NETWORK SEARCH: CLIMBING THE JOB LADDER FASTER

• Author: Dennis O'Dea,Marcelo Arbex,David Wiczer
• Published date: 01 May 2019
• Date: 01 May 2019
• DOI: http://doi.org/10.1111/iere.12375

INTERNATIONAL ECONOMIC REVIEW

Vol. 60, No. 2, May 2019 DOI: 10.1111/iere.12375

NETWORK SEARCH: CLIMBING THE JOB LADDER FASTER∗

BYMARCELO ARBEX,DENNIS O’DEA,AND DAVID WICZER1

University of Windsor, Canada; University of Washington, U.S.A.; Stony Brook University,

U.S.A .

We introduce an irregular network structure into a model of frictional, on-the-job search in which workers

find jobs through their network connections or directly from firms. We show network-found jobs have higher

wages, and thus better-connected workers climb the job ladder faster. The mean field approach allows us

to formulate heterogeneous workers’ recursive problem tractably. Our calibration is consistent with several

empirical findings because of a composition—not information—effect. We also introduce new model-consistent

evidence: Job-to-job switches at higher ladder rungs are more likely to use networks.

1. INTRODUCTION

How do workers find jobs and does it matter? In this article, we study an equilibrium search

model in which some jobs are found through network connections and workers differ in their

position within the network. In our framework, jobs may be found by direct contact with a firm

or through a searcher’s network, but by the latter method they draw fundamentally different

and better offers. There are two reasons. First, because we assume network connections pass

along wage offers like their own, offers passed through network search come from workers who

have already selected wages and climbed the ladder. The second reason comes endogenously

from the interaction between network structure and on-the-job search: Job referrals come dis-

proportionately from workers who are more central to the network, and these better-connected

workers have more access to better jobs and thus even higher wages. The heterogeneity pre-

sented in our model, workers’ network position, implies both differences in workers’ finding

rate and differences in the distribution from which they sample wage offers. Finding rates dif-

fer because better-connected workers have more connections from whom to draw a referral.

Draw distributions differ because better-connected workers samples more frequently from the

network distribution. To understand this mechanism, we show analytically how network search

affects equilibrium on-the-job search, then illustrate its features in a calibrated version of the

model and discuss how it relates to common empirical findings associated with job referrals and

network search.

The model is able to tractably incorporate two methods of search, direct contact and network

referral, along with heterogeneity associated with the latter. This tractability comes from our

ability to summarize a worker’s network position into a scalar state variable, her number of con-

nections. To do so, we employ the mean field approach, which amounts to a set of assumptions

over the network and information structure to omit local correlation and neighborhood effects.2

Rather, agents will use the global network structure to take expectations over their peers’ types.

∗Manuscript received August 2016; revised April 2018.

1We have benefited from comments and suggestions from Manolis Galenianos, George-Levi Gayle, Aaron Hedlund,

Henry Hyatt, Leo Kaas, Rasmus Lentz, Paulina Restrepo, Giorgio Topa, Weilong Zhang, and participants at various

seminars and conferences. Any errors are our own. Please address correspondence to: David Wiczer, Department of

Economics, Stony Brook University, 100 Nicolls Rd., Stony Brook, NY 11790. E-mail: wiczerd@gmail.com.

2The mean field approach is a technique to analyze the long run or average behavior of a complex system, and it

was first developed for use in statistical physics to analyze the Ising model of interacting charged particles (see Vega-

Redondo, 2007, for a summary). By replacing the effect of nearby particles with the “average” effect of the magnetic

693

C

(2018) by the Economics Department of the University of Pennsylvania and the Osaka University Institute of Social

and Economic Research Association

694 ARBEX,O’DEA,AND WICZER

This simplifies the state while still preserving the network’s influence in an equilibrium we term

a “Sufficient Recursive Equilibrium” because it uses a sufficient, scalar statistic to summarize

the worker’s network position.

Our principal theoretical result shows that the equilibrium wage offer distribution via so-

cial networks first-order stochastically dominates the wage offer distribution via direct search.

Underlying this result, we show that the reservation wage is decreasing and employment rate in-

creasing in the number of connections. Therefore, well-connected individuals are more likely to

be employed, and the distribution of referrals disproportionately weights these network-central

peers. The distribution of offers thus comes from a weighted average of wages of employed

workers, one that dominates the overall wage distribution of employed workers, which, in turn,

dominates the distribution of offers. Introducing network search also affects the equilibrium

wage and offer distributions by reducing firms’ profits. This occurs because the lowest-wage firm

shrinks as its workers, who are more likely well connected, are more quickly poached away.

With falling profits at the lowest-wage level, free entry implies they also fall throughout.

Because our model includes both network search and direct contact search, it is a useful

framework for understanding the differences between a job found via network search and

directly. Our model is consistent with many empirical findings about these differences.3In par-

ticular, jobs found through a worker’s network have (i) higher wages and (ii) longer employment

duration and (iii) workers experience shorter unemployment spells. These are all predictions

of our model. Network search brings higher wages both because network offers come from a

better distribution and also because those who find a job through their network tend to be better

connected, a composition effect. Employment duration is longer because these higher wages are

higher on the job ladder. And because of composition, unemployment spells are shorter; those

who find jobs through networks are better connected and have faster finding rates. Although

other studies have attributed these features to information frictions,4we show that network

heterogeneity and a job ladder are sufficient. We also contribute to the empirical literature by

introducing findings from the Survey of Consumer Expectations (SCE). A prediction central

to our model is that network search becomes increasingly important at higher rungs of the job

ladder. In fact, with the SCE, we find direct evidence of this: Higher-paid workers are more

likely to find their next job via a referral, and we use the rate that this likelihood increases as a

target in our calibration.

There is an extensive literature about frictional labor markets and wage dispersion. The basis

for our model, Burdett and Mortensen (1998), develops a wage-posting model in which firms

offer high wages to attract workers from other firms and to reduce worker turnover. Their

main result is an equilibrium with wage dispersion even if workers and firms are identical.

On-the-job search is crucial, allowing firms with different wages to attract different sets of

workers, depending on the wage in their current job. Although Burdett and Mortensen (1998)

also consider a version with reservation heterogeneity, this has quite different features than

our network heterogeneity and so, for simplicity, our benchmark comparison will be the pure-

dispersion version.

To match empirical wage distributions better than Burdett and Mortensen (1998), a list of

papers introduces additional heterogeneity. Postel-Vinay and Robin (2002) distinguish three

sources of wage dispersion: productive heterogeneity of workers, productive heterogeneity

of firms, and search frictions. They find all three sources are quantitatively important, and

field, one can solve for the equilibrium field without need to consider the potentially intractable interaction of local

effects.

3A large empirical literature has shown that labor market networks play an important role in matching workers to

employers. Even though estimates of the percentage of jobs found through social contacts vary across location and

profession, they consistently range between 25% and 80% of jobs in a given profession (see, for instance, Holzer, 1988;

Ioannides and Datcher Loury, 2004; Bayer et al., 2008; Hellerstein et al., 2011).

4Many principally empirical papers ascribe their findings, that network search leads to longer-lasting matches at

higher wages, to network connections ameliorating information frictions (see Datcher, 1983; Simon and Warner, 1992;

Granovetter, 1995; Marmaros and Sacerdote, 2002; Kugler, 2003; Dustmann et al., 2016).