dec04_Article 2 12/14/04 2:26 PM Page 1056

Journal of Economic Literature Vol. XLII (December 2004) pp. 1056–1093

Job Information Networks, Neighborhood Effects, and Inequality

1 YANNIS M. IOANNIDES and LINDA DATCHER LOURY

1. Introduction indicates, however, that access to information is heavily influenced by social structure and onsiderable interest has emerged that individuals use connections with others, recently in the economic literature C such as friends and social and professional about social interactions and the ways in acquaintances, to build and maintain infor- which social norms and structures condition mation networks. Albert Rees (1966) first individual behavior. The treatment of labor- drew attention to differences among workers market transactions as very different from in their use of the variety of available infor- trading in goods reflects the importance of mational outlets. In this context, formal idiosyncrasies due to social effects. sources of information include state and pri- One prominent example of where such vate employment agencies, newspaper idiosyncrasies play a prominent role is job- advertisements, union hiring halls, and market search. Search theory formally mod- school and college placement services. els frictions associated with job-seekers’ Informal sources include referrals from access to information about availability of employees and other employers, direct jobs of different types and about the condi- inquiries by job seekers, and indirect ones tions of employment (George Stigler 1961, through social connections. Since then a bur- 1962; Christopher Pissarides 2001). Until geoning literature in economics has devel- relatively recently, the job-search literature oped about the details of social interactions has focused on individuals making decisions that affect the job-search process. This liter- on a one-to-one basis. Everyday experience ature complements the more extensive soci- ological analysis of networks. One of the 1 Ioannides and Loury: Department of Economics, objectives of this article is to explore the roles Tufts University. This is an outgrowth of an invited paper presented at the American Economic Association social interactions and social norms play in Meetings, New York, January 5, 1999. We are grateful to the context of this new literature. There is an two referees and to the editor, John McMillan, and to important richness that the term “networks” Anna Hardman for generous comments and suggestions. We also thank Chris Pissarides for his insightful discussion connotes in sociology which to a considerable in New York and Giulio Zanella for useful suggestions. extent has entered economics as well. A sec- Ioannides acknowledges support from the National ond objective is to explain its salience within Science Foundation and the John D. and Catherine T. MacArthur Foundation. We thank Pamela Jia for expert both theoretical and empirical economics assistance. research.

1056 dec04_Article 2 12/14/04 2:26 PM Page 1057

Ioannides and Datcher Loury: Job Information Networks 1057

Section 2 of this paper attempts to organ- friends and relatives to search for jobs in ize what we have learned from the empirical the preceding four weeks (Thomas literature about how individuals go about Bradshaw 1973). The 1991 and 1992 CPS collecting information for the purpose of figures were higher at 23 percent (Steven finding jobs and how the outcomes are Bortnick and Michelle Ports 1992; and influenced by their social connections. The Ports 1993). Unemployed and employed conventional wisdom that can be garnered workers were equally likely to use friends from much of the empirical literature on job and relatives during periods of job search, information networks and neighborhood according to David Blau (1992) and Blau effects is organized into a number of broad and Philip Robins (1990). Our own compu- categories of seven stylized facts. Section 3 tations using PSID data for 1993,2 reported starts by reviewing the sociology literature in appendix table 1, show that 15.5 percent on job information networks. It then turns, of the unemployed and 8.5 percent of the in subsections 3.2 to 3.5, to models of employed check with friends and relatives. exogenous job information networks, ones This is in a sense surprising because econo- in which individuals obtain job-related mists tend to think of the U.S. economy as information through a given social structure being increasingly penetrated by markets, and, in subsection 3.6, to the consequences and yet at the same time reliance on friends for job information networks of the recent suggests persistence of personalized literature on evolutionary models of infor- exchange. mation transmission. Section 4 reviews The second stylized fact about job infor- models of endogenous information net- mation networks is that the use of friends works that result from individuals’ uncoor- and relatives to search for jobs often varies dinated action, starting in subsection 4.1 by location and by demographic character- with the recent literature on strategic net- istics. The 1971 Current Population work formation. We then examine, in sub- Surveys,3 analyzed by Bradshaw (1973), section 4.2, endogenous job information showed that unemployed women were less networks. Section 5 sketches the outline of likely to have checked with friends or rela- a model that integrates job information net- tives to find jobs in the preceding four works and the dynamics of human-capital weeks (12.5 percent) than were men (17.4 formation and thus provides an overarching percent). The same rough difference pre- theme for the purpose of examining earned vailed in 1992: 20.0 percent for women and income inequality. Section 6 summarizes 26.6 percent for men (Ports 1993). See also suggestions for future research and section J. E. Rosenbaum et al. (1999) and Sandra 7 concludes.

2 This is the same data set as used by Corcoran, Datcher, 2. Stylized Facts about Job Information and Duncan (1980). The job-search categories are those Networks and Neighborhood Effects employed by the Current Population Survey (Peter Kuhn and Mikal Skuterud 2000). Unlike the European data (see The first generation of empirical work on Michelle Pellizzari 2004a), they are not mutually exclusive. Unfortunately, in 1993, relevant questions were not asked job information networks established sever- of all respondents in the PSID, but of only those actively al stylized facts about such networks. The engaged in job search, who were either unemployed, (5.8 first stylized fact is that there is widespread percent of the sample), or on-the-job searchers (8.1 percent of the sample). use of friends, relatives, and acquaintances 3 Calculations based on data from surveys of individuals to search for jobs and this has increased are likely to understate the importance of referrals, over time. About 15 percent of unemployed because they consider only one side of the job market. In fact, the numbers discussed above are affected by adverse workers interviewed in the 1970 and 1971 selection, as employers are likely to receive referrals from monthly Current Population Surveys used current employees. dec04_Article 2 12/14/04 2:26 PM Page 1058

1058 Journal of Economic Literature, Vol. XLII (December 2004)

Smith (2000).4 More educated job-seekers searching through friends or relatives (23.9 were also less likely to use friends and rela- for whites and 21.5 for blacks in 1992; Ports tives. From appendix table 1, 20.8 percent 1993). Hispanics, however, used such meth- of the unemployed job-searchers who check ods more extensively at 32.8 percent. with friends and relatives have more than Methods of job search also vary by location. twelve years of schooling, and 23.3 percent James Elliott (1999) looks at workers with no had at most eight years. The respective per- more than twelve years of schooling and finds centages for those searching on the job are that those in high-poverty neighborhoods 44.8 and 11.6 percent. Clearly, these partic- were substantially more likely (88 percent) to ular patterns are unlikely to be two sides of use informal job-search methods than those the same phenomenon. Indeed, different from low-poverty neighborhoods (74 per- forces may be affecting workers with more cent). More generally, our own counts with schooling. the PSID data show that use of informal con- Differences in using informal contacts by tacts increases with the size of the largest city age, race, and ethnicity show conflicting pat- in the county where the household resides. terns. According to Ports (1993), about 18 While 40 percent of the respondents live in percent of 16–19 year-old job-seekers and 22 SMSAs with a largest city of at least 100,000 percent of 20–24 year-olds checked with inhabitants, 65 percent of unemployed job- friends or relatives, compared to about 26.5 searchers and 45 percent of employed job- percent of 45–55 year-olds and 55–64 year- searchers who used friends and relatives olds in 1992. On the other hand, others resided in such areas (appendix table 2). (Mary Corcoran, Linda Datcher, and Greg The third stylized fact about job informa- Duncan 1980; Peter Marsden and Karen tion networks is that job search through Campbell 1990, and Marsden and Jeanne friends and relatives is generally productive. Hurlbert 1988) reported that use of informal Both employed and unemployed workers contacts declines with age and/or work expe- who used friends to search for jobs received rience. Harry Holzer (1987a) found only more offers per contact and accepted more small racial gaps for younger workers. In his offers per contact than did workers who data, roughly 69 percent of white job-seek- used other sources of information about job ers ages 16 to 23 used friends and relatives to openings (Blau and Robins 1990). This could look for jobs in the previous year (1982) explain first why about half of all workers compared to 67 percent of black male job- heard about their current job through a seekers.5 More generally, across all age friend or relative (Corcoran et al. 1980). groups, there were small racial differences in Summarizing the results of 24 studies, Truman Bewley (1999) estimated that 30 to 4 The gender differences may, however, vary according 60 percent of jobs were found through to the type of contact. Marmaros and Sacerdote (2002) friends or relatives. report results on the effects of peer and social network on job search using a sample of Dartmouth College seniors. Using friends and family may be produc- Individuals who were randomly assigned as roommates tive, not only in finding jobs, but also in when freshmen were asked how they use social network- improving the quality of the match between ing in their job search later on when they reached their senior year. Women were less likely to get fraternity/soror- firms and workers. Those who found jobs ity help, equally likely to get help from relatives, and more through personal contacts were generally likely to use help from professors. less likely to quit (Datcher 1983, and 5 As in the case of gender difference, racial disparity in using informal sources may depend on the type of contact. Theresa Devine and Nicholas Kiefer 1991) Marmaros and Sacerdote (2002) report that whites were and had longer tenure on their jobs (Curtis more likely to report that fraternity/sorority members, rel- Simon and John Warner 1992). On the other atives, and professors were influential in helping them find a full-time job or career. Racial differences are especially hand, however, the estimated effects of job large for the last two categories. contacts on wages vary considerably across dec04_Article 2 12/14/04 2:26 PM Page 1059

Ioannides and Datcher Loury: Job Information Networks 1059

studies. Using their Dartmouth College data, The fifth stylized fact is that many differ- David Marmaros (2001) and Marmaros and ences in productivity of job search by age, Bruce Sacerdote (2002) found large positive gender, race, and ethnic group cannot be correlations between getting help from fra- completely accounted for by differences in ternity/sorority contacts and obtaining pres- usage. Consider employment effects first. tigious, high-paying jobs. Rosenbaum et al. According to Bortnick and Ports (1992), (1991) reported that workers with contacts men who were unemployed in a given enjoyed a wage advantage that increased month in 1991 and who used informal con- with age. In contrast, other work has found tacts in that month were slightly more likely that an initial wage advantage declined over than their female counterparts to have found time (Corcoran et al. 1980; Doug Staiger jobs (24 compared to 21 percent). The dif- 1990; Simon and Warner 1992) or found no ferences between blacks and whites who general initial or persistent earnings effects used informal contacts were more substan- (William Bridges and Wayne Willemez 1986; tial: 15 percent of blacks who were unem- Holzer 1987b; Marsden and Hulbert 1988). ployed in a given month in 1991 and who Adding to the spectrum of results, Loury used informal contacts in that month found (2003) and Elliott (1999) showed that at least jobs, compared to 24 percent of whites. some types of job contacts were correlated According to Sanders Korenman and Susan with lower wages. Turner (1996) (see also Rosenbaum et al. The fourth stylized fact about job infor- 1999), employed young black inner-city men mation networks is that part of the variation were less likely to have found their jobs in the productivity of job search by demo- through friends and relatives, even though graphic group simply reflects differences in other analysts report few racial differences usage. As indicated above, women were less in the incidence of use of friends and rela- likely to use friends and relatives during job tives by job seekers. More specifically, search. This could explain why, according to Holzer (1987a) showed that, in 1981, 25 per- Corcoran et al. (1980), 52 percent of white cent of previously unemployed African- men and 47 percent of all women who were Americans ages 16–23, compared to 32 household heads or wives ages 45 and under percent for similar whites, obtained jobs found out about their current job from a through friends and relatives. Most of this friend or relative (see also Smith 2000). discrepancy was due to differences in the Hispanic men report more frequent use of likelihood of receiving offers from jobs friends and relatives for job search than non- heard about through friends and relatives. In Hispanic whites, and are also significantly fact, almost one-fifth of the total difference more likely to have found out about their in probability of gaining employment most recent job through personal contacts between black and white youth resulted (Smith 2000). Looking at differences as from racial differences in this probability neighborhood income varies, Elliott (1999) (see also Smith 2000). showed that less-well educated workers in In addition to variation in the relationship high-poverty neighborhoods were more like- between informal contacts and employment ly to use informal contacts and that these across groups, there are many demographic contacts were also the main avenue by which differences in the effects on wages of job these individuals found work. About 73 per- search through contacts. Korenman and cent of jobs in neighborhoods with poverty Turner (1996) reported that, among young rates of 40 percent of more were found workers in inner-city Boston, whites who through informal means, compared to 52 found jobs through contacts received much percent of jobs in neighborhoods with larger wage gains, 19 percent higher, than poverty rates less than 20 percent. blacks with similar characteristics. Smith dec04_Article 2 12/14/04 2:26 PM Page 1060

1060 Journal of Economic Literature, Vol. XLII (December 2004)

(2000) showed that gender wage differences new technology. We argue in section 6 that were small for those using formal job-search this is an important issue for future research. methods. In contrast, she found larger wage The seventh stylized fact is also new but differences between Hispanics and whites arguably more robust than the sixth. There who used personal contacts to find jobs com- appear to be important differences across pared to those who used more formal means. countries in the use of personal contacts by Korenman and Turner (1996) replicated the both firms and workers. Pellizzari (2004a) results on Hispanics for a nationally repre- explores the empirical evidence for the sentative sample of urban youth. Elliott countries of the European Union as of 2003 (1999) reported that for less-educated work- (with the exception of Sweden)6 using the ers, the use of informal contacts results in European Community Household Panel and significantly lower wages. compares with the United States using the The sixth stylized fact is new and needs to National Longitudinal Survey of Youth be treated as tentative. The internet is being (NLSY). Pellizzari finds large cross-country used increasingly for the purpose of job and cross-industry variation in the wage dif- search, at least according to anecdotal evi- ferentials between jobs found through for- dence. This area is little explored because mal and informal methods. Across countries data are only just starting to become avail- and industries, premiums and penalties are able. According to Kuhn and Skuterud equally frequent. Pellizzari attributes these (2000), who use data from a special supple- differences to different recruitment strate- ment to the December 1998 Current gies by firms. However, such differences Population Survey, which asked respondents could be attributable to both different insti- about computer and internet use, 13 percent tutional and social practices which may com- of unemployed Americans and 7 percent of pound the impact of differences in industrial employed Americans looked for a new job compositions of economies. via the internet. However, there appears to Taken together, these stylized facts imply be a “digital divide” for the unemployed: that the role of information networks in the only 7 percent of unemployed Hispanic job- job search process is not straightforward. seekers looked for jobs online in December Neither is it always clear a priori why some 1998, compared with 9 percent of blacks and groups rely more on informal methods than more than 16 percent of whites. For those others, nor why the pattern of employment employed, the respective figures are 4, 6, and earnings payoffs to networks varies and 7 percent. The gender divide is not across groups. Moreover, recent research nearly as stark. Unemployed women used indicates that these stylized facts do not the internet for job search at the same rate exhaust the range of effects of job-search as men. Among employed women, 6.7 per- networks. For example, what effects do con- cent looked for jobs on the internet in tacts have on wage and employment inequal- December as opposed to 7.6 percent of ity, the duration of unemployment, and employed men. However, comparison of the labor-market withdrawal? trends in use of traditional methods of job More generally, a common problem search (ibid, table 8, p. 10) suggests that use underlies much of the literature on job con- of public employment agencies has declined tact use and the correlation of job contacts from 1994 to 1999, although it is overrepre- with labor-market outcomes. It often sented among internet job-seekers in (though not always) fails to be well-grounded December 1998 (op. cit., table 7, p. 9). While we will not pursue this angle further 6 The exception of Sweden is unfortunate because in this paper, we underscore that models of including it might have provided an important benchmark: job search will need to accommodate this all job openings are centrally registered in Sweden. dec04_Article 2 12/14/04 2:26 PM Page 1061

Ioannides and Datcher Loury: Job Information Networks 1061

in economic or other theory about how net- That is, do white workers earn more with or works form or under what circumstances without using contacts due to higher levels networks are likely to have their largest of unobserved productivity? Does the corre- effects. In filling this vacuum, much of the lation arise simply because contacts are more recent detailed research into job net- more likely to pass on job information to works (discussed below) points to four such workers? Alternatively, do whites gain important considerations—employer, rela- more from contacts than blacks because tional, contact, and worker heterogeneity. their contacts have access to more informa- Employer characteristics determine the tion about better-quality jobs? As the context in which job search methods oper- remainder of this article indicates, all four of ate. For some employers, desired applicant these elements—employer, relational, con- characteristics may be easily discernible, tact, and worker heterogeneity—are critical while for other employers recommenda- to understanding this myriad of commonly tions from trusted sources may provide bet- observed findings as well as to extending the ter information. Contact and relational role of job networks to account for other heterogeneity respectively denote varia- labor-market outcomes. tions in the resource endowments of one’s associates and the social relationships that 3. Job Information Networks allow individuals to claim access to resources possessed by their associates. One of the objectives of this essay is to Worker heterogeneity refers to differences actually identify and explore mutual influ- in worker productivity or other characteris- ences and patterns of interplay between the tics. It interacts with all three of the other literatures on social networks in economics areas in determining access to contacts and and sociology, with particular emphasis on employers. social networks that have consequences for Without the appropriate theoretical the job market. In sociology, the concept of grounding that pinpoints the role of these a network and use of network models is stan- four considerations, it is difficult, for exam- dard (Ronald Burt 1980). Usage of the term ple, to interpret demographic variation in networks is perhaps as ubiquitous as that of contacts use. Does lower contact use by markets in economics and is used in a com- women constitute a problem of access to the parably broad range of contexts. In econom- best job opportunities, or does it reflect dif- ics, one the other hand, network refers to ferences in the ease of observing the types of “personalized exchange among many agents” skills with which many women enter the (Samuel Kortum 2003). The central impor- labor market? Does higher contact use by tance of networks to sociology has forced it Hispanics than blacks signal high-quality to define economic networks more precisely contacts and, therefore, greater returns to as: 1) particular patterns in economic informal compared to formal methods for exchanges; 2) indications of “primordial” Hispanics? On the other hand, does it imply relationships among agents, and 3) “struc- an absence of job information alternatives tures of mutual orientation” (Ezra for Hispanics who are forced to rely on Zuckerman 2003). The third definition informal sources as, in some cases, the only encompasses the previous two as special means of finding jobs? Interpreting differ- cases: it allows for social ties that differ in ences in effects of contacts on labor-market terms of “type, valence and strength of con- outcomes generates similar ambiguity. Is the nection” (ibid). Networks in sociology are larger positive correlation between wages not just an important descriptive device but and using contacts for whites compared to also a tool to explore direction of causality blacks reported by some analysts spurious? and the role of unobserved heterogeneity. dec04_Article 2 12/14/04 2:26 PM Page 1062

1062 Journal of Economic Literature, Vol. XLII (December 2004)

In contrast to the extensive treatment of empirical findings in support of the role of networks by sociologists, economists’ formal weak contacts are widely cited in both the as well as qualitative understanding of the economics and sociology literature. The sec- role of networks is much less developed. ond edition (1995) of this book contains a There has also been influence in the other review of the evidence since the 1974 study. direction, however, from economics to sociol- Granovetter argues that the voluminous evi- ogy. In at least one instance, articulation of dence that has accumulated since 1974 con- patterns in interpersonal interactions rele- tinues to support his book’s original thesis vant to economic outcomes by economists that social networks are very important in has influenced sociological methodology. helping people to find jobs and employers to This is the case with the concept of social find prospective employees. In particular, capital, which is particularly relevant in our local contacts, “weak ties,” help women “in context. Social capital was originally defined women’s occupations” locate jobs, and that by Glenn Loury (1977) as the set of resources “networking is crucial for expanding resulting from family relationships and com- women’s opportunities in male-dominated munity social organization that affect the cog- occupations” (ibid, p. 170). He concludes nitive and social development of the young. that in spite of “an outpouring of research,” It has found much fertile ground in sociology. we still know little about the complex net- James Coleman (1990), who helped popular- work processes by which “inequities are ize the concept, treats social capital as a form produced and reproduced.” of social organization created when the struc- This paper is focused on job information ture of relations among persons facilitates networks as patterns of exchange of job-relat- action making possible the achievement of ed information. We look at these patterns certain ends that in its absence would not be either as given (“primordial”), or as the out- possible (ibid, p. 300). The term social capital come of deliberate decisions by individuals to is now used quite commonly to refer to the help us untangle the complex and seemingly web of connections individuals use in daily contradictory findings about use of job infor- life and is therefore important to mention mation networks in the labor-market context here.7 Some economists use the terms social and their effects on employment and wages. capital and social networks almost inter- The paper shows how network considera- changeably, which for reasons of clarity we do tions imply different outcomes than the sim- not adopt in the remainder of this essay. ple one-to-one search models that typify Another instance of interplay between most economic analyses of job acquisition. economics and sociology is found in the 3.1 The Sociology Literature on Job most recent economics literature on job Information Networks information networks and reflects the influence of the empirical findings of Mark The sociological literature includes a rich Granovetter (1974:1995). Granovetter’s array of analysis of three of the categories of job network effects—employer, contact, and relational heterogeneity. Much of the 7 The notion of social capital has been used extensively to account for a wide variety of outcomes (Partha initial research to explain the operation of Dasgupta and Ismail Serageldin 2000; Robert Putnam job networks focused on relational hetero- 2000; Edward Glaeser et al. 2000). The two key elements geneity. Granovetter (1974:1995) argued of social capital include the resource endowments of one’s associates and the “social relationship itself that allows that a key characteristic determining the individuals to claim access to resources possessed by their effect of job networks on finding employ- associates” (Alejandro Portes 1998; Glaeser et al. 2000; ment is the strength of social ties. Roughly Putnam 2000). Additional evidence of the popularity of the concept is given by Markus Möbius (2001), who uses the speaking, strong links join close friends and terms social capital and social networks interchangeably. weak links join acquaintances. Strong links dec04_Article 2 12/14/04 2:26 PM Page 1063

Ioannides and Datcher Loury: Job Information Networks 1063

tend to traverse a society “slowly.” If you nomics literature on exogenous job informa- start with an arbitrary person and develop tion networks and on neighborhood effects the network of links to her close friends, and that have consequences for job markets. then to the close friends of her close friends The notion of weak versus strong ties is, of and continue in this manner, then the over- course, one of the most important concepts all size of the group grows slowly. The close that have motivated the literature on the friends of my friends are likely to be my effects of social contacts on job outcomes. close friends too. Societies of closely knit Burt (1992) argues that, while tie strength is relations are likely to develop a large num- correlated with information benefits, the ber of closely knit groups. If, on the other true causal agent in social networks is the hand, we track my acquaintances and their structural hole. Burt defines a structural acquaintances in turn, it is less likely that hole as the “gap,” the separation, between the acquaintances of my acquaintances are nonredundant contacts. Regardless of tie also my own acquaintances. As sociological strength, individuals located at structural research has measured (Stanley Milgram holes provide a bridge for information to 1967) and random graph theory has demon- flow between groups that would otherwise strated (E. M. Palmer 1985; Mark Newman not have access to each other. For example, 2003) any two individuals in the United suppose group A consists of individuals States can be connected by as few as six who often contact each other and would be weak links. So, the overall size of the group considered to have strong ties in the of interconnected individuals will grow Granovetter framework. The same is true for faster with weak ties. Sheridan Dodd, Roby the members of group B. Group A is linked Muhamad, and Duncan Watts (2003) show, to group B only through the individual at a based on a recent global internet-based structural hole in group A. This link allows social search experiment, that successful information to flow between otherwise inac- social search is conducted primarily through cessible groups. Note, however, that the intermediate-to-weak strength ties, relies relationship between the group A individual disproportionately on professional relation- at a structural hole and his nonredundant ships, and reaches its destination in a medi- contact in group B may be strong or weak. an five-to-seven steps, depending on the The key factor is not the strength of the tie actual distance between source and target. but that the individual occupying the struc- Granovetter’s original work and the 1995 tural hole bridges the gap between groups A edition of the 1974 study have been read and B. The structural holes argument8 very widely. Within sociology the notion of implies that: 1) a network with more nonre- tie strength has been incorporated into dundant contacts can provide more informa- more general analyses of job networks and tion than the same size network with social capital. Yet, Granovetter’s work has redundant contacts; and, 2) a network with a been much less influential in promoting given number of nonredundant contacts modelling of job information networks with- provides more information if, in turn, those in economics. This is not so surprising: econ- contacts “reach separate and therefore more omists are typically more likely to be diverse social worlds” (ibid, p. 21 ). interested in why a network forms, whereas 8 sociologists take the existence of networks as As an aside, to economists this argument is akin to an arbitrage-type argument and thus amenable, in principle, given and study their effects. In the first to analytical treatment. A very recent paper by Sanjeev subsection below, we explore our under- Goyal and Fernando Vega-Redondo (2004) applies the standing of job information networks from strategic network formation approach to the concept of a structural hole. A noteworthy result here is, as we discuss the interplay between the sociology and eco- further below, that if the cost for forming links is not very nomics literatures. Then we turn to the eco- small, the unique equilibrium network is a star. dec04_Article 2 12/14/04 2:26 PM Page 1064

1064 Journal of Economic Literature, Vol. XLII (December 2004)

Burt (2001) also argues that the nonre- theoretic model of networks of relational dundant information that structural holes contracts. They examine networks of rela- possess is better communicated to, and acted tions under different informational regimes, upon, in networks with closure. He defines paying special attention to differences closure as “social capital” that is created by a between circular and noncircular architec- network of strongly interconnected ele- tures. If agents cannot discipline themselves ments. With high levels of closure, everyone within a certain relation and are allowed to is cohesively connected to everyone else have relations with only two other agents within the network. Information passed to a (“neighbors”), a “circular pooling of asym- subgroup within the network quickly finds its metries” made possible by a social network way to the remainder of the network. When may end up sustaining all the relationships there is closure, referred workers may have in equilibrium. higher job productivity because their reputa- There are properties of social structures tion within the network is more contingent that have been attributed to closure that may on their performance. instead be explained by symmetry in the net- Why does closure, whereby agents are work that represents social structure, rather effectively interconnected in a cyclical fash- than closure.9 For example, this is the case ion (Coleman 1990), and not just social con- when individuals who act similarly are more nectedness play such a key role in likely to be connected with one another than sociological thinking? Here is a guess. with others who act differently. Such patterns Consider a social setting where people tend of correlation in the behavior of connected to imitate one another, as in the spread of a individuals resemble spatial waves and exhib- fashion. When everyone is directly connect- it clustering. Clearly, this is an area where ed with everyone else, different individuals additional research would be very fruitful. starting fads are unlikely to have decisive Other sociological research contends that influence on each other. Alternatively, con- the role of relational heterogeneity cannot sider individuals connected along a path— be fully understood without taking into where each individual is connected to two account the role of social setting or contact others and there are two end agents—in heterogeneity. Nan Lin’s (2001) “strength of which case everyone is at least indirectly position” proposition argues that individuals connected with everyone else. Such an are more likely to associate with others in arrangement confers a modicum of influ- similar social and occupational positions; in ence to the two end agents, whose own other words, they sort. This proposition actions are influenced by one other agent implies that social networks develop along only. If those two end individuals were to be dimensions such as race, ethnicity, religious connected so that the topology of the social affiliation, and education. The emphasis in structure becomes a wheel, there is no Lin’s work is on the characteristics of the longer any influence to be conferred by the contacts themselves. Lin’s “social capital” topology of the social structure as such, and proposition claims that contacts who possess, this is accomplished with the minimum total or have access to, more highly valued number of connections. In spite of its popularity, the sociological 9 Ioannides (2001) shows that symmetry—technically concept of closure is to the best of our speaking, all individuals’ being connected to the same number of other individuals—ensures that the maximal knowledge not very rigorously developed. eigenvalue of the corresponding graph is equal to the An exception is Steffen Lippert and number of neighbors. Therefore, if individuals are influ- Giancarlo Spagnolo (2004), who provide rig- enced by the average action among those they are con- nected with, then the respective dynamical system has an orous support for the disciplining role of eigenvalue of 1 and an associated eigenvector that consists closure in sociology by means of a game- of 1’s, which ensures (relative) persistence. dec04_Article 2 12/14/04 2:26 PM Page 1065

Ioannides and Datcher Loury: Job Information Networks 1065

resources improve outcomes for job-seekers more information about the nonpecuniary more than other less well-placed contacts. aspects of employment and, therefore, are Lin, Vaughn, and Ensel (1981) and Lin, potentially better matches. Finally, connec- Ensel, and Vaughn (1981) report that the tions between new hires and incumbent family background, education, and early employees can make the job transition occupational status of job-seekers all influ- smoother, as well as create additional loyal- ence the occupational status of the contacts ties and attachments to the job. More gen- they can use. Furthermore, the occupational erally, Peter Marsden and Elizabeth status of these contacts alters the prestige Gorman (2001) argue that information pro- that job-seekers obtain. In related work, F. vided to employers through referrals may Carson Mencken and Idee Winfield (2000) reduce employer uncertainty about the emphasize that women who found their jobs prospective worker’s productivity. through informal contacts and who used The effects of job contact information are, male contacts were less likely to work in therefore, likely to be higher in firms and female-dominated occupations. Echoing industries where high-quality information Mencken and Winfield, John Beggs and about workers’ likely performance is impor- Hurlbert (1997) indicate that the gender of tant. These include conditions “when per- the informal contact tie affects occupational formance and skills are difficult to observe, status. Women whose contacts are other when staffing strategy is flexible, when the women work in occupations with lower use of networks is a central component of socioeconomic index scores. Evidence of cor- performance, and when selection errors are relation between contact characteristics and costly (p. 108).” Consistent with this predic- job contact effects is not, however, uniform. tion, Marsden and Gorman present evidence While Marsden and Hurlbert agree with Lin, that contacts are more likely to be used when Vaughn, and Ensel (1981) and with Lin, filling managerial, professional, or sale/serv- Ensel, and Vaughn (1981) with respect to ice positions and are less likely to be used in occupational prestige, they conclude that the public sector and in establishments that “the net effects of the social resource vari- are part of multi-site firms. Complementary ables (for other outcomes such as wages) can work by Elliott (1999) found differences in be summarized simply: there are none.” They informal contacts by occupation. General conclude that these findings are consistent laborers (defined as all nontechnical posi- with those of Bridges and Villemez (1986), tions lacking managerial authority, 1990 who find no effects of tie strength on income. Census Occupation Codes 243-902) and The third branch of sociological research those in managerial positions who found examines the role of employer heterogene- their jobs through informal contacts had sig- ity on contact effects. It points to important nificantly lower wages than those who found differences in the use of referrals in differ- their jobs without an active search. ent industries, perhaps reflecting differ- Intrafirm analyses of job contact effects ences in corporate culture. The sociology also point to the importance of context. research in this area appears to be ahead of Trond Petersen, Ishak Saporta, and Marc- its counterpart in economics in assessing the David Seidel (2000) use data on all 35,229 consequences of employer heterogeneity. job applicants to a mid-sized high-technolo- According to Roberto Fernandez and gy organization and find that all differences Emilio Castilla (2001), employers reap in job offers that are attributed to gender returns from referrals for three reasons. disappear once age and education are Referrals provide a large pool of qualified accounted for. Similarly, all effects of race applicants so that less screening is required disappear once the referral method is taken to fill positions. Referred applicants have into account. That is, when one controls only dec04_Article 2 12/14/04 2:26 PM Page 1066

1066 Journal of Economic Literature, Vol. XLII (December 2004)

for age, education, and rating at first inter- and small-scale retail trade, and Dominicans view, race has a strong impact on the likeli- in garments, restaurants, hotels, and a few hood of having a second interview and on other light manufacturing and retail trades. the increase in the salary offer. In their data, These ethnic-group niches have important personal and professional contacts account implications for the usefulness of job con- for 60.4 percent of applicants and 80.8 of tacts and connections for ethnic-group those receiving offers, and there is also an members entering the labor market or effect of being contacted by headhunters. changing jobs. Consider, for example, the Thus, the apparent and perhaps real meri- large concentration of African-Americans in tocracy characterizing the high-technology public service. On the plus side, it reduces industry raises additional questions on the their experience of discrimination and thus role of access to job information networks as raises black earnings relative to whites. a force in persistent inequality. Resulting job contacts and connections may Using data from a high-technology firm, lead to a greater likelihood of job offers and Joel Podolny and James Baron (1997) find to more rapid career advancement upon that intra-organizational mobility is accepting the offers. On the minus side, the enhanced by having a large network of infor- large concentration of African-Americans in mal ties that supply access to information the public sector is mirrored by their declin- and resources. Yet, availability of a small ing presence in the private sector and thus is dense network of social contacts with high associated with lower access to contacts, net- closure and cohesiveness is no less important works, and training opportunities. Moreover, in helping shape one’s organizational identity if the public sector niches held by African- and career goal expectations. This highlights Americans require relatively high levels of the importance of how social network struc- skill, job information about these niches may ture and content interact in determining not be especially useful for low-skilled careers within organizations. African-Americans. Differences between industries and 3.2 Models of Exogenous Job Information employers may also account for ethnic and Networks race variations in contact effects. Roger Waldinger (1996) shows that, both historical- Models derived from research by econo- ly and currently, ethnic groups have estab- mists on exogenous job networks have out- lished specific occupational and employment lined the specific implications of social niches that facilitate employment and train- structure in more detail than some of the ing of members of their group and that limit sociological work described above. By exoge- access of outsiders. Ethnic newcomers to nous, we mean that the network of connec- New York found their way to the bottom of tions (or, in more mathematical language, the the job ladder associated with the niche and graph that describes these connections) then gradually work their way upward among individuals is given. Dale Mortensen through the specialized economic activities and Tara Vishwanath (1994) show that the associated with the niche. Early examples of equilibrium wage distribution increases with ethnic niches include Jews in commerce and the probability that the offer is from a con- clothing manufacture and Italians in labor- tact. Their argument is based on the premise ing jobs in construction and longshoreman that wages received from jobs found through work. More recently, occupational niches contacts reflect the distribution of wages include African-Americans in the public sec- earned by individuals who are in contact with tor, West Indians in hospitals, nursing one another. This distribution, in turn, sto- homes, and health services, Chinese in chastically dominates the distribution of restaurants, laundries, the garment trade, wage offers across employers. Like Bridges dec04_Article 2 12/14/04 2:26 PM Page 1067

Ioannides and Datcher Loury: Job Information Networks 1067

and Villemez (1986) referred to earlier, those who do not are hired through the for- James Montgomery (1992) examines the link mal market. Firms may choose technology between wages and tie strength and con- that makes either type fully productive, cludes that, while weak ties increase reserva- except that choice along with that of the tion wages, this does not imply a similar wage rate must be made before the worker’s “relationship between wages and the type of type is known. tie actually used to find a job.” Intuitively, The model is closed by equating the per- weak ties are more likely to generate offers centage of those educated with the percent- than strong ties. Workers who accept weak- age of those facing education costs who find tie offers are likely therefore to have received it advantageous to acquire information. The fewer total offers and be less selective in the possibility of multiple equilibria depends jobs that they choose than those who accept critically upon the properties of the distribu- strong-tie offers. This indicates that the tion of education costs across the population. empirical finding of no relationship between A key element in Montgomery’s theory is the tie strength and wages in the Bridges and derivation of the probability that an individ- Villemez study does not imply that tie ual with a referral accepts a job offer, as a strength is irrelevant for determining job function of the offer: a(w)e– n(1–F(w)), outcomes. Montgomery’s work is a notable where F(w) is the distribution function of example of research by an economist that has referral wage offers and n the steady-state helped bridge the gap between economics fraction of educated workers. In this model, and sociology in this area. a higher probability of a social tie and a A number of more general models exam- higher percentage of educated workers ine additional detail about the interaction decrease the probability of acceptance but between contact and relational heterogene- increase wage dispersion. While the former ity. They highlight the specific characteris- is a straightforward supply effect, the latter tics of relations between contacts and job is subtle. Increased inbreeding by a group is seekers that alter contact effects. An exam- shown to be associated with larger differ- ple of such a model is Montgomery (1991), ences from other groups. Selection operates who suggests that the main social compo- over time via network density parameters nent is inbreeding social bias. That is, each and inbreeding. Individuals pass on their person is more likely to have a social tie to advantages to kin and social acquaintances. a younger person of the same type as her- These factors work to perpetuate and self. Thus, a social tie implies that a referral strengthen inequality over time. Kenneth possesses informational value. In Arrow and Ron Borzekowski (2004) show by Montgomery’s model, each individual lives means of simulations that differences in the for two periods, making an education deci- number of ties workers have with firms can sion in the first period, which is observable, induce substantial inequality and can explain and working in the second. Individuals may roughly 15 percent of the unexplained varia- be of two types. Each individual knows at tion in wages and a substantial part of the most one person in the older (and currently disparity between black and white income employed) generation, possessing a social distributions. tie with probability . Conditional on hold- Kaivan Munshi (2002) studies transmission ing a social tie, a worker knows someone of of job information among Mexican migrants 1 the same type with probability , 2 . to the U.S. labor market. He uses a model of Some young persons may have several referrals similar to Montgomery’s and exam- social ties while others have none. Those ines changes in the size and vintage within who do have social ties receive offers from given destination communities in order to the employers of their acquaintances, but determine whether those characteristics dec04_Article 2 12/14/04 2:26 PM Page 1068

1068 Journal of Economic Literature, Vol. XLII (December 2004)

affect migrant employment. His findings con- employed with probability less than or equal firm that community-based social interactions to the average employment rate in the popu- are important in matching Mexican migrants lation. The combination of a rigid social and U.S. firms, and appear to improve labor- structure with inbreeding bias in employ- market outcomes among migrants, with ment status implies that random matching is smaller and younger networks substantially less useful to individuals and the employment reducing the employment probabilities of status of one’s dyad partner is critical. Using Mexican migrants. this model, Montgomery shows that a higher Montgomery (1994) models the impact of proportion of weak-tie interactions reduces social interaction on employment transitions employment inequality. It also increases the and inequality in a way that links the notion steady state employment rate, provided that of strong versus weak ties to the social struc- inbreeding by employment status among ture. In his model, social structure consists of weak ties is sufficiently small. a large number of small groups, or dyads, It would be interesting to generalize the groups of two connected individuals. In each model of social structure employed by dyad, both members are employed, or only Montgomery, by assuming groups of differ- one is employed, or both are unemployed. ent sizes. For example, one may invoke a The relationship between two members of a random graphs setting (Paul Erdös and dyad is a strong tie. Individuals interact with Alfred Renyi 1960; Ioannides 1997), where a others at a rate per unit of time, and such fraction of the entire economy may be in interaction may lead to a match with one’s groups whose sizes are denumerable but dyad partner with probability , or it may possibly large. However, unless membership lead to a random match with someone else to groups of different sizes confers particular from the entire population, with probability advantages of a permanent nature with con- 1. In the latter case, because of random sequences for economic inequality, such as matching, the individual has an infinitesimal permanent effects on the incidence of unem- probability of interacting again with the same ployment, a model with a more complicated person. For this reason, random matches are size distribution of social groups will not add construed as weak ties. By varying the proba- much to the story. The literature must also bility of interaction with one’s strong tie, one address the fact that the availability of differ- may, in effect, model different social settings. ent informational technologies might pro- Jobs break up randomly at a rate , so that vide incentives for strategic behavior by either a strong or weak tie may be employed individuals in possession of job-related infor- or unemployed. The job finding rate for each mation. Such strategic issues in information person depends on the employment status of dissemination are not well understood. This her contact and on her general ability to col- is particularly evident, as we see further lect information on job openings. Although below, when social networks are the outcome individuals who make up each dyad do not of individual uncoordinated decisions.11 change over time, their employment status 3.3 The Work of Calvó-Armegnol and does change and so does the employment sit- Jackson uation in each dyad.10 As in Montgomery’s earlier work, social interactions are character- Two recent path-breaking papers, Antoni ized by inbreeding bias, whereby an unem- Calvó-Armegnol and Matthew Jackson (2002; ployed individual’s random match contact is 11 In a world where individuals have established rela- 10 Calvó-Armegnol, Thierry Verdier, and Yves Zenou tions with one another, the resulting social structure may (2004) explore the concept of a dyad in the context of the display certain “holes.” Such structural holes offer entre- labor market and, in addition, introduce criminal activities preneurial opportunities for information access, timing as an option within a given social structure. referrals and control. See Burt (1992). dec04_Article 2 12/14/04 2:26 PM Page 1069

Ioannides and Datcher Loury: Job Information Networks 1069

2003), explore the implications of exogenous the number of employment opportunities information networks. Both papers use the the agent has in hand. This function allows following model of the transmission of job for substitution among previous wage status information among workers. A network of and current employment prospects. contacts among n individuals, the social struc- Turnover in the job market is ensured by ture, is defined by means of an nn matrix random breakup of jobs. The paper distin- G, of intensities of social attachments: gij 0, guishes the effect of social connectedness if i is linked to j; gij 0, if i is not linked to j. (one agent’s expected job offer is sensitive to This formulation combines the notion of an another agent’s past wage and vice versa), adjacency matrix in graphs with the notion of and the passing of job-related information varying intensities of social contacts and between agents. allows for the network to be directed. That is, Using this model, Calvó-Armegnol and if individual i hears of a job opening, she tells Jackson develop explanations for several individual j if gij 0 . However, unless gji 0 , important stylized facts about labor markets. individual j will not pass on such information First, information passed from employed to i. The transmission of job information to individuals to their unemployed acquain- each worker through the network at the tances makes it more likely that their beginning of each period results in new acquaintances will become employed. This employment. An unemployed individual who generates positive correlation between hears of a job opening keeps it to himself. An employment and wages of networked indi- employed individual who hears of a job open- viduals within and across periods. Second, ing passes it on to each of her social contacts duration dependence and persistence in with probabilities that are proportional to the unemployment may be explained by recog- respective relative weights. For example, an nizing that when an individual’s direct and individual i’s unemployed contact j will hear indirect social contacts are unemployed, the from i with probability equal to the product likelihood of obtaining information about of the probability that i hears directly of a job jobs through contacts is reduced. Such dura- opening, which is assumed to be a function of tion dependence is well-documented; see, all agents’ wages in the previous period, times for example, Devine and Kiefer (1991), Lisa the relative weight of j’s social strength, which Lynch (1989), and Gerard van der Berg and is equal to gij divided by the sum of the Jan van Ours (1996). Third, the likelihood of weights of all contacts who are unemployed. dropping out of the labor force is higher for The weights of social attachment express a an individual whose social contacts have continuous counterpart of Granovetter’s poor employment experience. This can lead notion of strong versus weak ties, and thus to substantial differences in drop-out rates generalize it. across groups. Moreover, small differences Calvó-Armegnol and Jackson (2002) in initial conditions of different individuals assume that the expected number of offers and in network structure can lead to large that agent i receives is a nondecreasing differences in drop-out rates. Fourth, higher function of the wages of that agent’s con- initial drop-out rates for a set of networked tacts in the previous period and a nonin- individuals imply that its short-run as well as creasing one in agent i’s own wage. These its steady state distribution of unemploy- assumptions imply a set of “altruistic” values ment and wages will be worse (in the sense about social exchange that influences the of first-order stochastic dominance). passing around of job-related information. Contagion effects then cause inequality in Calvó-Armegnol and Jackson determine an wages and employment, because those agent’s wage by assuming that it is a nonde- remaining in the labor force will have fewer creasing function of the past wage and of direct and indirect acquaintances on the job, dec04_Article 2 12/14/04 2:26 PM Page 1070

1070 Journal of Economic Literature, Vol. XLII (December 2004)

and that in turn will hamper their job infor- leave that state. For similar reasons, parts of mation prospects. It is hard to explain such the economy can experience a boom while an effect outside of a social network model. simultaneously other parts of the economy Particularly noteworthy are results that are experiencing a bust. Calvó-Armegnol and Jackson (2002) obtain Differences in initial conditions combine about the key role of drop-out rates. To do with differences in the collective employ- so, they work with a specific case of the ment histories and with different network model in Calvó-Armegnol and Jackson dynamics of two otherwise identical net- (2003). In it, all jobs are identical and they works to produce sustained inequality of offer equal wage rates. Only one agent wages and drop-out rates that feed on each receives information about a job opening other. So, in the model of the two Calvó- and only if she is unemployed. The paper Armegnol and Jackson papers, history mat- presents a number of examples, each with ters and is responsible for producing income different explicit structures, that help inequality for reasons that are very different demonstrate the subtle role of networks. from those due to inequalities in human-cap- For example, if three agents are connected ital investments (Loury 1981), or in access to according to a path, the two end agents are community-based opportunities (Steven competitors for job information in the short Durlauf 1996a,b). This implies that interven- run, but their outcomes are positively corre- tions in the labor market, such as providing lated in the long run. That is so because incentives for individuals not to drop out, are their presence helps the center agent return likely to have long-lasting effects. to employment if she becomes unemployed. The design of such interventions should With more complex social network topolo- reflect the topology of the network. It would gies, an agent’s likelihood of being unem- be more effective to target groups of agents ployed depends on her position within the who are highly connected, taking advantage network. The average unemployment rate of social attachment effects among agents, increases with “close-knitness,” because instead of targeting the same number of uni- more extensive social ties make possible formly connected individuals.12 Similarly, greater diversification of information institutions that seek to “network” otherwise sources. Calvó-Armegnol and Jackson prove isolated individuals can potentially bring that, under certain general conditions, about socially desirable outcomes.13 These employment status across any arbitrary results depend on essentially altruistic periods is correlated among all intercon- assumptions about social exchange in the nected agents and that there is duration context of social interactions, in ways that are dependence in unemployment. conceptually similar to spatial interactions as These authors explain duration depend- modelled by Thomas Schelling (1971, 1978). ence as a social effect: the longer an individ- 3.4 Proximity, Information Sharing, and ual is unemployed the more likely it is that Neighborhood Effects her “social environment is poor,” making future employment prospects unfavorable. Several recent economic studies empha- This explanation for duration dependence size network effects as neighborhood complements the more commonly stated ones, such as unobserved heterogeneity and 12 Unfortunately, the technical problem of finding the like. This effect, essentially a network which individuals should be targeted is computationally externality, is also responsible for stickiness very difficult in general (David Kempe, Jon Kleinberg, and in aggregate employment dynamics. The Éva Tardos 2003). 13 The importance of network effects on the drop-out closer the economy is to very high employ- rate is also argued by D. Lee Heavner and Lance Lochner ment (or unemployment), the harder it is to (2002), though the results are not as dramatic. dec04_Article 2 12/14/04 2:26 PM Page 1071

Ioannides and Datcher Loury: Job Information Networks 1071

effects: they examine whether it is appro- interactions is consistent with the Calvó- priate to associate geographic proximity Armegnol and Jackson (2003) explanation with facilitation of information flow. This of duration dependence in unemployment question, of course, is identical to the ques- discussed above. tion addressed by the empirical literature A conceptually related study by Peter on the economics of social interactions. Hedström, Ann-Sofie Kolm, and Yvonne Even if we have compelling evidence of Åberg (2003) emphasizes transitions out of correlations in the behavior of individuals unemployment, using the Pissarides model who are in physical and social proximity to (Pissarides 2000) with data on all 20- to 24- one another, we wish to know what explains year-olds living in Stockholm during the such correlations. That is, we wish to know 1990s. Both of those papers produce evi- whether we see correlations among such dence for partial identification of social individuals because they share the same interactions in unemployment. sources of information (a correlated Bruce Weinberg, Patricia Reagan, and effect), because they share individual char- Jeffrey Yankow (2000) provide some evi- acteristics as a result of self-selection (a dence of non-monotonic neighborhood contextual, or exogenous social, effect), or effects on labor-market outcomes. They link because they learn from one another’s confidential street address data from the behavior (an endogenous social effect) NLSY79 with measures of neighborhood (Charles Manski 2000). social characteristics at the census tract level Georgio Topa (2001) finds geographic for 1990 and measures of job proximity correlations in patterns of unemployment based on the 1987 censuses of manufactur- across neighborhoods and cites them as evi- ing, retail trade, and services. They show that dence of positive correlation between one standard deviation increase in neighbor- employment and wages of networked indi- hood social characteristics and in job proxim- viduals. He points out that high unemploy- ity raises individuals’ hours worked by 6 ment rates were concentrated in relatively percent and 4 percent in the average, respec- few areas of Chicago in 1980 and 1990. tively. Such social interactions have nonlin- Using census tracts as units of observation, ear effects. The greatest impact is in the he assumes that residents of adjacent tracts worst neighborhoods. Being in a disadvan- exchange job information. He finds that taged neighborhood is more important than high unemployment in one tract is associat- the labor activity of one’s neighbors per se. ed with more unemployment in neighbor- Social interaction effects are also larger for ing tracts than can be explained by the less-educated individuals and for Hispanics, characteristics of the neighboring tracts but not for blacks compared to whites. alone. Timothy Conley and Topa (2003a) While these works are in broad agree- find that socioeconomic characteristics (and ment, recent research by Philip Oreopoulos in particular ethnic and occupational dis- (2003) finds that when neighborhoods are tance) explain a substantial component of not selected, neighborhood quality plays lit- the spatial dependence in unemployment. tle role in determining a youth’s eventual Using similar data for the Los Angeles earnings, likelihood of unemployment, and Standard Metropolitan Statistical Area welfare participation. In contrast, family dif- (SMSA), Conley and Topa (2003b) show ferences, as measured by sibling outcome that local interactions perform well in correlations in a relatively homogeneous explaining the spatial correlation patterns sample of low-income families living in present. In addition, using data on the dis- Toronto public housing, account for up to 30 tribution of individual unemployment spells percent of the total variance in earnings. in-progress, they show that their model of These findings are particularly significant dec04_Article 2 12/14/04 2:26 PM Page 1072

1072 Journal of Economic Literature, Vol. XLII (December 2004)

because the respondents in that study had influence employment locational choices at been administratively assigned to different the block-group level. For example, this con- public-housing residences, and therefore the trols for features of the urban transportation assignment process should have removed network that might induce clustering in both much of the selection across neighborhood residence and work location. Also, worker types, according to the author. characteristics might be correlated with both The recent study that makes the strongest residential location preferences and work and most compelling case to date for the location, if firms sort along the same variables. effects of geographical proximity on job mar- Their empirical strategy addresses several ket outcomes is Patrick Bayer, Stephen Ross, additional potential pitfalls, including possible and Georgio Topa (2004). They document sorting below the block level and the possibil- that people who live close to each other, ity of reverse causation. After they estimate defined as living in the same census block, the social interactions effect they consider also tend to work together, defined as work- whether the quality of the matches available ing in the same census block: the baseline in an individual’s block affects employment, probability of working together is 0.93 per- labor-force participation, and wage outcomes. cent compared to 0.51 percent at the block- Bayer, Ross, and Topa use data from the group level (a collection of ten contiguous 1990 U.S. Census for the Boston metropoli- blocks). Their findings are robust to the intro- tan area for all households that responded to duction of individual controls in the form of a the long U.S. Census form, and choose indi- number of socio-demographic characteristics viduals who did not reside in the same and block-group fixed effects. More specifi- household, were U.S.-born and aged cally, these authors examine the hypothesis between 25 to 59, and employed at the time, that agents interact very locally with their thus ending up with 110,000 observations. social contacts, exchanging information about From these data, about four million observa- jobs. Let: i and j be individuals who reside in tions on matched pairs were constructed by the same census block group but not in the matching up individuals in pairs, in a city b same household; Wij a dummy variable that is with 2,565 block groups with an average of equal to one, if i and j work in the same cen- ten blocks each. b sus block; Rij is a dummy variable that is Bayer et al. find that social interactions are equal to one, if i and j reside in the same cen- stronger when a pair of individuals are more

sus block: Xij a vector of socio-demographic likely to interact because of education, age, characteristics for a matched pair (a concept and the presence of children14; interactions to be clarified shortly below) (i,j); and, g a are stronger when one of the two individuals residential block-group fixed effect which is strongly attached to the labor market, and serves as the baseline probability of an are weaker when both are drop-outs, young, employment match for individuals living in or married females. In terms of the magni- the same block group. Then their hypothesis tude of the impact of match quality, a one may be examined in terms of a regression: standard deviation increase in referral b b opportunities raises labor-force participation WX=+′ ++() ′ XR⋅+ε . (1) ij gijijij01 ij by one percentage point, weeks worked by These authors’ test for the presence of social about two thirds of one week, and earnings interactions due to proximity boils down to by about two percentage points. This study is testing for the statistical significance of the also significant for its reliance on different term ( ˆ 0 ˆ 1X ij). They include both the base- geographical scales for identification. line probability ˆg and matched pair’s covari- ates in levels, Xij, to control for any 14 Assortative matching of this type in social networks observed and unobserved factors that may has been documented by Peter Marsden (1987). dec04_Article 2 12/14/04 2:26 PM Page 1073

Ioannides and Datcher Loury: Job Information Networks 1073

Two other studies, which preceded Bayer group in terms of their own use of public et al., sought to identify “network effects” by prenatal-care programs. Such use is highly using concepts of proximity similar to those of correlated within groups defined using Bayer et al. Both seek to establish that indi- race/ethnicity and neighborhoods, and per- viduals in close proximity share information. sists even after accounting for unobserved Therefore, they are not only methodological- characteristics by including zip code-year ly similar to Bayer et al. but also relevant to fixed effects. However, the richness of their those interested in establishing information data (which are abstracted from birth certifi- sharing among individuals in close proximity cates and include more than 3.5 million to one another. Marianne Bertrand, Erzo observations) allows them to test whether Luttmer, and Sendhil Mullainathan (2000) such estimated effects represent informa- consider the impact of social networks on tion sharing within groups. This is accom- welfare participation. They emphasize meth- plished by including fixed effects for the ods that allow them to distinguish between hospital of delivery interacted with the year the effects of networks from those of unob- of delivery. The results on the estimated servable characteristics of individuals and of effects of “networks” are then either the communities in which they live. Like reduced or eliminated and thus cast doubt Bayer et al., they attempt to distinguish on the idea that the observed correlations between the effects of mere geographic prox- can be interpreted as evidence of informa- imity and of information transmission made tion sharing. They point instead to differ- possible by proximity. In particular, they rely ences in the behavior of the women involved on language spoken and examine whether and of the institutions serving different being surrounded by others who speak the groups of low-income women as the primary same language increases welfare use more explanation for group-level differences in for individuals who belong to high welfare- the take-up of this important public pro- using groups. Individuals interact more with gram. They examine the role of institutions others who speak the same language and are by comparing the behavior of foreign-born therefore more likely to be influenced by with that of native-born Hispanic women. other members of that group. They obtain They find that “network effects” are quite highly significant and positive coefficients on similar for both those groups of Hispanic the interaction between contact availability women, in contrast to their expectation that and mean welfare participation of one’s lan- foreign-born women will have greater infor- guage group and interpret these findings as mational requirements. They conclude, evidence of network effects. therefore, that it is differences in the behav- Anna Aizer and Janet Currie (2002) exam- ior of institutions and not information shar- ine “network effects” in the utilization of ing that explains the established correlations publicly funded prenatal care. These authors between neighborhood and ethnic group consider women as belonging to a network if membership in prenatal care use. they live in the same neighborhood (defined Rafael Lalive (2003) also examines social as the areas of five-digit zip code) and interactions among unemployed individuals. belong to the same racial or ethnic group. This empirical study exploits an unusual They use data on take-up of publicly funded quasi-experimental setting created by selec- prenatal care, which originate in vital statis- tive extension of unemployment benefits tics from California’s Birth Public Use files from the Austrian government to individuals from 1989 to 2000. They find evidence in who resided in certain regions and were favor of their hypothesis that pregnant employed (or had been employed) by a cer- women are most likely to be influenced by tain group of industries. Evidence of social new mothers from the same area and ethnic interactions in unemployment behavior dec04_Article 2 12/14/04 2:26 PM Page 1074

1074 Journal of Economic Literature, Vol. XLII (December 2004)

measured in terms of length of unemploy- Employees may refer their own acquain- ment spells takes the form of both direct tances they believe to be qualified for effects among individuals who are entitled to employment. If those are hired, they in turn the extended benefits and indirect effects refer their acquaintances. Each step in the through the impact of the program’s exten- referral process is stochastic and reflects a sion on the behavior within the reference whole host of factors describing the labor group. Their methodology is interesting market. Their central result is that the den- because their identification of social interac- sity of referral linkages exhibits threshold tions uses a linear model with the dummy behavior. Above a threshold, workers variable indicating that an individual is aged throughout the hierarchy are referred for at least fifty years and has continuous work employment with probability one; below it, history as an individual effect and with its workers’ probability of referrals falls expo- corresponding group mean as a contextual nentially as one moves down the network so effect. Their approach is made possible that average income falls and workers at very because of the Austrian government’s “par- low levels are referred with probability zero. tial-population intervention.” This route to Referrals are more valuable than anonymous identification of social interactions model matching because they convey information was discussed by Robert Moffitt (2001); that about employees’ qualifications, reduce is, the endogenous social interactions coeffi- recruiting and training costs, and lower cient is disentangled from the direct effect monitoring costs. Finneran and Kelly estab- of extended benefits by distinguishing two lish the value of referrals in a general net- groups of the population and by comparing work, where the number of potential ties by reduced forms estimated with data for those each worker and the probability of forming affected with estimates for those not affect- ties all differ across the network. Given a set ed by the policy. This result is also particu- of potential direct referrals with a large larly relevant in explaining spatial variation number of potential connections n and a set in unemployment, an issue of great interest of workers who are always guaranteed in large economies. employment, the measure of paths that con- nect those who are always guaranteed 3.5 The Role of Job Referrals employment with other individuals in a net- Krauth (2000) focuses on spatial proximity work who are far away from them (in terms effects by studying the consequences of of the number of required referrals) is employers’ using their social ties with their increasing in the number of actual referrals employees to make inferences about unob- made. As the size of the economy grows, served components of the productivity of there exists a critical number of linkages, their workers’ social contacts. He shows that which varies slowly with n, such that the there is a critical value for neighborhood probability for any group of workers who are human capital below which long-run far in terms of linkages from those who are employment at equilibrium is low and above always guaranteed employment to be linked which it is high. The critical value depends with them through referrals tends to one, on the strength of social contacts. when the number is above the critical value, Lisa Finneran and Morgan Kelly (2003) and to zero, when it is below it. As in the examine theoretically the role of job refer- result of Calvó-Armegnol and Jackson, this is rals with special emphasis in the persistence due to the statistical properties of networks of inequality. Workers differ in terms of composed of workers who are otherwise skills, and such differences are ex ante unob- identical. servable by employers. Each individual is a Krauth (2003b) models search for jobs, member of a hierarchical referral network. where individuals’ social acquaintances dec04_Article 2 12/14/04 2:26 PM Page 1075

Ioannides and Datcher Loury: Job Information Networks 1075

provide referrals to potential employers. for the job-seeker. Simon and Warner (1992) The model assumes a social structure, rep- presented evidence that those who found resented by a directed graph, which is out about their jobs through an acquaintance exogenously given but may possibly be time inside the firm had higher starting salaries, varying, and may vary stochastically as well, while those who found out through an but remains exogenous. Firms hire workers acquaintance outside the firm had lower but do not observe their quality directly. A starting salaries. They attributed this finding worker’s past employment and social con- to reductions in employer uncertainty about nections affect her current employment worker productivity. prospects. Krauth proposes a model of Economics research has explored the strong versus weak ties that is intended to salient aspects of contact and relational het- express Granovetter’s concept of weak ver- erogeneity that influence job-contact sus strong ties as follows: a social tie from prospects for individuals, but there is little agent i to agent j is defined as strong if j also comparable work on the effects of employer has a social tie to one or more of i’s other characteristics. In a notable exception, the friends; it is weak, otherwise. The model varying role of referrals in the U.S. industri- involves starting from a network with only al sector has been studied directly by strong ties and switching some of the social Adrianna Kugler (2002). She finds that the connections randomly with probability p. observed positive correlation between For large networks, the probability that this industry wage premia and use of employee process will generate another strong tie is referrals when industry-level data are used close to 0, and the fraction of weak ties in disappears when she controls for sector of the resulting network is approximately p. employment using micro data from the Krauth uses simulations to show that the National Longitudinal Survey of Youth. long-run probability of employment is Relying on data from the European increasing with the proportion of weak ties. Community Household Panel, Pellizzari Weak ties appear to be a way for an individ- (2004a) reports substantial cross-country vari- ual to diversify her social resources. When ation in the effects of contacts on earnings. In individuals are friends of one another (they some cases, contacts result in wage premiums are connected through strong ties), their and in others workers who found their jobs employment statuses are correlated and through contacts earn less than those using this increases the variance in the number of formal sources. The latter largely occur in employed friends. Therefore, a network industries where firms invest substantially in with more weak ties is associated with formal recruitment activities. Firms are more smaller inequality in the distribution of likely to undertake such investments for high employed friends and thus with a higher productivity jobs where the costs of turnover overall employment rate. Troy Tassier and are substantial. When large investments are Filippo Menczer (2002) also study the role made, workers found through formal recruit- of referrals in labor-market outcomes. ment average higher productivity than those Empirical evidence on the interaction found through other means. Pellizzari among individuals and their social contacts (2004b) confirms this by using data from the and employers through job referrals is not as Survey of Employers’ Recruitment Practices extensive. According to Loury (2003), who for the United Kingdom in 1992. works with the National Longitudinal Survey 3.6 Evolutionary Models of Social Structure of Youth, personal contacts have significant wage effects for young men only when their Next we discuss briefly some connections contacts are older-generation male relatives of the neighborhood-effects literature with who know the boss or arranged an interview the literature on evolutionary models of dec04_Article 2 12/14/04 2:26 PM Page 1076

1076 Journal of Economic Literature, Vol. XLII (December 2004)

social structure. Specifically, this literature ascribed to weak ties in the interpersonal flow studies interactions through games individu- of job-related information. Clearly additional als play with members of a population where research is needed. a social state is defined as adoption of a norm or other institutions. First, it is interesting to 4. Strategic Network Formation and contrast the notion of weak versus strong ties Endogenous Job Information Networks with global versus local interaction in evolu- If social networks are determined by tionary models of social structure. Groups uncoordinated actions of individuals, then whose members are connected with strong several of their features are of interest. First, ties are typically small and close-knit. how many others is an individual in contact Second, as emphasized by Robert Dietz with to exchange job-related information? (2002), the evolutionary learning literature Second, how does this number vary among involves ideas that are conceptually related individuals within the economy, and how to social interactions. might it depend upon individual characteris- A particularly noteworthy finding in this tics? And, third, what is the topology of the literature is a result of Glenn Ellison (1993), associated network? Is everyone directly according to which if individuals interact with connected with everyone else, is there a sin- members of a large population then existing gle individual through whom everyone is behavior (or social conventions, historical connected, or is there some other stylized forces, etc.) are likely to dominate. On the pattern of connections that is discernible in other hand, if they interact with a small num- social networks? We review the existing lit- ber of neighbors, then cooperative outcomes erature while recognizing that it has not, to and thus coordination is more likely. So, date, succeeded in delivering endogenous interactions involving small neighborhoods outcomes with respect to all these features are more likely to be determined by evolu- simultaneously. The fast-developing litera- tionary forces. The results of Ellison have ture on strategic network formation, which been generalized by Peyton Young (1998) as seeks to motivate the creation of social con- follows.15 When individuals interact mainly tacts in terms of optimizing behavior by with small groups of neighbors, then “shifts of individuals, has not, until recently, empha- regime can occur exponentially faster than in sized job-information networks, although, as the case of uniform interaction ... . All else we see shortly, important progress has been being equal, the smaller the size of the neigh- made. The strategic network formation liter- borhood groups, and the more close-knit the ature has been eloquently reviewed recently groups are, the faster the transition time for by Bhaskar Dutta and Matthew Jackson the whole population (Young 1998, pp. (2002), Goyal (2003), and Jackson (2003). So 98–99). This suggests an important role of we will touch on it only very briefly. close-knit groups for social learning, which in the context of the literature reviewed here 4.1 Strategic Models of Network Formation may be interpreted as that of strong ties. This The principal contributions in this litera- contrasts with the role that Granovetter has ture, including Matthew Jackson and Alison Watts (2002) and Watts (2001), aim at 15 Regimes are described with stochastically stable axiomatic descriptions of network-based states. The concept of stochastically stable dynamic equi- concepts. Important such concepts are: effi- librium (Young 1993, 1998) is particularly appropriate to ciency, with a network being efficient if there circumstances where a system is subject to repeated shocks, that is when the system is shocked repeatedly is no other network that leads to higher pay- and before it has a chance to recover, it is shocked again, off for all of the members; stability, with a and again. In such settings, some states will occur more network being stable if no individual would frequently than other states in the long run. dec04_Article 2 12/14/04 2:26 PM Page 1077

Ioannides and Datcher Loury: Job Information Networks 1077

benefit by severing a link and no two players that are obtained endogenously, are impor- would benefit by forming a new link; and the tant for the functioning of job-information notion of a Nash network, where all mem- networks. That is, the information flows bers are playing Nash equilibrium. The mod- associated with two different chains of con- els in the strategic network formation tacts of identical length but in different literature make links endogenous by means topologies are generally different. We return of strategic considerations; that is, mainte- to this work in detail further below. nance (or creation) of links between two Bala and Goyal (2000) also model endoge- individuals requires that they both consent to nous network formation, where each individ- it, whereas severance can be done unilateral- ual derives utility that is proportional to the ly. The utility each member derives from number of other agents she is connected with being a member of a network, that is, from directly and indirectly, net of the costs of being connected with others who may them- maintaining those connections. Unlike selves be connected with others, typically Jackson and Wolinsky (1996), however, theirs depends additively upon the number of involves directed links. They define one-way other agents each agent is connected with, communication as your having access to minus the costs of maintaining connections. another person’s information. An one-way Some authors make an allowance for proxim- link does not imply that other person has ity, by means of a decay factor that depends access to yours, which would be two-way on the number of intervening agents. Others, communication (undirected links). Bala and like Venkatesh Bala and Sanjeev Goyal Goyal show that, with one-way communica- (2000), discussed in more detail below, dis- tion, Nash networks are either minimally con- tinguish between one- or two-way communi- nected and form a wheel (each player forms cation. As Jan Brueckner (2003) and Jackson exactly one link) or empty. In other words, (2003) demonstrate, these assumptions are information is either shared with everyone, or quite crucial for the results. there is no sharing. Bala and Goyal also study Matthew Jackson and Asher Wolinsky the dynamics of link formation by assuming a (1996) show the equilibrium network is naive best-response rule with inertia, that is, empty if linkage costs are high; it is a wheel an agent may choose, with fixed probabilities, network (where each agent is connected with either a myopic pure strategy best-response, two other agents thus forming a wheel, or a or the same action as in the previous period. circle) if linkage costs are moderate; and it is They show that irrespective of the number of a complete network if the linkage costs are agents and from any initial starting pattern of low. The two extreme outcomes are quite interconnections, the dynamic process self- intuitive. The wheel outcome is also intuitive organizes, by converging in finite time with when one recognizes that it is associated with probability 1 to the unique limit network. The two connections per person, which is the limit is the wheel if the linkage costs are minimal symmetric outcome. This demon- small, or either the wheel or the empty net- strates the sensitivity of network topology to work if the linkage costs are large.16 With parameter values and suggests that in prac- two-way communication the results are quite tice different topologies may emerge in a given economy for different sets of problems. 16 These results may be generalized by restricting the The recent resurgence of interest in job information available to agents, that is, by assuming only local information—each agent knows the residual set of all information networks has benefitted by those she is connected with, that is those her neighbors can extending several of the concepts proposed access without using links to her—and by allowing obser- by this literature. In particular, Calvó- vation of successful agents—there is some chance that she receives information from a “successful” agent, that is a Armegnol (2004) suggests that conditions person who observes the largest subset of people in the that lead to different network topologies, economy without assistance from her own links. dec04_Article 2 12/14/04 2:26 PM Page 1078

1078 Journal of Economic Literature, Vol. XLII (December 2004)

different: Nash networks are either “center- real-life job information networks. Different sponsored” stars (where an agent forms the individuals may value differently the payoffs network by connecting himself with all oth- from being connected with others relative to ers, occupies the central position, and pays the costs of doing so. for all links), or the empty network. 4.2 Endogenous Job Information Networks It is interesting that experimental evi- dence that has been obtained recently pro- Social interactions associated with endoge- vides support for the Bala and Goyal model. nous job information networks may be quite Armin Falk and Michael Kosfeld (2002) different from others, such as those inspired report that the prediction based on strict by constraint, preference or expectations Nash equilibrium works well in the one-way interactions, to use the terminology of communication model, with subjects form- Manski (2000). The theoretical predictions of ing the wheel or the empty network in a Calvó-Armegnol and Jackson (2002, 2003)17 majority of cases. In contrast, the predictions depend on ad hoc assumptions about social based on Nash and on strict Nash fail for the exchange: the network of social interactions two-way communication model: the subjects is given. By varying network characteristics do not form the center-sponsored star nor and parameters, one may explore potential the empty network in any of the experi- consequences of primitive behavioral ments. The authors attribute their results to assumptions about individuals’ valuations of the subjects’ sensitivity to fairness consider- social interactions, especially with respect to ations, which is a well-known factor affecting different types of individuals’ propensities to game outcomes in experimental settings. transmit job-related information. Brueckner (2003) extends the basic What do we know about the formation of assumptions of this literature by endogeniz- the networks used for acquiring and dissem- ing the probability of a link and of linking inating job information? Scott Boorman costs. This extension assumes that individu- (1975) was the first to ask formally how social als value only their direct connections, groups accommodate the transmission of involves a simple mathematical structure, job-related information. He presents an ana- and derives some properties that would seem lytical model of transmission of job-related to be relevant outside the confines of those information through contacts, where individ- particular models. For example, identical uals choose how to allocate effort over main- individuals will spread their efforts uniform- taining strong and weak contacts. Strong ly to create social acquaintances, thus bring- contacts are given priority in the transmission ing about symmetric outcomes and ruling of job-related information but require more out inherently asymmetric network topolo- time than weak ones to maintain. Choices gies like that of the star. However, asymmet- over strong versus weak contacts by all mem- ric outcomes are possible if individuals’ social bers of the society determine a rudimentary attractiveness differs—individuals can have social structure. magnetic personalities—or if individuals Let and denote the probability that a have different sets of social acquaintances. person will need a job at a particular point in The assumptions these models employ do time, and that she hears directly of a vacant not make them readily applicable to the job, respectively. If S and W denote the study of exchange of information about jobs. Still, the experimental evidence on network 17 Pierre Cahuc and Francois Fontaine (2002) allow for formation and the sensitive dependence of individuals to choose between job matching through social endogenous network topology on costs rela- networks and (costly) individual search methods. Such an extension appears to overturn several standard results. tive to benefits suggests that cultural factors Competitive search may be over- or under-utilized and may be important in the determination of multiple equilibria are possible. dec04_Article 2 12/14/04 2:26 PM Page 1079

Ioannides and Datcher Loury: Job Information Networks 1079

number of strong and weak contacts, respec- This is quite surprising in view of the fact tively, then the probability that a person will that it is a model of endogenous (job) infor- get a job through her contacts may be com- mation network, where agents choose the puted as the one minus the probability that number of links, which does not require the person will not get a job either from any strategic considerations. Curiously, the of her strong or her weak ties. The latter two development of strategic models more gen- probabilities may be obtained from elemen- erally seemed to have provided an impetus tary but tedious combinatoric computa- for the development of non-strategic mod- tions.18 The probability that a person will get els, as well. Calvó-Armegnol (2004) devel- a job is written as: ops the first precise model of a contact S  11−−() S  network deliberately aimed at passing job- −− −  11() 1 related information and Calvó-Armegnol  S  and Yves Zenou (2001) explore its implica- W  W  tions for labor-market-wide matching. − 11−−()  −−S 1  11() . (2) These papers, unlike Granovetter and  W  Boorman, do not distinguish between strong Maximizing this probability, subject to a time and weak contacts.19 constraint, WST, where denotes the In Calvó-Armegnol (2004), all workers are extra units of time required to maintain a initially employed; they may lose their jobs strong tie, allows us to study whether strong with a constant probability b, and hear of a or weak contacts are more likely to be cho- new job opportunity with probability a. If a sen for different values of the parameters. worker hearing of a job is unemployed she Boorman’s probabilistic approach has takes it; if she is employed, she passes the recently been taken up by Calvó-Armegnol information on to her unemployed direct (2004), discussed below. contacts. A worker is employed and hears of It turns out that if the probability of need- a job with probability a(1b), and is ing a job is small, 1, Boorman’s model unemployed and does not hear from her con- implies that the equilibrium where all indi- tacts about jobs with probability b(1a). viduals choose weak ties is stable. The oppo- Let g denote a network of contacts, gN the set site is true when ≈1 . These results have an of all subsets of the set of all individuals of intuitive appeal. The only reason for invest- size 2, that is the complete graph among N ing in strong contacts is the concern that individuals, and G{g|ggN}. Given g∈G, one’s weak contacts will be preempted by the the set of an individual’s direct contacts in g

demands of their own needy contacts. As the is denoted by Ni(g). The probability that an probability that anyone needs a job decreas- es this contingency becomes less important. 19 The earliest model of word-of-mouth communica- Boorman’s model does not use tradeoffs fac- tion that we are aware of is Strand (1983), who is interest- ed in intrafirm wage dispersion. The concept of ing workers searching for jobs, but it does word-of-mouth communication used by Calvó-Armegnol portray the properties of the communication and Zenou is similar to Glenn Ellison and Drew network in a uniform symmetric setting, Fudenberg (1995), although the latter stresses the effi- ciency of social learning. The latter paper’s finding that where each individual maintains both strong social learning is often more efficient when communica- and weak ties. tion between agents is fairly limited. In a model of word- Boorman’s path-breaking work took a long of-mouth communication, this is understood as sampling from a given sample of other participants, with some frac- time to influence the economics literature. tion of players ignoring the information and not changing their decisions and the remainder adopting the choice that appears to be best based on their own nonoptimal sample. 18 See ibid. and Calvó-Armegnol (2004), Proposition 1, This property of limited communication is conceptually for details. Actually, the latter does not distinguish relevant to the role of weak ties in social networks, as we between strong and weak ties. discussed in subsection 3.6. dec04_Article 2 12/14/04 2:26 PM Page 1080

1080 Journal of Economic Literature, Vol. XLII (December 2004)

individual i receives job information who have the same degree, . In that case, the ∈ from one of her direct contacts j, j Ni(g), tradeoffs between the number of direct and who is assumed to be informed, is given by indirect contacts on an agent’s employment ng() 11−−()b i probability implies that it attains a maxi- , where ni(g) |Ni(g)|. (This deri- bni () g mum over the set of different degrees, (1, vation is, of course, closely related to ). Where an agent’s employment probabil- Boorman’s, op. cit.) Therefore, the probabil- ity increases, it is also the case the marginal ity that i does not actually find a job thanks effect of increasing the network’s degree is ng() 11−−()b i negative. While increasing the number of to j is given by q(ni(g)) 1 , and bni () g direct links improves the employment prob- the probability that individual i actually gets ability, it also does so for everyone, thus also a job through her contacts is: increasing the number of indirect links, which are detrimental to the likelihood of =−∏ Pgii()1(()) qng . (3) employment for sufficiently high values of ∈ jNgi() the network’s degree. In other words, direct This probability is larger, the greater an indi- contacts are beneficial because they vidual’s set of direct contacts. A larger set of improve an individual’s information sources, contacts broadens the information channels but contacts that are two-links away are available to i, but not the number of direct detrimental because they create competi- contacts as such. This probability decreases tors for the information possessed by a the larger is an individual’s indirect contacts direct contact. This rivalry is also the reason

through any of his direct contacts, nj(g), that the sign and intensity of the payoff ∈ j Ni(g). That is, having more indirect con- spillovers that agents exert on one another tacts increases the competition for informa- are very much dependent on the geometry tion. This basic relationship determines the of the network. This is also a reason more return to adding and severing links among general analyses are difficult. any two individuals. Doing so affects the Calvó-Armegnol also considers asymmetric information flow for those directly affected as networks. In fact, with the same parameters, well as for their direct contacts. This expres- both symmetric and asymmetric networks are sion also implies an aggregate unemployment possible, but they cannot be “too asymmet- rate given by: ric.” This model is the only one to date that   has employed successfully a model of strate- ug()=− 1 1 ∑ Pg() . gic network formation to the job market con-  n iN∈ i  text. Interestingly, individual payoffs in this Calvó-Armegnol (2004) uses this model model do not, unlike the specific models in to examine properties of symmetric equilib- Bala and Goyal, contain a component that is rium networks, when a link between any linear in the number of other agents each two agents, which is undirected and costly agent is connected with. This could explain to both of them, is initiated only if it is why it is so much more difficult to study mutually advantageous. The model also endogenous network topology in the general implies tradeoffs associated with the topolo- case in Calvó-Armegnol’s model. gies of the networks of contacts. Networks Calvó-Armegnol and Zenou (2001) explore with the same total number of contacts but the implications of the model in Calvó- different topologies imply different aggre- Armegnol (2004) for aggregate matching gate unemployment rates. It is particularly properties of an economy. Aggregate match- simple to consider topologies of regular ing is increasing and concave in both the graphs, where all individuals have the same unemployment and vacancy rates. However, number of direct contacts, that is, all nodes hearing through both direct and indirect dec04_Article 2 12/14/04 2:26 PM Page 1081

Ioannides and Datcher Loury: Job Information Networks 1081

contacts implies that the impact of network suggest that it is important to research fur- size is not monotonic. This is so, as we argued ther the aggregate the properties of differ- above, because such an increase increases ent types of social networks with respect to the potential number of unemployed work- matching.20 ers who are connected directly to an employed informed worker. 5. Towards an Integration of Job informa- Specifically, if 1u the probability that a tion Networks and Sorting worker is employed and s is the number of The interconnection of individuals other (randomly drawn) workers each through job information networks clearly worker is in direct contact with (the degree gives rise to sorting phenomena in both the of the symmetric, or balanced, social net- economic and social spheres. Sorting, on the work), then each worker meets us unem- other hand, has typically been investigated ployed workers and (1u)s employed by the economists in a life-cycle context. workers in each period. Under the assump- This section examines the interplay of those tion that information is passed by two forces. employed to unemployed workers only and A well-established literature has investi- the vacancy rate represents the probability gated how the intertemporal evolution of that an unemployed individual will hear of human capital is affected by the human cap- a job vacancy directly, than the individual ital of parents, of the ethnic group to which probability of finding a job through social the individual belongs, and of the individ- contacts is given by: s ual’s neighborhood. The natural relationship  s  11−−()u between parents and children may affect Psuv(),, =−11 − v () 1 − u  . The  us  variables (such as “innate ability”) that play an important role in the perpetuation of matching function is given by m(s,u,v) inequality, but are not subject to choice. On u[v(1v)P(s,u,v)]. the other hand, mating does involve choice, The properties of the matching function and the human capital of spouses is not inde- with respect to unemployment and vacancy pendent of one another across the popula- rates are the same as in the earlier job tion. Parents’ choice of human-capital matching literature; they differ only with investment for their children is affected by respect to network degree. In this case, the their own human capital. matching function is not only not homoge- A prominent example of this literature is neous of degree one, as in Pissarides Michael Kremer (1997), who studies indi- (2000), but not even monotonic. When the viduals’ schooling as a function of that of network degree increases, unemployed the parents and of the mean schooling in workers hear about more vacancies through one’s neighborhood of upbringing. He their social network. At the same time, it is finds strong neighborhood effects; that is, more likely that information about multiple mean schooling in the neighborhood of vacancies will reach the same unemployed one’s upbringing has a coefficient that is worker. It is therefore important to see roughly the same as that of the parents’ whether this non-monotonicity is present in schooling. This effect is not sufficient, how- the data, which is an open question empir- ever, to explain a large role for residential ically, to the best of our knowledge. The sorting in the inequality of earnings across properties of the matching function affect the equilibrium level of unemployment in the economy. Matching is increasing in net- 20 The recent exhaustive review of the literature on the matching function by Barbara Petrongolo and Christopher work size, for sparse networks, and Pissarides (2001) does not discuss at all matching aspects decreasing for dense ones. These findings of social networks. dec04_Article 2 12/14/04 2:26 PM Page 1082

1082 Journal of Economic Literature, Vol. XLII (December 2004)

the population.21 George Borjas (1992) labor income as the product of human capital, regresses individuals’ schooling against which represents income earning capacity schooling of parents and the average school- measured in efficiency units, and the appro- ing of the ethnic group to which an individual priate wage rate, which represents its current belongs. Borjas, too, finds strong ethnic market price per efficiency unit, we may sepa- effects. Borjas (1995) allows, in addition, for a rate intertemporal effects, such as the parental potential link between parental and ethnic effect, the ethnic group effect and the neigh- capital, on one hand, and residential segrega- borhood of upbringing effect, from social net- tion, on the other, and again finds a strong work effects on individuals’ access to ethnic effect. Borjas (1998) finds that, in addi- job-related information. The former effects tion, schooling is affected by the presence operate at life-cycle frequencies, indexed by and skill levels of other ethnic groups in the period t 0,1,2, …, while the latter operate at neighborhood of one’s residence, with lower nearer to business-cycle frequencies, indexed segregation by ethnicity associated with more by t, which indicates intervals of time within schooling for an individual and for her ethnic each discrete life cycle period t. They are group. He also allows for potentially complex much shorter than life cycle frequency and interactions between one’s ethnic group and account for employment-related events. the distribution of socioeconomic characteris- These two sets of effects would typically not be tics in one’s neighborhood of upbringing and independent. The social connections of one’s obtains statistically significant results. parents are related to the pattern of social con- Job information networks can have com- nections within one’s ethnic group and may plex neighborhood effects, too. They affect also influence one’s own social connections. different individuals’ access to information The intertemporal evolution of income- about job opportunities and thus have bear- earning ability, human capital for short, of ing for individuals’ abilities to market their individual i who belongs to ethnic group labor services. Such effects imply a depend- e, hi,e,t, reflects such intertemporal effects as ence among individuals’ neighborhood of the parental effect, which operates through residence, social connections, and status and the natural parental relationship between terms of employment. These effects operate individuals i e,t1 and i e,t1, the parents of within much shorter time frames than the individual ie,t, assuming for simplicity that time span between the influence of upbring- both of individual i’s parents belong to the ing and of parents’ characteristics and off- same ethnic group, the ethnic effect, he,t, spring human capital. It is thus appropriate to and the neighborhood effect, h(i),e,t1, allow for different time scales.22 By defining where (i) indicates the neighborhood of individual i’s upbringing. We may combine Kremer (1997) and Borjas (1992; 1995; 21 Working with the same data, Ioannides (2002) finds 1998) and write an equation for the law of strong nonlinear effects, which imply multiple equilibria. Unfortunately, like Kremer’s, Ioannides’ results are not motion of human capital as follows: causal, in that the endogeneity of neighborhood choice by parents is not accounted for. Hi,e,t H(hi,e,t1, hi,e,t1; he,t, h(i),e,t1), (4) 22 In modelling physical and engineering processes, one often distinguishes between different time scales. See which allows the ethnic effect and the also, Young (1998) on the notion of “event” time, which neighborhood effects to interact, h(i),e,t1. marks numerous distinct events that may be compressed A complete description of the intertemporal into short intervals of “real” time. In the absence of differ- ent time scales, these considerations would produce a con- evolution of income earning ability requires temporaneous cross-sectional effect. We thank Giulio description of the dynamic evolution of the Zanella for directing our attention to James Meade (1976), ethnic effect. for a discussion of transmission of inequality based on a concept of effects of social contacts operating in different Let Si,e,t denote the event that individual i frequencies. is employed, Si,e,t 1 , or unemployed, dec04_Article 2 12/14/04 2:26 PM Page 1083

Ioannides and Datcher Loury: Job Information Networks 1083

Si,e,t 0, at a business-cycle point t of her life research earlier in the paper suggests that cycle period t. Following Calvó-Armegnol and the efficiency wage rate, the set of job Jackson, op. cit., we normalize wages so that opportunities and the employment probabil- Wi,e,t 0, if i is unemployed. Wage setting may ity all depend on one’s ethnic group. It is fair be expressed as a function of one’s wage in the to say that the empirical literature suggests previous (business cycle) period and of the strong persistence in ethnic composition and number of new job opportunities individual i income distribution of neighborhoods. has as of time t, Oi,e,t, Wi,e,t w(Wi,e,t1, The empirical research reviewed above Oi,e,t), where the index t 1 is defined with- offers separate glimpses on the joint distri- in the time scale of business cycle frequen- bution. Eq. (5), along with (4) and a descrip- cies. We may now write an equation for the tion of the dynamics of human capital by intertemporal evolution of labor income, ethnic groups encapsulate the joint effects of human capital and access to job opportuni- Yi,e,t Hi,e,tWi,e,tSi,e,t . (5) ties on the distribution of earned income We may obtain the full dynamic flavor of Eq. while accounting for the fact that those (5) by contemplating the dynamic evolution effects operate at different time scales. of the probability of employment and of the Conceptually, one may adapt the method number of new opportunities. For example, pioneered by Loury (1981) and the tools of recall the probability of i’s getting a job Carl Futia (1982) to describe the equilibri- through her contacts according to Calvó- um joint distribution of these characteristics Armegnol (2004), Eq. 3) above. That theory as an invariant distribution associated with implies that the employment probability the law of motion of income-earning ability, depends on the size of the set of contacts, ethnic effects, neighborhood effects, and ni=|Ni,t(gt)|, where gt denotes the graph job-information effects. It is unlikely that describing a particular realization of social this description can ever be reduced to a sin- networks at time t. gle dimension. However, it would also be The graph describing individual i’s social interesting to adapt Loury’s essentially deter- contacts will not necessarily be regular (all ministic approach to a stochastic one, incor- individuals having the same number of con- porating the tools of evolutionary stability tacts with others) and will evolve over (busi- employed by Young (1998). Further theoret- ness cycle) time in a way that exhibits ical research is likely to allow deeper analysis dynamic dependence. Therefore, in princi- of the impact of job-information networks on ple, we can describe the evolution of the inequality. This entirely new approach is cru- probability of getting a job through one’s cial for deeper understanding of the lifetime contacts as a function of the social networks income distribution. the individuals have access to. These will There has been a fair amount of success in emerge as a result of individual incentives describing the equilibrium distribution of that reflect strategic considerations associat- income, with an emphasis on the impact of

ed with network formation. That is, Si,t must sorting on human capital formation in life 23 be derived as a function of Ni,t(gt). We cycle frequencies. Our understanding of the note that the discussion of empirical impact of job-information networks and their role as a force of inequality is much less developed. They are potentially very impor- 23 In view of Nieke Oomes (2003), who shows that if labor can be hired in continuous quantities, the long run tant, in part because network formation distribution of employment in spatially separated mar- depends on sorting of individuals’ own char- kets is uniform, we conjecture that persistence of non- acteristics. It is interesting to speculate how trivial effects of social networks on job-related outcomes is intimately related to the presence of a labor-market sorting by own characteristics in choosing participation decision. whom to associate with, rather than passively dec04_Article 2 12/14/04 2:26 PM Page 1084

1084 Journal of Economic Literature, Vol. XLII (December 2004)

reacting to standard norms of behavior in immediate neighborhood is associated with groups, underlies an independent role of lower dropout rate. ethnicity as conjectured by Borjas. Kremer (1997, p. 135) notes that “to the 6. Suggestions for Future Research extent that people learn from classmates and co-workers, sorting by an individual’s own The specific model proposed in the previ- academic ability or productivity in schools or ous section provides an overarching theme workplaces may have larger effects on that helps integrate life-cycle and business- inequality than sorting by parental charac- cycle forces into a model of job-information teristics, since an individual’s future charac- networks and sorting. At the same time, a teristics are presumably more highly number of other concrete issues may be correlated with his or her current character- addressed directly and may thus comple- istics than with parental characteristics.” As ment the overarching theme. They are Kremer also notes, this is the kind of sorting briefly discussed next in this section. that most people would regard as egalitari- Theoretical issues needing further atten- an, presumably because it is based on indi- tion abound. The issue of strong versus weak viduals’ free choice of association,” but it social ties is a natural one. It has already may be most likely to significantly ‘increase been generalized by the exogenous job- inequality.’” The phenomenon of segrega- information networks literature. It may be tion by skill in firms, that Kremer and Eric pursued further along the lines of the meth- Maskin (1996) have analyzed, is particularly ods employed by the endogenous job-infor- relevant in this context. Firms are, mation networks literature, where recent metaphorically speaking, neighborhoods. contributions have not distinguished social Referrals among individuals who have been tie strength. Similarly, global versus local coworkers involves selection through one’s interactions among individuals provide a own characteristics and is closely related to tempting parallel to strong versus weak ties the properties of technologies used by firms. and therefore deserve additional theoretical It thus provides a route for technology to attention. The findings of this paper point to affect equality of earnings. Further work in a need to understand better the information- this area appears to be particularly fruitful. al and social infrastructure of the modern Research on the dynamics of inequality to economy. In this context, a particularly glar- date has emphasized the role of neighbor- ing weakness of the theoretical and empirical hood effects in either the intergenerational literature is almost total lack of research on dynamics of human capital or the labor sup- the role of professional intermediaries ply decisions separately from one another. (“headhunters”) throughout the job market, The framework we sketched above suggests in spite of anecdotal evidence of the increas- that it would be fruitful to look at both sets ing importance of such intermediaries, and of decisions jointly. Henry Overman (2002) especially outside its traditional territory of is a rare example that allows for both types of executive search. An economy’s social and spillovers. That paper uses data on a sample informational infrastructure is also important of Australian teenagers to test for neighbor- for understanding the role of institutions in hood effects on school dropout rates at two facilitating individuals’ access to resources. different spatial scales. Overman finds that We stress the fact that network topologies educational composition of the larger neigh- as equilibrium outcomes are very sensitive to borhood can influence the dropout rate, pos- parameter values and therefore it is impor- sibly reflecting the structure of local labor tant for a model to express the particular cir- market demand. He also finds, more surpris- cumstances of the problem. The strategic ingly, that low socioeconomic status of the network formation literature has examined dec04_Article 2 12/14/04 2:26 PM Page 1085

Ioannides and Datcher Loury: Job Information Networks 1085

how the economy self-organizes under dif- belief that network-based models are indis- ferent assumptions about expectations. The pensable for modelling more than just trivial exogenous social interactions literature has social interactions. As Harrison White explored the dynamics of different topolo- (1995) emphasizes, further theorizing on gies. Combining these two approaches social networks is likely to pay off even with- deserves further attention. in sociology, where in spite of technical Research on the impact of the information achievements in social-network measure- technology revolution on the job market is ments, modelling “network constructs have only just beginning. Richard Freeman had little impact so far on the main lines of (2002) provides evidence that computeriza- sociocultural theorizing …” (p. 1059). White tion and use of the Internet are associated sees an important role for studying social with greater hours as well as higher wages. interactions through interlinking of different Evidence reported by Kuhn and Skuterud individual-based networks associated with (2000) suggest increasing use of the internet social discourse. Granovetter (2000) urges for the purpose of job search. We know very sociologists to go beyond merely emphasiz- little about the impact of the internet on the ing “the embeddedness of action in social economy generally, let alone on recruitment networks” and states that a “focus on the and job search. mechanics of networks alone …” is insuffi- Both theoretical and empirical research cient to “lead us toward the more complex on firms’ recruitment practices also seems synthesis that we seek in understanding the likely to yield big payoffs. This research to economy” (p. 23). In view of this, it appears date has explored existing models of match- that distinguishing between the motives that ing to considerable advantage. However, the prompt individuals to engage in social net- methodology of strategic network formation working deserve attention in future lends itself equally well to such a task. Wage research. premiums and wage penalties associated with finding jobs through personal contacts 7. Conclusions are the joint outcome of firms’ recruitment efforts and individuals’ job search. We can- Most of the initial economics research on not identify the role of information through informal contacts aggregated the effects of social interactions without accounting for informal contacts and networks. It has both sides of the market. This is particularly assessed the role of job contacts on out- important if one recognizes the conceptual comes by comparing outcomes with to out- links between neighborhood effects and net- comes without job contacts. The new strands work effects and the econometric issues in theoretical and empirical economic they pose when it comes to the identifica- research examined in this paper build on tion of endogenous versus exogenous sociological analyses and point out that con- effects, which we discussed earlier in this tact effects are complex and vary due to indi- paper. Institutional and cultural differences vidual, contact, relational, and employer across different countries in how social con- heterogeneity. The new literature identifies tacts facilitate job contacts and how use of the specific ways in which the effects of intermediaries and formal sources differ informal networks depend on differences need to be better measured and understood. among job seekers themselves, on the char- An important benefit here would be for acteristics of the contacts they use, on the economists to learn from the multidimen- relationship between the job seekers and sional picture that the network-based theo- their contacts, and on features of the work ries of mathematical sociology confer. At the environments where individuals are seeking heart of the sociological literature is the jobs. The research makes clear that these dec04_Article 2 12/14/04 2:26 PM Page 1086

1086 Journal of Economic Literature, Vol. XLII (December 2004)

components do not operate in a vacuum but mechanism behind correlations in observed instead interact with each other to produce outcomes within social groups. Thus the variation we observe in the use and research has already used network analysis effects of informal networks. This has been to elucidate the origins of previous unex- confirmed by the new theoretical research plained similarities in outcomes by race, eth- that emphasizes the emergence of social net- nicity, and gender. Furthermore, it identifies works from individuals’ uncoordinated the source of some neighborhood correla- actions and finds that the resulting networks tions in labor-market outcomes. At the same are very sensitive to parameter values. time, there are a number of promising areas The research reviewed here suggests that where research is needed. In particular, the heterogeneity in network effects is impor- importance of employer characteristics and tant in a variety of contexts. It can help the role of the internet in altering the role of account for changes in wage and employ- informal contacts in the future are topics ment inequality across time. It clarifies the that deserve special attention. dec04_Article 2 12/14/04 2:26 PM Page 1087

Ioannides and Datcher Loury: Job Information Networks 1087

Appendix

Table 1, Table 2

TABLE 1 Education and Methods of Job Search

Category 12345678 Full All

did pub. priv. curr. other friend ads other nothing agency agency emplr emplr acquai. activ. searchers PSID

Unemployed Sample frequencies 11.5 21.8 9.5 5.7 29.4 15.5 33.5 31.0 5.8 100 Years 8 17.9 20.5 18.2 24.2 21.0 23.3 16.9 14.4 17.0 16.7 8 Years 11 41.8 25.2 25.5 30.3 28.7 28.9 27.2 30.6 31.6 17.9 Years 12 26.9 32.2 25.5 30.3 28.7 27.8 32.3 28.9 30.8 31.0 Years 12 nonac. 7.5 18.7 25.5 15.2 17.0 12.2 15.4 17.8 14.6 18.7 13 Years 15 4.5 9.5 5.5 0.0 3.5 6.7 6.7 5.0 4.3 9.4 BA adv. 1.5 6.2 0.0 0.0 1.2 1.1 1.5 3.3 1.7 6.2 Searching on-the-job Sample frequency 60.5 6.3 3.0 2.0 9.4 8.5 16.0 17.0 8.1 100 Years 8 7.1 9.8 16.7 6.3 13.2 11.6 13.2 6.6 8.3 16.7 8 Years 11 17.2 17.7 12.5 12.5 13.2 14.5 11.6 8.0 14.3 17.9 Years 12 33.7 41.2 25.0 31.3 34.2 29.0 38.0 38.7 34.9 31.0 Years 12 nonac. 27.2 25.5 25.0 43.8 22.4 21.7 18.6 22.6 25.3 18.7 13 Years 15 9.4 3.9 16.7 6.3 15.8 13.0 14.0 18.3 11.9 9.4 BA adv. 5.3 2.0 4.2 0.0 1.32 10.1 4.7 5.8 5.2 6.2

Notes: The categories are whether: 1. did nothing; 2. searched with a public employment agency; 3. searched with a private employment agency; 4. checked with the current employer; 5. checked with other employer; 6. checked with friend or relative; 7. placed or answered ads; or, 8. engaged in other activity. The results are sum- marized in the following table. The entries in the lines labelled "sample frequencies" are not mutually exclusive— some respondents may be engaged in more than one method—and thus do not add up to 100. The entries for educational attainment sum up to 100 in each column. The column labelled "Full" gives the educational attain- ments for the respective subsample of unemployed and those searching on the job in the 1993 sample of the PSID. The column labelled "All" gives the educational attainments for the entire 1993 sample of the PSID. dec04_Article 2 12/14/04 2:26 PM Page 1088

1088 Journal of Economic Literature, Vol. XLII (December 2004)

TABLE 2 Urban Size and Methods of Job Search

Category 12345678 Full All

did pub. priv. curr. other friend ads other nothing agency agency emplr emplr acquai. activ. searchers PSID

Unemployed Sample frequencies 11.5 21.8 9.5 5.7 29.4 15.5 33.5 31.0 5.8 100 500,000 38.20 27.17 15.60 35.10 13.17 51.11 22.21 23.69 23.86 16.02 [100000, 500000) 12.05 19.73 10.95 3.20 34.59 14.48 19.07 21.66 20.99 24.15 [50000, 100000) 1.69 7.28 15.34 21.16 7.12 8.37 15.61 6.86 9.39 11.76 [25000, 50000) 17.14 7.88 30.43 26.91 7.53 9.87 13.58 16.94 14.36 13.44 [10000, 25000) 3.31 14.57 22.81 5.88 17.94 7.39 20.86 18.36 14.30 15.74 10,000 27.62 23.38 4.87 5.21 17.14 8.78 8.11 12.16 15.85 17.38 Employed Sample frequency 60.5 6.3 3.0 2.0 9.4 8.5 16.0 17.0 8.1 100 500,000 16.19 6.36 7.85 .00 12.28 22.43 14.03 10.17 14.79 16.02 [100000, 500000) 30.77 22.63 22.92 19.27 27.47 22.24 28.45 27.48 30.09 24.15 [50000, 100000) 11.67 13.15 41.89 29.72 25.93 17.68 18.38 18.94 13.43 11.76 [25000, 50000) 12.21 13.29 .00 21.76 12.43 17.49 18.04 10.90 12.90 13.44 [10000, 25000) 13.68 5.98 2.18 1.33 6.73 5.81 10.42 16.27 13.08 15.74 10,000 12.10 38.60 25.16 27.93 15.17 14.34 10.68 16.25 13.44 17.38

Notes: The categories are whether: 1. did nothing; 2. searched with a public employment agency; 3. searched with a private employment agency; 4. checked with the current employer; 5. checked with the other employer; 6. checked with friend or relative; 7. placed or answered ads; or, 8. engaged in other activity. The entries in the lines labelled "sample frequency" are not mutually exclusive—some respondents may be engaged in more than one methods—and thus do not add up to the number in column "Full". The column labelled "Full" gives the relative geographical distribution of the two respective categories, unemployed and employed looking for job, for the entire 1993 sample of the PSID. "All" gives the geographical distribution of the entire 1993 sample of the PSID. All calculations are weighted by means of the latest weight in PSID. The geographical categories are defined in terms of the size of the largest city in the county of a household's resi- dence. The categories are: SMSA with largest city 500,000 or more; SMSA with largest city between 100,000 and 499,000; SMSA with largest city 50,000 to 99,999; non SMSA with largest city 25,000 to 49,999; non-SMSA with largest city 10,000 to 24,999; non SMSA with largest city less than 10,000. dec04_Article 2 12/14/04 2:26 PM Page 1089

Ioannides and Datcher Loury: Job Information Networks 1089

REFERENCES Brueckner, Jan; Jacques-Francois Thisse and Yves Zenou. 1999. “Why Is Central Paris Rich and Aizer, Anna and Janet Currie. 2002. “Networks or Downtown Detroit Poor? An Amenity-Based Neighborhoods? Correlations in the Use of Publicly Theory,” Europ. Econ. Rev. 43:1, pp. 91–107. Funded Maternity Care in California,” NBER work. Burt, Ronald S. 1980. “Models of Network Structure,” pap. 9209. Ann. Rev. Sociol. 6, pp. 79–141. Anderson, Carolyn J.; Stanley Wasserman and Bradley ———. 1992. Structural Holes. Cambridge, MA: Crouch. 1999. “A p* Primer: Logit Models for Social Harvard U. Press. Networks,” Soc. Networks 21, pp. 37–66. ———. 2000. “Network Structure of Social Capital,” Arrow, Kenneth J. and Ron Borzekowski. 2004. in Research in Organizational Behavior, vol. 22. “Linited Network Connections and the Distribution Robert Sutton and Barry Staw, eds. Elsevier Science. of Wages,” work pap., Departmemt of Economics, ———. 2001. “Structural Holes versus Network Stanford U. Closure as Social Capital,” in Social Capital: Theory Bala, Venkatesh and Sanjeev Goyal. 2000. “A and Research. Nan Lin, Karen Cook and R. S. Burt, Noncooperative Model of Network Formation,” eds. Aldine de Gruyter. Econometrica 68:5, pp. 1181–229. Cahuc, Pierre and Francois Fontaine. 2002. “On The Bayer, Patrick; Stephen L. Ross and Giorgio Topa. Efficiency of Job Search with Social Networks,” dis- 2004. “Place of Work and Place of Residence: cus. pap. 583, U. Paris 1-EUREQua. Informal Hiring Networks and Labor Market Calvó-Armegnol, Antoni. 2004. “Job Contact Outcomes,” work. pap. Yale U. Networks,” J. Econ. Theory 115:1, pp. 191–206. Beggs, John and Hurlbert, Jeanne. 1997. “The Social Calvó-Armegnol, Antoni and Matthew O. Jackson. Context of Men’s and Women’s Job Search Ties: 2002. “Networks in Labor Markets: Wage and Membership in Voluntary Organizations, Social Employment Dynamics and Inequality,” work. pap. Resources, and Job Search Outcomes,” Sociol. Caltech and U. Autonoma de Barcelona; rev. April Perspect. 40:4, pp. 601–22. 2003. Bertrand, Marianne; Erzo F. P. Luttmer and Sendhil ———. 2003. “The Effects of Social Networks on Mullainathan. 2000. “Network Effects and Welfare Employment and Inequality,” work. pap. Caltech Cultures,” Quart. J. Econ. 115:3, pp. 1019–55. and U. Autonoma de Barcelona; forthcoming Amer. Bewley, Truman F. 1999. Why Wages Don’t Fall During Econ. Rev. a Recession. Cambridge, MA: Harvard U. Press. Calvó-Armegnol, Antoni; Thierry Verdier and Yves Blau, David. 1992. “An Empirical Analysis of Zenou. 2004. “Strong and Weak Ties in Employment Employed and Unemployed Job Search Behavior,” and Crime,” IUI Conference on Networks: Theory Ind. Lab. Relat. Rev. 45, pp. 738–52. and Applications; Vaxholm, Sweden. Blau, David and Philip Robins. 1992. “Job Search Calvó-Armegnol, Antoni and Yves Zenou. 2001. “Job Outcomes for the Employed and Unemployed,” J. Matching, Social Network and Word-of-Mouth Polit. Econ. 98:3 pp. 637–55. Communication,” work. pap., U. Southampton, May; Boorman, Scott A. 1975. “A Combinatorial IZA discus. pap. 771, 2003. Optimization Model for Transmission of Job Coleman, James S. 1990. Foundations of Social Theory. Information through Contact Networks,” Bell J. Cambridge, MA: Harvard U. Press. Econ. Manage. Sci. 6, pp. 216– 49. Conley, Timothy and Georgio Topa. 2003a. “Socio- Borjas, George J. 1992. “Ethnic Capital and Economic Distance and Spatial Patterns in Intergenerational Mobility,” Quart. J. Econ. 107:1, Unemployment,” J. Appl. Econometrics, forthcoming. pp. 123–50. ———. 2003b. “Dynamic Properties of Local ———. 1995. “Ethnicity, Neighborhoods, and Human Interaction Models,” forthcoming in The Economy Capital Externalities,” Amer. Econ. Rev. 85:3, pp. as an Evolving Complex System, III. Laurence 365–90. Blume and Steven Durlauf, eds. Menlo Park, CA: ———. 1998. “To Ghetto or Not to Ghetto: Ethnicity Addison-Wesley. and Residential Segregation,” J. Urban Econ. 44, pp. Corcoran, Mary; Linda Datcher and Greg Duncan. 228–53. 1980. “Information and Influence Networks in Labor Bortnick, Steven and Michelle Ports. 1992. “Job Search Markets,” in Five Thousand American Families, vol. Methods and Results: Tracking the Unemployed,” VIII. Greg Duncan and James Morgan, eds. Institute Monthly Lab. Rev. 115:12, pp. 29–35. Social Research, U. Michigan, pp. 1–37. Bradshaw, Thomas. 1973. “Jobseeking Methods Used Coulson, N. Edward; Derek Laing and Ping Wang. by Unemployed Workers,” Monthly Lab. Rev. 96:2, 1997. “Spatial Mismatch in Search Equilibrium,” pp. 35–40. work. pap. Pennsylvania State U. Bridges, William and Wayne Villemez. 1986. “Informal Crane, Jonathan. 1991. “The Epidemic Theory of Hiring and Income in the Labor Market,” Amer. Ghettos and Neighborhood Effects on Dropping Ot Sociol. Rev. 51:4, pp. 574–82. and Teenage Childbearing,” Amer. J. Sociol. 96:5, pp. Brooks-Gunn, Jeanne; Greg J. Duncan, Pamela Kato 1226–59. Klebanov and Naomi Sealand. 1993. “Do Cutler, David M. and Edward L. Glaeser. 1997. “Are Neighborhoods Influence Child and Adolescent Ghettos Good or Bad?” Quart. J. Econ. pp. 827–72. Development?” Amer. J. Sociol. 99:2, pp. 353–95. Dasgupta, Partha and Ismail Serageldin. 2000. Social Brueckner, Jan. 2003. “Friendship Networks,” work. Capital, A Multifaceted Perspective. Washington pap. econ. dept. U. Illinois. DC: World Bank. dec04_Article 2 12/14/04 2:26 PM Page 1090

1090 Journal of Economic Literature, Vol. XLII (December 2004)

Datcher, Linda. 1982. “Effects of Community and 1999. “The Influence of Neighborhoods on Family Background on Achievement,” Rev. Econ. Children’s Development: A Theoretical Perspective Statist. 64:1, pp. 32–41. and a Research Agenda,” in Neighborhood Poverty: ———. 1983. “The Impact of Informal Networks on Policy Implications in Studying Neighborhoods, vol. Quit Behavior,” Rev. Econ. Statist. 65:3, pp. 491–95. II. Jeanne Brooks-Gunn, Greg J. Duncan and J. de Bartolome, Charles A. M. 1990. “Equilibrium and Lawrence Aber, eds. NY: Russell Sage. Inefficiency in a Community Model with Peer Futia, Carl A. 1982. “Invariant Distributions and the Group Effects,” J. Polit. Econ. 98:1, pp. 110–33. Limiting Behavior of Markovian Economic Models,” Devine, Theresa and Nicholas Kiefer. 1991. Empirical Econometrica 50:2, pp. pp. 377–408. Labor Economics: The Search Approach. NY: Oxford Glaeser, Edward; David Laibson, Jose Scheinkman and U. Press. Christine Soutter. 2000. “Measuring Trust,” Quart. J. Dietz, Robert D. 2002. “The Estimation of Econ. 115:3, pp. 811–46. Neighborhood Effects in the Social Sciences,” Soc. Goyal, Sanjeev. 2003. “Learning in Networks: a Sci. Res. 31, pp. 539–75. Survey,” forthcoming in Group Formation in Econ.: Dodds, Sheridan; Roby Muhamad and Duncan J. Networks, Clubs, and Coalitions. Myrna Wooders Watts. 2003. “An Experimental Study of Global and Gabrielle Demange, eds. Cambridge U. Press. Social Networks,” Science 301:5634, pp. 827–29. Goyal, Sanjeev and Fernando Vega–Redondo. 2004. Durlauf, Steven N. 1994. “Bilateral Interactions and “Structural Holes in Social Networks,” work. pap. Aggregate Fluctuations,” in Proceedings of the First econ. dept. U. Essex. World Congress of Nonlinear Analysts. Granovetter, Mark. 1973. “The Strength of Weak Ties,” ———. 1996a. “A Theory of Persistent Income Amer. J. Sociol. 78:6, pp. 1360–80. Inequality,” J. Econ. Growth 1:1, pp. 75–93. ———. 1983. “The Strength of Weak Ties: A Network ———. 1996b. “Neighborhood Feedbacks, Theory Revisited,” Sociol. Theory 1, pp. 201–33. Endogenous Stratification, and Income Inequality,” ———. 1974(1995). Getting a Job: A Study of in Dynamic Disequilibrium Modelling.W.Barnett, Contacts and Careers, first ed., Harvard U. Press; G. Gandolfo and C. Hillinger, eds. Cambridge second ed., 1995, Chicago: U. Chicago Press. U. Press. ———. 2000. “A Theoretical Agenda for Economic Dutta, Bhaskar and Matthew O. Jackson. 2002. Models Sociology,” forthcoming in New Directions in of the Strategic Formation of Networks and Groups. Economic Sociology. Mauro F. Guillen, Randall Berlin: Springer Verlag, forthcoming. Collins, Paula England, and Marshall Meyer, eds. ———.2003. “On the Formation of Networks and NY: Russell Sage. Groups,” in Networks and Groups. Dutta and Haynie, Dana L. 2001. “Delinquent Peers Revisited: Jackson, eds. Berlin: Springer, pp. 1–15. Does Network Structure Matter?” Amer. J. Sociology Elliott, James. 1999. “Social Isolation and Labor 106:4, pp. 1013–57. Market Isolation: Network and Neighborhood Heavner, D. Lee and Lance Lochner. 2002. “Social Effects on Less-Educated Urban Workers,” Sociol. Networks and the Aggregation of Individual Quart. 40:2, pp. 199–216. Decisions,” NBER work. pap. w8979. Ellison, Glenn. 1993. “Learning, Local Interaction, and Hedström, Peter; Ann-Sofie Kolm and Yvonne Åberg. Coordination,” Econometrica, 61, pp. 1047–61. 2003. “Social Interactions and Unemployment,” Ellison, Glenn and Drew Fudenberg. 1995. “Word of work. pap. 2003:15, Institute Labor Market Policy Mouth Communication and Social Learning,” Quart. Evaluation, Uppsala. J. Econ. 110:1, pp. 93–125. Hogan, Dennis and Evelyn Kitagawa. 1985. “The Erdös, Paul and Alfred Renyi. 1960. “On the Evolution Impact of Social Status, Family Structure, and of Random Graphs,” Math. Institute Hungarian Neighborhood on the Fertility of Black Academy Sci. 5, pp. 17–61. Adolescents,” Amer. J. Sociology 90, pp. 825–55. Evans, William N.; Wallace E. Oates and Robert M. Holzer, Harry. 1987a. “Informal Job Search and Black Schwab. 1992. “Measuring Peer Group Effects: A Youth Unemployment,” Amer. Econ. Rev. 77:3, pp. Study of Teenage Behavior,” J. Polit. Econ. 100:5, pp. 446–52. 966–91. ———. 1987b. “Job Search by Employed and Falk, Armin and Michael Kosfeld. 2003. “It’s All about Unemployed Youth,” Ind. Lab. Relat. Rev. 40:4, pp. Connections: Evidence on Network Formation,” 601–11. work. pap. Institute Empirical Research Econ., ———. 1988. “Job Search Methods Used by Zurich. Unemployed Youth,” J. Lab. Econ. 6:1, pp. 1–20. Fernandez, Roberto and Emilio Castilla. 2001. “How Ioannides, Yannis M. 1997. “Evolution of Trading Much Is That Network Worth? Social Capital in Structures,” in The Economy as an Evolving Employee Referral Networks,” in Social Capital: Complex System II. W. Brian Arthur, Steven Durlauf Theory and Research. Nan Lin, Karen Cook and and David Lane, eds. Menlo Park, CA: Addison- Ronald Burt, eds. NY: Aldine De Gruyter, pp. 85–104. Wesley, pp. 129–67. Finneran, Lisa and Morgan Kelly. 2003. “Social ———. 2001. “Topologies of Social Interactions,” Networks and Inequality,” J. Urban Econ. 53:2, pp. Tufts U., econ. dept. work. pap., revised June 2004. 282–99. ———. 2002. “Nonlinear Neighborhood Interactions Freeman, Richard B. 2002. “The Labour Market in the and Intergenerational Transmission of Human New Information Economy,” NBER work. pap. 9254. Capital,” in Essays in Economic Theory, Growth and Furstenberg, Frank F. and Mary Elizabeth Hughes. Labour Markets: A Festschrift in Honour of dec04_Article 2 12/14/04 2:26 PM Page 1091

Ioannides and Datcher Loury: Job Information Networks 1091

Emmanuel Drandakis. George Bitros and Yannis Kuhn, Peter, and Mikal Skuterud. 2000. “Job Search Katsoulacos, eds. Cheltenham, UK: Edward Elgar, Methods: Internet versus Traditional,” Monthly pp. 75–112. Labor Rev. pp. 3–11. Jackson, Matthew O. 2003. “A Survey of Models of Lalive, Rafael. 2003. “Social Interactions in Network Formation: Stability and Efficiency,” work. Unemployment,” CESifo work. pap. 1077. pap. Caltech; forthcoming in Group Formation in Lin, Nan. 2001. Social Capital: A Theory of Social Economics: Networks, Clubs and Coalitions. Structure and Action. Cambridge and New York: Gabrielle Demange and Myrna Wooders, eds. Cambridge U. Press. Cambridge: Cambridge U. Press. Lin, N.; W.M. Ensel and J.C. Vaughn. 1981. “Social Jackson, Matthew O. and Alison Watts. 2002. “The Resources and Strength of Ties; Structural Factors in Evolution of Social and Economic Networks,” J. Occupational Status Attainment,” Amer. Soc. Rev. Econ. Theory 106:2, pp. 265–95. 46, pp. 393–405. Jackson, Matthew O. and Asher Wolinsky. 1996. “A Lin, N.; J.C. Vaughn and W.M. Ensel. 1981. “Social Strategic Model of Social and Economic Networks,” Resources and Occupational Status Attainment,” J. Econ. Theory 71:1, pp. 44–74. Social Forces 59, pp. 1163–81. Jencks, Christopher and Marsha Brown. 1975. “The Lippert, Steffen, and Giancarlo Spagnolo. 2004. Effects of High Schools on Their Students,” Harvard “Networks of Relations,” U. Mannheim econ. dept. Ed. Rev. 45, pp. 273–324. work. pap. Jencks, Christopher and Susan E. Mayer. 1990. “The Loury, Glenn C. 1977. “A Dynamic Theory of Racial Social Consequences of Growing Up in a Poor Income Differences,” in Women, Minorities and Neighborhood: A Review,” in Concentrated Urban Employment Discrimination. P.A Wallace and A. Le Poverty in America. M. McGeary and L. Lynn, eds. Mund, eds. Lexington, MA: Lexington Books, ch. 8. Washington DC: Nat. Academy. ———. 1981. “Intergenerational Transfers and the Kempe, David; Jon Kleinberg and Éva Tardos. 2003. Distribution of Earnings,” Econometrica 49:4, pp. “Maximizing the Spread of Influence through a 843–67. Social Network,” SIGKDD ’03 Washington, DC. Loury, Linda. 2003. “Some Contacts Are More Equal Kirman, Alan. 1983. “Communication in Markets: A than Others: Earnings and Job Information Suggested Approach,” Econ. Letters 12, pp. 101–108. Networks,” Tufts U. work. pap. ———. 1997. “The Economy as an Interactive Lynch Lisa. 1989. “The Youth Labor Market in the System,” in The Economy as an Evolving Complex Eighties: Determinants of Re-employment System II. Brian Arthur, Steven Durlauf and David Probabilities for Young Men and Women,” Rev. Lane, eds. Menlo Park, CA: Addison-Wesley, pp. Econ. Statist. 71:1, pp. 37–45. 491–531. Manski, Charles F. 2000. “Economic Analysis of Social Kirman, Alan; Claude Oddou and Shlomo Weber. Interactions,” J. Econ. Perspect. 14:3, pp. 115–36. 1986. “Stochastic Communication and Coalition Marmaros, David. 2001. “Analysis of Peer and Social Formation,” Econometrica 54, pp. 129–38. Effects on Employment Opportunities,” Senior Korenman, Sanders and Susan Turner. 1996. Honor Thesis, Department of Economics, “Employment Contacts and Minority–White Wage Dartmouth College, June. Differences,” Industrial Relations, 35:1, pp. 106–22. Marmaros, David and Bruce Sacerdote. 2002. “Peer Kortum, Samuel. 2003. “Book Review: Networks and and Social Networks in Job Search,” Europ. Econ. Markets, by James E. Rauch and Alessandra Casella, Rev. 46:4–5, pp. 870–79. eds.” Russell Sage Foundation, 2001. J. Int. Econ. Marsden, Peter V. 1987. “Core Discussion Networks of 61:1, pp. 249–52. Americans,” Amer. Sociol. Rev. 52, pp. 122–31. Krauth, Brian V. 2000. “Social Interactions, ———. 2001. “Interpersonal Ties, Social Capital, Thresholds, and Unemployment in Neighborhoods,” and Employer Staffing Practices,” in Social econ. dept. Simon Fraser U. work. pap. Capital: Theory and Research. Nan Lin, Karen ———. 2001a. “The Macroeconomics of Job Cook and Ronald Burt, eds. NY: Aldine De Networking,” econ. dept. Simon Fraser U. work. pap. Gruyter, pp. 105–26. ———. 2003a. “A Dynamic Model of Job Networking Marsden, Peter and Karen Campbell. 1990. and Persistent Inequality,” forthcoming in “Recruitment and Selection Processes: The Heterogeneous Agents, Interactions, and Economic Organizational Side of Job Searches,” in Social Performance. Cowan and Jonard, eds. Springer. Mobility and Social Structure. Ronald Breiger, ed. ———. 2003b. “A Dynamic Model of Job Networking NY: Cambridge U. Press, pp. 59–79. and Social Influences in Employment,” econ. dept. Marsden, Peter and Elizabeth H. Gorman. 2001. Simon Fraser U. work. pap.; forthcoming in J. Econ. “Social Networks, Job Changes, and Recruitment,” Dynam. Control. in Sourcebook of Labor Markets: Evolving Structure Kremer, Michael. 1997. “How Much Does Sorting and Processes. Ivar Berg and Arne L. Kalleberg, eds. Increase Inequality,” Quart. J. Econ. 112:1, pp. NY: Kluwer Academic/Plenum Publishers, pp. 115–39. 467–502. Kremer, Michael and Eric Maskin. 1996. “Wage Marsden, Peter and Hurlbert, Jeanne. 1988. “Social Inequality and Segregation by Skill,” NBER work. Resources and Mobility Outcomes: A Replication pap. 5718. and Extension,” Social Forces 66, pp. 1038–59. Kugler, Adriana. 2002. “Employee Referrals and Meade, James. 1976. The Just Economy. London: Efficiency Wages,” IZA discus. pap. 633. George Allen and Unwin. dec04_Article 2 12/14/04 2:26 PM Page 1092

1092 Journal of Economic Literature, Vol. XLII (December 2004)

Mencken, F. Carson and Idee Winfield. 2000. “Job “Resources and Relationships: Social Networks and Search and Sex Segregation: Does Sex of Social Mobility in the Workplace,” Amer. Soc. Rev. 62:5, pp. Contact Matter?” Sex Roles: A Journal of Research, 673–93. 42:9–10, pp. 847–64. Portes, Alejandro. 1998. “Social Capital: Its Origins and Milgram, Stanley. 1967. “The Small World Problem,” Applications in Modern Sociology,” Ann. Rev. Soc. Psych. Today 2, pp. 60–67. 24, pp. 1–24. Miyao, Takahiro. 1978a. “Dynamic Instability of a Ports, Michelle. 1993. “Trends in Job Search Methods: Mixed City in the Presence of Neighborhood 1970–1992,” Monthly Lab. Rev. Oct. pp. 63–67. Externalities,” Amer. Econ. Rev. 68:3, pp. 454–63. Putnam, Robert D. 2000. Bowling Alone. NY: Simon & Möbius, Markus. 2001. “Why Should Theorists Care Schuster. about Social Capital?” unpub. manuscript, Harvard Rees, Albert. 1966. “Information Networks in Labor U. econ. dept. Markets,” Amer. Econ. Rev. Pap. Proceed. 56:2, pp. Moffitt, Robert. A. 2001. “Policy Interventions, Low- 559–66. Level Equilibria, and Social Interactions,” in Social Reid, Graham. “Job Search and the Effectiveness of Dynamics. Steven N. Durlauf and H. Peyton Young, Job-Finding Methods,” Ind. Lab. Rel. Rev. 25, pp. eds. pp. 45–82. 479–95. Montgomery, James D. 1991. “Social Networks and Rosenbaum, J.E.; S. DeLuca, S.R. Miller and K. Roy. Labor Market Outcomes: Towards an Economic 1999. “Pathways into Work: Short- and Long-Term Analysis,” Amer. Econ. Rev. 81:5, pp. 1408–18. Effects of Personal and Institutional Ties,” Sociol. ———. 1992. “Social Networks and Persistent Ed. 72:3, pp. 179–96. Inequality in the Labor Market,” econ. dept. Schelling, Thomas C. 1971. “Dynamic Models of Northwestern U. mimeo. Segregation,” J. Math. Soc. 1, pp. 143–86. ———. 1994. “Weak Ties, Employment and ———. 1978. Micromotives and Macrobehavior. NY: Inequality: An Equilibrium Analysis,” Amer. J. Norton. Sociol. 99:5, pp. 1212–36. Simon, Curtis and Warner, John. 1992. “Matchmaker, Mortensen, Dale and Tara Vishwanath. 1994. “Personal Matchmaker: The Effect of Old-Boy Networks on Contacts and Earnings: It Is Who You Know,” Lab. Job Match Quality, Earnings, and Tenure,” J. Lab. Econ. 1:1, pp. 187–201. Econ. 10:3, pp. 306–30. Munshi, Kaivan. 2002. “The Identification of Network Smith, Sandra. 2000. “Mobilizing Social Resources: Effects: Mexican Migrants in the U.S. Labor Race, Ethnic, and Gender Differences in Social Market,” econ. dept. U. Penn. work. pap. Capital and Persisting Wage Inequalities,” Sociol. Myerson, Robert B. 1977. “Graphs and Cooperation in Quart. 41 4: pp. 509–37. Games,” Math. Op. Res. 2, pp. 225–29. Staiger, Doug. 1990. “The Effects of Connections on Newman, Mark E.J. 2003. “The Structure and the Wages and Mobility of Young Workers,” work. Function of Complex Networks,” SIAM Rev. 45, pp. pap. Harvard U. Kennedy School Govt. 167–256. Stigler, George J. 1961. “The Economics of Oomes, Nienke. 2003. “Local Trade Networks and Information,” J. Polit. Econ. 69:5, pp. 213–25. Spatially Persistent Unemployment,” J. Econ. ———. 1962. “Information in the Labor Market,” J. Dynam. Control 27:11, pp. 2115–49. Polit. Econ. 70:5, pp. 94–105. Oreopoulos, Philip. 2003. “The Long-Run Strand, Jon. 1983. “Equilibrium Wage Distributions Consequences of Living in a Poor Neighborhood,” with Word-of-Mouth Worker Communication,” U. Quart. J. Econ. 118:4, pp. 1533–75. Oslo mimeo. Overman, Henry G. 2002. “Neighborhood Effects in Topa, Georgio. 2001. “Social Interactions, Local Large and Small Neighborhoods,” Urban Stud. 39:1, Spillovers and Unemployment,” Rev. Econ. Stud. 68, pp. 117–30. pp. 261–95. Palmer, E.M. 1985. Graphical Evolution. NY: Wiley Tassier, Troy and Filippo Menczer. 2002. “Social Interscience. Network Structure, Equality and Segregation in a Pellizzari, Michele. 2004a. “Do Friends and Relatives Labor Market with Referral Hiring,” work. pap. U. Really Help in Getting a Good Job?” CEP discus. Iowa econ. dept. Pap. 623, London School Econ. van den Berg, Gerard and Jan van Ours. 1996. ———. 2004b. “Employers; Search and the Efficiency “Unemployment and Duration Dependence,” J. of Matching,” preliminary, London School Econ. Lab. Econ. 14:1, pp. 100–25. Petersen, Trond, Ishak Saporta, and Marc-David L. Waldinger, Roger D. 1996. Still the Promised City? Seidel. 2000. “Offering a Job: Meritocracy and Social Cambridge MA: Harvard U. Press. Networks,” Amer. J. Sociol. 106:3, pp. 763–816. Wasmer, Etienne and Yves Zenou. 1997. “Equilibrium Petrongolo, Barbara and Christopher A. Pissarides. Urban Unemployment,” LSE Centre Econ. Perform. 2001. “Looking into the Black Box: A Survey of the DP 368. Matching Function,” J. Econ. Lit. 39:2, pp. 390–431. Wasserman, Stanley and Katherine Faust. 1994. Social Pissarides, Christopher A. 2000. Equilibrium Network Analysis: Methods and Applications. Unemployment Theory. Cambridge: MIT Press. Cambridge and NY: Cambridge U. Press. ———. 2001. “The Economics of Search,” in Wasserman, Stanley and Garry Robins. 2001. “An International Encyclopedia of the Social and Introduction to Random Graphs, Dependence Behavioural Sciences, Amsterdam: Elsevier. Graphs, and p*,” forthcoming in Models and Podolny, Joel M. and James N. Baron. 1997. Methods in Social Networks Analysis. Peter dec04_Article 2 12/14/04 2:26 PM Page 1093

Ioannides and Datcher Loury: Job Information Networks 1093

Carrington, John Scott and Stanley Wasserman, eds. Bayesian Forks: Reconstructing the Social and Cambridge and NY: Cambridge U. Press. Behavioral Sciences,” Soc. Res. 62:4, pp. 1035–63. Watts, Alison. 2001. ”A Dynamic Model of Network Young, H. Peyton. 1993. “The Evolution of Formation,” Games Econ. Behav. 34, pp. 331–41. Conventions, “ Econometrica 61:1, pp. 57–84. Weinberg, Bruce A.; Patricia B. Reagan and Jeffrey J. ———. 1998. Individual Strategy and Social Yankow. 2000. “Do Neighborhoods Affect Work Structure. Princeton, NJ: Princeton University Press. Behavior? Evidence from the NLSY79,” forthcom- Zuckerman, Ezra W. 2003. “On Networks and Markets ing, J. Lab. Econ. by Rauch and Casella, eds,” J. Econ. Lit. 41:3, pp. White, Harrison C. 1995. “Network Switchings and 545–65.