Job Information Networks, Neighborhood Effects, and Inequality · studies. Using their Dartmouth College data, David Marmaros (2001) and Marmaros and Bruce Sacerdote (2002) found

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

Job Information Networks,Neighborhood Effects, and Inequality

YANNIS M. IOANNIDES and LINDA DATCHER LOURY1

1056

1 Ioannides and Loury: Department of Economics,Tufts University. This is an outgrowth of an invited paperpresented at the American Economic AssociationMeetings, New York, January 5, 1999. We are grateful totwo referees and to the editor, John McMillan, and toAnna Hardman for generous comments and suggestions.We also thank Chris Pissarides for his insightful discussionin New York and Giulio Zanella for useful suggestions.Ioannides acknowledges support from the NationalScience Foundation and the John D. and Catherine T.MacArthur Foundation. We thank Pamela Jia for expertassistance.

1. Introduction

Considerable interest has emergedrecently in the economic literature

about social interactions and the ways inwhich social norms and structures conditionindividual behavior. The treatment of labor-market transactions as very different fromtrading in goods reflects the importance ofidiosyncrasies due to social effects.

One prominent example of where suchidiosyncrasies play a prominent role is job-market search. Search theory formally mod-els frictions associated with job-seekers’access to information about availability ofjobs of different types and about the condi-tions of employment (George Stigler 1961,1962; Christopher Pissarides 2001). Untilrelatively recently, the job-search literaturehas focused on individuals making decisionson a one-to-one basis. Everyday experience

indicates, however, that access to informationis heavily influenced by social structure andthat individuals use connections with others,such as friends and social and professionalacquaintances, to build and maintain infor-mation networks. Albert Rees (1966) firstdrew attention to differences among workersin their use of the variety of available infor-mational outlets. In this context, formalsources of information include state and pri-vate employment agencies, newspaperadvertisements, union hiring halls, andschool and college placement services.Informal sources include referrals fromemployees and other employers, directinquiries by job seekers, and indirect onesthrough social connections. Since then a bur-geoning literature in economics has devel-oped about the details of social interactionsthat affect the job-search process. This liter-ature complements the more extensive soci-ological analysis of networks. One of theobjectives of this article is to explore the rolessocial interactions and social norms play inthe context of this new literature. There is animportant richness that the term “networks”connotes in sociology which to a considerableextent has entered economics as well. A sec-ond objective is to explain its salience withinboth theoretical and empirical economicsresearch.

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

Ioannides and Datcher Loury: Job Information Networks 1057

Section 2 of this paper attempts to organ-ize what we have learned from the empiricalliterature about how individuals go aboutcollecting information for the purpose offinding jobs and how the outcomes areinfluenced by their social connections. Theconventional wisdom that can be garneredfrom much of the empirical literature on jobinformation networks and neighborhoodeffects is organized into a number of broadcategories of seven stylized facts. Section 3starts by reviewing the sociology literatureon job information networks. It then turns,in subsections 3.2 to 3.5, to models ofexogenous job information networks, onesin which individuals obtain job-relatedinformation through a given social structureand, in subsection 3.6, to the consequencesfor job information networks of the recentliterature on evolutionary models of infor-mation transmission. Section 4 reviewsmodels of endogenous information net-works that result from individuals’ uncoor-dinated action, starting in subsection 4.1with the recent literature on strategic net-work formation. We then examine, in sub-section 4.2, endogenous job informationnetworks. Section 5 sketches the outline ofa model that integrates job information net-works and the dynamics of human-capitalformation and thus provides an overarchingtheme for the purpose of examining earnedincome inequality. Section 6 summarizessuggestions for future research and section7 concludes.

2. Stylized Facts about Job InformationNetworks and Neighborhood Effects

The first generation of empirical work onjob information networks established sever-al stylized facts about such networks. Thefirst stylized fact is that there is widespreaduse of friends, relatives, and acquaintancesto search for jobs and this has increasedover time. About 15 percent of unemployedworkers interviewed in the 1970 and 1971monthly Current Population Surveys used

2 This is the same data set as used by Corcoran, Datcher,and Duncan (1980). The job-search categories are thoseemployed by the Current Population Survey (Peter Kuhnand Mikal Skuterud 2000). Unlike the European data (seeMichelle Pellizzari 2004a), they are not mutually exclusive.Unfortunately, in 1993, relevant questions were not askedof all respondents in the PSID, but of only those activelyengaged in job search, who were either unemployed, (5.8percent of the sample), or on-the-job searchers (8.1 percentof the sample).

3 Calculations based on data from surveys of individualsare likely to understate the importance of referrals,because they consider only one side of the job market. Infact, the numbers discussed above are affected by adverseselection, as employers are likely to receive referrals fromcurrent employees.

friends and relatives to search for jobs inthe preceding four weeks (ThomasBradshaw 1973). The 1991 and 1992 CPSfigures were higher at 23 percent (StevenBortnick and Michelle Ports 1992; andPorts 1993). Unemployed and employedworkers were equally likely to use friendsand relatives during periods of job search,according to David Blau (1992) and Blauand Philip Robins (1990). Our own compu-tations using PSID data for 1993,2 reportedin appendix table 1, show that 15.5 percentof the unemployed and 8.5 percent of theemployed check with friends and relatives.This is in a sense surprising because econo-mists tend to think of the U.S. economy asbeing increasingly penetrated by markets,and yet at the same time reliance on friendssuggests persistence of personalizedexchange.

The second stylized fact about job infor-mation networks is that the use of friendsand relatives to search for jobs often variesby location and by demographic character-istics. The 1971 Current PopulationSurveys,3 analyzed by Bradshaw (1973),showed that unemployed women were lesslikely to have checked with friends or rela-tives to find jobs in the preceding fourweeks (12.5 percent) than were men (17.4percent). The same rough difference pre-vailed in 1992: 20.0 percent for women and26.6 percent for men (Ports 1993). See alsoJ. E. Rosenbaum et al. (1999) and Sandra


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

4 The gender differences may, however, vary accordingto the type of contact. Marmaros and Sacerdote (2002)report results on the effects of peer and social network onjob search using a sample of Dartmouth College seniors.Individuals who were randomly assigned as roommateswhen freshmen were asked how they use social network-ing in their job search later on when they reached theirsenior year. Women were less likely to get fraternity/soror-ity help, equally likely to get help from relatives, and morelikely to use help from professors.

5 As in the case of gender difference, racial disparity inusing informal sources may depend on the type of contact.Marmaros and Sacerdote (2002) report that whites weremore likely to report that fraternity/sorority members, rel-atives, and professors were influential in helping them finda full-time job or career. Racial differences are especiallylarge for the last two categories.

Smith (2000).4 More educated job-seekerswere also less likely to use friends and rela-tives. From appendix table 1, 20.8 percentof the unemployed job-searchers who checkwith friends and relatives have more thantwelve years of schooling, and 23.3 percenthad at most eight years. The respective per-centages for those searching on the job are44.8 and 11.6 percent. Clearly, these partic-ular patterns are unlikely to be two sides ofthe same phenomenon. Indeed, differentforces may be affecting workers with moreschooling.

Differences in using informal contacts byage, race, and ethnicity show conflicting pat-terns. According to Ports (1993), about 18percent of 16–19 year-old job-seekers and 22percent of 20–24 year-olds checked withfriends or relatives, compared to about 26.5percent of 45–55 year-olds and 55–64 year-olds in 1992. On the other hand, others(Mary Corcoran, Linda Datcher, and GregDuncan 1980; Peter Marsden and KarenCampbell 1990, and Marsden and JeanneHurlbert 1988) reported that use of informalcontacts declines with age and/or work expe-rience. Harry Holzer (1987a) found onlysmall racial gaps for younger workers. In hisdata, roughly 69 percent of white job-seek-ers ages 16 to 23 used friends and relatives tolook for jobs in the previous year (1982)compared to 67 percent of black male job-seekers.5 More generally, across all agegroups, there were small racial differences in

searching through friends or relatives (23.9for whites and 21.5 for blacks in 1992; Ports1993). Hispanics, however, used such meth-ods more extensively at 32.8 percent.

Methods of job search also vary by location.James Elliott (1999) looks at workers with nomore than twelve years of schooling and findsthat those in high-poverty neighborhoodswere substantially more likely (88 percent) touse informal job-search methods than thosefrom low-poverty neighborhoods (74 per-cent). More generally, our own counts withthe PSID data show that use of informal con-tacts increases with the size of the largest cityin the county where the household resides.While 40 percent of the respondents live inSMSAs with a largest city of at least 100,000inhabitants, 65 percent of unemployed job-searchers and 45 percent of employed job-searchers who used friends and relativesresided in such areas (appendix table 2).

The third stylized fact about job informa-tion networks is that job search throughfriends and relatives is generally productive.Both employed and unemployed workerswho used friends to search for jobs receivedmore offers per contact and accepted moreoffers per contact than did workers whoused other sources of information about jobopenings (Blau and Robins 1990). This couldexplain first why about half of all workersheard about their current job through afriend or relative (Corcoran et al. 1980).Summarizing the results of 24 studies,Truman Bewley (1999) estimated that 30 to60 percent of jobs were found throughfriends or relatives.

Using friends and family may be produc-tive, not only in finding jobs, but also inimproving the quality of the match betweenfirms and workers. Those who found jobsthrough personal contacts were generallyless likely to quit (Datcher 1983, andTheresa Devine and Nicholas Kiefer 1991)and had longer tenure on their jobs (CurtisSimon and John Warner 1992). On the otherhand, however, the estimated effects of jobcontacts on wages vary considerably across



studies. Using their Dartmouth College data,David Marmaros (2001) and Marmaros andBruce Sacerdote (2002) found large positivecorrelations between getting help from fra-ternity/sorority contacts and obtaining pres-tigious, high-paying jobs. Rosenbaum et al.(1991) reported that workers with contactsenjoyed a wage advantage that increasedwith age. In contrast, other work has foundthat an initial wage advantage declined overtime (Corcoran et al. 1980; Doug Staiger1990; Simon and Warner 1992) or found nogeneral initial or persistent earnings effects(William Bridges and Wayne Willemez 1986;Holzer 1987b; Marsden and Hulbert 1988).Adding to the spectrum of results, Loury(2003) and Elliott (1999) showed that at leastsome types of job contacts were correlatedwith lower wages.

The fourth stylized fact about job infor-mation networks is that part of the variationin the productivity of job search by demo-graphic group simply reflects differences inusage. As indicated above, women were lesslikely to use friends and relatives during jobsearch. This could explain why, according toCorcoran et al. (1980), 52 percent of whitemen and 47 percent of all women who werehousehold heads or wives ages 45 and underfound out about their current job from afriend or relative (see also Smith 2000).Hispanic men report more frequent use offriends and relatives for job search than non-Hispanic whites, and are also significantlymore likely to have found out about theirmost recent job through personal contacts(Smith 2000). Looking at differences asneighborhood income varies, Elliott (1999)showed that less-well educated workers inhigh-poverty neighborhoods were more like-ly to use informal contacts and that thesecontacts were also the main avenue by whichthese individuals found work. About 73 per-cent of jobs in neighborhoods with povertyrates of 40 percent of more were foundthrough informal means, compared to 52percent of jobs in neighborhoods withpoverty rates less than 20 percent.

The fifth stylized fact is that many differ-ences in productivity of job search by age,gender, race, and ethnic group cannot becompletely accounted for by differences inusage. Consider employment effects first.According to Bortnick and Ports (1992),men who were unemployed in a givenmonth in 1991 and who used informal con-tacts in that month were slightly more likelythan their female counterparts to have foundjobs (24 compared to 21 percent). The dif-ferences between blacks and whites whoused informal contacts were more substan-tial: 15 percent of blacks who were unem-ployed in a given month in 1991 and whoused informal contacts in that month foundjobs, compared to 24 percent of whites.According to Sanders Korenman and SusanTurner (1996) (see also Rosenbaum et al.1999), employed young black inner-city menwere less likely to have found their jobsthrough friends and relatives, even thoughother analysts report few racial differencesin the incidence of use of friends and rela-tives by job seekers. More specifically,Holzer (1987a) showed that, in 1981, 25 per-cent of previously unemployed African-Americans ages 16–23, compared to 32percent for similar whites, obtained jobsthrough friends and relatives. Most of thisdiscrepancy was due to differences in thelikelihood of receiving offers from jobsheard about through friends and relatives. Infact, almost one-fifth of the total differencein probability of gaining employmentbetween black and white youth resultedfrom racial differences in this probability(see also Smith 2000).

In addition to variation in the relationshipbetween informal contacts and employmentacross groups, there are many demographicdifferences in the effects on wages of jobsearch through contacts. Korenman andTurner (1996) reported that, among youngworkers in inner-city Boston, whites whofound jobs through contacts received muchlarger wage gains, 19 percent higher, thanblacks with similar characteristics. Smith



(2000) showed that gender wage differenceswere small for those using formal job-searchmethods. In contrast, she found larger wagedifferences between Hispanics and whiteswho used personal contacts to find jobs com-pared to those who used more formal means.Korenman and Turner (1996) replicated theresults on Hispanics for a nationally repre-sentative sample of urban youth. Elliott(1999) reported that for less-educated work-ers, the use of informal contacts results insignificantly lower wages.

The sixth stylized fact is new and needs tobe treated as tentative. The internet is beingused increasingly for the purpose of jobsearch, at least according to anecdotal evi-dence. This area is little explored becausedata are only just starting to become avail-able. According to Kuhn and Skuterud(2000), who use data from a special supple-ment to the December 1998 CurrentPopulation Survey, which asked respondentsabout computer and internet use, 13 percentof unemployed Americans and 7 percent ofemployed Americans looked for a new jobvia the internet. However, there appears tobe a “digital divide” for the unemployed:only 7 percent of unemployed Hispanic job-seekers looked for jobs online in December1998, compared with 9 percent of blacks andmore than 16 percent of whites. For thoseemployed, the respective figures are 4, 6,and 7 percent. The gender divide is notnearly as stark. Unemployed women usedthe internet for job search at the same rateas men. Among employed women, 6.7 per-cent looked for jobs on the internet inDecember as opposed to 7.6 percent ofemployed men. However, comparison of thetrends in use of traditional methods of jobsearch (ibid, table 8, p. 10) suggests that useof public employment agencies has declinedfrom 1994 to 1999, although it is overrepre-sented among internet job-seekers inDecember 1998 (op. cit., table 7, p. 9).While we will not pursue this angle furtherin this paper, we underscore that models ofjob search will need to accommodate this

6 The exception of Sweden is unfortunate becauseincluding it might have provided an important benchmark:all job openings are centrally registered in Sweden.

new technology. We argue in section 6 thatthis is an important issue for future research.

The seventh stylized fact is also new butarguably more robust than the sixth. Thereappear to be important differences acrosscountries in the use of personal contacts byboth firms and workers. Pellizzari (2004a)explores the empirical evidence for thecountries of the European Union as of 2003(with the exception of Sweden)6 using theEuropean Community Household Panel andcompares with the United States using theNational Longitudinal Survey of Youth(NLSY). Pellizzari finds large cross-countryand cross-industry variation in the wage dif-ferentials between jobs found through for-mal and informal methods. Across countriesand industries, premiums and penalties areequally frequent. Pellizzari attributes thesedifferences to different recruitment strate-gies by firms. However, such differencescould be attributable to both different insti-tutional and social practices which may com-pound the impact of differences in industrialcompositions of economies.

Taken together, these stylized facts implythat the role of information networks in thejob search process is not straightforward.Neither is it always clear a priori why somegroups rely more on informal methods thanothers, nor why the pattern of employmentand earnings payoffs to networks variesacross groups. Moreover, recent researchindicates that these stylized facts do notexhaust the range of effects of job-searchnetworks. For example, what effects do con-tacts have on wage and employment inequal-ity, the duration of unemployment, andlabor-market withdrawal?

More generally, a common problemunderlies much of the literature on job con-tact use and the correlation of job contactswith labor-market outcomes. It often(though not always) fails to be well-grounded



in economic or other theory about how net-works form or under what circumstancesnetworks are likely to have their largesteffects. In filling this vacuum, much of themore recent detailed research into job net-works (discussed below) points to fourimportant considerations—employer, rela-tional, contact, and worker heterogeneity.Employer characteristics determine thecontext in which job search methods oper-ate. For some employers, desired applicantcharacteristics may be easily discernible,while for other employers recommenda-tions from trusted sources may provide bet-ter information. Contact and relationalheterogeneity respectively denote varia-tions in the resource endowments of one’sassociates and the social relationships thatallow individuals to claim access toresources possessed by their associates.Worker heterogeneity refers to differencesin worker productivity or other characteris-tics. It interacts with all three of the otherareas in determining access to contacts andemployers.

Without the appropriate theoreticalgrounding that pinpoints the role of thesefour considerations, it is difficult, for exam-ple, to interpret demographic variation incontacts use. Does lower contact use bywomen constitute a problem of access to thebest job opportunities, or does it reflect dif-ferences in the ease of observing the types ofskills with which many women enter thelabor market? Does higher contact use byHispanics than blacks signal high-qualitycontacts and, therefore, greater returns toinformal compared to formal methods forHispanics? On the other hand, does it implyan absence of job information alternativesfor Hispanics who are forced to rely oninformal sources as, in some cases, the onlymeans of finding jobs? Interpreting differ-ences in effects of contacts on labor-marketoutcomes generates similar ambiguity. Is thelarger positive correlation between wagesand using contacts for whites compared toblacks reported by some analysts spurious?

That is, do white workers earn more with orwithout using contacts due to higher levelsof unobserved productivity? Does the corre-lation arise simply because contacts aremore likely to pass on job information tosuch workers? Alternatively, do whites gainmore from contacts than blacks becausetheir contacts have access to more informa-tion about better-quality jobs? As theremainder of this article indicates, all four ofthese elements—employer, relational, con-tact, and worker heterogeneity—are criticalto understanding this myriad of commonlyobserved findings as well as to extending therole of job networks to account for otherlabor-market outcomes.

3. Job Information Networks

One of the objectives of this essay is toactually identify and explore mutual influ-ences and patterns of interplay between theliteratures on social networks in economicsand sociology, with particular emphasis onsocial networks that have consequences forthe job market. In sociology, the concept ofa network and use of network models is stan-dard (Ronald Burt 1980). Usage of the termnetworks is perhaps as ubiquitous as that ofmarkets in economics and is used in a com-parably broad range of contexts. In econom-ics, one the other hand, network refers to“personalized exchange among many agents”(Samuel Kortum 2003). The central impor-tance of networks to sociology has forced itto define economic networks more preciselyas: 1) particular patterns in economicexchanges; 2) indications of “primordial”relationships among agents, and 3) “struc-tures of mutual orientation” (EzraZuckerman 2003). The third definitionencompasses the previous two as specialcases: it allows for social ties that differ interms of “type, valence and strength of con-nection” (ibid). Networks in sociology arenot just an important descriptive device butalso a tool to explore direction of causalityand the role of unobserved heterogeneity.



7 The notion of social capital has been used extensivelyto account for a wide variety of outcomes (ParthaDasgupta and Ismail Serageldin 2000; Robert Putnam2000; Edward Glaeser et al. 2000). The two key elementsof social capital include the resource endowments of one’sassociates and the “social relationship itself that allowsindividuals to claim access to resources possessed by theirassociates” (Alejandro Portes 1998; Glaeser et al. 2000;Putnam 2000). Additional evidence of the popularity of theconcept is given by Markus Möbius (2001), who uses theterms social capital and social networks interchangeably.

In contrast to the extensive treatment ofnetworks by sociologists, economists’ formalas well as qualitative understanding of therole of networks is much less developed.There has also been influence in the otherdirection, however, from economics to sociol-ogy. In at least one instance, articulation ofpatterns in interpersonal interactions rele-vant to economic outcomes by economistshas influenced sociological methodology.This is the case with the concept of socialcapital, which is particularly relevant in ourcontext. Social capital was originally definedby Glenn Loury (1977) as the set of resourcesresulting from family relationships and com-munity social organization that affect the cog-nitive and social development of the young.It has found much fertile ground in sociology.James Coleman (1990), who helped popular-ize the concept, treats social capital as a formof social organization created when the struc-ture of relations among persons facilitatesaction making possible the achievement ofcertain ends that in its absence would not bepossible (ibid, p. 300). The term social capitalis now used quite commonly to refer to theweb of connections individuals use in dailylife and is therefore important to mentionhere.7 Some economists use the terms socialcapital and social networks almost inter-changeably, which for reasons of clarity we donot adopt in the remainder of this essay.

Another instance of interplay betweeneconomics and sociology is found in themost recent economics literature on jobinformation networks and reflects the influence of the empirical findings of MarkGranovetter (1974:1995). Granovetter’s

empirical findings in support of the role ofweak contacts are widely cited in both theeconomics and sociology literature. The sec-ond edition (1995) of this book contains areview of the evidence since the 1974 study.Granovetter argues that the voluminous evi-dence that has accumulated since 1974 con-tinues to support his book’s original thesisthat social networks are very important inhelping people to find jobs and employers tofind prospective employees. In particular,local contacts, “weak ties,” help women “inwomen’s occupations” locate jobs, and that“networking is crucial for expandingwomen’s opportunities in male-dominatedoccupations” (ibid, p. 170). He concludesthat in spite of “an outpouring of research,”we still know little about the complex net-work processes by which “inequities areproduced and reproduced.”

This paper is focused on job informationnetworks as patterns of exchange of job-relat-ed information. We look at these patternseither as given (“primordial”), or as the out-come of deliberate decisions by individuals tohelp us untangle the complex and seeminglycontradictory findings about use of job infor-mation networks in the labor-market contextand their effects on employment and wages.The paper shows how network considera-tions imply different outcomes than the sim-ple one-to-one search models that typifymost economic analyses of job acquisition.

3.1 The Sociology Literature on JobInformation Networks

The sociological literature includes a richarray of analysis of three of the categories ofjob network effects—employer, contact,and relational heterogeneity. Much of theinitial research to explain the operation ofjob networks focused on relational hetero-geneity. Granovetter (1974:1995) arguedthat a key characteristic determining theeffect of job networks on finding employ-ment is the strength of social ties. Roughlyspeaking, strong links join close friends andweak links join acquaintances. Strong links



tend to traverse a society “slowly.” If youstart with an arbitrary person and developthe network of links to her close friends, andthen to the close friends of her close friendsand continue in this manner, then the over-all size of the group grows slowly. The closefriends of my friends are likely to be myclose friends too. Societies of closely knitrelations are likely to develop a large num-ber of closely knit groups. If, on the otherhand, we track my acquaintances and theiracquaintances in turn, it is less likely thatthe acquaintances of my acquaintances arealso my own acquaintances. As sociologicalresearch has measured (Stanley Milgram1967) and random graph theory has demon-strated (E. M. Palmer 1985; Mark Newman2003) any two individuals in the UnitedStates can be connected by as few as sixweak links. So, the overall size of the groupof interconnected individuals will growfaster with weak ties. Sheridan Dodd, RobyMuhamad, and Duncan Watts (2003) show,based on a recent global internet-basedsocial search experiment, that successfulsocial search is conducted primarily throughintermediate-to-weak strength ties, reliesdisproportionately on professional relation-ships, and reaches its destination in a medi-an five-to-seven steps, depending on theactual distance between source and target.

Granovetter’s original work and the 1995edition of the 1974 study have been readvery widely. Within sociology the notion oftie strength has been incorporated intomore general analyses of job networks andsocial capital. Yet, Granovetter’s work hasbeen much less influential in promotingmodelling of job information networks with-in economics. This is not so surprising: econ-omists are typically more likely to beinterested in why a network forms, whereassociologists take the existence of networks asgiven and study their effects. In the firstsubsection below, we explore our under-standing of job information networks fromthe interplay between the sociology and eco-nomics literatures. Then we turn to the eco-

8 As an aside, to economists this argument is akin to anarbitrage-type argument and thus amenable, in principle,to analytical treatment. A very recent paper by SanjeevGoyal and Fernando Vega-Redondo (2004) applies thestrategic network formation approach to the concept of astructural hole. A noteworthy result here is, as we discussfurther below, that if the cost for forming links is not verysmall, the unique equilibrium network is a star.

nomics literature on exogenous job informa-tion networks and on neighborhood effectsthat have consequences for job markets.

The notion of weak versus strong ties is, ofcourse, one of the most important conceptsthat have motivated the literature on theeffects of social contacts on job outcomes.Burt (1992) argues that, while tie strength iscorrelated with information benefits, thetrue causal agent in social networks is thestructural hole. Burt defines a structuralhole as the “gap,” the separation, betweennonredundant contacts. Regardless of tiestrength, individuals located at structuralholes provide a bridge for information toflow between groups that would otherwisenot have access to each other. For example,suppose group A consists of individuals who often contact each other and would beconsidered to have strong ties in theGranovetter framework. The same is true forthe members of group B. Group A is linkedto group B only through the individual at astructural hole in group A. This link allowsinformation to flow between otherwise inac-cessible groups. Note, however, that therelationship between the group A individualat a structural hole and his nonredundantcontact in group B may be strong or weak.The key factor is not the strength of the tiebut that the individual occupying the struc-tural hole bridges the gap between groups Aand B. The structural holes argument8

implies that: 1) a network with more nonre-dundant contacts can provide more informa-tion than the same size network withredundant contacts; and, 2) a network with agiven number of nonredundant contactsprovides more information if, in turn, thosecontacts “reach separate and therefore morediverse social worlds” (ibid, p. 21 ).



Burt (2001) also argues that the nonre-dundant information that structural holespossess is better communicated to, and actedupon, in networks with closure. He definesclosure as “social capital” that is created by anetwork of strongly interconnected ele-ments. With high levels of closure, everyoneis cohesively connected to everyone elsewithin the network. Information passed to asubgroup within the network quickly finds itsway to the remainder of the network. Whenthere is closure, referred workers may havehigher job productivity because their reputa-tion within the network is more contingenton their performance.

Why does closure, whereby agents areeffectively interconnected in a cyclical fash-ion (Coleman 1990), and not just social con-nectedness play such a key role insociological thinking? Here is a guess.Consider a social setting where people tendto imitate one another, as in the spread of afashion. When everyone is directly connect-ed with everyone else, different individualsstarting fads are unlikely to have decisiveinfluence on each other. Alternatively, con-sider individuals connected along a path—where each individual is connected to twoothers and there are two end agents—inwhich case everyone is at least indirectlyconnected with everyone else. Such anarrangement confers a modicum of influ-ence to the two end agents, whose ownactions are influenced by one other agentonly. If those two end individuals were to beconnected so that the topology of the socialstructure becomes a wheel, there is nolonger any influence to be conferred by thetopology of the social structure as such, andthis is accomplished with the minimum totalnumber of connections.

In spite of its popularity, the sociologicalconcept of closure is to the best of ourknowledge not very rigorously developed.An exception is Steffen Lippert andGiancarlo Spagnolo (2004), who provide rig-orous support for the disciplining role ofclosure in sociology by means of a game-

9 Ioannides (2001) shows that symmetry—technicallyspeaking, all individuals’ being connected to the samenumber of other individuals—ensures that the maximaleigenvalue of the corresponding graph is equal to thenumber of neighbors. Therefore, if individuals are influ-enced by the average action among those they are con-nected with, then the respective dynamical system has aneigenvalue of 1 and an associated eigenvector that consistsof 1’s, which ensures (relative) persistence.

theoretic model of networks of relationalcontracts. They examine networks of rela-tions under different informational regimes,paying special attention to differencesbetween circular and noncircular architec-tures. If agents cannot discipline themselveswithin a certain relation and are allowed tohave relations with only two other agents(“neighbors”), a “circular pooling of asym-metries” made possible by a social networkmay end up sustaining all the relationshipsin equilibrium.

There are properties of social structuresthat have been attributed to closure that mayinstead be explained by symmetry in the net-work that represents social structure, ratherthan closure.9 For example, this is the casewhen individuals who act similarly are morelikely to be connected with one another thanwith others who act differently. Such patternsof correlation in the behavior of connectedindividuals resemble spatial waves and exhib-it clustering. Clearly, this is an area whereadditional research would be very fruitful.

Other sociological research contends thatthe role of relational heterogeneity cannotbe fully understood without taking intoaccount the role of social setting or contactheterogeneity. Nan Lin’s (2001) “strength ofposition” proposition argues that individualsare more likely to associate with others insimilar social and occupational positions; inother words, they sort. This propositionimplies that social networks develop alongdimensions such as race, ethnicity, religiousaffiliation, and education. The emphasis inLin’s work is on the characteristics of thecontacts themselves. Lin’s “social capital”proposition claims that contacts who possess,or have access to, more highly valued



resources improve outcomes for job-seekersmore than other less well-placed contacts.

Lin, Vaughn, and Ensel (1981) and Lin,Ensel, and Vaughn (1981) report that thefamily background, education, and earlyoccupational status of job-seekers all influ-ence the occupational status of the contactsthey can use. Furthermore, the occupationalstatus of these contacts alters the prestigethat job-seekers obtain. In related work, F.Carson Mencken and Idee Winfield (2000)emphasize that women who found their jobsthrough informal contacts and who usedmale contacts were less likely to work infemale-dominated occupations. EchoingMencken and Winfield, John Beggs andHurlbert (1997) indicate that the gender ofthe informal contact tie affects occupationalstatus. Women whose contacts are otherwomen work in occupations with lowersocioeconomic index scores. Evidence of cor-relation between contact characteristics andjob contact effects is not, however, uniform.While Marsden and Hurlbert agree with Lin,Vaughn, and Ensel (1981) and with Lin,Ensel, and Vaughn (1981) with respect tooccupational prestige, they conclude that“the net effects of the social resource vari-ables (for other outcomes such as wages) canbe summarized simply: there are none.” Theyconclude that these findings are consistentwith those of Bridges and Villemez (1986),who find no effects of tie strength on income.

The third branch of sociological researchexamines the role of employer heterogene-ity on contact effects. It points to importantdifferences in the use of referrals in differ-ent industries, perhaps reflecting differ-ences in corporate culture. The sociologyresearch in this area appears to be ahead ofits counterpart in economics in assessing theconsequences of employer heterogeneity.According to Roberto Fernandez andEmilio Castilla (2001), employers reapreturns from referrals for three reasons.Referrals provide a large pool of qualifiedapplicants so that less screening is requiredto fill positions. Referred applicants have

more information about the nonpecuniaryaspects of employment and, therefore, arepotentially better matches. Finally, connec-tions between new hires and incumbentemployees can make the job transitionsmoother, as well as create additional loyal-ties and attachments to the job. More gen-erally, Peter Marsden and ElizabethGorman (2001) argue that information pro-vided to employers through referrals mayreduce employer uncertainty about theprospective worker’s productivity.

The effects of job contact information are,therefore, likely to be higher in firms andindustries where high-quality informationabout workers’ likely performance is impor-tant. These include conditions “when per-formance and skills are difficult to observe,when staffing strategy is flexible, when theuse of networks is a central component ofperformance, and when selection errors arecostly (p. 108).” Consistent with this predic-tion, Marsden and Gorman present evidencethat contacts are more likely to be used whenfilling managerial, professional, or sale/serv-ice positions and are less likely to be used inthe public sector and in establishments thatare part of multi-site firms. Complementarywork by Elliott (1999) found differences ininformal contacts by occupation. Generallaborers (defined as all nontechnical posi-tions lacking managerial authority, 1990Census Occupation Codes 243-902) andthose in managerial positions who foundtheir jobs through informal contacts had sig-nificantly lower wages than those who foundtheir jobs without an active search.

Intrafirm analyses of job contact effectsalso point to the importance of context.Trond Petersen, Ishak Saporta, and Marc-David Seidel (2000) use data on all 35,229job applicants to a mid-sized high-technolo-gy organization and find that all differencesin job offers that are attributed to genderdisappear once age and education areaccounted for. Similarly, all effects of racedisappear once the referral method is takeninto account. That is, when one controls only



for age, education, and rating at first inter-view, race has a strong impact on the likeli-hood of having a second interview and onthe increase in the salary offer. In their data,personal and professional contacts accountfor 60.4 percent of applicants and 80.8 ofthose receiving offers, and there is also aneffect of being contacted by headhunters.Thus, the apparent and perhaps real meri-tocracy characterizing the high-technologyindustry raises additional questions on therole of access to job information networks asa force in persistent inequality.

Using data from a high-technology firm,Joel Podolny and James Baron (1997) findthat intra-organizational mobility isenhanced by having a large network of infor-mal ties that supply access to informationand resources. Yet, availability of a smalldense network of social contacts with highclosure and cohesiveness is no less importantin helping shape one’s organizational identityand career goal expectations. This highlightsthe importance of how social network struc-ture and content interact in determiningcareers within organizations.

Differences between industries andemployers may also account for ethnic andrace variations in contact effects. RogerWaldinger (1996) shows that, both historical-ly and currently, ethnic groups have estab-lished specific occupational and employmentniches that facilitate employment and train-ing of members of their group and that limitaccess of outsiders. Ethnic newcomers toNew York found their way to the bottom ofthe job ladder associated with the niche andthen gradually work their way upwardthrough the specialized economic activitiesassociated with the niche. Early examples ofethnic niches include Jews in commerce andclothing manufacture and Italians in labor-ing jobs in construction and longshoremanwork. More recently, occupational nichesinclude African-Americans in the public sec-tor, West Indians in hospitals, nursinghomes, and health services, Chinese inrestaurants, laundries, the garment trade,

and small-scale retail trade, and Dominicansin garments, restaurants, hotels, and a fewother light manufacturing and retail trades.

These ethnic-group niches have importantimplications for the usefulness of job con-tacts and connections for ethnic-groupmembers entering the labor market orchanging jobs. Consider, for example, thelarge concentration of African-Americans inpublic service. On the plus side, it reducestheir experience of discrimination and thusraises black earnings relative to whites.Resulting job contacts and connections maylead to a greater likelihood of job offers andto more rapid career advancement uponaccepting the offers. On the minus side, thelarge concentration of African-Americans inthe public sector is mirrored by their declin-ing presence in the private sector and thus isassociated with lower access to contacts, net-works, and training opportunities. Moreover,if the public sector niches held by African-Americans require relatively high levels ofskill, job information about these niches maynot be especially useful for low-skilledAfrican-Americans.

3.2 Models of Exogenous Job InformationNetworks

Models derived from research by econo-mists on exogenous job networks have out-lined the specific implications of socialstructure in more detail than some of thesociological work described above. By exoge-nous, we mean that the network of connec-tions (or, in more mathematical language, thegraph that describes these connections)among individuals is given. Dale Mortensenand Tara Vishwanath (1994) show that theequilibrium wage distribution increases withthe probability that the offer is from a con-tact. Their argument is based on the premisethat wages received from jobs found throughcontacts reflect the distribution of wagesearned by individuals who are in contact withone another. This distribution, in turn, sto-chastically dominates the distribution ofwage offers across employers. Like Bridges



and Villemez (1986) referred to earlier,James Montgomery (1992) examines the linkbetween wages and tie strength and con-cludes that, while weak ties increase reserva-tion wages, this does not imply a similar“relationship between wages and the type oftie actually used to find a job.” Intuitively,weak ties are more likely to generate offersthan strong ties. Workers who accept weak-tie offers are likely therefore to have receivedfewer total offers and be less selective in thejobs that they choose than those who acceptstrong-tie offers. This indicates that theempirical finding of no relationship betweentie strength and wages in the Bridges andVillemez study does not imply that tiestrength is irrelevant for determining joboutcomes. Montgomery’s work is a notableexample of research by an economist that hashelped bridge the gap between economicsand sociology in this area.

A number of more general models exam-ine additional detail about the interactionbetween contact and relational heterogene-ity. They highlight the specific characteris-tics of relations between contacts and jobseekers that alter contact effects. An exam-ple of such a model is Montgomery (1991),who suggests that the main social compo-nent is inbreeding social bias. That is, eachperson is more likely to have a social tie to a younger person of the same type as her-self. Thus, a social tie implies that a referralpossesses informational value. InMontgomery’s model, each individual livesfor two periods, making an education deci-sion in the first period, which is observable,and working in the second. Individuals maybe of two types. Each individual knows atmost one person in the older (and currentlyemployed) generation, possessing a socialtie with probability �. Conditional on hold-ing a social tie, a worker knows someone ofthe same type with probability �, �� .Some young persons may have severalsocial ties while others have none. Thosewho do have social ties receive offers fromthe employers of their acquaintances, but

12

those who do not are hired through the for-mal market. Firms may choose technologythat makes either type fully productive,except that choice along with that of thewage rate must be made before the worker’stype is known.

The model is closed by equating the per-centage of those educated with the percent-age of those facing education costs who findit advantageous to acquire information. Thepossibility of multiple equilibria dependscritically upon the properties of the distribu-tion of education costs across the population.A key element in Montgomery’s theory is thederivation of the probability that an individ-ual with a referral accepts a job offer, as afunction of the offer: a(w)�e–�n(1–F(w)),where F(w) is the distribution function ofreferral wage offers and n the steady-statefraction of educated workers. In this model,a higher probability of a social tie and ahigher percentage of educated workersdecrease the probability of acceptance butincrease wage dispersion. While the formeris a straightforward supply effect, the latteris subtle. Increased inbreeding by a group isshown to be associated with larger differ-ences from other groups. Selection operatesover time via network density parametersand inbreeding. Individuals pass on theiradvantages to kin and social acquaintances.These factors work to perpetuate andstrengthen inequality over time. KennethArrow and Ron Borzekowski (2004) show bymeans of simulations that differences in thenumber of ties workers have with firms caninduce substantial inequality and can explainroughly 15 percent of the unexplained varia-tion in wages and a substantial part of thedisparity between black and white incomedistributions.

Kaivan Munshi (2002) studies transmissionof job information among Mexican migrantsto the U.S. labor market. He uses a model ofreferrals similar to Montgomery’s and exam-ines changes in the size and vintage withingiven destination communities in order todetermine whether those characteristics



10 Calvó-Armegnol, Thierry Verdier, and Yves Zenou(2004) explore the concept of a dyad in the context of thelabor market and, in addition, introduce criminal activitiesas an option within a given social structure.

affect migrant employment. His findings con-firm that community-based social interactionsare important in matching Mexican migrantsand U.S. firms, and appear to improve labor-market outcomes among migrants, withsmaller and younger networks substantiallyreducing the employment probabilities ofMexican migrants.

Montgomery (1994) models the impact ofsocial interaction on employment transitionsand inequality in a way that links the notionof strong versus weak ties to the social struc-ture. In his model, social structure consists ofa large number of small groups, or dyads,groups of two connected individuals. In eachdyad, both members are employed, or onlyone is employed, or both are unemployed.The relationship between two members of adyad is a strong tie. Individuals interact withothers at a rate � per unit of time, and suchinteraction may lead to a match with one’sdyad partner with probability �, or it maylead to a random match with someone elsefrom the entire population, with probability1��. In the latter case, because of randommatching, the individual has an infinitesimalprobability of interacting again with the sameperson. For this reason, random matches areconstrued as weak ties. By varying the proba-bility of interaction with one’s strong tie, onemay, in effect, model different social settings.Jobs break up randomly at a rate �, so thateither a strong or weak tie may be employedor unemployed. The job finding rate for eachperson depends on the employment status ofher contact and on her general ability to col-lect information on job openings. Althoughindividuals who make up each dyad do notchange over time, their employment statusdoes change and so does the employment sit-uation in each dyad.10 As in Montgomery’searlier work, social interactions are character-ized by inbreeding bias, whereby an unem-ployed individual’s random match contact is

11 In a world where individuals have established rela-tions with one another, the resulting social structure maydisplay certain “holes.” Such structural holes offer entre-preneurial opportunities for information access, timingreferrals and control. See Burt (1992).

employed with probability less than or equalto the average employment rate in the popu-lation. The combination of a rigid socialstructure with inbreeding bias in employ-ment status implies that random matching isless useful to individuals and the employmentstatus of one’s dyad partner is critical. Usingthis model, Montgomery shows that a higherproportion of weak-tie interactions reducesemployment inequality. It also increases thesteady state employment rate, provided thatinbreeding by employment status amongweak ties is sufficiently small.

It would be interesting to generalize themodel of social structure employed byMontgomery, by assuming groups of differ-ent sizes. For example, one may invoke arandom graphs setting (Paul Erdös andAlfred Renyi 1960; Ioannides 1997), where afraction of the entire economy may be ingroups whose sizes are denumerable butpossibly large. However, unless membershipto groups of different sizes confers particularadvantages of a permanent nature with con-sequences for economic inequality, such aspermanent effects on the incidence of unem-ployment, a model with a more complicatedsize distribution of social groups will not addmuch to the story. The literature must alsoaddress the fact that the availability of differ-ent informational technologies might pro-vide incentives for strategic behavior byindividuals in possession of job-related infor-mation. Such strategic issues in informationdissemination are not well understood. Thisis particularly evident, as we see furtherbelow, when social networks are the outcomeof individual uncoordinated decisions.11

3.3 The Work of Calvó-Armegnol andJackson

Two recent path-breaking papers, AntoniCalvó-Armegnol and Matthew Jackson (2002;



2003), explore the implications of exogenousinformation networks. Both papers use thefollowing model of the transmission of jobinformation among workers. A network ofcontacts among n individuals, the social struc-ture, is defined by means of an n�n matrixG, of intensities of social attachments: gij0,if i is linked to j; gij�0, if i is not linked to j.This formulation combines the notion of anadjacency matrix in graphs with the notion ofvarying intensities of social contacts andallows for the network to be directed. That is,if individual i hears of a job opening, she tellsindividual j if gij0 . However, unless gji0 ,individual j will not pass on such informationto i. The transmission of job information toeach worker through the network at thebeginning of each period results in newemployment. An unemployed individual whohears of a job opening keeps it to himself. Anemployed individual who hears of a job open-ing passes it on to each of her social contactswith probabilities that are proportional to therespective relative weights. For example, anindividual i’s unemployed contact j will hearfrom i with probability equal to the productof the probability that i hears directly of a jobopening, which is assumed to be a function ofall agents’ wages in the previous period, timesthe relative weight of j’s social strength, whichis equal to gij divided by the sum of theweights of all contacts who are unemployed.The weights of social attachment express acontinuous counterpart of Granovetter’snotion of strong versus weak ties, and thusgeneralize it.

Calvó-Armegnol and Jackson (2002)assume that the expected number of offersthat agent i receives is a nondecreasingfunction of the wages of that agent’s con-tacts in the previous period and a nonin-creasing one in agent i’s own wage. Theseassumptions imply a set of “altruistic” valuesabout social exchange that influences thepassing around of job-related information.Calvó-Armegnol and Jackson determine anagent’s wage by assuming that it is a nonde-creasing function of the past wage and of

the number of employment opportunitiesthe agent has in hand. This function allowsfor substitution among previous wage statusand current employment prospects.Turnover in the job market is ensured byrandom breakup of jobs. The paper distin-guishes the effect of social connectedness(one agent’s expected job offer is sensitive toanother agent’s past wage and vice versa),and the passing of job-related informationbetween agents.

Using this model, Calvó-Armegnol andJackson develop explanations for severalimportant stylized facts about labor markets.First, information passed from employedindividuals to their unemployed acquain-tances makes it more likely that theiracquaintances will become employed. Thisgenerates positive correlation betweenemployment and wages of networked indi-viduals within and across periods. Second,duration dependence and persistence inunemployment may be explained by recog-nizing that when an individual’s direct andindirect social contacts are unemployed, thelikelihood of obtaining information aboutjobs through contacts is reduced. Such dura-tion dependence is well-documented; see,for example, Devine and Kiefer (1991), LisaLynch (1989), and Gerard van der Berg andJan van Ours (1996). Third, the likelihood ofdropping out of the labor force is higher foran individual whose social contacts havepoor employment experience. This can leadto substantial differences in drop-out ratesacross groups. Moreover, small differencesin initial conditions of different individualsand in network structure can lead to largedifferences in drop-out rates. Fourth, higherinitial drop-out rates for a set of networkedindividuals imply that its short-run as well asits steady state distribution of unemploy-ment and wages will be worse (in the senseof first-order stochastic dominance).Contagion effects then cause inequality inwages and employment, because thoseremaining in the labor force will have fewerdirect and indirect acquaintances on the job,



and that in turn will hamper their job infor-mation prospects. It is hard to explain suchan effect outside of a social network model.

Particularly noteworthy are results thatCalvó-Armegnol and Jackson (2002) obtainabout the key role of drop-out rates. To doso, they work with a specific case of themodel in Calvó-Armegnol and Jackson(2003). In it, all jobs are identical and theyoffer equal wage rates. Only one agentreceives information about a job openingand only if she is unemployed. The paperpresents a number of examples, each withdifferent explicit structures, that helpdemonstrate the subtle role of networks.For example, if three agents are connectedaccording to a path, the two end agents arecompetitors for job information in the shortrun, but their outcomes are positively corre-lated in the long run. That is so becausetheir presence helps the center agent returnto employment if she becomes unemployed.With more complex social network topolo-gies, an agent’s likelihood of being unem-ployed depends on her position within thenetwork. The average unemployment rateincreases with “close-knitness,” becausemore extensive social ties make possiblegreater diversification of informationsources. Calvó-Armegnol and Jackson provethat, under certain general conditions,employment status across any arbitraryperiods is correlated among all intercon-nected agents and that there is durationdependence in unemployment.

These authors explain duration depend-ence as a social effect: the longer an individ-ual is unemployed the more likely it is thather “social environment is poor,” makingfuture employment prospects unfavorable.This explanation for duration dependencecomplements the more commonly statedones, such as unobserved heterogeneity andthe like. This effect, essentially a networkexternality, is also responsible for stickinessin aggregate employment dynamics. Thecloser the economy is to very high employ-ment (or unemployment), the harder it is to

12 Unfortunately, the technical problem of findingwhich individuals should be targeted is computationallyvery difficult in general (David Kempe, Jon Kleinberg, andÉva Tardos 2003).

13 The importance of network effects on the drop-outrate is also argued by D. Lee Heavner and Lance Lochner(2002), though the results are not as dramatic.

leave that state. For similar reasons, parts ofthe economy can experience a boom whilesimultaneously other parts of the economyare experiencing a bust.

Differences in initial conditions combinewith differences in the collective employ-ment histories and with different networkdynamics of two otherwise identical net-works to produce sustained inequality ofwages and drop-out rates that feed on eachother. So, in the model of the two Calvó-Armegnol and Jackson papers, history mat-ters and is responsible for producing incomeinequality for reasons that are very differentfrom those due to inequalities in human-cap-ital investments (Loury 1981), or in access tocommunity-based opportunities (StevenDurlauf 1996a,b). This implies that interven-tions in the labor market, such as providingincentives for individuals not to drop out, arelikely to have long-lasting effects.

The design of such interventions shouldreflect the topology of the network. It wouldbe more effective to target groups of agentswho are highly connected, taking advantageof social attachment effects among agents,instead of targeting the same number of uni-formly connected individuals.12 Similarly,institutions that seek to “network” otherwiseisolated individuals can potentially bringabout socially desirable outcomes.13 Theseresults depend on essentially altruisticassumptions about social exchange in thecontext of social interactions, in ways that areconceptually similar to spatial interactions asmodelled by Thomas Schelling (1971, 1978).

3.4 Proximity, Information Sharing, andNeighborhood Effects

Several recent economic studies empha-size network effects as neighborhood



effects: they examine whether it is appro-priate to associate geographic proximitywith facilitation of information flow. Thisquestion, of course, is identical to the ques-tion addressed by the empirical literatureon the economics of social interactions.Even if we have compelling evidence ofcorrelations in the behavior of individualswho are in physical and social proximity toone another, we wish to know what explainssuch correlations. That is, we wish to knowwhether we see correlations among suchindividuals because they share the samesources of information (a correlatedeffect), because they share individual char-acteristics as a result of self-selection (acontextual, or exogenous social, effect), orbecause they learn from one another’sbehavior (an endogenous social effect)(Charles Manski 2000).

Georgio Topa (2001) finds geographiccorrelations in patterns of unemploymentacross neighborhoods and cites them as evi-dence of positive correlation betweenemployment and wages of networked indi-viduals. He points out that high unemploy-ment rates were concentrated in relativelyfew areas of Chicago in 1980 and 1990.Using census tracts as units of observation,he assumes that residents of adjacent tractsexchange job information. He finds thathigh unemployment in one tract is associat-ed with more unemployment in neighbor-ing tracts than can be explained by thecharacteristics of the neighboring tractsalone. Timothy Conley and Topa (2003a)find that socioeconomic characteristics (andin particular ethnic and occupational dis-tance) explain a substantial component ofthe spatial dependence in unemployment.Using similar data for the Los AngelesStandard Metropolitan Statistical Area(SMSA), Conley and Topa (2003b) showthat local interactions perform well inexplaining the spatial correlation patternspresent. In addition, using data on the dis-tribution of individual unemployment spellsin-progress, they show that their model of

interactions is consistent with the Calvó-Armegnol and Jackson (2003) explanationof duration dependence in unemploymentdiscussed above.

A conceptually related study by PeterHedström, Ann-Sofie Kolm, and YvonneÅberg (2003) emphasizes transitions out ofunemployment, using the Pissarides model(Pissarides 2000) with data on all 20- to 24-year-olds living in Stockholm during the1990s. Both of those papers produce evi-dence for partial identification of socialinteractions in unemployment.

Bruce Weinberg, Patricia Reagan, andJeffrey Yankow (2000) provide some evi-dence of non-monotonic neighborhoodeffects on labor-market outcomes. They linkconfidential street address data from theNLSY79 with measures of neighborhoodsocial characteristics at the census tract levelfor 1990 and measures of job proximitybased on the 1987 censuses of manufactur-ing, retail trade, and services. They show thatone standard deviation increase in neighbor-hood social characteristics and in job proxim-ity raises individuals’ hours worked by 6percent and 4 percent in the average, respec-tively. Such social interactions have nonlin-ear effects. The greatest impact is in theworst neighborhoods. Being in a disadvan-taged neighborhood is more important thanthe labor activity of one’s neighbors per se.Social interaction effects are also larger forless-educated individuals and for Hispanics,but not for blacks compared to whites.

While these works are in broad agree-ment, recent research by Philip Oreopoulos(2003) finds that when neighborhoods arenot selected, neighborhood quality plays lit-tle role in determining a youth’s eventualearnings, likelihood of unemployment, andwelfare participation. In contrast, family dif-ferences, as measured by sibling outcomecorrelations in a relatively homogeneoussample of low-income families living inToronto public housing, account for up to 30percent of the total variance in earnings.These findings are particularly significant



because the respondents in that study hadbeen administratively assigned to differentpublic-housing residences, and therefore theassignment process should have removedmuch of the selection across neighborhoodtypes, according to the author.

The recent study that makes the strongestand most compelling case to date for theeffects of geographical proximity on job mar-ket outcomes is Patrick Bayer, Stephen Ross,and Georgio Topa (2004). They documentthat people who live close to each other,defined as living in the same census block,also tend to work together, defined as work-ing in the same census block: the baselineprobability of working together is 0.93 per-cent compared to 0.51 percent at the block-group level (a collection of ten contiguousblocks). Their findings are robust to the intro-duction of individual controls in the form of anumber of socio-demographic characteristicsand block-group fixed effects. More specifi-cally, these authors examine the hypothesisthat agents interact very locally with theirsocial contacts, exchanging information aboutjobs. Let: i and j be individuals who reside inthe same census block group but not in thesame household; Wij

b a dummy variable that isequal to one, if i and j work in the same cen-sus block; Rij

b is a dummy variable that isequal to one, if i and j reside in the same cen-sus block: Xij a vector of socio-demographiccharacteristics for a matched pair (a conceptto be clarified shortly below) (i,j); and, g aresidential block-group fixed effect whichserves as the baseline probability of anemployment match for individuals living inthe same block group. Then their hypothesismay be examined in terms of a regression:

(1)

These authors’ test for the presence of socialinteractions due to proximity boils down totesting for the statistical significance of theterm (�̂0��̂�1X ij). They include both the base-line probability ̂g and matched pair’s covari-ates in levels, �Xij, to control for anyobserved and unobserved factors that may

W X X Rijb

g ij ij ijb

ij= + ′ + + ′ ⋅ + � �( )0 1 ε .

14 Assortative matching of this type in social networkshas been documented by Peter Marsden (1987).

influence employment locational choices atthe block-group level. For example, this con-trols for features of the urban transportationnetwork that might induce clustering in bothresidence and work location. Also, workercharacteristics might be correlated with bothresidential location preferences and worklocation, if firms sort along the same variables.Their empirical strategy addresses severaladditional potential pitfalls, including possiblesorting below the block level and the possibil-ity of reverse causation. After they estimatethe social interactions effect they considerwhether the quality of the matches availablein an individual’s block affects employment,labor-force participation, and wage outcomes.

Bayer, Ross, and Topa use data from the1990 U.S. Census for the Boston metropoli-tan area for all households that responded tothe long U.S. Census form, and choose indi-viduals who did not reside in the samehousehold, were U.S.-born and agedbetween 25 to 59, and employed at the time,thus ending up with 110,000 observations.From these data, about four million observa-tions on matched pairs were constructed bymatching up individuals in pairs, in a citywith 2,565 block groups with an average often blocks each.

Bayer et al. find that social interactions arestronger when a pair of individuals are morelikely to interact because of education, age,and the presence of children14; interactionsare stronger when one of the two individualsis strongly attached to the labor market, andare weaker when both are drop-outs, young,or married females. In terms of the magni-tude of the impact of match quality, a onestandard deviation increase in referralopportunities raises labor-force participationby one percentage point, weeks worked byabout two thirds of one week, and earningsby about two percentage points. This study isalso significant for its reliance on differentgeographical scales for identification.



Two other studies, which preceded Bayeret al., sought to identify “network effects” byusing concepts of proximity similar to those ofBayer et al. Both seek to establish that indi-viduals in close proximity share information.Therefore, they are not only methodological-ly similar to Bayer et al. but also relevant tothose interested in establishing informationsharing among individuals in close proximityto one another. Marianne Bertrand, ErzoLuttmer, and Sendhil Mullainathan (2000)consider the impact of social networks onwelfare participation. They emphasize meth-ods that allow them to distinguish betweenthe effects of networks from those of unob-servable characteristics of individuals and ofthe communities in which they live. LikeBayer et al., they attempt to distinguishbetween the effects of mere geographic prox-imity and of information transmission madepossible by proximity. In particular, they relyon language spoken and examine whetherbeing surrounded by others who speak thesame language increases welfare use morefor individuals who belong to high welfare-using groups. Individuals interact more withothers who speak the same language and aretherefore more likely to be influenced byother members of that group. They obtainhighly significant and positive coefficients onthe interaction between contact availabilityand mean welfare participation of one’s lan-guage group and interpret these findings asevidence of network effects.

Anna Aizer and Janet Currie (2002) exam-ine “network effects” in the utilization ofpublicly funded prenatal care. These authorsconsider women as belonging to a network ifthey live in the same neighborhood (definedas the areas of five-digit zip code) andbelong to the same racial or ethnic group.They use data on take-up of publicly fundedprenatal care, which originate in vital statis-tics from California’s Birth Public Use filesfrom 1989 to 2000. They find evidence infavor of their hypothesis that pregnantwomen are most likely to be influenced bynew mothers from the same area and ethnic

group in terms of their own use of publicprenatal-care programs. Such use is highlycorrelated within groups defined usingrace/ethnicity and neighborhoods, and per-sists even after accounting for unobservedcharacteristics by including zip code-yearfixed effects. However, the richness of theirdata (which are abstracted from birth certifi-cates and include more than 3.5 millionobservations) allows them to test whethersuch estimated effects represent informa-tion sharing within groups. This is accom-plished by including fixed effects for thehospital of delivery interacted with the yearof delivery. The results on the estimatedeffects of “networks” are then eitherreduced or eliminated and thus cast doubton the idea that the observed correlationscan be interpreted as evidence of informa-tion sharing. They point instead to differ-ences in the behavior of the women involvedand of the institutions serving differentgroups of low-income women as the primaryexplanation for group-level differences inthe take-up of this important public pro-gram. They examine the role of institutionsby comparing the behavior of foreign-bornwith that of native-born Hispanic women.They find that “network effects” are quitesimilar for both those groups of Hispanicwomen, in contrast to their expectation thatforeign-born women will have greater infor-mational requirements. They conclude,therefore, that it is differences in the behav-ior of institutions and not information shar-ing that explains the established correlationsbetween neighborhood and ethnic groupmembership in prenatal care use.

Rafael Lalive (2003) also examines socialinteractions among unemployed individuals.This empirical study exploits an unusualquasi-experimental setting created by selec-tive extension of unemployment benefitsfrom the Austrian government to individualswho resided in certain regions and wereemployed (or had been employed) by a cer-tain group of industries. Evidence of socialinteractions in unemployment behavior



measured in terms of length of unemploy-ment spells takes the form of both directeffects among individuals who are entitled tothe extended benefits and indirect effectsthrough the impact of the program’s exten-sion on the behavior within the referencegroup. Their methodology is interestingbecause their identification of social interac-tions uses a linear model with the dummyvariable indicating that an individual is agedat least fifty years and has continuous workhistory as an individual effect and with itscorresponding group mean as a contextualeffect. Their approach is made possiblebecause of the Austrian government’s “par-tial-population intervention.” This route toidentification of social interactions modelwas discussed by Robert Moffitt (2001); thatis, the endogenous social interactions coeffi-cient is disentangled from the direct effectof extended benefits by distinguishing twogroups of the population and by comparingreduced forms estimated with data for thoseaffected with estimates for those not affect-ed by the policy. This result is also particu-larly relevant in explaining spatial variationin unemployment, an issue of great interestin large economies.

3.5 The Role of Job Referrals

Krauth (2000) focuses on spatial proximityeffects by studying the consequences ofemployers’ using their social ties with theiremployees to make inferences about unob-served components of the productivity oftheir workers’ social contacts. He shows thatthere is a critical value for neighborhoodhuman capital below which long-runemployment at equilibrium is low and abovewhich it is high. The critical value dependson the strength of social contacts.

Lisa Finneran and Morgan Kelly (2003)examine theoretically the role of job refer-rals with special emphasis in the persistenceof inequality. Workers differ in terms ofskills, and such differences are ex ante unob-servable by employers. Each individual is amember of a hierarchical referral network.

Employees may refer their own acquain-tances they believe to be qualified foremployment. If those are hired, they in turnrefer their acquaintances. Each step in thereferral process is stochastic and reflects awhole host of factors describing the labormarket. Their central result is that the den-sity of referral linkages exhibits thresholdbehavior. Above a threshold, workersthroughout the hierarchy are referred foremployment with probability one; below it,workers’ probability of referrals falls expo-nentially as one moves down the network sothat average income falls and workers at verylow levels are referred with probability zero.Referrals are more valuable than anonymousmatching because they convey informationabout employees’ qualifications, reducerecruiting and training costs, and lowermonitoring costs. Finneran and Kelly estab-lish the value of referrals in a general net-work, where the number of potential ties byeach worker and the probability of formingties all differ across the network. Given a setof potential direct referrals with a largenumber of potential connections n and a setof workers who are always guaranteedemployment, the measure of paths that con-nect those who are always guaranteedemployment with other individuals in a net-work who are far away from them (in termsof the number of required referrals) isincreasing in the number of actual referralsmade. As the size of the economy grows,there exists a critical number of linkages,which varies slowly with n, such that theprobability for any group of workers who arefar in terms of linkages from those who arealways guaranteed employment to be linkedwith them through referrals tends to one,when the number is above the critical value,and to zero, when it is below it. As in theresult of Calvó-Armegnol and Jackson, this isdue to the statistical properties of networkscomposed of workers who are otherwiseidentical.

Krauth (2003b) models search for jobs,where individuals’ social acquaintances



provide referrals to potential employers.The model assumes a social structure, rep-resented by a directed graph, which isexogenously given but may possibly be timevarying, and may vary stochastically as well,but remains exogenous. Firms hire workersbut do not observe their quality directly. Aworker’s past employment and social con-nections affect her current employmentprospects. Krauth proposes a model ofstrong versus weak ties that is intended toexpress Granovetter’s concept of weak ver-sus strong ties as follows: a social tie fromagent i to agent j is defined as strong if j alsohas a social tie to one or more of i’s otherfriends; it is weak, otherwise. The modelinvolves starting from a network with onlystrong ties and switching some of the socialconnections randomly with probability p.For large networks, the probability that thisprocess will generate another strong tie isclose to 0, and the fraction of weak ties inthe resulting network is approximately p.Krauth uses simulations to show that thelong-run probability of employment isincreasing with the proportion of weak ties.Weak ties appear to be a way for an individ-ual to diversify her social resources. Whenindividuals are friends of one another (theyare connected through strong ties), theiremployment statuses are correlated andthis increases the variance in the number ofemployed friends. Therefore, a networkwith more weak ties is associated withsmaller inequality in the distribution ofemployed friends and thus with a higheroverall employment rate. Troy Tassier andFilippo Menczer (2002) also study the roleof referrals in labor-market outcomes.

Empirical evidence on the interactionamong individuals and their social contactsand employers through job referrals is not asextensive. According to Loury (2003), whoworks with the National Longitudinal Surveyof Youth, personal contacts have significantwage effects for young men only when theircontacts are older-generation male relativeswho know the boss or arranged an interview

for the job-seeker. Simon and Warner (1992)presented evidence that those who foundout about their jobs through an acquaintanceinside the firm had higher starting salaries,while those who found out through anacquaintance outside the firm had lowerstarting salaries. They attributed this findingto reductions in employer uncertainty aboutworker productivity.

Economics research has explored thesalient aspects of contact and relational het-erogeneity that influence job-contactprospects for individuals, but there is littlecomparable work on the effects of employercharacteristics. In a notable exception, thevarying role of referrals in the U.S. industri-al sector has been studied directly byAdrianna Kugler (2002). She finds that theobserved positive correlation betweenindustry wage premia and use of employeereferrals when industry-level data are useddisappears when she controls for sector ofemployment using micro data from theNational Longitudinal Survey of Youth.

Relying on data from the EuropeanCommunity Household Panel, Pellizzari(2004a) reports substantial cross-country vari-ation in the effects of contacts on earnings. Insome cases, contacts result in wage premiumsand in others workers who found their jobsthrough contacts earn less than those usingformal sources. The latter largely occur inindustries where firms invest substantially informal recruitment activities. Firms are morelikely to undertake such investments for highproductivity jobs where the costs of turnoverare substantial. When large investments aremade, workers found through formal recruit-ment average higher productivity than thosefound through other means. Pellizzari(2004b) confirms this by using data from theSurvey of Employers’ Recruitment Practicesfor the United Kingdom in 1992.

3.6 Evolutionary Models of Social Structure

Next we discuss briefly some connectionsof the neighborhood-effects literature withthe literature on evolutionary models of



15 Regimes are described with stochastically stablestates. The concept of stochastically stable dynamic equi-librium (Young 1993, 1998) is particularly appropriate tocircumstances where a system is subject to repeatedshocks, that is when the system is shocked repeatedlyand before it has a chance to recover, it is shocked again,and again. In such settings, some states will occur morefrequently than other states in the long run.

social structure. Specifically, this literaturestudies interactions through games individu-als play with members of a population wherea social state is defined as adoption of a normor other institutions. First, it is interesting tocontrast the notion of weak versus strong tieswith global versus local interaction in evolu-tionary models of social structure. Groupswhose members are connected with strongties are typically small and close-knit.Second, as emphasized by Robert Dietz(2002), the evolutionary learning literatureinvolves ideas that are conceptually relatedto social interactions.

A particularly noteworthy finding in thisliterature is a result of Glenn Ellison (1993),according to which if individuals interact withmembers of a large population then existingbehavior (or social conventions, historicalforces, etc.) are likely to dominate. On theother hand, if they interact with a small num-ber of neighbors, then cooperative outcomesand thus coordination is more likely. So,interactions involving small neighborhoodsare more likely to be determined by evolu-tionary forces. The results of Ellison havebeen generalized by Peyton Young (1998) asfollows.15 When individuals interact mainlywith small groups of neighbors, then “shifts ofregime can occur exponentially faster than inthe case of uniform interaction ... . All elsebeing equal, the smaller the size of the neigh-borhood groups, and the more close-knit thegroups are, the faster the transition time forthe whole population (Young 1998, pp.98–99). This suggests an important role ofclose-knit groups for social learning, which inthe context of the literature reviewed heremay be interpreted as that of strong ties. Thiscontrasts with the role that Granovetter has

ascribed to weak ties in the interpersonal flowof job-related information. Clearly additionalresearch is needed.

4. Strategic Network Formation andEndogenous Job Information Networks

If social networks are determined byuncoordinated actions of individuals, thenseveral of their features are of interest. First,how many others is an individual in contactwith to exchange job-related information?Second, how does this number vary amongindividuals within the economy, and howmight it depend upon individual characteris-tics? And, third, what is the topology of theassociated network? Is everyone directlyconnected with everyone else, is there a sin-gle individual through whom everyone isconnected, or is there some other stylizedpattern of connections that is discernible insocial networks? We review the existing lit-erature while recognizing that it has not, todate, succeeded in delivering endogenousoutcomes with respect to all these featuressimultaneously. The fast-developing litera-ture on strategic network formation, whichseeks to motivate the creation of social con-tacts in terms of optimizing behavior byindividuals, has not, until recently, empha-sized job-information networks, although, aswe see shortly, important progress has beenmade. The strategic network formation liter-ature has been eloquently reviewed recentlyby Bhaskar Dutta and Matthew Jackson(2002), Goyal (2003), and Jackson (2003). Sowe will touch on it only very briefly.

4.1 Strategic Models of Network Formation

The principal contributions in this litera-ture, including Matthew Jackson and AlisonWatts (2002) and Watts (2001), aim ataxiomatic descriptions of network-basedconcepts. Important such concepts are: effi-ciency, with a network being efficient if thereis no other network that leads to higher pay-off for all of the members; stability, with anetwork being stable if no individual would



benefit by severing a link and no two playerswould benefit by forming a new link; and thenotion of a Nash network, where all mem-bers are playing Nash equilibrium. The mod-els in the strategic network formationliterature make links endogenous by meansof strategic considerations; that is, mainte-nance (or creation) of links between twoindividuals requires that they both consent toit, whereas severance can be done unilateral-ly. The utility each member derives frombeing a member of a network, that is, frombeing connected with others who may them-selves be connected with others, typicallydepends additively upon the number ofother agents each agent is connected with,minus the costs of maintaining connections.Some authors make an allowance for proxim-ity, by means of a decay factor that dependson the number of intervening agents. Others,like Venkatesh Bala and Sanjeev Goyal(2000), discussed in more detail below, dis-tinguish between one- or two-way communi-cation. As Jan Brueckner (2003) and Jackson(2003) demonstrate, these assumptions arequite crucial for the results.

Matthew Jackson and Asher Wolinsky(1996) show the equilibrium network isempty if linkage costs are high; it is a wheelnetwork (where each agent is connected withtwo other agents thus forming a wheel, or acircle) if linkage costs are moderate; and it isa complete network if the linkage costs arelow. The two extreme outcomes are quiteintuitive. The wheel outcome is also intuitivewhen one recognizes that it is associated withtwo connections per person, which is theminimal symmetric outcome. This demon-strates the sensitivity of network topology toparameter values and suggests that in prac-tice different topologies may emerge in agiven economy for different sets of problems.

The recent resurgence of interest in jobinformation networks has benefitted byextending several of the concepts proposedby this literature. In particular, Calvó-Armegnol (2004) suggests that conditionsthat lead to different network topologies,

16 These results may be generalized by restricting theinformation available to agents, that is, by assuming onlylocal information—each agent knows the residual set of allthose she is connected with, that is those her neighbors canaccess without using links to her—and by allowing obser-vation of successful agents—there is some chance that shereceives information from a “successful” agent, that is aperson who observes the largest subset of people in theeconomy without assistance from her own links.

that are obtained endogenously, are impor-tant for the functioning of job-informationnetworks. That is, the information flowsassociated with two different chains of con-tacts of identical length but in differenttopologies are generally different. We returnto this work in detail further below.

Bala and Goyal (2000) also model endoge-nous network formation, where each individ-ual derives utility that is proportional to thenumber of other agents she is connected withdirectly and indirectly, net of the costs ofmaintaining those connections. UnlikeJackson and Wolinsky (1996), however, theirsinvolves directed links. They define one-waycommunication as your having access toanother person’s information. An one-waylink does not imply that other person hasaccess to yours, which would be two-waycommunication (undirected links). Bala andGoyal show that, with one-way communica-tion, Nash networks are either minimally con-nected and form a wheel (each player formsexactly one link) or empty. In other words,information is either shared with everyone, orthere is no sharing. Bala and Goyal also studythe dynamics of link formation by assuming anaive best-response rule with inertia, that is,an agent may choose, with fixed probabilities,either a myopic pure strategy best-response,or the same action as in the previous period.They show that irrespective of the number ofagents and from any initial starting pattern ofinterconnections, the dynamic process self-organizes, by converging in finite time withprobability 1 to the unique limit network. Thelimit is the wheel if the linkage costs aresmall, or either the wheel or the empty net-work if the linkage costs are large.16 Withtwo-way communication the results are quite



different: Nash networks are either “center-sponsored” stars (where an agent forms thenetwork by connecting himself with all oth-ers, occupies the central position, and paysfor all links), or the empty network.

It is interesting that experimental evi-dence that has been obtained recently pro-vides support for the Bala and Goyal model.Armin Falk and Michael Kosfeld (2002)report that the prediction based on strictNash equilibrium works well in the one-waycommunication model, with subjects form-ing the wheel or the empty network in amajority of cases. In contrast, the predictionsbased on Nash and on strict Nash fail for thetwo-way communication model: the subjectsdo not form the center-sponsored star northe empty network in any of the experi-ments. The authors attribute their results tothe subjects’ sensitivity to fairness consider-ations, which is a well-known factor affectinggame outcomes in experimental settings.

Brueckner (2003) extends the basicassumptions of this literature by endogeniz-ing the probability of a link and of linkingcosts. This extension assumes that individu-als value only their direct connections,involves a simple mathematical structure,and derives some properties that would seemto be relevant outside the confines of thoseparticular models. For example, identicalindividuals will spread their efforts uniform-ly to create social acquaintances, thus bring-ing about symmetric outcomes and rulingout inherently asymmetric network topolo-gies like that of the star. However, asymmet-ric outcomes are possible if individuals’ socialattractiveness differs—individuals can havemagnetic personalities—or if individualshave different sets of social acquaintances.

The assumptions these models employ donot make them readily applicable to thestudy of exchange of information about jobs.Still, the experimental evidence on networkformation and the sensitive dependence ofendogenous network topology on costs rela-tive to benefits suggests that cultural factorsmay be important in the determination of

17 Pierre Cahuc and Francois Fontaine (2002) allow forindividuals to choose between job matching through socialnetworks and (costly) individual search methods. Such anextension appears to overturn several standard results.Competitive search may be over- or under-utilized andmultiple equilibria are possible.

real-life job information networks. Differentindividuals may value differently the payoffsfrom being connected with others relative tothe costs of doing so.

4.2 Endogenous Job Information Networks

Social interactions associated with endoge-nous job information networks may be quitedifferent from others, such as those inspiredby constraint, preference or expectationsinteractions, to use the terminology ofManski (2000). The theoretical predictions ofCalvó-Armegnol and Jackson (2002, 2003)17

depend on ad hoc assumptions about socialexchange: the network of social interactionsis given. By varying network characteristicsand parameters, one may explore potentialconsequences of primitive behavioralassumptions about individuals’ valuations ofsocial interactions, especially with respect todifferent types of individuals’ propensities totransmit job-related information.

What do we know about the formation ofthe networks used for acquiring and dissem-inating job information? Scott Boorman(1975) was the first to ask formally how socialgroups accommodate the transmission ofjob-related information. He presents an ana-lytical model of transmission of job-relatedinformation through contacts, where individ-uals choose how to allocate effort over main-taining strong and weak contacts. Strongcontacts are given priority in the transmissionof job-related information but require moretime than weak ones to maintain. Choicesover strong versus weak contacts by all mem-bers of the society determine a rudimentarysocial structure.

Let � and � denote the probability that aperson will need a job at a particular point intime, and that she hears directly of a vacantjob, respectively. If S and W denote the



18 See ibid. and Calvó-Armegnol (2004), Proposition 1,for details. Actually, the latter does not distinguishbetween strong and weak ties.

number of strong and weak contacts, respec-tively, then the probability that a person willget a job through her contacts may be com-puted as the one minus the probability thatthe person will not get a job either from anyof her strong or her weak ties. The latter twoprobabilities may be obtained from elemen-tary but tedious combinatoric computa-tions.18 The probability that a person will geta job is written as:

(2)

Maximizing this probability, subject to a timeconstraint, W��S�T, where � denotes theextra units of time required to maintain astrong tie, allows us to study whether strongor weak contacts are more likely to be cho-sen for different values of the parameters.Boorman’s probabilistic approach hasrecently been taken up by Calvó-Armegnol(2004), discussed below.

It turns out that if the probability of need-ing a job is small, ��1, Boorman’s modelimplies that the equilibrium where all indi-viduals choose weak ties is stable. The oppo-site is true when �≈1 . These results have anintuitive appeal. The only reason for invest-ing in strong contacts is the concern thatone’s weak contacts will be preempted by thedemands of their own needy contacts. As theprobability that anyone needs a job decreas-es this contingency becomes less important.Boorman’s model does not use tradeoffs fac-ing workers searching for jobs, but it doesportray the properties of the communicationnetwork in a uniform symmetric setting,where each individual maintains both strongand weak ties.

Boorman’s path-breaking work took a longtime to influence the economics literature.

1 11 11− − − −

−� ��

�( )

( )SW

W

W.

1 1 11 1− − − − −

� �

�

�( )

( )SS

S

19 The earliest model of word-of-mouth communica-tion that we are aware of is Strand (1983), who is interest-ed in intrafirm wage dispersion. The concept ofword-of-mouth communication used by Calvó-Armegnoland Zenou is similar to Glenn Ellison and DrewFudenberg (1995), although the latter stresses the effi-ciency of social learning. The latter paper’s finding thatsocial learning is often more efficient when communica-tion between agents is fairly limited. In a model of word-of-mouth communication, this is understood as samplingfrom a given sample of other participants, with some frac-tion of players ignoring the information and not changingtheir decisions and the remainder adopting the choice thatappears to be best based on their own nonoptimal sample.This property of limited communication is conceptuallyrelevant to the role of weak ties in social networks, as wediscussed in subsection 3.6.

This is quite surprising in view of the factthat it is a model of endogenous (job) infor-mation network, where agents choose thenumber of links, which does not requirestrategic considerations. Curiously, thedevelopment of strategic models more gen-erally seemed to have provided an impetusfor the development of non-strategic mod-els, as well. Calvó-Armegnol (2004) devel-ops the first precise model of a contactnetwork deliberately aimed at passing job-related information and Calvó-Armegnoland Yves Zenou (2001) explore its implica-tions for labor-market-wide matching.These papers, unlike Granovetter andBoorman, do not distinguish between strongand weak contacts.19

In Calvó-Armegnol (2004), all workers areinitially employed; they may lose their jobswith a constant probability b, and hear of anew job opportunity with probability a. If aworker hearing of a job is unemployed shetakes it; if she is employed, she passes theinformation on to her unemployed directcontacts. A worker is employed and hears ofa job with probability ��a(1�b), and isunemployed and does not hear from her con-tacts about jobs with probability �b(1�a).Let g denote a network of contacts, gN the setof all subsets of the set of all individuals ofsize 2, that is the complete graph among Nindividuals, and G�{g|g�gN}. Given g∈G,the set of an individual’s direct contacts in gis denoted by Ni(g). The probability that an



individual i receives job information whofrom one of her direct contacts j, j∈Ni(g),who is assumed to be informed, is given by

, where ni(g)� |Ni(g)|. (This deri-

vation is, of course, closely related toBoorman’s, op. cit.) Therefore, the probabil-ity that i does not actually find a job thanks

to j is given by q(ni(g))�1�� , and

the probability that individual i actually getsa job through her contacts is:

(3)

This probability is larger, the greater an indi-vidual’s set of direct contacts. A larger set ofcontacts broadens the information channelsavailable to i, but not the number of directcontacts as such. This probability decreasesthe larger is an individual’s indirect contactsthrough any of his direct contacts, nj(g),j∈Ni(g). That is, having more indirect con-tacts increases the competition for informa-tion. This basic relationship determines thereturn to adding and severing links amongany two individuals. Doing so affects theinformation flow for those directly affected aswell as for their direct contacts. This expres-sion also implies an aggregate unemploymentrate given by:

.

Calvó-Armegnol (2004) uses this modelto examine properties of symmetric equilib-rium networks, when a link between anytwo agents, which is undirected and costlyto both of them, is initiated only if it ismutually advantageous. The model alsoimplies tradeoffs associated with the topolo-gies of the networks of contacts. Networkswith the same total number of contacts butdifferent topologies imply different aggre-gate unemployment rates. It is particularlysimple to consider topologies of regulargraphs, where all individuals have the samenumber of direct contacts, that is, all nodes

u g P gn ii N( ) = −

∈∑ 1 1 ( )

P g q n gi ij N gi

( ) .= −∈∏1 ( ( ))

( )

1 1− −( )( )

( )b

bn g

n gi

i

1 1− −( )( )

( )b

bn g

n gi

i

have the same degree, � . In that case, thetradeoffs between the number of direct andindirect contacts on an agent’s employmentprobability implies that it attains a maxi-mum over the set of different degrees, (1,� ). Where an agent’s employment probabil-ity increases, it is also the case the marginaleffect of increasing the network’s degree isnegative. While increasing the number ofdirect links improves the employment prob-ability, it also does so for everyone, thus alsoincreasing the number of indirect links,which are detrimental to the likelihood ofemployment for sufficiently high values ofthe network’s degree. In other words, directcontacts are beneficial because theyimprove an individual’s information sources,but contacts that are two-links away aredetrimental because they create competi-tors for the information possessed by adirect contact. This rivalry is also the reasonthat the sign and intensity of the payoffspillovers that agents exert on one anotherare very much dependent on the geometryof the network. This is also a reason moregeneral analyses are difficult.

Calvó-Armegnol also considers asymmetricnetworks. In fact, with the same parameters,both symmetric and asymmetric networks arepossible, but they cannot be “too asymmet-ric.” This model is the only one to date thathas employed successfully a model of strate-gic network formation to the job market con-text. Interestingly, individual payoffs in thismodel do not, unlike the specific models inBala and Goyal, contain a component that islinear in the number of other agents eachagent is connected with. This could explainwhy it is so much more difficult to studyendogenous network topology in the generalcase in Calvó-Armegnol’s model.

Calvó-Armegnol and Zenou (2001) explorethe implications of the model in Calvó-Armegnol (2004) for aggregate matchingproperties of an economy. Aggregate match-ing is increasing and concave in both theunemployment and vacancy rates. However,hearing through both direct and indirect



20 The recent exhaustive review of the literature on thematching function by Barbara Petrongolo and ChristopherPissarides (2001) does not discuss at all matching aspectsof social networks.

contacts implies that the impact of networksize is not monotonic. This is so, as we arguedabove, because such an increase increasesthe potential number of unemployed work-ers who are connected directly to anemployed informed worker.

Specifically, if 1�u the probability that aworker is employed and s is the number ofother (randomly drawn) workers eachworker is in direct contact with (the degreeof the symmetric, or balanced, social net-work), then each worker meets us unem-ployed workers and (1�u)s employedworkers in each period. Under the assump-tion that information is passed byemployed to unemployed workers only andthe vacancy rate represents the probabilitythat an unemployed individual will hear ofa job vacancy directly, than the individualprobability of finding a job through socialcontacts is given by:

. The

matching function is given by m(s,u,v)�u[v�(1�v)P(s,u,v)].

The properties of the matching functionwith respect to unemployment and vacancyrates are the same as in the earlier jobmatching literature; they differ only withrespect to network degree. In this case, thematching function is not only not homoge-neous of degree one, as in Pissarides(2000), but not even monotonic. When thenetwork degree increases, unemployedworkers hear about more vacancies throughtheir social network. At the same time, it ismore likely that information about multiplevacancies will reach the same unemployedworker. It is therefore important to seewhether this non-monotonicity is present inthe data, which is an open question empir-ically, to the best of our knowledge. Theproperties of the matching function affectthe equilibrium level of unemployment inthe economy. Matching is increasing in net-work size, for sparse networks, anddecreasing for dense ones. These findings

P s u v v u uus

ss

( ) ( ) ( ), , = − − −

− −1 1 1 1 1

suggest that it is important to research fur-ther the aggregate the properties of differ-ent types of social networks with respect tomatching.20

5. Towards an Integration of Job informa-tion Networks and Sorting

The interconnection of individualsthrough job information networks clearlygives rise to sorting phenomena in both theeconomic and social spheres. Sorting, on theother hand, has typically been investigatedby the economists in a life-cycle context.This section examines the interplay of thosetwo forces.

A well-established literature has investi-gated how the intertemporal evolution ofhuman capital is affected by the human cap-ital of parents, of the ethnic group to whichthe individual belongs, and of the individ-ual’s neighborhood. The natural relationshipbetween parents and children may affectvariables (such as “innate ability”) that playan important role in the perpetuation ofinequality, but are not subject to choice. Onthe other hand, mating does involve choice,and the human capital of spouses is not inde-pendent of one another across the popula-tion. Parents’ choice of human-capitalinvestment for their children is affected bytheir own human capital.

A prominent example of this literature isMichael Kremer (1997), who studies indi-viduals’ schooling as a function of that ofthe parents and of the mean schooling inone’s neighborhood of upbringing. Hefinds strong neighborhood effects; that is,mean schooling in the neighborhood ofone’s upbringing has a coefficient that isroughly the same as that of the parents’schooling. This effect is not sufficient, how-ever, to explain a large role for residentialsorting in the inequality of earnings across



22 In modelling physical and engineering processes,one often distinguishes between different time scales. Seealso, Young (1998) on the notion of “event” time, whichmarks numerous distinct events that may be compressedinto short intervals of “real” time. In the absence of differ-ent time scales, these considerations would produce a con-temporaneous cross-sectional effect. We thank GiulioZanella for directing our attention to James Meade (1976),for a discussion of transmission of inequality based on aconcept of effects of social contacts operating in differentfrequencies.

the population.21 George Borjas (1992)regresses individuals’ schooling againstschooling of parents and the average school-ing of the ethnic group to which an individualbelongs. Borjas, too, finds strong ethniceffects. Borjas (1995) allows, in addition, for apotential link between parental and ethniccapital, on one hand, and residential segrega-tion, on the other, and again finds a strongethnic effect. Borjas (1998) finds that, in addi-tion, schooling is affected by the presenceand skill levels of other ethnic groups in theneighborhood of one’s residence, with lowersegregation by ethnicity associated with moreschooling for an individual and for her ethnicgroup. He also allows for potentially complexinteractions between one’s ethnic group andthe distribution of socioeconomic characteris-tics in one’s neighborhood of upbringing andobtains statistically significant results.

Job information networks can have com-plex neighborhood effects, too. They affectdifferent individuals’ access to informationabout job opportunities and thus have bear-ing for individuals’ abilities to market theirlabor services. Such effects imply a depend-ence among individuals’ neighborhood ofresidence, social connections, and status andterms of employment. These effects operatewithin much shorter time frames than thetime span between the influence of upbring-ing and of parents’ characteristics and off-spring human capital. It is thus appropriate toallow for different time scales.22 By defining

labor income as the product of human capital,which represents income earning capacitymeasured in efficiency units, and the appro-priate wage rate, which represents its currentmarket price per efficiency unit, we may sepa-rate intertemporal effects, such as the parentaleffect, the ethnic group effect and the neigh-borhood of upbringing effect, from social net-work effects on individuals’ access tojob-related information. The former effectsoperate at life-cycle frequencies, indexed byperiod t�0,1,2, …, while the latter operate atnearer to business-cycle frequencies, indexedby �t, which indicates intervals of time withineach discrete life cycle period t. They aremuch shorter than life cycle frequency andaccount for employment-related events.These two sets of effects would typically not beindependent. The social connections of one’sparents are related to the pattern of social con-nections within one’s ethnic group and mayalso influence one’s own social connections.

The intertemporal evolution of income-earning ability, human capital for short, ofindividual i who belongs to ethnic groupe, hi,e,t, reflects such intertemporal effects asthe parental effect, which operates throughthe natural parental relationship betweenindividuals i�e,t�1 and i��e,t�1, the parents ofindividual ie,t, assuming for simplicity thatboth of individual i’s parents belong to thesame ethnic group, the ethnic effect, he,t,and the neighborhood effect, h� (i),e,t�1,where �(i) indicates the neighborhood ofindividual i’s upbringing. We may combineKremer (1997) and Borjas (1992; 1995;1998) and write an equation for the law ofmotion of human capital as follows:

Hi,e,t�H(hi�,e,t�1, hi��,e,t�1; he,t, h�(i),e,t�1), (4)

which allows the ethnic effect and theneighborhood effects to interact, h�(i),e,t�1.A complete description of the intertemporalevolution of income earning ability requiresdescription of the dynamic evolution of theethnic effect.

Let Si,e,�t denote the event that individual iis employed, Si,e,�t�1 , or unemployed,

21 Working with the same data, Ioannides (2002) findsstrong nonlinear effects, which imply multiple equilibria.Unfortunately, like Kremer’s, Ioannides’ results are notcausal, in that the endogeneity of neighborhood choice byparents is not accounted for.



23 In view of Nieke Oomes (2003), who shows that iflabor can be hired in continuous quantities, the long rundistribution of employment in spatially separated mar-kets is uniform, we conjecture that persistence of non-trivial effects of social networks on job-related outcomesis intimately related to the presence of a labor-marketparticipation decision.

Si,e,�t�0, at a business-cycle point �t of her lifecycle period t. Following Calvó-Armegnol andJackson, op. cit., we normalize wages so thatWi,e,t�0, if i is unemployed. Wage setting maybe expressed as a function of one’s wage in theprevious (business cycle) period and of thenumber of new job opportunities individual ihas as of time �t, Oi,e,�t, Wi,e,t�w(Wi,e,�t�1,Oi,e,�t), where the index �t�1 is defined with-in the time scale of business cycle frequen-cies. We may now write an equation for theintertemporal evolution of labor income,

Yi,e,t�Hi,e,tWi,e,�tSi,e,�t . (5)

We may obtain the full dynamic flavor of Eq.(5) by contemplating the dynamic evolutionof the probability of employment and of thenumber of new opportunities. For example,recall the probability of i’s getting a jobthrough her contacts according to Calvó-Armegnol (2004), Eq. 3) above. That theoryimplies that the employment probabilitydepends on the size of the set of contacts,ni=|Ni,t(gt)|, where gt denotes the graphdescribing a particular realization of socialnetworks at time �t.

The graph describing individual i’s socialcontacts will not necessarily be regular (allindividuals having the same number of con-tacts with others) and will evolve over (busi-ness cycle) time in a way that exhibitsdynamic dependence. Therefore, in princi-ple, we can describe the evolution of theprobability of getting a job through one’scontacts as a function of the social networksthe individuals have access to. These willemerge as a result of individual incentivesthat reflect strategic considerations associat-ed with network formation. That is, Si,�t mustbe derived as a function of Ni,t(gt).23 Wenote that the discussion of empirical

research earlier in the paper suggests thatthe efficiency wage rate, the set of jobopportunities and the employment probabil-ity all depend on one’s ethnic group. It is fairto say that the empirical literature suggestsstrong persistence in ethnic composition andincome distribution of neighborhoods.

The empirical research reviewed aboveoffers separate glimpses on the joint distri-bution. Eq. (5), along with (4) and a descrip-tion of the dynamics of human capital byethnic groups encapsulate the joint effects ofhuman capital and access to job opportuni-ties on the distribution of earned incomewhile accounting for the fact that thoseeffects operate at different time scales.Conceptually, one may adapt the methodpioneered by Loury (1981) and the tools ofCarl Futia (1982) to describe the equilibri-um joint distribution of these characteristicsas an invariant distribution associated withthe law of motion of income-earning ability,ethnic effects, neighborhood effects, andjob-information effects. It is unlikely thatthis description can ever be reduced to a sin-gle dimension. However, it would also beinteresting to adapt Loury’s essentially deter-ministic approach to a stochastic one, incor-porating the tools of evolutionary stabilityemployed by Young (1998). Further theoret-ical research is likely to allow deeper analysisof the impact of job-information networks oninequality. This entirely new approach is cru-cial for deeper understanding of the lifetimeincome distribution.

There has been a fair amount of success indescribing the equilibrium distribution ofincome, with an emphasis on the impact ofsorting on human capital formation in lifecycle frequencies. Our understanding of theimpact of job-information networks and theirrole as a force of inequality is much lessdeveloped. They are potentially very impor-tant, in part because network formationdepends on sorting of individuals’ own char-acteristics. It is interesting to speculate howsorting by own characteristics in choosingwhom to associate with, rather than passively



reacting to standard norms of behavior ingroups, underlies an independent role ofethnicity as conjectured by Borjas.

Kremer (1997, p. 135) notes that “to theextent that people learn from classmates andco-workers, sorting by an individual’s ownacademic ability or productivity in schools orworkplaces may have larger effects oninequality than sorting by parental charac-teristics, since an individual’s future charac-teristics are presumably more highlycorrelated with his or her current character-istics than with parental characteristics.” AsKremer also notes, this is the kind of sortingthat most people would regard as egalitari-an, presumably because it is based on indi-viduals’ free choice of association,” but itmay be most likely to significantly ‘increaseinequality.’ ” The phenomenon of segrega-tion by skill in firms, that Kremer and EricMaskin (1996) have analyzed, is particularlyrelevant in this context. Firms are,metaphorically speaking, neighborhoods.Referrals among individuals who have beencoworkers involves selection through one’sown characteristics and is closely related tothe properties of technologies used by firms.It thus provides a route for technology toaffect equality of earnings. Further work inthis area appears to be particularly fruitful.

Research on the dynamics of inequality todate has emphasized the role of neighbor-hood effects in either the intergenerationaldynamics of human capital or the labor sup-ply decisions separately from one another.The framework we sketched above suggeststhat it would be fruitful to look at both setsof decisions jointly. Henry Overman (2002)is a rare example that allows for both types ofspillovers. That paper uses data on a sampleof Australian teenagers to test for neighbor-hood effects on school dropout rates at twodifferent spatial scales. Overman finds thateducational composition of the larger neigh-borhood can influence the dropout rate, pos-sibly reflecting the structure of local labormarket demand. He also finds, more surpris-ingly, that low socioeconomic status of the

immediate neighborhood is associated withlower dropout rate.

6. Suggestions for Future Research

The specific model proposed in the previ-ous section provides an overarching themethat helps integrate life-cycle and business-cycle forces into a model of job-informationnetworks and sorting. At the same time, anumber of other concrete issues may beaddressed directly and may thus comple-ment the overarching theme. They arebriefly discussed next in this section.

Theoretical issues needing further atten-tion abound. The issue of strong versus weaksocial ties is a natural one. It has alreadybeen generalized by the exogenous job-information networks literature. It may bepursued further along the lines of the meth-ods employed by the endogenous job-infor-mation networks literature, where recentcontributions have not distinguished socialtie strength. Similarly, global versus localinteractions among individuals provide atempting parallel to strong versus weak tiesand therefore deserve additional theoreticalattention. The findings of this paper point toa need to understand better the information-al and social infrastructure of the moderneconomy. In this context, a particularly glar-ing weakness of the theoretical and empiricalliterature is almost total lack of research onthe role of professional intermediaries(“headhunters”) throughout the job market,in spite of anecdotal evidence of the increas-ing importance of such intermediaries, andespecially outside its traditional territory ofexecutive search. An economy’s social andinformational infrastructure is also importantfor understanding the role of institutions infacilitating individuals’ access to resources.

We stress the fact that network topologiesas equilibrium outcomes are very sensitive toparameter values and therefore it is impor-tant for a model to express the particular cir-cumstances of the problem. The strategicnetwork formation literature has examined



how the economy self-organizes under dif-ferent assumptions about expectations. Theexogenous social interactions literature hasexplored the dynamics of different topolo-gies. Combining these two approachesdeserves further attention.

Research on the impact of the informationtechnology revolution on the job market isonly just beginning. Richard Freeman(2002) provides evidence that computeriza-tion and use of the Internet are associatedwith greater hours as well as higher wages.Evidence reported by Kuhn and Skuterud(2000) suggest increasing use of the internetfor the purpose of job search. We know verylittle about the impact of the internet on theeconomy generally, let alone on recruitmentand job search.

Both theoretical and empirical researchon firms’ recruitment practices also seemslikely to yield big payoffs. This research todate has explored existing models of match-ing to considerable advantage. However, themethodology of strategic network formationlends itself equally well to such a task. Wagepremiums and wage penalties associatedwith finding jobs through personal contactsare the joint outcome of firms’ recruitmentefforts and individuals’ job search. We can-not identify the role of information throughsocial interactions without accounting forboth sides of the market. This is particularlyimportant if one recognizes the conceptuallinks between neighborhood effects and net-work effects and the econometric issuesthey pose when it comes to the identifica-tion of endogenous versus exogenouseffects, which we discussed earlier in thispaper. Institutional and cultural differencesacross different countries in how social con-tacts facilitate job contacts and how use ofintermediaries and formal sources differneed to be better measured and understood.

An important benefit here would be foreconomists to learn from the multidimen-sional picture that the network-based theo-ries of mathematical sociology confer. At theheart of the sociological literature is the

belief that network-based models are indis-pensable for modelling more than just trivialsocial interactions. As Harrison White(1995) emphasizes, further theorizing onsocial networks is likely to pay off even with-in sociology, where in spite of technicalachievements in social-network measure-ments, modelling “network constructs havehad little impact so far on the main lines ofsociocultural theorizing …” (p. 1059). Whitesees an important role for studying socialinteractions through interlinking of differentindividual-based networks associated withsocial discourse. Granovetter (2000) urgessociologists to go beyond merely emphasiz-ing “the embeddedness of action in socialnetworks” and states that a “focus on themechanics of networks alone …” is insuffi-cient to “lead us toward the more complexsynthesis that we seek in understanding theeconomy” (p. 23). In view of this, it appearsthat distinguishing between the motives thatprompt individuals to engage in social net-working deserve attention in futureresearch.

7. Conclusions

Most of the initial economics research oninformal contacts aggregated the effects ofinformal contacts and networks. It hasassessed the role of job contacts on out-comes by comparing outcomes with to out-comes without job contacts. The new strandsin theoretical and empirical economicresearch examined in this paper build onsociological analyses and point out that con-tact effects are complex and vary due to indi-vidual, contact, relational, and employerheterogeneity. The new literature identifiesthe specific ways in which the effects ofinformal networks depend on differencesamong job seekers themselves, on the char-acteristics of the contacts they use, on therelationship between the job seekers andtheir contacts, and on features of the workenvironments where individuals are seekingjobs. The research makes clear that these



components do not operate in a vacuum butinstead interact with each other to producethe variation we observe in the use andeffects of informal networks. This has beenconfirmed by the new theoretical researchthat emphasizes the emergence of social net-works from individuals’ uncoordinatedactions and finds that the resulting networksare very sensitive to parameter values.

The research reviewed here suggests thatheterogeneity in network effects is impor-tant in a variety of contexts. It can helpaccount for changes in wage and employ-ment inequality across time. It clarifies the

mechanism behind correlations in observedoutcomes within social groups. Thusresearch has already used network analysisto elucidate the origins of previous unex-plained similarities in outcomes by race, eth-nicity, and gender. Furthermore, it identifiesthe source of some neighborhood correla-tions in labor-market outcomes. At the sametime, there are a number of promising areaswhere research is needed. In particular, theimportance of employer characteristics andthe role of the internet in altering the role ofinformal contacts in the future are topicsthat deserve special attention.



TABLE 1Education and Methods of Job Search

Category 1 2 3 4 5 6 7 8 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. searchedwith 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 foreducational 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 thePSID. The column labelled "All" gives the educational attainments for the entire 1993 sample of the PSID.

Appendix

Table 1, Table 2



TABLE 2Urban Size and Methods of Job Search

Category 1 2 3 4 5 6 7 8 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. searchedwith 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 beengaged in more than one methods—and thus do not add up to the number in column "Full". The columnlabelled "Full" gives the relative geographical distribution of the two respective categories, unemployed andemployed looking for job, for the entire 1993 sample of the PSID. "All" gives the geographical distribution of theentire 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 and499,000; SMSA with largest city 50,000 to 99,999; non SMSA with largest city 25,000 to 49,999; non-SMSA withlargest city 10,000 to 24,999; non SMSA with largest city less than 10,000.



REFERENCES

Aizer, Anna and Janet Currie. 2002. “Networks orNeighborhoods? Correlations in the Use of PubliclyFunded Maternity Care in California,” NBER work.pap. 9209.

Anderson, Carolyn J.; Stanley Wasserman and BradleyCrouch. 1999. “A p* Primer: Logit Models for SocialNetworks,” Soc. Networks 21, pp. 37–66.

Arrow, Kenneth J. and Ron Borzekowski. 2004.“Linited Network Connections and the Distributionof Wages,” work pap., Departmemt of Economics,Stanford U.

Bala, Venkatesh and Sanjeev Goyal. 2000. “ANoncooperative Model of Network Formation,”Econometrica 68:5, pp. 1181–229.

Bayer, Patrick; Stephen L. Ross and Giorgio Topa.2004. “Place of Work and Place of Residence:Informal Hiring Networks and Labor MarketOutcomes,” work. pap. Yale U.

Beggs, John and Hurlbert, Jeanne. 1997. “The SocialContext of Men’s and Women’s Job Search Ties:Membership in Voluntary Organizations, SocialResources, and Job Search Outcomes,” Sociol.Perspect. 40:4, pp. 601–22.

Bertrand, Marianne; Erzo F. P. Luttmer and SendhilMullainathan. 2000. “Network Effects and WelfareCultures,” Quart. J. Econ. 115:3, pp. 1019–55.

Bewley, Truman F. 1999. Why Wages Don’t Fall Duringa Recession. Cambridge, MA: Harvard U. Press.

Blau, David. 1992. “An Empirical Analysis ofEmployed and Unemployed Job Search Behavior,”Ind. Lab. Relat. Rev. 45, pp. 738–52.

Blau, David and Philip Robins. 1992. “Job SearchOutcomes for the Employed and Unemployed,” J.Polit. Econ. 98:3 pp. 637–55.

Boorman, Scott A. 1975. “A CombinatorialOptimization Model for Transmission of JobInformation through Contact Networks,” Bell J.Econ. Manage. Sci. 6, pp. 216– 49.

Borjas, George J. 1992. “Ethnic Capital andIntergenerational Mobility,” Quart. J. Econ. 107:1,pp. 123–50.

———. 1995. “Ethnicity, Neighborhoods, and HumanCapital Externalities,” Amer. Econ. Rev. 85:3, pp.365–90.

———. 1998. “To Ghetto or Not to Ghetto: Ethnicityand Residential Segregation,” J. Urban Econ. 44, pp.228–53.

Bortnick, Steven and Michelle Ports. 1992. “Job SearchMethods and Results: Tracking the Unemployed,”Monthly Lab. Rev. 115:12, pp. 29–35.

Bradshaw, Thomas. 1973. “Jobseeking Methods Usedby Unemployed Workers,” Monthly Lab. Rev. 96:2,pp. 35–40.

Bridges, William and Wayne Villemez. 1986. “InformalHiring and Income in the Labor Market,” Amer.Sociol. Rev. 51:4, pp. 574–82.

Brooks-Gunn, Jeanne; Greg J. Duncan, Pamela KatoKlebanov and Naomi Sealand. 1993. “DoNeighborhoods Influence Child and AdolescentDevelopment?” Amer. J. Sociol. 99:2, pp. 353–95.

Brueckner, Jan. 2003. “Friendship Networks,” work.pap. econ. dept. U. Illinois.

Brueckner, Jan; Jacques-Francois Thisse and YvesZenou. 1999. “Why Is Central Paris Rich andDowntown Detroit Poor? An Amenity-BasedTheory,” Europ. Econ. Rev. 43:1, pp. 91–107.

Burt, Ronald S. 1980. “Models of Network Structure,”Ann. Rev. Sociol. 6, pp. 79–141.

———. 1992. Structural Holes. Cambridge, MA:Harvard U. Press.

———. 2000. “Network Structure of Social Capital,”in Research in Organizational Behavior, vol. 22.Robert Sutton and Barry Staw, eds. Elsevier Science.

———. 2001. “Structural Holes versus NetworkClosure as Social Capital,” in Social Capital: Theoryand Research. Nan Lin, Karen Cook and R. S. Burt,eds. Aldine de Gruyter.

Cahuc, Pierre and Francois Fontaine. 2002. “On TheEfficiency of Job Search with Social Networks,” dis-cus. pap. 583, U. Paris 1-EUREQua.

Calvó-Armegnol, Antoni. 2004. “Job ContactNetworks,” J. Econ. Theory 115:1, pp. 191–206.

Calvó-Armegnol, Antoni and Matthew O. Jackson.2002. “Networks in Labor Markets: Wage andEmployment Dynamics and Inequality,” work. pap.Caltech and U. Autonoma de Barcelona; rev. April2003.

———. 2003. “The Effects of Social Networks onEmployment and Inequality,” work. pap. Caltechand U. Autonoma de Barcelona; forthcoming Amer.Econ. Rev.

Calvó-Armegnol, Antoni; Thierry Verdier and YvesZenou. 2004. “Strong and Weak Ties in Employmentand Crime,” IUI Conference on Networks: Theoryand Applications; Vaxholm, Sweden.

Calvó-Armegnol, Antoni and Yves Zenou. 2001. “JobMatching, Social Network and Word-of-MouthCommunication,” work. pap., U. Southampton, May;IZA discus. pap. 771, 2003.

Coleman, James S. 1990. Foundations of Social Theory.Cambridge, MA: Harvard U. Press.

Conley, Timothy and Georgio Topa. 2003a. “Socio-Economic Distance and Spatial Patterns inUnemployment,” J. Appl. Econometrics, forthcoming.

———. 2003b. “Dynamic Properties of LocalInteraction Models,” forthcoming in The Economyas an Evolving Complex System, III. LaurenceBlume and Steven Durlauf, eds. Menlo Park, CA:Addison-Wesley.

Corcoran, Mary; Linda Datcher and Greg Duncan.1980. “Information and Influence Networks in LaborMarkets,” in Five Thousand American Families, vol.VIII. Greg Duncan and James Morgan, eds. InstituteSocial Research, U. Michigan, pp. 1–37.

Coulson, N. Edward; Derek Laing and Ping Wang.1997. “Spatial Mismatch in Search Equilibrium,”work. pap. Pennsylvania State U.

Crane, Jonathan. 1991. “The Epidemic Theory ofGhettos and Neighborhood Effects on Dropping Otand Teenage Childbearing,” Amer. J. Sociol. 96:5, pp.1226–59.

Cutler, David M. and Edward L. Glaeser. 1997. “AreGhettos Good or Bad?” Quart. J. Econ. pp. 827–72.

Dasgupta, Partha and Ismail Serageldin. 2000. SocialCapital, A Multifaceted Perspective. WashingtonDC: World Bank.



Datcher, Linda. 1982. “Effects of Community andFamily Background on Achievement,” Rev. Econ.Statist. 64:1, pp. 32–41.

———. 1983. “The Impact of Informal Networks onQuit Behavior,” Rev. Econ. Statist. 65:3, pp. 491–95.

de Bartolome, Charles A. M. 1990. “Equilibrium andInefficiency in a Community Model with PeerGroup Effects,” J. Polit. Econ. 98:1, pp. 110–33.

Devine, Theresa and Nicholas Kiefer. 1991. EmpiricalLabor Economics: The Search Approach. NY: OxfordU. Press.

Dietz, Robert D. 2002. “The Estimation ofNeighborhood Effects in the Social Sciences,” Soc.Sci. Res. 31, pp. 539–75.

Dodds, Sheridan; Roby Muhamad and Duncan J.Watts. 2003. “An Experimental Study of GlobalSocial Networks,” Science 301:5634, pp. 827–29.

Durlauf, Steven N. 1994. “Bilateral Interactions andAggregate Fluctuations,” in Proceedings of the FirstWorld Congress of Nonlinear Analysts.

———. 1996a. “A Theory of Persistent IncomeInequality,” J. Econ. Growth 1:1, pp. 75–93.

———. 1996b. “Neighborhood Feedbacks,Endogenous Stratification, and Income Inequality,”in Dynamic Disequilibrium Modelling. W. Barnett,G. Gandolfo and C. Hillinger, eds. CambridgeU. Press.

Dutta, Bhaskar and Matthew O. Jackson. 2002. Modelsof the Strategic Formation of Networks and Groups.Berlin: Springer Verlag, forthcoming.

———.2003. “On the Formation of Networks andGroups,” in Networks and Groups. Dutta andJackson, eds. Berlin: Springer, pp. 1–15.

Elliott, James. 1999. “Social Isolation and LaborMarket Isolation: Network and NeighborhoodEffects on Less-Educated Urban Workers,” Sociol.Quart. 40:2, pp. 199–216.

Ellison, Glenn. 1993. “Learning, Local Interaction, andCoordination,” Econometrica, 61, pp. 1047–61.

Ellison, Glenn and Drew Fudenberg. 1995. “Word ofMouth Communication and Social Learning,” Quart.J. Econ. 110:1, pp. 93–125.

Erdös, Paul and Alfred Renyi. 1960. “On the Evolutionof Random Graphs,” Math. Institute HungarianAcademy Sci. 5, pp. 17–61.

Evans, William N.; Wallace E. Oates and Robert M.Schwab. 1992. “Measuring Peer Group Effects: AStudy of Teenage Behavior,” J. Polit. Econ. 100:5, pp.966–91.

Falk, Armin and Michael Kosfeld. 2003. “It’s All aboutConnections: Evidence on Network Formation,”work. pap. Institute Empirical Research Econ.,Zurich.

Fernandez, Roberto and Emilio Castilla. 2001. “HowMuch Is That Network Worth? Social Capital inEmployee Referral Networks,” in Social Capital:Theory and Research. Nan Lin, Karen Cook andRonald Burt, eds. NY: Aldine De Gruyter, pp. 85–104.

Finneran, Lisa and Morgan Kelly. 2003. “SocialNetworks and Inequality,” J. Urban Econ. 53:2, pp.282–99.

Freeman, Richard B. 2002. “The Labour Market in theNew Information Economy,” NBER work. pap. 9254.

Furstenberg, Frank F. and Mary Elizabeth Hughes.

1999. “The Influence of Neighborhoods onChildren’s Development: A Theoretical Perspectiveand a Research Agenda,” in Neighborhood Poverty:Policy Implications in Studying Neighborhoods, vol.II. Jeanne Brooks-Gunn, Greg J. Duncan and J.Lawrence Aber, eds. NY: Russell Sage.

Futia, Carl A. 1982. “Invariant Distributions and theLimiting Behavior of Markovian Economic Models,”Econometrica 50:2, pp. pp. 377–408.

Glaeser, Edward; David Laibson, Jose Scheinkman andChristine Soutter. 2000. “Measuring Trust,” Quart. J.Econ. 115:3, pp. 811–46.

Goyal, Sanjeev. 2003. “Learning in Networks: aSurvey,” forthcoming in Group Formation in Econ.:Networks, Clubs, and Coalitions. Myrna Woodersand Gabrielle Demange, eds. Cambridge U. Press.

Goyal, Sanjeev and Fernando Vega–Redondo. 2004.“Structural Holes in Social Networks,” work. pap.econ. dept. U. Essex.

Granovetter, Mark. 1973. “The Strength of Weak Ties,”Amer. J. Sociol. 78:6, pp. 1360–80.

———. 1983. “The Strength of Weak Ties: A NetworkTheory Revisited,” Sociol. Theory 1, pp. 201–33.

———. 1974(1995). Getting a Job: A Study ofContacts and Careers, first ed., Harvard U. Press;second ed., 1995, Chicago: U. Chicago Press.

———. 2000. “A Theoretical Agenda for EconomicSociology,” forthcoming in New Directions inEconomic Sociology. Mauro F. Guillen, RandallCollins, Paula England, and Marshall Meyer, eds.NY: Russell Sage.

Haynie, Dana L. 2001. “Delinquent Peers Revisited:Does Network Structure Matter?” Amer. J. Sociology106:4, pp. 1013–57.

Heavner, D. Lee and Lance Lochner. 2002. “SocialNetworks and the Aggregation of IndividualDecisions,” NBER work. pap. w8979.

Hedström, Peter; Ann-Sofie Kolm and Yvonne Åberg.2003. “Social Interactions and Unemployment,”work. pap. 2003:15, Institute Labor Market PolicyEvaluation, Uppsala.

Hogan, Dennis and Evelyn Kitagawa. 1985. “TheImpact of Social Status, Family Structure, andNeighborhood on the Fertility of BlackAdolescents,” Amer. J. Sociology 90, pp. 825–55.

Holzer, Harry. 1987a. “Informal Job Search and BlackYouth Unemployment,” Amer. Econ. Rev. 77:3, pp.446–52.

———. 1987b. “Job Search by Employed andUnemployed Youth,” Ind. Lab. Relat. Rev. 40:4, pp.601–11.

———. 1988. “Job Search Methods Used byUnemployed Youth,” J. Lab. Econ. 6:1, pp. 1–20.

Ioannides, Yannis M. 1997. “Evolution of TradingStructures,” in The Economy as an EvolvingComplex System II. W. Brian Arthur, Steven Durlaufand David Lane, eds. Menlo Park, CA: Addison-Wesley, pp. 129–67.

———. 2001. “Topologies of Social Interactions,”Tufts U., econ. dept. work. pap., revised June 2004.

———. 2002. “Nonlinear Neighborhood Interactionsand Intergenerational Transmission of HumanCapital,” in Essays in Economic Theory, Growth andLabour Markets: A Festschrift in Honour of



Emmanuel Drandakis. George Bitros and YannisKatsoulacos, eds. Cheltenham, UK: Edward Elgar,pp. 75–112.

Jackson, Matthew O. 2003. “A Survey of Models ofNetwork Formation: Stability and Efficiency,” work.pap. Caltech; forthcoming in Group Formation inEconomics: Networks, Clubs and Coalitions.Gabrielle Demange and Myrna Wooders, eds.Cambridge: Cambridge U. Press.

Jackson, Matthew O. and Alison Watts. 2002. “TheEvolution of Social and Economic Networks,” J.Econ. Theory 106:2, pp. 265–95.

Jackson, Matthew O. and Asher Wolinsky. 1996. “AStrategic Model of Social and Economic Networks,”J. Econ. Theory 71:1, pp. 44–74.

Jencks, Christopher and Marsha Brown. 1975. “TheEffects of High Schools on Their Students,” HarvardEd. Rev. 45, pp. 273–324.

Jencks, Christopher and Susan E. Mayer. 1990. “TheSocial Consequences of Growing Up in a PoorNeighborhood: A Review,” in Concentrated UrbanPoverty in America. M. McGeary and L. Lynn, eds.Washington DC: Nat. Academy.

Kempe, David; Jon Kleinberg and Éva Tardos. 2003.“Maximizing the Spread of Influence through aSocial Network,” SIGKDD ’03 Washington, DC.

Kirman, Alan. 1983. “Communication in Markets: ASuggested Approach,” Econ. Letters 12, pp. 101–108.

———. 1997. “The Economy as an InteractiveSystem,” in The Economy as an Evolving ComplexSystem II. Brian Arthur, Steven Durlauf and DavidLane, eds. Menlo Park, CA: Addison-Wesley, pp.491–531.

Kirman, Alan; Claude Oddou and Shlomo Weber.1986. “Stochastic Communication and CoalitionFormation,” Econometrica 54, pp. 129–38.

Korenman, Sanders and Susan Turner. 1996.“Employment Contacts and Minority–White WageDifferences,” Industrial Relations, 35:1, pp. 106–22.

Kortum, Samuel. 2003. “Book Review: Networks andMarkets, by James E. Rauch and Alessandra Casella,eds.” Russell Sage Foundation, 2001. J. Int. Econ.61:1, pp. 249–52.

Krauth, Brian V. 2000. “Social Interactions,Thresholds, and Unemployment in Neighborhoods,”econ. dept. Simon Fraser U. work. pap.

———. 2001a. “The Macroeconomics of JobNetworking,” econ. dept. Simon Fraser U. work. pap.

———. 2003a. “A Dynamic Model of Job Networkingand Persistent Inequality,” forthcoming inHeterogeneous Agents, Interactions, and EconomicPerformance. Cowan and Jonard, eds. Springer.

———. 2003b. “A Dynamic Model of Job Networkingand Social Influences in Employment,” econ. dept.Simon Fraser U. work. pap.; forthcoming in J. Econ.Dynam. Control.

Kremer, Michael. 1997. “How Much Does SortingIncrease Inequality,” Quart. J. Econ. 112:1, pp.115–39.

Kremer, Michael and Eric Maskin. 1996. “WageInequality and Segregation by Skill,” NBER work.pap. 5718.

Kugler, Adriana. 2002. “Employee Referrals andEfficiency Wages,” IZA discus. pap. 633.

Kuhn, Peter, and Mikal Skuterud. 2000. “Job SearchMethods: Internet versus Traditional,” MonthlyLabor Rev. pp. 3–11.

Lalive, Rafael. 2003. “Social Interactions inUnemployment,” CESifo work. pap. 1077.

Lin, Nan. 2001. Social Capital: A Theory of SocialStructure and Action. Cambridge and New York:Cambridge U. Press.

Lin, N.; W.M. Ensel and J.C. Vaughn. 1981. “SocialResources and Strength of Ties; Structural Factors inOccupational Status Attainment,” Amer. Soc. Rev.46, pp. 393–405.

Lin, N.; J.C. Vaughn and W.M. Ensel. 1981. “SocialResources and Occupational Status Attainment,”Social Forces 59, pp. 1163–81.

Lippert, Steffen, and Giancarlo Spagnolo. 2004.“Networks of Relations,” U. Mannheim econ. dept.work. pap.

Loury, Glenn C. 1977. “A Dynamic Theory of RacialIncome Differences,” in Women, Minorities andEmployment Discrimination. P.A Wallace and A. LeMund, eds. Lexington, MA: Lexington Books, ch. 8.

———. 1981. “Intergenerational Transfers and theDistribution of Earnings,” Econometrica 49:4, pp.843–67.

Loury, Linda. 2003. “Some Contacts Are More Equalthan Others: Earnings and Job InformationNetworks,” Tufts U. work. pap.

Lynch Lisa. 1989. “The Youth Labor Market in theEighties: Determinants of Re-employmentProbabilities for Young Men and Women,” Rev.Econ. Statist. 71:1, pp. 37–45.

Manski, Charles F. 2000. “Economic Analysis of SocialInteractions,” J. Econ. Perspect. 14:3, pp. 115–36.

Marmaros, David. 2001. “Analysis of Peer and SocialEffects on Employment Opportunities,” SeniorHonor Thesis, Department of Economics,Dartmouth College, June.

Marmaros, David and Bruce Sacerdote. 2002. “Peerand Social Networks in Job Search,” Europ. Econ.Rev. 46:4–5, pp. 870–79.

Marsden, Peter V. 1987. “Core Discussion Networks ofAmericans,” Amer. Sociol. Rev. 52, pp. 122–31.

———. 2001. “Interpersonal Ties, Social Capital,and Employer Staffing Practices,” in SocialCapital: Theory and Research. Nan Lin, KarenCook and Ronald Burt, eds. NY: Aldine DeGruyter, pp. 105–26.

Marsden, Peter and Karen Campbell. 1990.“Recruitment and Selection Processes: TheOrganizational Side of Job Searches,” in SocialMobility and Social Structure. Ronald Breiger, ed.NY: Cambridge U. Press, pp. 59–79.

Marsden, Peter and Elizabeth H. Gorman. 2001.“Social Networks, Job Changes, and Recruitment,”in Sourcebook of Labor Markets: Evolving Structureand Processes. Ivar Berg and Arne L. Kalleberg, eds.NY: Kluwer Academic/Plenum Publishers, pp.467–502.

Marsden, Peter and Hurlbert, Jeanne. 1988. “SocialResources and Mobility Outcomes: A Replicationand Extension,” Social Forces 66, pp. 1038–59.

Meade, James. 1976. The Just Economy. London:George Allen and Unwin.



Mencken, F. Carson and Idee Winfield. 2000. “JobSearch and Sex Segregation: Does Sex of SocialContact Matter?” Sex Roles: A Journal of Research,42:9–10, pp. 847–64.

Milgram, Stanley. 1967. “The Small World Problem,”Psych. Today 2, pp. 60–67.

Miyao, Takahiro. 1978a. “Dynamic Instability of aMixed City in the Presence of NeighborhoodExternalities,” Amer. Econ. Rev. 68:3, pp. 454–63.

Möbius, Markus. 2001. “Why Should Theorists Careabout Social Capital?” unpub. manuscript, HarvardU. econ. dept.

Moffitt, Robert. A. 2001. “Policy Interventions, Low-Level Equilibria, and Social Interactions,” in SocialDynamics. Steven N. Durlauf and H. Peyton Young,eds. pp. 45–82.

Montgomery, James D. 1991. “Social Networks andLabor Market Outcomes: Towards an EconomicAnalysis,” Amer. Econ. Rev. 81:5, pp. 1408–18.

———. 1992. “Social Networks and PersistentInequality in the Labor Market,” econ. dept.Northwestern U. mimeo.

———. 1994. “Weak Ties, Employment andInequality: An Equilibrium Analysis,” Amer. J.Sociol. 99:5, pp. 1212–36.

Mortensen, Dale and Tara Vishwanath. 1994. “PersonalContacts and Earnings: It Is Who You Know,” Lab.Econ. 1:1, pp. 187–201.

Munshi, Kaivan. 2002. “The Identification of NetworkEffects: Mexican Migrants in the U.S. LaborMarket,” econ. dept. U. Penn. work. pap.

Myerson, Robert B. 1977. “Graphs and Cooperation inGames,” Math. Op. Res. 2, pp. 225–29.

Newman, Mark E.J. 2003. “The Structure andFunction of Complex Networks,” SIAM Rev. 45, pp.167–256.

Oomes, Nienke. 2003. “Local Trade Networks andSpatially Persistent Unemployment,” J. Econ.Dynam. Control 27:11, pp. 2115–49.

Oreopoulos, Philip. 2003. “The Long-RunConsequences of Living in a Poor Neighborhood,”Quart. J. Econ. 118:4, pp. 1533–75.

Overman, Henry G. 2002. “Neighborhood Effects inLarge and Small Neighborhoods,” Urban Stud. 39:1,pp. 117–30.

Palmer, E.M. 1985. Graphical Evolution. NY: WileyInterscience.

Pellizzari, Michele. 2004a. “Do Friends and RelativesReally Help in Getting a Good Job?” CEP discus.Pap. 623, London School Econ.

———. 2004b. “Employers; Search and the Efficiencyof Matching,” preliminary, London School Econ.

Petersen, Trond, Ishak Saporta, and Marc-David L.Seidel. 2000. “Offering a Job: Meritocracy and SocialNetworks,” Amer. J. Sociol. 106:3, pp. 763–816.

Petrongolo, Barbara and Christopher A. Pissarides.2001. “Looking into the Black Box: A Survey of theMatching Function,” J. Econ. Lit. 39:2, pp. 390–431.

Pissarides, Christopher A. 2000. EquilibriumUnemployment Theory. Cambridge: MIT Press.

———. 2001. “The Economics of Search,” inInternational Encyclopedia of the Social andBehavioural Sciences, Amsterdam: Elsevier.

Podolny, Joel M. and James N. Baron. 1997.

“Resources and Relationships: Social Networks andMobility in the Workplace,” Amer. Soc. Rev. 62:5, pp.673–93.

Portes, Alejandro. 1998. “Social Capital: Its Origins andApplications in Modern Sociology,” Ann. Rev. Soc.24, pp. 1–24.

Ports, Michelle. 1993. “Trends in Job Search Methods:1970–1992,” Monthly Lab. Rev. Oct. pp. 63–67.

Putnam, Robert D. 2000. Bowling Alone. NY: Simon &Schuster.

Rees, Albert. 1966. “Information Networks in LaborMarkets,” Amer. Econ. Rev. Pap. Proceed. 56:2, pp.559–66.

Reid, Graham. “Job Search and the Effectiveness ofJob-Finding Methods,” Ind. Lab. Rel. Rev. 25, pp.479–95.

Rosenbaum, J.E.; S. DeLuca, S.R. Miller and K. Roy.1999. “Pathways into Work: Short- and Long-TermEffects of Personal and Institutional Ties,” Sociol.Ed. 72:3, pp. 179–96.

Schelling, Thomas C. 1971. “Dynamic Models ofSegregation,” J. Math. Soc. 1, pp. 143–86.

———. 1978. Micromotives and Macrobehavior. NY:Norton.

Simon, Curtis and Warner, John. 1992. “Matchmaker,Matchmaker: The Effect of Old-Boy Networks onJob Match Quality, Earnings, and Tenure,” J. Lab.Econ. 10:3, pp. 306–30.

Smith, Sandra. 2000. “Mobilizing Social Resources:Race, Ethnic, and Gender Differences in SocialCapital and Persisting Wage Inequalities,” Sociol.Quart. 41 4: pp. 509–37.

Staiger, Doug. 1990. “The Effects of Connections onthe Wages and Mobility of Young Workers,” work.pap. Harvard U. Kennedy School Govt.

Stigler, George J. 1961. “The Economics ofInformation,” J. Polit. Econ. 69:5, pp. 213–25.

———. 1962. “Information in the Labor Market,” J.Polit. Econ. 70:5, pp. 94–105.

Strand, Jon. 1983. “Equilibrium Wage Distributionswith Word-of-Mouth Worker Communication,” U.Oslo mimeo.

Topa, Georgio. 2001. “Social Interactions, LocalSpillovers and Unemployment,” Rev. Econ. Stud. 68,pp. 261–95.

Tassier, Troy and Filippo Menczer. 2002. “SocialNetwork Structure, Equality and Segregation in aLabor Market with Referral Hiring,” work. pap. U.Iowa econ. dept.

van den Berg, Gerard and Jan van Ours. 1996.“Unemployment and Duration Dependence,” J.Lab. Econ. 14:1, pp. 100–25.

Waldinger, Roger D. 1996. Still the Promised City?Cambridge MA: Harvard U. Press.

Wasmer, Etienne and Yves Zenou. 1997. “EquilibriumUrban Unemployment,” LSE Centre Econ. Perform.DP 368.

Wasserman, Stanley and Katherine Faust. 1994. SocialNetwork Analysis: Methods and Applications.Cambridge and NY: Cambridge U. Press.

Wasserman, Stanley and Garry Robins. 2001. “AnIntroduction to Random Graphs, DependenceGraphs, and p*,” forthcoming in Models andMethods in Social Networks Analysis. Peter


Carrington, John Scott and Stanley Wasserman, eds.Cambridge and NY: Cambridge U. Press.

Watts, Alison. 2001. ”A Dynamic Model of NetworkFormation,” Games Econ. Behav. 34, pp. 331–41.

Weinberg, Bruce A.; Patricia B. Reagan and Jeffrey J.Yankow. 2000. “Do Neighborhoods Affect WorkBehavior? Evidence from the NLSY79,” forthcom-ing, J. Lab. Econ.

White, Harrison C. 1995. “Network Switchings and

Bayesian Forks: Reconstructing the Social andBehavioral Sciences,” Soc. Res. 62:4, pp. 1035–63.

Young, H. Peyton. 1993. “The Evolution ofConventions, “ Econometrica 61:1, pp. 57–84.

———. 1998. Individual Strategy and SocialStructure. Princeton, NJ: Princeton University Press.

Zuckerman, Ezra W. 2003. “On Networks and Marketsby Rauch and Casella, eds,” J. Econ. Lit. 41:3, pp.545–65.



Documents

Job Information Networks, Neighborhood Effects, and Inequality · studies. Using their Dartmouth College data, David Marmaros (2001) and Marmaros and Bruce Sacerdote (2002) found