0.3659** (0.0021) (0.0019) (0.0919) (0.0034) (0.0025) (0.0858) Age 10-12 0.0026 0.0005 0.2111* 0.0021 0.0006 0.1848† (0.0016) (0.0014) (0.1024) (0.0029) (0.0019) (0.0995) Age 13-15 0.0075* -0.0055† 0.2237** 0.0054 -0.0042 0.2134** (0.0033) (0.0030) (0.0696) (0.0057) (0.0044) (0.0666) Male 0.0060* -0.0021 -0.2784** 0.0043 0.0011 -0.2944** (0.0027) (0.0022) (0.0832) (0.0045) (0.0028) (0.0863) Female 0.0031 -0.0027 0.3399** 0.0026 -0.0044 0.3481** (0.0025) (0.0024) (0.0976) (0.0046) (0.0037) (0.0912) No education hh-head -0.0016 -0.0093* 0.2240* -0.0038 -0.0084 0.2854** (0.0049) (0.0042) (0.1009) (0.0089) (0.0062) (0.1036) Primary hh-head 0.0023 -0.0026 0.1589† -0.0077 0.0050 0.1768† (0.0035) (0.0027) (0.0912) (0.0067) (0.0040) (0.0986) Junior Sec. hh-head 0.0057† -0.0067* -0.0511 0.0017 -0.0042 -0.0506 (0.0032) (0.0028) (0.0433) (0.0049) (0.0039) (0.0447) Senior Sec. hh-head -0.0009 0.0015 -0.0583 0.0005 -0.0003 -0.0306 (0.0038) (0.0042) (0.0696) (0.0035) (0.0028) (0.0760) Notes: Coefficients from difference GMM estimates on tariffs and lagged schooling are presented in rows for different subsamples of children (by age, gender and household head’s education). All models include region and time interactions, average age, share of females (where applicable), rural share, adult literacy, and the share of unskilled headed families (where applicable). Standard errors (clustered at district level) are in parentheses. N = 522 for tariffs weighted by labor shares, and N = 488 for tariffs weighted by GRDP shares. **,*,† denote significance at the 1, 5, and 10% level.

inclusion of region-year interactions, the magnitudes of these effects cannot be interpreted directly. Reduction in child work due to tariff reductions are strongest in the age group of 13 to 15 years, which is not surprising given the low incidence of child work among primary school age children. There is no gender gap, as the effects are of comparable size for both girls and boys. In case of tariff exposure based on district labor shares, improvements in child labor are irrespective of household skill composition, while children from relatively low skill households seem to be the main beneficiaries of trade liberalization based on the GRDP sector shares of the local economy. When decomposing these effects for the rural and urban subsamples (cf. Table 6.12), it becomes apparent that these favorable effects are mainly rural, irrespective of the tariff weighting scheme. For schooling, GMM gives results opposite to the fixed effects models, as overall schooling seems to decrease due to trade liberalization (cf. Table 6.11). the effects of trade liberalization in indonesia 127

Table 6.12: Child market work and tariﬀ protection by district type (GMM)

Dependent (yt) Child market work (aged 10–15 years) Tariﬀs weighted by Labor shares GRDP shares

Tt Tt−1 yt−1 Tt Tt−1 yt−1 Rural districts: Age 10–15 0.0171** -0.0059 0.3884** 0.0170* -0.0085† 0.2794 (0.0066) (0.0044) (0.1461) (0.0086) (0.0050) (0.1774) Age 10-12 0.0088 -0.0057 0.3864* 0.0071 -0.0035 0.2465 (0.0055) (0.0040) (0.1726) (0.0072) (0.0041) (0.1885) Age 13-15 0.0264** -0.0081 0.2206** 0.0268* -0.0143* 0.1843 (0.0089) (0.0063) (0.0949) (0.0121) (0.0068) (0.1167) Male 0.0149† -0.0061 0.3568** 0.0104 -0.0089 0.2064 (0.0079) (0.0056) (0.1281) (0.0092) (0.0064) (0.1600) Female 0.0171* -0.0058 0.3162** 0.0221** -0.0086† 0.2693* (0.0075) (0.0054) (0.1228) (0.0092) (0.0052) (0.1214) No education hh-head 0.0226* -0.0088 0.2925* 0.0278* -0.0145* 0.1616 (0.0098) (0.0058) (0.1368) (0.0138) (0.0072) (0.1351) Primary hh-head 0.0129† -0.0025 0.2329* 0.0147 -0.0025 0.2315* (0.0072) (0.0059) (0.1043) (0.0090) (0.0060) (0.1126) Junior Sec. hh-head 0.0124† 0.0053 0.0247 0.0072 -0.0021 0.0392 (0.0069) (0.0070) (0.0883) (0.0093) (0.0077) (0.1028) Senior Sec. hh-head 0.0243** -0.0115 0.1085 0.0103 -0.0026 0.1361† (0.0093) (0.0078) (0.0697) (0.0068) (0.0044) (0.0799) Urban districts: Age 10–15 -0.0128 0.0058 0.1274 -0.0033 -0.0036 0.1015 (0.0159) (0.0120) (0.2058) (0.0068) (0.0040) (0.1550) Age 10-12 -0.0119 -0.0008 0.0547 -0.0006 -0.0023 0.0264 (0.0123) (0.0092) (0.1349) (0.0070) (0.0050) (0.1071) Age 13-15 -0.0275 0.0140 0.1648 -0.0087 -0.0050 0.1482 (0.0304) (0.0206) (0.1816) (0.0128) (0.0060) (0.1599) Male -0.0102 0.0061 -0.0151 -0.0037 -0.0015 -0.0250 (0.0204) (0.0110) (0.1801) (0.0091) (0.0048) (0.1628) Female -0.0110 0.0071 0.1765 -0.0009 -0.0056 0.1810 (0.0195) (0.0184) (0.1623) (0.0067) (0.0046) (0.1428) No education hh-head 0.0082 0.0090 0.3298** 0.0248 -0.0145 0.3180** (0.0200) (0.0135) (0.1157) (0.0255) (0.0117) (0.1206) Primary hh-head -0.0044 0.0020 -0.1555 -0.0058 -0.0002 -0.1647 (0.0120) (0.0132) (0.1232) (0.0082) (0.0064) (0.1195) Junior Sec. hh-head -0.0267 0.0179† -0.0760 -0.0185 0.0047 -0.0824 (0.0264) (0.0099) (0.0963) (0.0180) (0.0068) (0.0993) Senior Sec. hh-head -0.0112 0.0093 -0.1498 0.0049 -0.0037 -0.1804 (0.0255) (0.0159) (0.1131) (0.0081) (0.0048) (0.1134) Notes: Coefficients from difference GMM estimates on tariffs and lagged market work are presented in rows for different subsamples of children (by age, gender and household head’s education). Further controls are identical to those listed in Table 6.10. Standard errors (clustered at district level) are in parentheses. N = 522 for tariffs weighted by labor shares, and N = 488 for tariffs weighted by GRDP shares. **,*,† denote significance at the 1, 5, and 10% level. the effects of trade liberalization in indonesia 128

Table 6.13: Child schooling and tariﬀ protection by district type (GMM)

Dependent (yt) Schooling (aged 10–15 years) Tariﬀs weighted by Labor shares GRDP shares

Tt Tt−1 yt−1 Tt Tt−1 yt−1 Rural districts: Age 10–15 0.0036 -0.0024 0.4048** -0.0048 0.0013 0.4102** (0.0039) (0.0037) (0.1128) (0.0055) (0.0036) (0.1106) Age 10-12 0.0055* 0.0026 0.1844† 0.0002 0.0040 0.1743 (0.0028) (0.0025) (0.1086) (0.0048) (0.0028) (0.1099) Age 13-15 0.0050 -0.0039 0.2651** -0.0077 0.0008 0.2640** (0.0072) (0.0067) (0.0814) (0.0100) (0.0070) (0.0837) Male 0.0054 -0.0028 0.2940** -0.0071 0.0068† 0.3073** (0.0049) (0.0039) (0.1027) (0.0057) (0.0041) (0.1091) Female 0.0018 -0.0013 0.3551** -0.0037 -0.0054 0.4010** (0.0056) (0.0055) (0.1125) (0.0075) (0.0056) (0.1120) No education hh-head 0.0004 -0.0054 0.3362** -0.0163 0.0005 0.3852** (0.0066) (0.0057) (0.1259) (0.0113) (0.0078) (0.1239) Primary hh-head 0.0060 0.0005 0.1596 -0.0146† 0.0099* 0.1450 (0.0044) (0.0035) (0.0974) (0.0076) (0.0044) (0.1153) Junior Sec. hh-head 0.0087 -0.0073 -0.0580 -0.0014 -0.0010 -0.0619 (0.0055) (0.0045) (0.0429) (0.0086) (0.0064) (0.0467) Senior Sec. hh-head -0.0042 0.0022 -0.0851 -0.0036 0.0006 -0.0552 (0.0051) (0.0058) (0.0739) (0.0069) (0.0050) (0.0809) Urban districts: Age 10–15 0.0044 -0.0156* -0.0565 0.0118 -0.0063 0.0092 (0.0130) (0.0074) (0.1651) (0.0112) (0.0052) (0.1941) Age 10-12 0.0032 -0.0080 -0.2243 0.0072 -0.0060** -0.1831 (0.0101) (0.0066) (0.1824) (0.0063) (0.0025) (0.1615) Age 13-15 0.0127 -0.0231† 0.0714 0.0175 -0.0022 0.0589 (0.0278) (0.0134) (0.1509) (0.0188) (0.0099) (0.1470) Male 0.0189 -0.0214* 0.1338 0.0151 -0.0086 0.1360 (0.0182) (0.0103) (0.1456) (0.0180) (0.0085) (0.1936) Female 0.0171 -0.0149 -0.0916 0.0093 -0.0027 -0.0357 (0.0212) (0.0135) (0.1756) (0.0148) (0.0084) (0.2123) No education hh-head 0.0132 -0.0227 -0.2458 -0.0035 -0.0064 -0.1413 (0.0173) (0.0165) (0.1588) (0.0320) (0.0183) (0.1727) Primary hh-head 0.0168 -0.0277 -0.0307 -0.0076 0.0048 0.0802 (0.0320) (0.0171) (0.1832) (0.0209) (0.0096) (0.1952) Junior Sec. hh-head -0.0067 -0.0067 -0.0387 0.0174 -0.0106 -0.0071 (0.0183) (0.0120) (0.1626) (0.0138) (0.0084) (0.1931) Senior Sec. hh-head 0.0138 -0.0099 0.0515 -0.0013 -0.0012 0.0658 (0.0291) (0.0139) (0.1570) (0.0076) (0.0041) (0.1588) Notes: Coefficients from difference GMM estimates on tariffs and lagged schooling are presented in rows for different subsamples of children (by age, gender and household head’s education). Further controls are identical to those listed in Table 6.11. Standard errors (clustered at district level) are in parentheses. N = 522 for tariffs weighted by labor shares, and N = 488 for tariffs weighted by GRDP shares. **,*,† denote significance at the 1, 5, and 10% level. the effects of trade liberalization in indonesia 129

Table 6.14: Tariﬀ reductions and poverty, GMM estimates

Tariﬀs weighted by Labor shares GRDP shares

Tt Tt−1 yt−1 Tt Tt−1 yt−1

Panel A: Dependent (yt): Poverty headcount ratio (P0) Total sample 0.0118* -0.0036 -0.0988 0.0163 -0.0115† -0.0428 (0.0058) (0.0049) (0.0974) (0.0106) (0.0061) (0.0971) Rural districts 0.0146 0.0110 -0.1492 -0.0031 -0.0082 -0.0754 (0.0094) (0.0079) (0.1093) (0.0178) (0.0108) (0.1126) Urban districts 0.0404 -0.0249 -0.3854* 0.0454 -0.0153 -0.3549* (0.0407) (0.0258) (0.1841) (0.0327) (0.0165) (0.1556)

Panel B: Dependent (yt): Squared poverty gap (P2) Total sample 0.0021** -0.0010 0.0691 0.0025 -0.0017 0.0650 (0.0008) (0.0006) (0.1642) (0.0014) (0.0008) (0.1821) Rural districts 0.0036† -0.0015 0.0785 -0.0001 -0.0008 0.0650 (0.0020) (0.0013) (0.1565) (0.0030) (0.0014) (0.1821) Urban districts -0.0001 -0.0005 -0.5436** 0.0021 -0.0011 -0.5286** (0.0029) (0.0024) (0.1699) (0.0030) (0.0018) (0.1491) Notes: Coefficients from difference GMM estimates on tariffs and lagged dependent are presented in rows (for the total sample, rural and urban districts). All models include region and time interactions, rural share, lagged per capita GRDP and adult literacy. The lagged dependent variable, tariffs, rural share, lagged pc. GRDP and literacy are treated as endogenous and instrumented in GMM style. Standard errors (clustered at district level) are in parentheses. N is 508(474) in the total, 416(382) in the rural and 92(92) in the urban sample for tariffs weighted by labor (GRDP) shares. **,*,† denote significance at the 1, 5, and 10% level.

When decomposing this effect for rural and urban subsamples (cf. Table 6.13), these effects are only significant for 10–12 year old in rural districts, when tariffs are weighted by labor shares. By contrast, there are also sizable improvements of schooling with a one-period time lag, in urban areas, mainly for boys aged 13 to 15. The schooling results for urban areas are intuitively appealing: the effects of tariff reductions take some time to materialize in terms of households’ enrollment decisions, as these typically involve lumpy and sunk costs; but when effects on enrollment do materialize, they are especially relevant around the transition from primary to junior secondary schooling, and in urban areas where the supply and variety of education is relatively large. However, the positive coefficients for the rural sample are difficult to explain, in particular because of the decreases in rural child labor but also in the light of the strong initial negative correlation between enrollment and tariff exposure. Given the sensitivity of the schooling results to the choice of specification, interpretation of these results should be cautious since endogeneity and omitted variable bias can not be ruled out. This study remains largely a reduced form analysis: the main transmis- the effects of trade liberalization in indonesia 130

Table 6.15: Market work by age cohorts, GMM estimates

Dependent (yt) Market work in the given age group Tariﬀs weighted by Labor shares GRDP shares

Tt Tt−1 yt−1 Tt Tt−1 yt−1

Aged 10–12 0.0058** -0.0045* 0.3638** 0.0064† -0.0039 0.2366 (0.0023) (0.0022) (0.1474) (0.0034) (0.0027) (0.1655) Aged 13–15 0.0118** -0.0006 0.2368** 0.0161** -0.0088* 0.2608** (0.0040) (0.0037) (0.0813) (0.0063) (0.0043) (0.0845) Aged 16–18 0.0109** 0.0023 0.1331 0.0105 -0.0048 0.2161* (0.0041) (0.0040) (0.0934) (0.0079) (0.0053) (0.0960) Aged 19–24 0.0067 0.0022 0.0524 0.0007 -0.0015 0.0500 (0.0034) (0.0034) (0.0848) (0.0062) (0.0053) (0.0889) Aged 25–30 0.0002 0.0019 0.2121* -0.0017 -0.0006 0.2221** (0.0036) (0.0032) (0.0953) (0.0056) (0.0040) (0.0910) Aged 31–40 0.0004 -0.0003 0.2351** -0.0045 -0.0008 0.2597** (0.0026) (0.0022) (0.0994) (0.0043) (0.0030) (0.0967) Aged 41–60 -0.0011 0.0016 0.1968** -0.0069 0.0020 0.2180** (0.0027) (0.0021) (0.0823) (0.0043) (0.0028) (0.0849) Notes: Coefficients from difference GMM estimates on tariffs and lagged market work are presented in rows for different subsamples of children and adults (by age). All models include region and time interactions, average age, share of females, rural share, adult literacy, and the share of unskilled headed families. Standard errors (clustered at district level) are in parentheses. N = 522 for tariffs weighted by labor shares, and N = 488 for tariffs weighted by GRDP shares. **,*,† denote significance at the 1, 5, and 10% level. sion channels cannot be identified through which child work and schooling are affected by reduced tariff exposure. Nevertheless, some global indication of the main mechanisms at work can be seen by looking at the effects on district poverty profiles and adult employment. Tariff reductions have lead to a reduction in the extent and severity of poverty. Table 6.14 shows the estimated effects of reduced tariff exposure on the poverty head count ratio (Panel A) and the squared poverty gap (Panel B), where the model specification is similar to the earlier dynamic GMM. While the poverty head count merely records the fraction of the district population that cross an arbitrary level of consumption, the squared poverty gap reflects the curvature in the per capita expenditure distribution for the population living below the poverty line. The results show that a percentage point reduction in tariff exposure reduces the poverty headcount in districts by 1.2 percentage points, and also reduces inequality among the poor. In other words, the results seem to suggest that income effects play a role, in particular at the bottom end of the income distribution. Tariff reductions do not impact workforce participation of cohorts older than 18 (cf. results in Table 6.15). This would suggest that the effect of trade the effects of trade liberalization in indonesia 131

liberalization on child labor is not driven by substitution of adult for child labor, and that the observed income eﬀects are not due to a labor supply response and reduced unemployment. Rather, income eﬀects seem to be a result of relative wage increases, in particular for low skilled labor.

6.6 Conclusion

This chapter examined the effects of trade liberalization on child work and schooling in Indonesia. In the 1990s, Indonesia went through a major reduction in tariff barriers, as average import tariff lines decreased from around 17.2 percent in 1993 to 6.6 percent in 2002; a period which also saw reductions in child work and increased school enrollment. The empirical analysis identifies the effects of trade liberalization by combining geographic variation in sector composition of the economy with temporal variation in tariff lines by product category. This yields geographic variation in changes in average exposure to trade liberalization over time, hence identifying geographical differences in the effects of trade policy. The specification and sensitivity analysis confirms the robustness of the results for child labor, and it shows no evidence of remaining sources of bias. The schooling results, however, may still be capturing confounding effects, and we need to be careful interpreting these as causal effects. The main findings of this study suggest that Indonesia’s trade liberalization experience in the 1990s has contributed to a strong decline in child labor. Decreased tariff exposure is associated with a decrease in child work and an increase in enrollment among 10 to 15 year old children. The effects of tariff reductions increase with the age of children, and are strongest for children from low skill backgrounds and in rural areas. Through these human capital investments, trade liberalization will have long term welfare implications, in particular for low skill, and presumably poorer, households. Although this reduced form analysis can at best provide indirect evidence of the main transmission channels, there is strong support for the hypothesis that reduction of child labor is driven by positive income effects from trade liberalization for the poorest. This is consistent with other studies, which argue that trade liberalization in Indonesia brought about a relative wage increase for low skilled labor, although causal effects are hard to confirm (Suryahadi 2003, Arnold and Smarzynska Javorcik 2005, Sitalaksmi et al. 2009). Further analysis of this causal relationship would be an area of future research. The empirical evidence from this study and other country studies would suggest that the potential benefits to be gained from trade liberalization, and its distributional implications, are indeed context specific. The Indonesian context seems to have provided the pre-conditions needed to generate classic the effects of trade liberalization in indonesia 132

Stolper-Samuelson eﬀects, partly facilitated by a coinciding process of structural change in the 1990s that saw a reallocation of labor from agriculture to services and manufacturing. In particular the mobility of low skilled labor seems to play an important role, which, combined with increased productivity and competitiveness, has lead to better employment opportunities outside agriculture and increased returns to low skilled labor. Such cross-country heterogeneity may be underlying the weak average eﬀects of trade liberalization on child labor and human capital investments found at macro level, highlight- ing the importance of considering local economic contexts when propagating trade reforms and formulating subsequent social policy responses. chapter 7

Conclusion

7.1 Summary

This book addressed both supply and demand side determinants of child labor, and the potential effects of trade liberalization on child labor from a theoretical and empirical perspective. In the global fight against child labor the main challenge is not national regulation, but the effective enforcement of existing national legal norms. Most countries already have compulsory schooling laws as well as minimum working age requirements in place, but actual enforcement of these rules is weak and often nonexistent. As child labor is a very heterogeneous phenomenon (see Chapter 2), it calls for a multitude of policy approaches. From an ethical perspective, the worst, exploitative forms of child labor should be eradicated by all means, enforcement of a ban on such child labor should be unconditional and effectively administered. The appropriate response to other types of economic activity by children depends on the underlying causes of child labor. Thus, effective eradication of child labor requires an understanding of the supply and demand side forces that shape child labor outcomes, as well as of the political economy context that can help or hinder an effective domestic regulation. For a better understanding of the supply side determinants of child labor, it is important to investigate the differences in trade-offs faced by children and their families. Chapter 3 focused on gender differences in the determinants of various forms of work (market and domestic work) and their interaction with schooling and idleness. The empirical analysis estimated jointly the determinants of participation in market work, domestic work, and schooling for a sample of North-Indian children. By distinguishing between two types of work while investigating the work-school decision, a more differentiated picture of the work-school trade-off arose. As the estimation framework also allowed for children combining activities or staying idle, determinants of specialization in activities were explicitly addressed. The analysis has shown that factors traditionally linked with child labor (like parental preferences for education captured by adult literacy) are more closely linked to the trade-off between the gender specific activity (market work for boys and domestic work for girls) and thus their effects differ by gender. By contrast, economic incentives to work (arising from family business, land ownership, adult wages in the village or higher costs of schooling) are increasing market work participation of both

133 conclusion 134

sexes alike. Social norms (proxied by caste or religion) are better to explain the participation in the gender unspecific work activities. The overall richness of these results emphasized the importance of a more detailed view on child labor. In order to assess how economic policies affect child labor it is also crucial to understand the interactions between child labor supply and demand, especially because several factors reducing child labor supply increase demand at the same time. However, empirical studies based on household data often only focus on households’ labor supply decisions but cannot identify the role that differences in labor demand play in shaping child labor outcomes. The empirical study in Chapter 4 addressed specifically this issue of the interplay between demand and supply factors, using village level data on child labor in small scale manufacturing in Indonesia. The great geographic variation across (virtually) all Indonesian villages and urban neighborhoods allowed to distinguish between demand and supply side determinants of child labor. The empirical evidence has shown that the interplay of supply and demand side factors can also lead to seemingly counterintuitive results. Child labor might be unaffected by or increasing with credit access and school proximity as these factors not only reduce child labor supply but simultaneously constitute positive location factors for firms, thereby increasing the demand for child laborers. These results are of great policy importance as they show that growth enhancing policies might lead to increases in child labor in the short run. To counteract the labor demand effects, such growth enhancing policies should be complemented by policies enabling more flexible adult employment and promoting school attendance. The supply of and the demand for child labor can be both affected if a country opens up to international trade. The effects of trade liberalization are transmitted trough changes in relative goods prices as well as in relative returns to skilled and unskilled labor. As a result, income and substitution effects arise that can both increase or decrease child labor; a priori it is not clear whether the favorable income effects will outweigh the incentive effects of higher labor demand. However, as Chapter 5 has demonstrated, against general expectations, even the income effects of trade will not necessarily reduce child labor. The theoretical discussion deliberately abstracted from incentive effects and restricted its focus to child labor that arises out of subsistence needs. Focusing on income effects only resulted in the counterintuitive argument that under specific circumstances the inequality reducing effects of trade might increase child labor even in the absence of incentive effects. The distributional consequences of increasing specialization might raise aggregate child labor if the gains to low skilled families are not large enough to make them withdraw their children from work, while the need for child labor in capital owner families increases. Such circumstances are more likely to arise in very poor food exporting countries. Subsequently, descriptive support for this argument was conclusion 135

presented, based on a panel of countries extending over four decades: child labor decreased with increasing openness by significantly less within the group of the poorest food exporting countries than in food exporting countries on average. As the overall effect of trade liberalization on child labor and schooling is a priori undetermined, it remains an empirical issue whether trade liberalization succeeds at reducing child labor and increasing schooling. Chapter 6 addressed the causal effects of trade liberalization—measured in form of tariff reductions—on child work and schooling in the context of Indonesian regions. The analysis used a district panel—consisting of four rounds of regionally representative household data (from 1993 to 2002)—and related the district shares of 10 to 15 year old children performing market work or enrolled in school to a measure of trade liberalization exposure of that district. The extensive robustness checks as well as the dynamic panel data analysis helped to address endogeneity concerns with the estimation procedure. Although trade liberalization in Indonesia in the 1990s did not have a robust effect on child schooling, it robustly decreased market work of children. These effects were mainly restricted to rural districts, and were higher for the lower skilled households. Along with the evidence on the poverty reducing effects of trade liberalization in Indonesia, these findings lead to the conclusion that the favorable distributional effects of trade liberalization overturned potential incentive effects, and resulted in reductions in child labor.

7.2 Policy implications

Based on the existing literature as well as on the theoretical and empirical evidence presented in this book, a few central policy conclusions emerge.

Domestic measures in the ﬁght against child labor

The worst forms of child labor should be eradicated through domestic policies, relying both on consequent monitoring and strong sanctions. However, an overall ban on child labor in all likelihood will make families worse off, either by sending children to more clandestine and worse forms of child labor or by threatening the survival of families. Rather the underlying causes of child labor should be addressed. As child labor and schooling are conflicting alternatives (cf. Chapter 3), policies that shift the trade-off from work in favor of more schooling can help to reduce child labor at the same time. Policies that raise the net returns to education will be increasing the attractiveness of schooling for the families. Improvements in the quality of education can be achieved by improved school availability and facilities, but even more importantly, by an increased conclusion 136

quality of actual tuition; possible instruments include a better screening of teachers, better curricula, remedial education programs, and programs aiming at reducing teacher absenteeism (Orazem and King 2007). The attractiveness of schooling can also be increased by reducing the costs of schooling through providing free schoolbooks, free transportation or subsidizing school fees. The provision of additional benefits, school meals or cash transfers conditional on school attendance, has also been shown to be effective to increase schooling and to reduce child labor even if substitution between work and schooling is less than perfect (Parker et al. 2007). Poverty is clearly among the main underlying causes of child labor, so increasing the incomes of the poor will reduce the necessity to recur to child labor in order to cover subsistence needs. Moreover, it is not only lower permanent income that makes the poor to resort to child labor; their vulnerability to shocks and their limited access to credit for consumption smoothing are equally important. Providing the poor with better access to credit markets, which would allow them to save and borrow to smooth consumption over time and to insure against income shocks, can also reduce child labor supply, especially if subsidized credits are also coupled to school attendance. Introducing social security schemes can also help families to make the fertility transition from many uneducated to a few better educated children. Policies that foster regional economic activity, by offering credit to small businesses or redistributing land to poor households, typically also raise the demand for unskilled labor and can thus also increase child labor in the short run (see also the evidence in Chapter 4). In a similar vein, trade liberalization can also increase child labor by increasing the demand for unskilled labor and hence the incentives to send children to work (cf. Chapters 5 and 6). Nonethe- less, such policies are usually strongly beneficial as they promote growth and wealth accumulation. In the long run, the favorable income effects of such policies are most likely to dominate, and child labor can be expected to decrease. However, in order to avoid increases in child labor in the short run, such policies should be flanked by measures that increase both the attractiveness of schooling and reduce the need for child labor within the family. For this latter it is important to make the hiring of extra labor from outside of the family easier and thus reduce the complementarity between asset ownership and child labor.

The role of trade sanctions

In the international discussion, trade sanctions have been often proposed as means to address child labor standard violations abroad (see Chapter 1). How- ever, the effects of trade sanctions are theoretically not entirely clear as the resulting income and substitution effects point into different directions (cf. Chapters 5 and 6). Restricting trade flows with a developing (and most likely conclusion 137

unskilled labor abundant) country can be expected to reduce the demand for unskilled labor and hence unskilled wages in that economy. This will reduce the incentives to send children to work, while the expectation of increasing skill premia over time could further raise the incentives for educating children. This might even sound good, as this is the outcome one would prefer. How- ever, trade sanctions can only affect tradable sectors, they will thus not be able to address child labor in a country as such, but only its occurrence in the export sectors of the economy, shifting child labor into non-tradable sectors, and potentially even into more clandestine occupations. Moreover, it is not at all clear that the poverty increasing effects of such policies will not outweigh any favorable incentive effects, increasing thus child labor even further. Re- stricting trade flows means not only an average income loss in the developing country, but additionally, trade sanctions impart a heavy burden on unskilled households by reducing their incomes disproportionately. If the poor are the ones who benefit from more trade, trade restrictions will be counterproductive, increase poverty and thus child labor. Although the net effect of the income and substitution effects is a priori not clear, the existing empirical evidence should invoke caution. The trade liberalization experience of Indonesia (cf. Chapter 6) clearly emphasizes the potentially positive role of income effects of trade for fighting child labor. Nor should the findings of Chapter 5—although cautioning about the potentially unfavorable aggregate child labor response to trade liberalization—be used to motivate the use of trade sanctions against countries where child labor occurs. If not trade sanctions, what alternatives are there for international policy? Social labeling schemes intend to improve the welfare of families producing export goods without child labor, and hence to reduce the incentives to let children work. These schemes rely on the expectation that consumers should be willing to pay more for child labor free products if they care about how those products were made (Freeman 1994). The problem with such schemes is that they can push children towards uncertified activities, while exporters without a label will have to bear welfare losses; as a result, children will not be necessarily better off with the introduction of the ”child labor free” products (see Basu, Chau and Grote 2006, Baland and Duprez 2009). On their actual impacts there are no systematic empirical findings beyond anecdotal evidence. Child labor bans, induced by the threat of trade restrictions or otherwise, might be less effective in addressing the global child labor problem than other policy instruments. The forward-looking policy response to child labor should entail promoting consequential pro-poor policies, improving significantly the school system, and shifting the incentives in favor of education. If feeling compassion with working children, developed nations should rather help to build well-functioning school systems so that it becomes more attractive to send children to school. conclusion 138

This book has argued that globalization, more specifically trade liberalization, is neither a curse nor panacea; it can both increase and decrease child labor and thus its effects on working children are context specific. But if the poor can reap the benefits from the increasing production, while at the same time policies actively promote schooling, the reduction of trade barriers can contribute to the eradication of child labor. appendix A

Statistical Appendix

Table A.1: Age distribution of children by school-class attended

Class attended Age 1 2 3 4 5 6 7 8 9 10 11 12 Total

4 200000000000 2 5 1220000000000 14 6 13741 3 0 0 0 0 0 0 000 181 7 136 74 37 7 1 0 0 0 0 0 0 0 255 8 97 124 89 30 3 0 0 0 0 0 0 0 343 9 30 64 62 25 14 3 0 0 0 0 0 0 198 10 38 71 86 61 64 22 8 3 0 0 0 0 353 11 7 20 27 27 24 20 8 3 0 0 0 0 136 12 11 18 30 48 55 53 45 19 9 1 0 0 289 13 0 8 23 10 21 32 33 27 10 3 0 0 167 14 0 2 8 11 13 12 17 36 36 17 0 0 152 15 0 4 0 6 10 12 16 25 31 30 1 2 137 16 1 0 1 3 3 4 517193672 98 17 0 2 0 0 3 2 1 4 52432 46 10–17 57 125 175 166 193 157 133 134 110 111 11 6 1378 Source: LSMS Uttar Pradesh/Bihar 1997/98.

139 statistical appendix 140

Table A.2: Deﬁnitions of explanatory variables

Variable Description

Married Variable takes 1 if individual is married, 0 otherwise Household income The natural logarithm of yearly household income (in Rupees), excluding child wage income, per adult (18+) hh. member Family business Variable takes 1 if at least one adult hh. member is involved in small scale business activities, 0 otherwise Marginal/Small/ Variables take 1 if acres of land owned per adult (18+) hh. member Large land are 1. between 0.0025 and 0.5 ac., 2. between 0.5 and 2 ac., 3. larger than 2 ac, 0 otherwise. Comparis. group: no land owned. Female (Male) literacy Proportion of literate among adult (18+) female (male) hh. members Share infants (children) No. of hh. members aged 0–5 (6–9) relative to adult hh. members (18+) Share boys (girls) No. of male (female) hh. members aged 10–17 relative to adults (18+) Elderly share No. of hh. members aged 67 or above relative to adult hh. members (18+) Female share Share of females among adult hh. members (18+) Birth order Birth order among siblings of the same sex (first born: 1) Lower castes Variable takes 1 if hh. belongs to a backward (agricultural or other) caste, 0 otherwise (comp. group: higher/middle castes) Scheduled castes Variable takes 1 if hh. belongs to a scheduled caste or tribe, 0 otherwise (comp. group: higher/middle castes). Definition is based on The Scheduled Castes and the Scheduled Tribes Act, 1989 Muslim Variable takes 1 if hh. belongs to the Muslim religion, 0 otherwise Time to school Time to reach the nearest secondary school (in 10s of minutes) School costs (v.) The median of total yearly expenses per child in primary school (classes 1 to 5) in the village (in .000 Rupees), calculated from sample data Female work low/high Variables take 1 1 if based both on the LSMS sample (village level) and the NSS 1999/00 sample (district level) workforce participation rate of adult (18+) females is below/above the median. Comparis. group: middle levels of female work (where info from the two data sources lies on the two sides of the median values). Median wages (v.) Median of the daily wage rate of an adult male worker in the village (in 10s of Rupees) across different agricultural activities Tractor owners Variable takes 1 if hh. owns a tractor, 0 otherwise Thresher owners No. of threshers owned by the hh. Value of transfers Value of transfers received by the household during the past year (.000 Rupees) divided by the No. of adult (18+) hh. members Rooms per adult No. of rooms in the dwelling, divided by the no. of adult (18+) hh. members statistical appendix 141

Table A.3: Child labor incidence in small scale manufacturing by island

Isle (No. of villages) Child labor incidence % Small ﬁrms By industry (u.m.) (c.m.) % of vill. No.(c.m.)

Sumatera (21113) 11.0 31.3 35.2 5.6 Leather 0.2 17.9 1.0 4.0 Wood 3.2 25.2 12.6 3.9 Metal 0.5 21.0 2.5 4.2 Ceramic 2.2 23.2 9.6 14.0 weaving 1.3 19.7 6.7 10.8 Food 5.7 27.4 20.7 8.7 Other 2.1 27.4 7.5 14.2

Jawa (24952) 19.5 27.3 71.5 30.2 Leather 0.8 18.3 4.5 8.3 Wood 5.1 15.4 33.0 8.4 Metal 0.9 15.9 5.7 11.0 Ceramic 5.6 21.9 25.4 37.9 weaving 1.7 17.6 9.8 19.1 Food 10.6 21.9 48.4 18.4 Other 3.6 19.1 18.9 32.0

Bali/Nusa Tengg. (3974) 20.6 30.7 66.9 46.4 Leather 0.2 11.7 1.9 2.7 Wood 7.4 30.6 24.3 23.7 Metal 1.3 19.6 6.8 18.2 Ceramic 5.4 29.5 18.1 44.3 weaving 7.2 18.1 39.7 53.4 Food 6.6 27.4 24.3 19.9 Other 3.7 25.7 14.3 37.0

Kalimantan (6014) 15.4 39.3 39.2 7.7 Leather 0.1 12.9 0.5 2.7 Wood 3.8 31.7 12.0 6.7 Metal 0.9 23.5 3.7 4.9 Ceramic 7.1 44.4 16.0 21.8 weaving 1.3 35.3 3.7 17.4 Food 5.6 26.5 21.0 8.6 Continued on Next Page statistical appendix 142

Table A.3 — Continued

Isle (No. of villages) Child labor incidence % Small ﬁrms By industry (u.m.)a (c.m.) % of vill. No.(c.m.)

Other 2.9 35.5 8.0 8.9 Sulawesi (7659) 16.2 29.6 54.6 9.7 Leather 0.1 19.2 0.7 3.2 Wood 6.7 20.6 32.3 4.5 Metal 0.8 16.0 4.9 5.1 Ceramic 3.4 22.7 15.2 10.0 weaving 1.9 18.0 10.7 25.2 Food 6.5 27.1 23.9 10.1 Other 2.8 27.0 10.5 13.0

Maluku (1577) 7.4 23.2 31.7 4.4 Leather 0.3 23.5 1.1 2.1 Wood 2.6 35.3 7.4 3.9 Metal 0.6 35.7 1.8 2.1 Ceramic 2.8 20.0 14.0 8.5 weaving 1.3 17.7 7.2 13.6 Food 3.7 28.4 12.9 12.2 Other 1.0 20.0 4.8 6.1

Irian Jaya (3507) 3.2 35.1 9.1 1.6 Leather 0.1 7.5 1.5 11.3 Wood 1.2 28.6 4.2 15.5 Metal 0.2 31.8 0.6 2.3 Ceramic 0.9 40.0 2.3 5.7 weaving 0.2 37.5 0.5 1.9 Food 1.4 44.1 3.2 19.3 Other 0.5 50.0 0.9 6.8

a Notes: Provinces are as of 2002. The unconditional mean (u.m.) of child labor gives the average prevalence rate of child labor over all villages. Conditional means (c.m.) are calculated for the subsample of those villages where small scale businesses operate in the given industry. statistical appendix 143

Table A.4: Determinants of the number of small ﬁrms (w. sample selection)

(A) Outcome stage (B) Selection stage Dependent ln No. small firms Small firm presence Coeff. SE Coeff. SE dy/dx

Credit access 0.6493** (0.0594) 0.3766** (0.0294) 0.1457 Small business credit 0.2253** (0.0508) 0.1383** (0.0293) 0.0546 Epidemic death rate (%) -0.0568 (0.0790) 0.0009 (0.0320) 0.0028 Bad housing -0.1902† (0.1009) -0.1373** (0.0481) -0.0529 Family size -0.0618* (0.0245) -0.0317** (0.0119) -0.0122 Unemployment (%) -0.0000 (0.0038) -0.0003 (0.0016) -0.0002 Primary school presence 0.4929** (0.0685) 0.2451** (0.0370) 0.1001 Sec. school distance -0.0719** (0.0218) -0.0551** (0.0107) -0.0137 Village market 0.1749** (0.0317) 0.0795** (0.0175) 0.0256 Urban -0.0181 (0.0696) 0.0636† (0.0351) 0.0226 Coastal 0.1562* (0.0768) 0.0772* (0.0352) 0.0335 Lower altitudes 0.3670** (0.1282) 0.2148** (0.0603) 0.0837 Distance to market -0.0461** (0.0162) -0.0359** (0.0068) -0.0143 Families w. electricity 0.0730* (0.0295) 0.0240 Mill’s lambda 1.7901** (0.0460) Notes: The model is estimated by a two-step Heckman procedure. Estimations are based on 68345 observations, 35094 of which are uncensored. Both equations include a constant, province fixed effects, population and population squared. Robust standard errors (clustered at district level) are reported in parentheses. Marginal effects are evaluated at the sample mean. *,**,† denote significance at the 1, 5, and 10% level.

Table A.5: Countries in the sample (No. obs. per country)

Algeria (4), Argentina (4), Australia (2), Austria (2), Bangladesh (4), Barbados (2), Belgium (2), Belize (2), Benin (4), Bolivia (3), Botswana (4), Brazil (4), Burkina Faso (4), Burundi (4), Cameroon (3), Chad (4), Chile (3), China (3), Colombia (4), Congo, Rep. (4), Costa Rica (4), Cote d’Ivoire (4), Denmark (2), Dominican Republic (4), Ecuador (4), Egypt, Arab Rep. (4), El Salvador (4), Fiji (3), Finland (2), France (2), Gabon (4), Ghana (4), Greece (3), Guatemala (4), Guyana (3), Haiti (4), Honduras (4), Hong Kong, China (3), Hungary (3), Iceland (2), India (4), Indonesia (4), Ireland (3), Israel (2), Italy (4), Jamaica (4), Japan (2), Kenya (4), Korea, Rep. (3), Lesotho (4), Luxembourg (2), Madagascar (4), Malawi (4), Malaysia (4), Malta (3), Mauritania (4), Mexico (4), Morocco (4), Nepal (3), Netherlands (2), Nicaragua (4), Niger (4), Nigeria (4), Oman (3), Pakistan (3), Panama (2), Papua New Guinea (2), Paraguay (4), Peru (4), Philippines (4), Portugal (4), Puerto Rico (2), Rwanda (4), Senegal (4), Sierra Leone (3), South Africa (3), Spain (2), Sri Lanka (4), Sudan (2), Sweden (2), Switzerland (2), Syrian Arab Republic (4), Thailand (4), Togo (4), Trinidad and Tobago (3), United Kingdom (2), United States (2), Uruguay (4), Venezuela, RB (4), Zambia (4), Zimbabwe (2) statistical appendix 144

Table A.6: Deﬁnitions of variables

Variable Deﬁnition

∆Chlab Percentage point change in labor force participation rates of children (aged 10–14) over a decade Chlab1960/Chlab1970 Labor force participation rates of children (aged 10–14) in 1960/1970 ∆GDP Growth rate of per capita GDP (in constant 1985 USD) over a decade ∆Urban Percentage point change in the urbanization rate over a decade ∆Open Percentage point change in the openness indicator (value of exports and imports divided by GDP) over a decade ∆Fraser Change in the Freedom to Trade Internationally Index by the Fraser Institute (rescaled between 0 and 1) over a decade

IGroup Indicator variable for the group of n poorest countries (Group = n); Poverty deﬁned based on GDP per capita (in constant 1985 USD) in 1960 FoodExp Indicator variable that takes 1 for countries whose exports of food and agricultural raw materials made out more than 50% of their merchan- dise exports (in at least one of the years of 1960, 1970, 1980, 1990 or 2000). Notes: Except for the Fraser Institute Index, all variables are taken from or based on the World Development Indicators (World Bank 2004).

Table A.7: The originally poorest countries (Rank, Name, GDP p.c. in 1960)

1. China (94.3); 2. Malawi (97.8); 23. Burundi (128.1); 4. Lesotho (138.7); 5. Nepal (139.4); 6. Burkina Faso (172.7); 7. India (180.4); 8. Pakistan (180.8); 9. Kenya (201.2); 10. Nigeria (223.5); 11. Togo (228.7); 12. Sudan (230.9); 13. Bangladesh (239.2); 14. Indonesia (248.9); 15. Sierra Leone (263.7); 16. Rwanda (276.0); 17. Mauritania (281.1); 18. Sri Lanka (284.8); 19. Chad (290.3); 20. Botswana (313.1); 21. Syrian Arab Republic (319.1); 22. Benin (321.3); 23. Congo, Dem. Rep. (347.2); 24. Egypt, Arab Rep. (359.7); 25. Madagascar (382.6); 26. Niger (385.7); 27. Ghana (450.3); 28. Congo, Rep. 457.5; 29. Thailand (464.6); 30. Zimbabwe (467.4); 31. Honduras (513.2); 32. Cameroon (518.1); 33. Papua New Guinea (563.6); 34. Haiti (569.9); 35. Nicaragua (636.2) Notes: The values in parentheses refer to nominal GDP p.c. in year 1960, measured in 1985 USD. Source: World Bank (2004). statistical appendix 145

Table A.8: Child market work and tariﬀ protection, full GMM results

Dependent (yt) Child market work (aged 10–15 years) Tariﬀs weighted by Labor shares GRDP shares Districts All Rural Urban All Rural Urban (1) (2) (3) (4) (5) (6)

Tariﬀs 0.0086** 0.0171** -0.0128 0.0106* 0.0170* -0.0033 (0.0028) (0.0066) (0.0159) (0.0043) (0.0086) (0.0068) Lagged tariﬀs -0.0026 -0.0059 0.0058 -0.0060† -0.0085† -0.0036 (0.0026) (0.0044) (0.0120) (0.0032) (0.0050) (0.0040) Lagged dependent 0.3876** 0.3884** 0.1274 0.3338* 0.2794 0.1015 (0.1252) (0.1461) (0.2058) (0.1379) (0.1774) (0.1550) Average age 0.0817** 0.0895** 0.0228 0.0736** 0.0811** 0.0191 (0.0207) (0.0261) (0.0249) (0.0228) (0.0288) (0.0262) Share of girls -0.0968 -0.1625* 0.0182 -0.1025 -0.1544† 0.0077 (0.0611) (0.0828) (0.0871) (0.0629) (0.0819) (0.0953) Share hh-heads w/o ed. 0.1030* 0.0879 0.0863 0.1112* 0.0896 0.0869 (0.0484) (0.0563) (0.0729) (0.0483) (0.0563) (0.0814) Adult literacy -0.5251** -0.5621** -0.1906 -0.4095* -0.4322* -0.1963 (0.1482) (0.1712) (0.2233) (0.1624) (0.1819) (0.2289) Rural share 0.0686* 0.0841* -0.0263 0.0645† 0.0713† -0.0192 (0.0310) (0.0381) (0.0402) (0.0332) (0.0394) (0.0375)

Region×year interact. Yes Yes Yes Yes Yes Yes Observations 522 418 104 488 384 104 No. districts 261 209 52 244 192 52 Hansen p-value 0.332 0.159 0.050† 0.269 0.162 0.090† Notes: All models include time and region interactions and are estimated by difference GMM, treating tariffs, the lagged dependent and adult literacy rates as endogenous. Standard errors (clustered at district level) are in parentheses. **,*,† denote significance at the 1, 5, and 10% level. statistical appendix 146

Table A.9: Child schooling and tariﬀ protection, full GMM results

Dependent (yt) Schooling of children (aged 10–15) Tariﬀs weighted by Labor shares GRDP shares Districts All Rural Urban All Rural Urban (1) (2) (3) (4) (5) (6)

Tariﬀs 0.0046** 0.0036 0.0044 0.0034 -0.0048 0.0118 (0.0021) (0.0039) (0.0130) (0.0034) (0.0055) (0.0112) Lagged tariﬀs -0.0023 -0.0024 -0.0156* -0.0014 0.0013 -0.0063 (0.0019) (0.0037) (0.0074) (0.0025) (0.0036) (0.0052) Lagged dependent 0.3672** 0.4048** -0.0565 0.3659** 0.4102** 0.0092 (0.0919) (0.1128) (0.1651) (0.0858) (0.1106) (0.1941) Average age -0.0679** -0.0740** -0.0356 -0.0724** -0.0740** -0.0397 (0.0178) (0.0202) (0.0306) (0.0192) (0.0220) (0.0305) Share of girls -0.0019 -0.0399 -0.0046 0.0018 -0.0263 0.0255 (0.0505) (0.0605) (0.0786) (0.0514) (0.0636) (0.0806) Share hh-heads w/o ed. -0.1368** -0.139** -0.1243† -0.1322** -0.130** -0.1273† (0.0365) (0.0421) (0.0661) (0.0368) (0.0447) (0.0751) Adult literacy 0.3337** 0.3583** 0.1633 0.3807** 0.4195** 0.1425 (0.1070) (0.1227) (0.2202) (0.1118) (0.1348) (0.2179) Rural share -0.0233 -0.0169 -0.0006 -0.0266 -0.0190 -0.0545 (0.0223) (0.0247) (0.0486) (0.0231) (0.0251) (0.0468)

Region×year interact. Yes Yes Yes Yes Yes Yes Observations 522 418 104 488 384 104 No. districts 261 209 52 244 192 52 Hansen p-value 0.330 0.435 0.449 0.521 0.597 0.545 Notes: All models include time and region interactions and are estimated by difference GMM, treating tariffs, the lagged dependent and adult literacy rates as endogenous. Standard errors (clustered at district level) are in parentheses. **,*,† denote significance at the 1, 5, and 10% level. appendix B

Mathematical Appendix

B.1 First order eﬀects on the number of child laborers

Proof of dLC = dLCθ: Diﬀerentiate (5.7) totally and get 1 dL = dηN + [S dp + p dS − dY − γL dw] C wγ θ θ Aθ Cθ

s s Substitute for dη = g(ks)dks and dθ = g(k )dk − g(ks)dks by noticing that s s s dSθ = Ns dθ, dYAθ = [Nw + rdKθ] dθ+θN dw+Kθ dr, and dKθ = N[k g(k )dk − ksg(ks)dks]. It follows that

s s s p dSθ − dYAθ = N{(ps − w − rk ) g(k )dk − (ps − w − rks] g(ks)dks} =0 =wγ

−θN| dw −{zKθ dr,} | {z } and hence,

1 pS − rK dL = S dp − K dr − θ θ dw = dL . C wγ θ θ w Cθ

B.2 Relative changes in aggregate child labor

Equation (5.10a) follows from totally differentiating (5.7). It can be further rewritten by using the magnification effects of Jones (1965), which can be derived by totally differentiating the zero profit conditions (5.1) and noting raK2 waL2 that unit input coefficients minimize unit costs: w = δ p and r = − δ p , wa ra where δ = δ 2 − δ 1 = δ 1 − δ 2, while δ = Li , and δ = Ki (cf. K K L L Li pi Ki pi Wong 1995). Accordingly, (5.10a) becomes: b b b b 1 1 LC = r(Kθ − aK1Sθ) p δ YC c 1 1 b = − w(θN + γLCθ − aL1Sθ) p (5.10b) δ YC

147 b mathematical appendix 148

The two expressions are equivalent as child income serves to close the “income gap”, rKθ = pSθ −w(θN +γLCθ), and the zero proﬁt conditions apply, raK1 = p − waL1.

B.3 Child labor supply response under universal subsistence

If no families live below the poverty line, η = 0 and LCθ = LC . Since the relatively labor intensive subsistence good is exported (X1 > sN), it holds that γ N + L ≥ L = a 1X1 + a 2X2 > a 1sN. θ Cθ L L L The relationship on the left hand side follows as average hours of child work LCθ LCθ among the poor families are higher as within the whole population ( θN > N ) Hence, θN + γLCθ > aL1Sθ.

B.4 Suﬃcient conditions for an increase in aggregate child labor

If η > 0, a suﬃcient condition for an increase in hours worked by children is given by γ N + L

The prerequisites for such an outcome are

LCθ LC 1. θN < N : average hours of child work among the families where children work some time are lower than average hours of child work within the whole economy.

2. aL2X2 < aL1(sηN − Sη): labor inputs needed to produce the amount of food that is missing for those who live below the poverty line are higher than the labor inputs used to produce the manufacturing good. This is more likely if the comparative advantage in the labor intensive good is relatively small, and if subsistence poverty is large. Bibliography

Acemoglu, D. and Angrist, J.: 2000, How large are human-capital externalities? Evidence from compulsory schooling laws, NBER/Macroeconomics Annual 15(1), 9–59.

Amiti, M. and Konings, J.: 2007, Trade liberalization, intermediate inputs and productivity: Evidence from Indonesia, American Economic Review 97(5), 1611–1638.

Arellano, M. and Bond, S.: 1991, Some tests of speciﬁcation for panel data: Monte Carlo evidence and an application to employment equations, Re- view of Economic Studies 58(2), 277–297.

Arnold, J. and Smarzynska Javorcik, B.: 2005, Gifted kids or pushy parents? Foreign acquisitions and plant performance in Indonesia, CEPR Discus- sion Papers 5065, Centre for Economic Policy Research, London.

Attanasio, O., Fitzsimons, E., Gomez, A., Lopez, D., Meghir, C. and Mesnard, A.: 2006, Child education and work choices in the presence of a conditional cash transfer programme in rural Colombia, CEPR Discussion Papers 5792, Centre for Economic Policy Research, London.

Bacolod, M. P. and Ranjan, P.: 2008, Why children work, attend school, or stay idle: The roles of ability and household wealth, Economic Development and Cultural Change 56(4), 791–828.

Baland, J.-M. and Duprez, C.: 2009, Are fair trade labels eﬀective against child labour?, Journal of Public Economics 93(11–12), 1125–1130.

Baland, J.-M. and Robinson, J. A.: 2000, Is child labor ineﬃcient?, Journal of Political Economy 108(4), 663–679.

Barro, R. J.: 1974, Are government bonds net wealth?, Journal of Political Economy 82(6), 1095–1117.

Bartus, T.: 2005, Estimation of marginal eﬀects using margeﬀ, The Stata Journal 5(3), 309–329.

Basri, M. C. and Hill, H.: 2004, Ideas, interests and oil prices: The political economy of trade reform during Soeharto’s Indonesia, The World Econ- omy 27(5), 633–655.

Basu, A. K., Chau, N. H. and Grote, U.: 2006, Guaranteed manufactured without child labor: The economics of consumer boycotts, social labeling and trade sanctions, Review of Development Economics 10(3), 466–491.

149 BIBLIOGRAPHY 150

Basu, K., Das, S. and Dutta, B.: 2009, Child labor and household wealth: Theory and empirical evidence of an inverted-U, Journal of Development Economics 91(1), 8–14.

Basu, K. and Van, P. H.: 1998, The economics of child labor, American Eco- nomic Review 88(3), 412–427.

Becker, G. S.: 1974, A theory of social interactions, Journal of Political Econ- omy 82(6), 1063–1093.

Becker, G. S. and Lewis, H. G.: 1973, On the interaction between the quantity and quality of children, Journal of Political Economy 81(2), 279–88.

Beegle, K., Dehejia, R. and Gatti, R.: 2005, Why should we care about child labor? The education, labor market, and health consequences of child labor, Policy Research Working Paper Series 3479, The World Bank, Washing- ton D.C.

Beegle, K., Dehejia, R. H. and Gatti, R.: 2006, Child labor and agricultural shocks, Journal of Development Economics 81(1), 80–96.

Behrman, J.: 1988, Intrahousehold allocation of nutrients in rural India: Are boys favored? Do parents exhibit inequality aversion?, Oxford Economic Papers 40(1), 32–54.

Bergstrom, T. C.: 1989, A fresh look at the rotten kid theorem – and other household mysteries, Journal of Political Economy 97(5), 1138–1159.

Bhagwati, J.: 1995, Trade liberalisation and ’fair trade’ demands: Adress- ing the environmental and labour standards issues, The World Economy 18(6), 745–759.

Bhagwati, J. and Srinivasan, T. N.: 2002, Trade and poverty in the poor countries, American Economic Review 92(2), 180–183.

Bhalotra, S.: 2003, Is child work necessary?, Bristol Economics Discussion Papers 03/554, Department of Economics, University of Bristol, UK.

Bhalotra, S.: 2004, Parent altruism, cash transfers and child poverty, Bristol Economics Discussion Papers 04/561, Department of Economics, Univer- sity of Bristol, UK.

Bhalotra, S. and Heady, C.: 2003, Child farm labor: The wealth paradox, World Bank Economic Review 17(2), 197–227.

Bhalotra, S. and Tzannatos, Z.: 2003, Child labor: What have we learnt?, Social Protection Discussion Paper Series 317, World Bank, Washington D.C. BIBLIOGRAPHY 151

Bourguignon, F., Ferreira, F. and Leite, P. G.: 2003, Conditional cash transfers, schooling, and child labor: Micro-simulating Brazil’s Bolsa Escola Program, World Bank Economic Review 17(2), 229–254.

Brown, D. K., Deardorﬀ, A. V. and Stern, R. M.: 1996, International labor standards and trade: A theoretical analysis, in J. Bhagwati and R. Hudec (eds), Fair Trade and Harmonization: Prerequisites for Free Trade?, Vol. 1. Economic Analysis, Cambridge and London: MIT Press, pp. 227–280.

Brown, D. K., Deardorﬀ, A. V. and Stern, R. M.: 2003, Child labor: The- ory, evidence, and policy, in K. Basu, H. Horn, L. Rom´an and J. Shapiro (eds), International Labor Standards: History, Theory, and Policy Op- tions, Blackwell, Malden, Mass.; Oxford and Carlton, Australia, chapter 3, pp. 195–247.

Cameron, L. A.: 2001, The impact of the Indonesian ﬁnancial crisis on children: An analysis using the 100 villages survey, Bulletin of Indonesian Economic Studies 37(1), 43–64.

Canagarajah, S. and Coulombe, H.: 1997, Child labor and schooling in Ghana, Policy Research Working Paper Series 1844, The World Bank, Washing- ton D.C.

Cappellari, L. and Jenkins, S. P.: 2003, Multivariate probit regression using simulated maximum likelihood, The Stata Journal 3(3), 278–294.

Cappellari, L. and Jenkins, S. P.: 2006, Calculation of multivariate normal probabilities by simulation, with applications to maximum simulated likelihood estimation, ISER Working Paper 2006–16, University of Essex, Colchester.

Cardoso, E. and Souza, A.: 2004, The impact of cash transfers on child labour and school attendance in Brazil, Working Paper 407, Department of Eco- nomics, Vanderbilt University.

Cartwright, K.: 1998, Child labor in Colombia, in C. Grootaert and H. A. Patrinos (eds), The Policy Analysis of Child Labor: A Comparative Study, St Martin’s Press, London, chapter 4.

Census of India: 2001, Primary Census Abstract, http://www.censusindia. net/_00_003.html, retrieved on November 10, 2006.

Chamarbagwala, R. and Tchernis, R.: 2010, Exploring the spatial determinants of children’s activities: Evidence from India, Empirical Economics 39(2), 593–617.

Chiquiar, D.: 2008, Globalization, regional wage diﬀerentials and the Stolper- Samuelson Theorem: Evidence from Mexico, Journal of International Economics 74(1), 70–93. BIBLIOGRAPHY 152

Cigno, A. and Rosati, F. C.: 2000, Why do Indian children work, and is it bad for them?, IZA Discussion Papers 115, IZA, Bonn.

Cigno, A. and Rosati, F. C.: 2005, The Economics of Child Labour, Oxford University Press, Oxford, chapter 5, pp. 83–103.

Cigno, A., Rosati, F. C. and Guarcello, L.: 2002, Does globalisation increase child labour?, World Development 30(9), 1579–1589.

Davis, D. R. and Mishra, P.: 2007, Stolper-Samuelson is dead: And other crimes of both theory and data, in A. Harrison (ed.), Globalization and Poverty, NBER Chapters, National Bureau of Economic Research, Inc, pp. 87–108. de Soto, H.: 2000, The Mistery of Capital: Why Capitalism Triumphs in the West and Fails Everywhere Else, Basic Books, New York.

Dehejia, R. H. and Gatti, R.: 2005, Child labor: The role of ﬁnancial development and income variability across countries, Economic Development and Cultural Change 53(4), 913–932.

Dessy, S. E. and Pallage, S.: 2001, Child labor and coordination failures, Journal of Development Economics 65(2), 469–476.

Dinopoulos, E. and Zhao, L.: 2007, Child labor and globalization, Journal of Labor Economics 25(3), 553–579.

Dixit, A. K.: 2000, Comment on ’A transaction cost politics analysis of international child labor standards’ (Brown, D. K.), in A. V. Deardorﬀ and R. M. Stern (eds), Social Dimensions of U.S. Trade Policies, University of Michigan Press, Ann Arbor, pp. 267–270.

Doepke, M. and Zilibotti, F.: 2005, The macroeconomics of child labor regulation, American Economic Review 95(5), 1492–1524.

Doepke, M. and Zilibotti, F.: 2009, International labor standards and the political economy of child-labor regulation, Journal of the European Economic Association 7(2-3), 508–518.

Dr`eze, J. and Kingdon, G. G.: 2001, School participation in rural India, Review of Development Economics 5(1), 1–24.

Duﬂo, E.: 2001, Schooling and labor market consequences of school construction in Indonesia: Evidence from an unusual policy experiment, American Economic Review 91(4), 795–813.

Dumas, C.: 2007, Why do parents make their children work? A test of the poverty hypothesis in rural areas of Burkina Faso, Oxford Economic Pa- pers 59(2), 301–329. BIBLIOGRAPHY 153

Duraisamy, M.: 2000, Child schooling and child work in India, Economet- ric Society World Congress 2000 Contributed Papers 0837, Econometric Society.

Duryea, S. and Arends Kuenning, M.: 2003, School attendance, child labor and local labor market ﬂuctuations in urban Brazil, World Development 31(7), 1165–1178.

Duryea, S., Lam, D. and Levison, D.: 2007, Eﬀects of economic shocks on children’s employment and schooling in Brazil, Journal of Development Economics 84(1), 188–214.

Edmonds, E. V.: 2005, Does child labor decline with improving economic status?, Journal of Human Resources 40(1), 77–99.

Edmonds, E. V.: 2006, Understanding sibling diﬀerences in child labor, Journal of Population Economics 19(4), 795–821.

Edmonds, E. V.: 2007, Child labor, in T. P. Schultz and J. A. Strauss (eds), Handbook of Development Economics, Vol. 4 of North Holland Handbooks in Economics, Elsevier, chapter 57, pp. 3607–3709.

Edmonds, E. V. and Pavcnik, N.: 2005a, Child labor in a global economy, Journal of Economic Perspectives 8(1), 199–220.

Edmonds, E. V. and Pavcnik, N.: 2005b, The eﬀect of trade liberalization on child labor, Journal of International Economics 65(2), 401–419.

Edmonds, E. V. and Pavcnik, N.: 2006, International trade and child labor: Cross–country evidence, Journal of International Economics 68(1), 115– 140.

Edmonds, E. V., Pavcnik, N. and Topalova, P.: 2007, Trade adjustment and human capital investments: Evidence from Indian tariﬀ reform, NBER Working Papers 12884, National Bureau of Economic Research, Inc, Cam- bridge, Mass.

Edmonds, E. V. and Schady, N.: 2008, Poverty alleviation and child labor, Pol- icy Research Working Paper Series 4702, The World Bank, Washington D.C.

Edmonds, E. V. and Turk, C.: 2004, Child labor in transition in Vietnam, in P. Glewwe, N. Agrawal and D. Dollar (eds), Economic Growth, Poverty and Household Welfare in Vietnam, The World Bank, Washington D.C., pp. 505–550.

Emerson, P. M. and Souza, A. P.: 2002, Birth order, child labor and school attendance in Brazil, Working Papers 0212, Department of Economics, Vanderbilt University. BIBLIOGRAPHY 154

Emerson, P. M. and Souza, A. P.: 2003, Is there a child labor trap? Intergen- erational persistence of child labor in Brazil, Economic Development and Cultural Change 51(2), 375–398.

Engerman, S. L.: 2003, Child labor: Theory, evidence, and policy, in K. Basu, H. Horn, L. Rom´an and J. Shapiro (eds), The History and Political Econ- omy of International Labor Standards, Blackwell, Malden, Mass.; Oxford and Carlton, Australia, chapter 1, pp. 9–83.

Ersado, L.: 2005, Child labor and schooling decisions in urban and rural areas: Comparative evidence from Nepal, Peru, and Zimbabwe, World Develop- ment 33(3), 455–480.

Fafchamps, M. and Wahba, J.: 2006, Child labor, urban proximity, and household composition, Journal of Development Economics 79(2), 374–397.

Fane, G.: 1999, Indonesian economic policies and performance, 1960–98, The World Economy 22(5), 651–668.

Flug, K., Spilimbergo, A. and Wachtenheim, E.: 1998, Investment in education: Do economic volatility and credit constraints matter?, Journal of Development Economics 55(2), 465–481.

Foster, A. D. and Rosenzweig, M. R.: 2004, Technological change and the distribution of schooling: Evidence from green–revolution India, Journal of Development Economics 74(1), 87 – 111.

Foster, A. and Rosenzweig, M.: 1996, Technical change and human-capital returns and investments: Evidence from the green revolution, American economic review 86(4), 931–953.

Frankel, J. A. and Rose, A. K.: 2002, An estimate of the eﬀect of common currencies on trade and income, Quarterly Journal of Economics. 117(2), 437–466.

Freeman, R. B.: 1994, A hard-headed look at labor standards, in Sengenberger and Campbell (eds), International Labour Standards and Economic Inter- dependence, International Institute for Labour Studies, Geneva, pp. 79–92.

Greene, W. H.: 2003, Econometric Analysis, ﬁfth edn, Prentice Hall, Upper Saddle River, New Jersey.

Grootaert, C.: 1998, Child labour in Cote d’Ivoire: Incidence and determinants, Policy Research Working Paper Series 1905, World Bank, Wash- ington D.C.

Grootaert, C. and Kanbur, R.: 1995, Child labour: An economic perspective, International Labour Review 134(2), 187–203. BIBLIOGRAPHY 155

Grote, U., Basu, A. and Weinhold, D.: 1998, Child labor and the international policy debate: The education/child labor trade-oﬀ and the consequences of trade sanctions, ZEF Discussion Papers on Development Policy 1, Cen- ter of Development Research, Bonn.

Guarcello, L., Mealli, F. and Rosati, F. C.: 2009, Household vulnerability and child labor: The eﬀect of shocks, credit rationing, and insurance, Journal of Population Economics 23(1), 169–198.

Gunnarsson, V., Orazem, P. F. and S´anchez, M. A.: 2006, Child labor and school achievement in Latin America, World Bank Economic Review 20(1), 31–54.

Gwartney, J. D., Lawson, R. and Gartzke, E.: 2005, Economic Freedom of the World, 2005 Annual Report, Fraser Institute, Vancouver, B.C.

Harrison, A. (ed.): 2007, Globalization and Poverty, NBER Books, University of Chicago Press for NBER, Chicago.

Hazan, M. and Berdugo, B.: 2002, Child labor, fertility, and economic growth, Economic Journal 112(482), 810–828.

Hazarika, G. and Bedi, A.: 2003, Schooling costs and child work in rural Pakistan, Journal of Development Studies 39(5), 29–64.

Heady, C.: 2003, The eﬀect of child labor on learning achievement, World Development 31(2), 385–398.

Hertel, T. W., Ivanic, M., Preckel, P. V. and Cranﬁeld, J. A. L.: 2004, The earnings eﬀects of multilateral trade liberalization: Implications for poverty, World Bank Economic Review 18(2), 205–236.

Hill, H., Resosudarmo, B. P. and Vidyattama, Y.: 2008, Indonesia’s changing economic geography, Bulletin of Indonesian Economic Studies 44(3), 407– 435.

Hilton, M. (ed.): 2003, Monitoring International Labor Standards: Quality of Information, Summary of a Workshop, Division of Behavioral and So- cial Sciences and Education and Policy and Global Aﬀairs Division, The National Academic Press, Washington DC.

Ilahi, N., Orazem, P. F. and Sedlacek, G.: 2005, How does working as a child aﬀect wage, income and poverty as an adult?, Social Protection Discussion Paper Series 514, World Bank, Washington D.C.

ILO: 1932, C33 Minimum Age (Non–Industrial Employment) Convention, On- line Document (ILOLEX), International Labour Organization. http:// www.ilo.org/ilolex/english/convdisp1.htm, retrieved on October 30, 2009. BIBLIOGRAPHY 156

ILO: 1973, C138 Minimum Age Convention, Online Document (ILOLEX), International Labour Organization. http://www.ilo.org/ilolex/english/ convdisp1.htm, retrieved on October 30, 2009.

ILO: 1998, ILO Declaration on Fundamental Principles and Rights at Work, Online Document, International Labour Organization. http://www.ilo. org/declaration/thedeclaration/textdeclaration/lang--en/index.htm, retrieved on October 30, 2009.

ILO: 1999a, C182 Worst Forms of Child Labour Convention, Online Docu- ment (ILOLEX), International Labour Organization. http://www.ilo. org/ilolex/english/convdisp1.htm, retrieved on October 30, 2009.

ILO: 1999b, R190 Worst Forms of Child Labour Recommendation, Online Document (ILOLEX), International Labour Organization. http://www. ilo.org/ilolex/english/recdisp1.htm, retrieved on October 30, 2009.

ILO: 2000, Estimates and Projections of the Economically Active Population 1950–2010, Vol. 10 of Sources and Methods: Labour Statistics, Interna- tional Labour Oﬃce, Geneva.

ILO: 2002a, Core labour standards and globalization, Statement by Kari Tapi- ola, Executive Director, International Labour Oﬃce, Online Document, International Labour Organization. http://www.ilo.org/wcmsp5/groups/ public/---ed_norm/---declaration/documents/statement/wcms_099696. pdf, retrieved on October 10th, 2009.

ILO: 2002b, Every Child Counts: New Global Estimates on Child Labour, In- ternational Labour Oﬃce, Geneva.

ILO: 2006, The End of Child Labour: Within Reach, Global Report under the Follow-up to the ILO Declaration on Fundamental Principles and Rights at Work, International Labour Oﬃce, Geneva.

ILO: 2008, Report of the 18th International Conference of Labour Statisticians, International Labour Oﬃce, Geneva.

Jacoby, H. G. and Skouﬁas, E.: 1997, Risk, ﬁnancial markets, and human capital in a developing country, Review of Economic Studies 64(3), 311– 335.

Jafarey, S. and Lahiri, S.: 2002, Will trade sanctions reduce child labour?, Journal of Development Economics 68(1), 137–156.

Jones, G. W. and Hagul, P.: 2001, Schooling in Indonesia: Crisis-related and longer-term issues, Bulletin of Indonesian Economic Studies 37(2), 207– 231. BIBLIOGRAPHY 157

Jones, R. W.: 1965, The structure of simple general equilibrium models, Jour- nal of Political Economy 73(6), 557–572.

Kambhampati, U. S. and Rajan, R.: 2004, The ’nowhere’ children: Patri- archy and the role of girls in India’s rural economy, Discussion paper, University of Reading, Centre for Institutional Performance, Department of Economics.

Kambhampati, U. S. and Rajan, R.: 2006, Economic growth: A Panacea for child labor?, World Development 34(3), 426–445.

Kambhampati, U. S. and Rajan, R.: 2008, The ’nowhere’ children: Patriarchy and the role of girls in India’s rural economy, Journal of Development Studies 44(9), 1309–1341.

Kanbargi, R. and Kulkarni, P.: 1991, Child work, schooling and fertility in rural Karnataka, in R. Kanbargi (ed.), Child Labor in the Indian Subcontinent: Dimensions and Implications, Sage Publications, New Delhi, pp. 125–163.

Kingdon, G. G.: 1998, Does the labour market explain lower female schooling in India?, Journal of Development Studies 35(1), 39–65.

Kis-Katos, K.: 2007, Does globalization reduce child labor?, Journal of Inter- national Trade and Economic Development 16(1), 71–92.

Kis-Katos, K. and Schulze, G.: fth, Child labor in Indonesian small industries, Journal of Development Studies .

Kis-Katos, K. and Schulze, G. G.: 2002, Labour standards and international trade, World Economics 3(4), 101–129.

Kis-Katos, K. and Schulze, G. G.: 2005, Regulation of child labour, Economic Aﬀairs 25(3), 24–30.

Kis-Katos, K. and Sparrow, R.: 2009, Child labor and trade liberalization in Indonesia, IZA Discussion Papers 4376, Institute for the Study of Labor (IZA), Bonn.

Kochar, A.: 2004, Urban inﬂuences on rural schooling in India, Journal of Development Economics 74(1), 113–136.

Komiya, R.: 1967, Non-traded goods and the pure theory of international trade, International Economic Review 8(2), 132–152.

Kondylis, F. and Manacorda, M.: 2006, School proximity and child labor: Evidence from rural Tanzania, Unpublished manuscript, London School of Economics, London, England. BIBLIOGRAPHY 158

Krueger, A. B.: 1996, Observations on international labor standards and trade, NBER Working Papers 5632, National Bureau of Economic Research, Cambridge, Mass.

Krueger, D. and Tjornhom Donohue, J.: 2005, On the distributional consequences of child labor legislation, International Economic Review 46(3), 785–815.

Kruger, D. I.: 2007, Coﬀee production eﬀects on child labor and schooling in rural Brazil, Journal of Development Economics 82(2), 448–463.

Kucera, D.: 2002, Core labour standards and foreign direct investment, Inter- national Labour Review 141(1–2), 31–69.

Lanjouw, P., Pradhan, M., Saadah, F., Sayed, H. and Sparrow, R.: 2002, Poverty, education and health in Indonesia: Who beneﬁts from public spending?, in C. Morrisson (ed.), Education and Health Expenditures, and Development: The cases of Indonesia and Peru, OECD Development Centre, Paris, pp. 17–78.

Lapenu, C.: 1999, Indonesia’s rural financial system: The role of the state and financial institutions, World bank microfinance case studies, World Bank, Washington D.C.

Levison, D.: 2000, Children as economic agents, Feminist Economics 6(1), 125– 134.

Levison, D., Moe, K. and Knaul, F.: 2001, Youth education and work in Mexico, World Development 29(1), 167–188.

Lewis, B. and Pattinasarany, D.: 2010, The cost of primary education in In- donesia, Technical report, The World Bank Country Oﬃce in Indonesia, Jakarta.

Liang, K.-Y. and Zeger, S. L.: 1986, Longitudinal data analysis using generalized linear models, Biometrika 73(1), 13–22.

L´opez-Calva, L. F.: 2003, Social norms, coordination, and policy issues in the ﬁght against child labor, in L. R. K. Basu, H. Horn and J. Shapiro (eds), International Labor Standards, History, Theory, and Policy Op- tions, Blackwell, Oxford, pp. 256–269.

Maﬀei, A., Raabe, N. and Ursprung, H. W.: 2006, Political repression and child labour: Theory and empirical evidence, The World Economy 29(2), 211– 239.

Maitra, P. and Ray, R.: 2002, The joint estimation of child participation in schooling and employment: Comparative evidence from three continents, Oxford Development Studies 30(1), 41–62. BIBLIOGRAPHY 159

Manacorda, M. and Rosati, F. C.: 2007, Local labor demand and child labor, UCW Working Paper 34, Understanding Children’s Work (UCW Project).

Manning, C.: 2000, The economic crisis and child labor in Indonesia, ILO/IPEC Working Paper, International Labour Oﬃce, Geneva.

Maskus, K. E.: 1997, Should core labour standards be imposed through international trade policy?, Policy Research Working Paper Series 1817, World Bank, Washington D.C.

Maskus, K. E. and Holman, J. A.: 1996, The economics of child labor standards, Discussion Paper 96/10, Department of Economics, University of Colorado at Boulder, Boulder, CO.

Menon, N.: 2005, Why might investment credit reduce schooling and increase child labor?, mimeo, Brandeis University.

Moehling, C.: 1999, State child labor laws and the decline of child labor, Explorations in Economic History 36(1), 72–106.

Morduch, J.: 2000, Sibling rivalry in Africa, American Economic Review 90(2), 405–409.

Neumayer, E. and de Soysa, I.: 2005, Trade openness, foreign direct investment and child labor, World Development 33(1), 43–63.

Nielsen, H. S.: 1998, Child labor and school attendance: Two joint decisions, CLS Working Paper 98–015, University of Aarhus.

Orazem, P. F. and King, E. M.: 2007, Schooling in developing countries: The roles of supply, demand and government policy, in T. P. Schultz and J. A. Strauss (eds), Handbook of Development Economics, Vol. 4 of North Holland Handbooks in Economics, Elsevier, chapter 55, pp. 3475–3559.

Orazem, P. and Gunnarsson, L. V.: 2004, Child labour, school attendance and performance: A review, Working Paper 04001, Department of Economics, Iowa State University, Ames, Iowa.

Pal, S.: 2004, How much of the gender diﬀerence in child school enrolment can be explained? Evidence from rural India, Bulletin of Economic Research 56(2), 133–158.

Paqueo, V. and Sparrow, R.: 2006, Free basic education in Indonesia: Policy scenarios and implications for school enrolment, Mimeo, The World Bank, Jakarta. BIBLIOGRAPHY 160

Parikh, A. and Sadoulet, E.: 2005, The eﬀect of parents’ occupation in child labor and school attendance in Brazil, CUDARE Working Papers 1000, Department of Agricultural and Resource Economics, UCB, University of California, Berkeley.

Parish, W. L. and Willis, R. J.: 1993, Daughters, education and family budgets: Taiwan experiences, Journal of Human Resources 28(4), 863–898.

Parker, S. W., Rubalcava, L. and Teruel, G.: 2007, Evaluating conditional schooling and health programs, in T. P. Schultz and J. A. Strauss (eds), Handbook of Development Economics, Vol. 4 of North Holland Handbooks in Economics, Elsevier, pp. 3963–4035.

Parsons, D. and Goldin, C.: 1989, Parental altruism and self interest: Child labor among late Nineteenth–century families, Economic Inquiry 27(4), 637–659.

Pradhan, M.: 1998, Enrolment and delayed enrolment of secondary school age children in Indonesia, Oxford Bulletin of Economics and Statistics 60(4), 413–430.

Pradhan, M., Suryahadi, A., Sumarto, S. and Pritchett, L.: 2001, Eating like which ”Joneses”? An iterative solution to the choice of a poverty line ”reference group”, Review of Income and Wealth 47(4), 473–487.

Psacharopoulos, G. and Patrinos, H. A.: 1997, Family size, schooling and child labor in Peru – An empirical analysis, Journal of Population Economics 10(4), 387–405.

Ranjan, P.: 2001, Credit constraints and the phenomenon of child labor, Jour- nal of Development Economics 64(1), 81–102.

Ravallion, M. and Wodon, Q.: 2000, Does child labour displace schooling? Evidence on behavioural responses to an enrollment subsidy, Economic Journal 110(462), C158–175.

Rivers, D. and Vuong, Q. H.: 1988, Limited information estimators and exogeneity tests for simultaneous probit models, Journal of Econometrics 39(3), 347–366.

Rodrik, D.: 1996, Labor standards in international trade, in R. Z. Lawrence, D. Rodrik and J. Whalley (eds), Emerging Agenda for Global Trade: High Stakes for Developing Countries, Overseas Development Council, Wash- ington D.C., pp. 37–80.

Rogers, C. A. and Swinnerton, K. A.: 2001, Inequality, productivity, and child labor: Theory and evidence, Working paper, Georgetown University, Georgetown University Department of Economics Washington, DC. BIBLIOGRAPHY 161

Roodman, D.: 2003, XTABOND2: Stata module to extend xtabond dynamic panel data estimator, Statistical Software Components S435901, Boston College Department of Economics, Boston.

Rose, E.: 1999, Consumption smoothing and excess female mortality in rural India, Review of Economics and Statistics 81(1), 41–49.

Rosenzweig, M. R.: 1981, Household and non-household activities of youths: Issues of modeling, data, and estimation strategies, in G. Rodgers and G. Standing (eds), Child Work, Poverty and Underdevelopment, ILO Press, Geneva.

Rosenzweig, M. R. and Schultz, T. P.: 1982, Market opportunities, genetic endowments, and intrafamily resource distribution: Child survival in rural India, American Economic Review 72(4), 803–15.

Rybczynski, T. M.: 1955, Factor endowment and relative commodity prices, Economica 22(88), 336–341.

Sachs, J. D. and Warner, A.: 1995, Economic reform and the process of global integration, Brookings Papers on Economic Activity 26(1995–1), 1–118.

Satz, D.: 2003, Child labor: A normative perspective, World Bank Economic Review 17(2), 297–309.

Schultz, T. P.: 2004, School subsidies for the poor: Evaluating the Mex- ican Progresa poverty program, Journal of Development Economics 74(1), 199–250.

Sen, A. K.: 1992, Missing women, British Medical Journal 304(6827), 1–17.

Shelburne, R. C.: 2001, An explanation of the international variation in the prevalence of child labour, The World Economy 24(3), 359–378.

Sitalaksmi, S., Ismalina, P., Fitrady, A. and Robertson, R.: 2009, Globaliza- tion and working conditions: Evidence from Indonesia, in R. Robertson, D. Brown, G. Pierre and L. Sanchez-Puerta (eds), Globalization, Wages, and the Quality of Jobs: Five Country Studies, World Bank Publications, Washington D.C., pp. 203–236.

Sparrow, R.: 2007, Protecting education for the poor in times of crisis: An evaluation of a scholarship programme in Indonesia, Oxford Bulletin of Economics and Statistics 69(1), 99–122.

Stolper, W. F. and Samuelson, P. A.: 1941, Protection and real wages, Review of Economic Studies 9(1), 58–73. BIBLIOGRAPHY 162

Suryahadi, A.: 2003, International economic integration and labor markets: The case of Indonesia, in R. Hasan and D. Mitra (eds), The Impact of Trade on Labor: Issues, Perspectives, and Experiences from Developing Asia, Elsevier Science, Amsterdam.

Suryahadi, A., Priyambada, A. and Sumarto, S.: 2005, Poverty, school and work: Children during the economic crisis in Indonesia, Development and Change 36(2), 351–373.

Suryahadi, A., Sumarto, S. and Pritchett, L.: 2003, Evolution of poverty during the crisis in Indonesia, Asian Economic Journal 17(3), 221–241.

Suryahadi, A., Suryadarma, D. and Sumarto, S.: 2009, The eﬀects of location and sectoral components of economic growth on poverty: Evidence from Indonesia, Journal of Development Economics 89(1), 109–117.

Swaminathan, M.: 1998, Economic growth and the persistence of child labor: Evidence from an Indian city, World Development 26(8), 1513–1528.

Swinnerton, K. A. and Rogers, C. A.: 1999, The economics of child labor: Comment, American Economic Review 89(5), 1382–1385.

Thomas, D., Beegle, K., Frankenberg, E., Sikoki, B., Strauss, J. and Teruel, G.: 2004, Education in a crisis, Journal of Development Economics 74(1), 53– 85.

Topalova, P.: 2005, Trade liberalization, poverty, and inequality: Evidence from Indian districts, NBER Working Papers 11614, National Bureau of Economic Research, Inc, Cambridge, Mass.

UN: 1989, Convention on the Rights of the Child, Online Document (Oﬃce of the United Nations High Commissioner on Human Rights). http://www2. ohchr.org/english/law/crc.htm, retrieved on October 30, 2009.

UNICEF: 1997, The State of the World’s Children, Oxford University Press for UNICEF, London.

UNICEF/ILO: 2004, Addressing child labour in the Bangladesh garment industry 1995–2001: A synthesis of UNICEF and ILO evaluation studies of the Bangladesh garment sector projects, UNICEF and ILO, New York and Geneva.

US Senate: 2007a, A bill to amend the Tariﬀ Act of 1930 to eliminate the consumptive demand exception relating to the importation of goods made with forced labor. (Introduced in Senate), Online Document, The Library of Congress (THOMAS). http://thomas.loc.gov/cgi-bin/query/z?c110: S.1157:, retrieved on October 1, 2009. BIBLIOGRAPHY 163

US Senate: 2007b, Child Labor Deterrence Act of 1999. (Introduced in Senate), Online Document, The Library of Congress (THOMAS). http://thomas. loc.gov/cgi-bin/query/z?c106:S.1551:, retrieved on October 1, 2009. USDOL: 2008, U.S. Department of Labor’s 2008 Findings on the Worst Forms of Child Labor, U.S. Department of Labor, Bureau of International Labor Affairs, Washington, D.C. Verhoogen, E.: 2008, Trade, quality upgrading, and wage inequality in the Mexican manufacturing sector, Quarterly Journal of Economics 123(2), 489–530. Wahba, J.: 2006, The influence of market wages and parental history on child labour and schooling in Egypt, Journal of Population Economics 19(4), 823–852. Weiner, M.: 1991, The Child and the State in India: Child Labor and Education Policy in Comparative Perspective, Princeton University Press, Oxford. Windmeijer, F.: 2005, A finite sample correction for the variance of linear efficient two-step GMM estimators, Journal of Econometrics 126(1), 25– 51. Wong, K.-Y.: 1995, International Trade in Goods and Factor Mobility, The MIT Press, Cambridge, Mass. Wooldridge, J. M.: 2002, Econometric Analysis of Cross Section and Panel Data, The MIT Press, Cambridge, Mass. World Bank: 2004, World Development Indicators, CD-ROM. World Bank: 2006, Making Indonesia Work for the Poor, World Bank Office Jakarta. World Bank: 2009, World Development Indicators Online, Online Database. http://go.worldbank.org/6HAYAHG8H0, retrieved on September 30, 2009. WTO: 1998, Trade Policy Review Indonesia, World Trade Organization, Geneva. WTO: 2001, Doha WTO Ministerial Declaration (WT/MIN(01)/DEC/1), On- line Document, The World Trade Organization. http://www.wto.org/ english/thewto_e/minist_e/min01_e/mindecl_e.htm, retrieved on Octo- ber 1, 2009. WTO: 2003, Trade Policy Review Indonesia, World Trade Organization, Geneva. Wydick, B.: 1999, The effect of microenterprise lending on child schooling in Guatemala, Economic Development and Cultural Change 47(4), 853–69.