Abstract

Excerpted From: Amanda Agan and Sonja Starr, Employers' Neighborhoods and Racial Discrimination, 53 J. Journal of Legal Studies 115 (January, 2024) (33 Footnotes) (Full Document)

This study investigates the role of neighborhoods in shaping employment discrimination and disparity. We use data from a large field experiment in which we sent over 15,000 fictitious job applications from Black and white pairs of job seekers to businesses throughout New Jersey and New York City. This paper makes three key contributions. First, we show that employers in whiter and less-Black neighborhoods discriminate much more heavily in favor of white applicants, a result that is robust to the inclusion of a rich set of controls. Second, we build on our prior finding that when employers lose access to criminal records, they discriminate on the basis of race more (Agan and Starr 2018); we now show that this effect is driven by employers in neighborhoods with low populations of Black residents, who appear especially prone to stereotype Black applicants as criminal. Third, we provide evidence of racial disparities in locations of job opportunities and, using simulations, show how these disparities combine with local variation in discrimination patterns to shape racial gaps in employment. Each of these findings may help to motivate policies designed to bring jobs into Black neighborhoods, and the second also informs expectations about the likely effects on employment disparities of laws restricting access to criminal records. These insights may be useful to policy makers seeking to address persistent employment discrimination (see Quillian et al. [2017] for a meta-analysis of field experiments) and high Black unemployment rates (DeSilver 2013).

Our research method, known as an audit study, has long been a key tool for studying discrimination in employment and other areas (Neumark, Bank, and Van Nort 1996; Riach and Rich 2002; Pager 2003; Bertrand and Mullainathan 2004; Lahey 2008; Oreopoulos 2011; Kroft, Lange, and Notowidigdo 2013; Deming et al. 2016). To our knowledge, this is the first audit study to closely examine the effects of neighborhood racial composition on employment discrimination, though two others (Bertrand and Mullainathan 2004; Kline, Rose, and Walters 2022) briefly consider the question. Our study is particularly well designed to explore this relationship. The targeted positions are overwhelmingly service jobs at employers with localized customer bases, distributed across two large jurisdictions with wide variation in racial composition and other neighborhood characteristics. We tailored applications to be competitive in all those localities, carefully choosing addresses for applicants in nearby neighborhoods.

We find robust evidence that the racial composition of employers' neighborhoods predicts discrimination. In our simplest specification, we estimate a 2.7-percentage-point callback advantage for Black applicants in entirely Black neighborhoods (a 25 percent increase over the baseline callback rate for Black applicants) and a 3.4-percentage-point advantage for white applicants in entirely non-Black neighborhoods (a similar 25 percent increase over the baseline callback rate for white applicants). Because job openings are mainly in non-Black neighborhoods, white applicants have a clear net advantage.

There are several theoretical reasons to expect such a relationship. These include possible manifestations of in-group bias: hiring managers might cater to the perceived in-group preferences of local customers, indulge their own in-group biases or those of existing staff, or anticipate in-group preferences of the applicant to try to increase yield or retention. Employers with less frequent contact with Black individuals may be more likely to stereotype Black applicants negatively. There is also a possible reverse-causal pathway; firms inclined to discriminate could choose to locate in less-Black neighborhoods.

While we cannot necessarily pinpoint the exact mechanisms that underpin our main results, we can rule some pathways in or out. Our results hold conditional on the inclusion of fixed effects for the chain to which an establishment belongs, which (discussed below) the reverse-causal pathway cannot explain; rather, a neighborhood's racial composition appears to shape the hiring choices of local managers in some way. One mechanism that seems to contribute is that employers in less-Black neighborhoods are more likely to stereotype Black applicants as criminal. In Agan and Starr (2018) we provide evidence that ban-the-box (BTB) laws (which hide criminal records from employers) caused a large spike in the callback gap between Black and white applicants (hereafter, the Black-white callback gap), which suggests that employers make negative assumptions about Black applicants' records. These assumptions appear very exaggerated relative to real-world conviction-rate differences--evidence of stereotyping. Now we show that the BTB effect is driven entirely by neighborhoods with low shares of Black residents, which suggests that such stereotypes are more prevalent there. Still, crime-related stereotyping cannot explain most of the interaction between neighborhoods' racial composition and applicants' race, which is very strong even when employers have information about individuals' criminal records. Other place-based mechanisms must also be at play.

Many studies document in-group bias, including lab experiments (for reviews, see Hewstone, Rubin, and Willis 2002; Anderson, Fryer, and Holt 2006), implicit-bias studies (for example, Quillian and Pager 2001), and surveys (Greenwald and Krieger 2006). But such studies do not explain how biases translate into real-world employment decisions. Observational studies find suggestive real-world evidence of in-group bias--for example, surveys of employers find that customers' demographics, hiring managers' race, and urban location are all correlated with the probability that the most recent hire is Black (Holzer and Ihlanfeldt 1998; Raphael, Stoll, and Holzer 2000). Such studies face identification challenges, especially because researchers typically lack much information about the pool of applicants, which might vary by race in ways that correlate with the race of customers, managers, and existing staff. Without that information, it is hard to say in which direction(s) discrimination runs. For example, if employers in Black neighborhoods hire more Black applicants, multiple explanations are possible: those employers could be biased in Black applicants' favor, employers in other neighborhoods could be biased against them, all could be biased in the same direction but to different degrees, or employers could be unbiased but face different pools of candidates across neighborhoods.

The auditing approach, in contrast, allows otherwise identical pools of candidates to be compared, which produces well-identified evidence of discrimination in a real-world setting. To be sure, we experimentally identify only racial discrimination itself; neighborhoods' racial composition cannot be experimentally manipulated. Thus, while we can say definitively that racial discrimination patterns vary sharply with neighborhoods' racial composition, our ability to explain why depends on a selection-on-observables assumption. However, our result is robust to the inclusion of a rich set of controls for other characteristics of neighborhoods and employers, which strongly suggests a causal role of neighborhoods' racial composition in shaping discrimination.

Our causal interpretation is not definitive (although we think it is well supported), and our conclusions about specific mechanisms are even less decisive. But all studies of the effects of neighborhoods' racial composition face similar identification constraints plus often others that our experimental design avoids. Racial composition generally changes slowly and is not readily manipulated, and any shocks to it typically change neighborhoods in too many ways to serve as natural experiments. Moreover, for many purposes, it is not important to identify the mechanism or even whether neighborhoods' racial composition plays a causal role. What we do strongly identify-- discrimination patterns strongly predicted by neighborhoods' racial composition--shapes disparities in employment opportunities, with policy implications (discussed below) that do not turn on whether the relationship is mediated by unobserved characteristics of neighborhoods.

Beyond examining how discrimination patterns vary by neighborhood, we also assess how those patterns translate into racial gaps in employment opportunities. Importantly, the reversal of discrimination patterns in Black neighborhoods does not mean that their effects even out. There are more non-Black neighborhoods, job postings are concentrated there, and overall callback rates are higher there. Our study produced data about where jobs are located, because the market defined our sample. Using popular online job search sites, we applied for every job we could find in New Jersey and New York City within our search constraints (which targeted entry-level, low-skill jobs). So this study allows us to investigate how local variation in employment discrimination and job availability combine to produce racial disparities in employment.

This combination may produce much larger disparities in real-world settings than the average 23 percent callback advantage for white applicants in the experimental sample. In our experiment, we artificially made the fictitious Black and white applicants come from identical neighborhoods and apply to identical jobs. This was important for causal identification, but real-world disparities are also affected by the fact that applicants of different races tend to have different nearby job opportunities. One might hope that this could have a positive effect--perhaps, for real-world populations, geographic self-sorting (applying to jobs close to home) could mitigate disparities, because Black applicants would tend to apply in Black neighborhoods, where employers seem to favor them. If so, it could illustrate the prediction of Becker (1957) that job applicants generally sort toward firms less likely to discriminate against them. But even if that is so, the net impact of such self-sorting on hiring disparities depends on the geographic distribution of job openings. When jobs are scarce where Black people live, such sorting can exacerbate disparity; the effects of the reversed discrimination pattern in those neighborhoods can easily be overwhelmed by the fact that there are few vacancies and few callbacks overall.

We illustrate this point through simulations that reweight our data such that the distribution of Black and white applicants by neighborhood mirrors the real-world population, incorporating commuting-time data to define job search parameters. In New York City, where job availability heavily favors white neighborhoods, all our simulations predict racial disparities far exceeding those in the experiment: white job seekers are projected to receive between 68 percent and 190 percent more callbacks per capita than equally qualified Black job seekers. In New Jersey, the job distribution pattern is more nuanced (white neighborhoods have more jobs in them, but Black neighborhoods are in denser regions and may have more jobs near them), and the simulation results vary according to our assumptions about job searches. Overall, the evidence that geographic self-sorting can alleviate racial disparities is weak, and in New York the opposite is clearly predicted.

Our findings on discrimination and job accessibility are relevant to a large literature that evaluates how geography shapes underemployment for Black individuals (for reviews, see Ihlanfeldt 1994; Kain 2004; Gobillon, Selod, and Zenou 2007). Much of that literature centers on the spatial-mismatch hypothesis first developed by Kain (1968) (see Stoll and Covington [2012] for a more recent example). Kain hypothesizes that housing segregation reduces job opportunities for Black residents and contributes to employment gaps between Black and white individuals. A key premise is that fewer jobs are located in Black neighborhoods; the mismatch is between the location of jobs and residential location by race (Kain 1968; Ihlanfeldt 1994). Another premise is that people tend to seek jobs close to home to reduce commutes and search costs (Kain 1968; Ihlanfeldt 1994; Gobillon, Selod, and Zenou 2007). Indeed, Black residents may be particularly constrained from pursuing distant employment partly because of lower car-ownership rates and less access to public transit (Kain 1968; Mouw 2000; Raphael and Stoll 2002; Johnson 2006; Gautier and Zenou 2010).

Most of the literature finds that job distributions disfavor Black communities (Raphael and Stoll 2002; Stoll 2006; Stoll and Covington 2012), and some find that this gap has expanded over time (Ihlanfeldt 1994; Mouw 2000; Stoll 2006; Gobillon, Selod, and Zenou 2007; Miller 2023; for a contrary view, see Kneebone and Holmes 2015). But many scholars conclude that employment discrimination is a more important explanation than spatial mismatch for persistent disparities (see Ellwood [1986] and Leonard [1986] for seminal examples of this theory of “race, not space”). Our findings show how the two work in tandem: space heavily mediates the role of race. Kain (1968) and others suggest that local variN ation in employment discrimination could exacerbate spatial mismatch, but this idea has rarely been tested empirically. The literature on spatial mismatch mainly has relied on economic survey data about residential location and job location; such surveys typically lack both data on job applications (for example, Hellerstein, Neumark, and McInerney 2008; Stoll and Covington 2012; Johnson 2006 is an exception) and information about pools of applicants, which limits their utility in assessing discrimination (see Mouw 2000).

Here, in contrast, we directly test local variation in discrimination, and our simulations show how it combines with variation in job access to produce disparities. Our study also complements the literature on spatial mismatch in other ways. We focus not on overall distribution of jobs but on low-skill job vacancies, identified by typical modern job search methods. By assessing the geographic distribution of callbacks, we implicitly account for differences in applicant pools. Compared with most of the literature, we use more recent data and more realistic rush-hour driving and public transit commuting-time data rather than aerial or driving distance.

Our findings inform debates about legal and policy interventions designed to improve employment in Black communities. In general, our findings tend to support place-based interventions that address spatial mismatch by incentivizing employers to locate and hire in Black neighborhoods. Such interventions can mitigate hurdles for Black job seekers in two ways: reducing employment discrimination and increasing physical access to jobs. Moreover, while many factors shape specific policy choices, our findings suggest that it may be more effective to redress spatial mismatch by bringing jobs into Black communities than by bringing residents of those communities to jobs (for example, via transit assistance). This is because even if Black residents can travel or move to less-Black neighborhoods to seek work, they will likely face more racial discrimination when they get there. These policy implications do not necessarily depend on the causal nature of the relationship between racial composition and discrimination, much less particular mechanisms; even if it were driven by other race-correlated characteristics of neighborhoods, it would still be the case that shifting job opportunities to Black neighborhoods would be expected to reduce the level of discrimination that Black applicants face.

Policy makers have many potential tools for incentivizing firms to locate in neighborhoods with more Black residents--for example, economic enterprise zones and other tax-incentive-based programs, economic development subsidies, brownfield redevelopment, changes to zoning and other regulatory requirements to facilitate business development, and place-based wage subsidies. While specific policy recommendations are beyond this paper's scope, Section 4 briefly examines the policy landscape and existing literature on the efficacy of these options. In addition, our findings are relevant to the ongoing debate about BTB laws, adding nuance to our prior findings. Ban-the-box laws are one of the most important policy tools for reducing employment barriers to people with criminal records, but their unintended racial consequences are potentially a major drawback. Our new findings suggest that the cost-benefit analysis could differ substantially on the basis of the racial composition of the adopting jurisdiction.

[. . .]

Finally, our findings may have other implications for law and policy beyond BTB laws and place-based policies. First, they add to the existing body of experimental evidence documenting employment discrimination in favor of white applicants and against Black applicants and thus generally support interventions to improve enforcement of existing antidiscrimination laws. But successfully improving enforcement is not an easy task. Enforcing hiring-discrimination prohibitions is particularly difficult (see Neumark, Song, and Button 2017), because if one is not called back for a job, it is usually essentially impossible to prove why: an employer could point to numerous distinctions among candidates. Indeed, even armed with experimental evidence like ours (which is difficult and costly to produce), one could only compellingly demonstrate a pattern of discrimination in the broader market, not that any one particular callback decision was based on race, as an individual plaintiff would need to prove. Moreover, hard-to-quantify damages can further deter enforcement lawsuits (Neumark, Song, and Button 2017). Empiricists have even debated whether antidiscrimination laws reduce hiring of protected groups, because they are more enforceable at subsequent stages (for example, wrongful termination), and firms may make hiring decisions designed to reduce their long-term exposure (see Bloch 1994; Posner 1995; Epstein 1992; Neumark, Song, and Button [2017] find no evidence of this unintended consequence in the age-discrimination context). In any case, there is little reason to be sanguine about improving hiring-stage enforcement, given the persistence of these problems. Quillian et al. (2017) and Quillian and Lee (2023) conduct meta-analyses of US and international audit studies, respectively, and find no reduction in hiring discrimination over the course of decades.

We think that it is worth considering novel strategies for preventing hiring discrimination ex ante instead of focusing on ex post legal remedies. In a strategy inspired by BTB laws, for example, employers could remove from application forms names and other fields (like home address) that tend to identify race but are not job relevant. Although employers would presumably acquire information about race later (for example, via interviews), most applicants (nearly 90 percent in our study) are screened out at the initial application stage, so reducing discrimination at this stage is important. The difference in discrimination patterns between New Jersey and New York City, which the observable variables we analyzed does not explain, also merits further study; it may stem from some policy intervention in New York City that other jurisdictions could emulate.

amanda agan is Associate Professor of Economics at Rutgers University.

sonja starr is Julius Kreeger Professor of Law and Criminology at University of Chicago.