Some empirical evidence of the efficacy of job matching in urban labor markets.(ORIGINAL PAPER)

Abstract Theory predicts that workers in cities are more likely to engage in job search, ceteris paribus, due to market efficiencies associated with greater job density. However, if job search is more efficient in urban markets, then the quality of a given job match should also tend to be higher in cities, ceteris paribus. Employed workers living in cities might then be expected to search less than their nonurban counterparts. In this latter instance, it is not city residency itself that makes search less likely, but rather the positive correlation between city residency and job match quality. Using data from the National Longitudinal Survey of Youth 1979, this prediction is confirmed: The estimated coefficient on an indicator of urban residency is found to be near zero and statistically insignificant in models of employed search that omit proxies for job match quality. When job match proxies are included in the models, the estimated coefficient on urban residency becomes positive and highly significant. This result suggests that workers are not only more likely to engage in employed search in urban labor markets, but also tend to find more productive job matches in cities over time.

Keywords Job search * Job matching * Urban labor markets * Coordination hypothesis

JEL J0 * J62 * R0 * R23

Introduction

It is well established that urban workers collect a significant wage premium over characteristically similar workers in living nonurban areas. For example, Glaeser and Mare (2001) report that prime-age workers in dense metropolitan areas earn 25% more when controlling for basic observable characteristics, while Yankow (2006) identifies a 19% wage advantage under similar circumstances for younger workers. Several competing theories purport to explain this empirical regularity. Two of the leading theories put forth by urban economists are the learning hypothesis which posits externalities in learning and human capital production (Rauch 1993; Glaeser 1999; Moretti 2004), and the coordination hypothesis which emphasizes efficiencies in job search and matching between workers and firms (Kim 1990; Helsey and Strange 1990; Sato 2001). Although recent empirical work has begun to highlight the importance of the latter theory as it relates to the urban wage premium (Wheeler 2006; Yankow 2006; Andersson et al. 2007), direct empirical evidence of superior job matching in urban labor markets has been elusive.

The purpose of this study is to explore the relationship between job search and match quality in urban labor markets. Formal search theory suggests that drawing from a superior wage offer distribution, realizing a higher arrival rate of job offers, or having lower search costs increases the expected net benefits of search, making both search and job change more likely (Mortensen 1986). Because each is more likely to be true for workers searching in dense urban labor markets, a straightforward prediction of the model is that workers in cities have a greater incentive to engage in job search, ceteris paribus. Over time, however, as workers move into superior job matches, the expected benefit of continued search declines. If job search is indeed more efficient in urban labor markets as hypothesized, then the quality of a given job match should also tend to be higher in cities, ceteris paribus. Hence, employed workers living in cities might be expected to search less than their nonurban counterparts if the average job match is better in cities. In this latter instance, it is not city residency itself that makes search less likely, but rather the positive correlation between city residency and job match. Using data on employed search behavior, this notion is exploited to test whether workers are more likely to engage in job search in cities and whether the quality of the job match tends to be higher in cities.

Empirical Strategy

The labor market coordination hypothesis suggests at least two advantages to searching in urban rather than nonurban labor markets. First, a greater number of job openings in any time period allow more jobs to be sampled more quickly and increases the probability of receiving an acceptable offer (Burdett 1978). Second, the greater density of employment opportunities and more efficient spatial configuration in urban areas reduces search time and travel costs per job vacancy. The likely outcome for workers searching in urban labor markets is more productive job turnover and greater wage growth when changing employers. In other words, workers will tend to find better job matches in less time than workers living outside of urban areas. This tendency generates a positive correlation between urban residency and the quality of the job match. Consequently, regression models that omit measures of the quality of the match between the worker and firm will produce a biased coefficient on the covariate indicating urban residency, since the coefficient confounds the (positive) impact of the urban labor market itself on the propensity to search and the (negative) influence of a good job match.

To see why, suppose that the true model relating both city residency C and the quality of the job match M to the likelihood of employed job search S is:

S = [[beta].sub.1]C + [[beta].sub.2]M + [epsilon] (1)

where [epsilon] is a random error term. Due to the urban labor market efficiencies discussed above [[beta].sub.1] is hypothesized to be positive, while [[beta].sub.2] is hypothesized to be negative. Also from the above discussion, we know that C and M are likely to be positively correlated. Now suppose that M is omitted from the estimating equation. The estimate of [[beta].sub.1] obtained from OLS regression is then

[[^.[beta]].sub.1] = [[SIGMA]CY/[SIGMA][C.sup.z]] = [[[SIGMA]C([[beta].sub.1]C + [[beta].sub.z]M + [epsilon])]/[SIGMA][C.sup.z]] = [[beta].sub.1] + [[beta].sub.2] [[SIGMA]CM/[SIGMA][C.sup.z]] + [[SIGMA][C.sup.[epsilon]]/[SIGMA][C.sup.z]] (2)

Since E([SIGMA][C.[epsilon]] = 0, we get

E([[^.[beta]].sub.1]) = [b.sub.1] + [b.sub.MC][[beta].sub.2] (3)

where [b.sub.MC] is the regression coefficient from a regression of M on C. Thus, [[^.[beta]].sub.1] is a biased estimator for [[beta].sub.1]. More specifically, we can see that omission of the control for match quality M will bias the estimated coefficient on the city variable C (expected to be positive) toward zero since [[beta].sub.2] is negative and [b.sub.MC] is positive.

Using this idea to advantage, this study conducts a straightforward test of the coordination hypothesis. First, a model of the probability of employed search is estimated while omitting key measures of the quality of the job match between the worker and her current employer. If job matching is important and positively correlated with city employment, the estimated coefficient on an indicator of urban residency is likely to be close to zero (or equivalently, statistically insignificant). Upon adding controls for the quality of the job match to the specification, however, the estimated coefficient on the indicator of urban residence is expected to be positive, larger by an order of magnitude (compared to the model without the job match controls), and statistically significant (assuming that job search is more likely to occur in dense urban areas as the theory suggests). Moreover, the proxies for job match quality should exhibit a significant negative impact on the propensity to engage in employed job search.

The decision to search reflects an underlying marginal net benefit calculation (Burdett 1978; Mortensen 1986). Let the difference between the benefit and cost of employed search be represented by the latent variable, S*, such that

S* = [X.sub.it][beta] + [C.sub.it][gamma] + [M.sub.it][delta] + [a.sub.i] + [[epsilon].sub.it] (4)

The individual's observable characteristics at time t are given by [X.sub.it]. Information on city residency at time t is given by [C.sub.it], while information on the quality of the job match at time t is contained in [M.sub.it]. Both [[alpha].sub.i] and [[epsilon].sub.it] are components of the error specification. The former can be thought of as an unobservable time-invariant, person-specific factor that affects search propensity, while the latter is a purely random disturbance.

Although we do not observe the latent variable, S*, we do observe whether employed job search has occurred. That is, we observe

S = 1 iff S* > 0

S = 0 iff S* [less than or equal to] 0 (5)

The probability that employed search occurs can then be modeled as

Prob(S* > 0) = Prob([X.sub.it][beta] + [C.sub.it][gamma] + [M.sub.it][delta] + [[alpha].sub.i] + [[epsilon].sub.it] > 0). (6)

Assuming that [[alpha].sub.i] is equal to zero and [[epsilon].sub.it] is distributed (identically and independently) normal with mean zero and variance one, the model can be estimated as a Probit (suitably adjusted for panel data). However, if [[alpha].sub.i] (unobservable individual facility in or preferences towards employed job search) is not zero, standard Probit methodology may produce biased parameter estimates. To account for this possibility, one can treat the person-specific error component [[alpha].sub.i] as a random disturbance that varies across individuals but not over time and estimate Eq. (6) as a random-effects Probit model. The proportion of the total variance contributed by the random-effect variance component can then be examined in order to test the appropriateness of the random-effects assumption.

Data and Sample Selection

The data for this analysis come from the National Longitudinal Survey of Youth 1979 (NLSY79). The original NLSY79 cohort contains 12,686 men and women born between 1957-1964. Personal and employment information was collected from respondents on an annual basis from 1979 through 1994 and biennially thereafter. Unfortunately, questions regarding employed search are not asked in every year of the survey. Data on employed search is available annually from 1980 through 1984 but then not again until the 1996 interview. As a consequence, the analysis is concentrated on a limited number of survey years. In all, measures of on-the-job search are constructed corresponding to the 1980, 1984, and 1996 survey waves of the NLSY79. (1)

Print Get the Mag Weekly Updates