Discrimination in Low-Wage Labor Markets:
Evidence from an Experimental Audit Study in New York City
Devah Pager
Bruce Western
Princeton University

Submission to the Population Association of America Annual Meetings 2005

Discrimination in Low-Wage Labor Markets:
Evidence from an Experimental Audit Study in New York City
This study considers the implications of three distinct trends for the prevalence of
discrimination against young men in low-wage labor markets. Rising inequality, sharply
increasing incarceration rates, and growing numbers of immigrants each contribute to a
population of low-wage workers with characteristics very different from those who may
employ them. The low levels of trust characteristic of these employment relationships
may be associated with discriminatory hiring practices that limit the employment
opportunities available to certain groups of workers.
Little is known about the extent of employment discrimination against minorities and
criminal offenders. Vast disparities can be observed in employment rates by race,
ethnicity, and incarceration status (Freeman & Holzer, 1986; Western & Pettit, 2000), but
the causes of these disparities remain controversial. Unobserved skill differences and the
self-selection of workers into segregated labor markets prevent us from directly
comparing the outcomes of various groups using standard data sources. Likewise, most
recent research debating the effects of race discrimination or criminal stigma has studied
the labor market fortunes of workers, rather than the hiring behavior of employers.
Without observing discrimination directly, it is difficult to make causal statements about
the nature or magnitude of the barriers to employment for disadvantaged groups.
Our paper reports new results from a novel study of employers, reporting the reactions to
minority and formerly-incarcerated job seekers. Using an experimental audit
methodology, we sent teams of male testers with equivalent resumes to apply for entrylevel jobs in New York City. Testers were matched on the basis of age and appearance;
after selection, they participated in extensive training to ensure consistency in their
interactions with employers. The testers used fictitious matched resumes reflecting equal
levels of education and work experience. In several of the teams, the resumes also
reflected evidence of an 18-month term of incarceration. Testers within teams rotated
which member of the team served in the criminal record condition to control for
unobserved differences within tester pairs that could affect hiring outcomes. Because
testers are given equivalent resumes, and criminal conviction status is randomly assigned,
the unobserved heterogeneity that typically plagues studies of workers is minimized in
this experimental setting.
The audit study began in February 2004, employing a dozen different black, white, and
Latino testers, in 14 experimental conditions. By the completion of our data collection
(October 2004), will have audited over 1000 employers. After each visit to an employer,
testers complete a detailed debriefing form to record their interactions. Voicemail boxes
also record whether employers called back testers to make job offers or schedule secondround interviews, with additional voicemail boxes set up to record calls to references.
This study represents the largest and most complex audit experiment ever conducted in a
single field site.

1

With a majority of our data collection complete, we are able to report some preliminary
results for four of our six audit teams. Table 1 reports the percentage of callbacks and/or
job offers received from employers following job interviews with matched white, Latino,
and black applicants. These results demonstrate a clear racial hierarchy with white
applicants at the top, followed by Latinos, and blacks at a distant third (the black-white
difference in response rates (7.9 percentage points) is statistically significant).
This first set of results tests a standard racial hierarchy, with the white applicant serving
as a benchmark against which to measure variation in racial ethnic discrimination. The
results presented in Table 2 now change the benchmark to a white applicant with a felony
conviction. In this set of audits, our white tester presented evidence of a recent felony
drug conviction; his black and Latino test partners presented equal qualifications but no
criminal history. This table presents the striking result that a felony conviction confers
roughly the same penalty to job applicants as does minority status. The positive response
rates received by each of these groups are statistically indistinguishable.
Finally, table 3 compares the effect of a felony conviction for black and white job
applicants. To study criminal stigma, we sent teams of two whites or two blacks to apply
for jobs, randomly assigning a resume with a criminal conviction within each team. The
results indicate a significantly lower fraction of callbacks and job offers were received by
testers presenting a criminal record. Further, the criminal record effect appears
substantially larger for blacks than whites (11.2 compared to 6.1 percentage), although
this difference in the magnitude of effects is not statistically significant.
These results indicate that employers do treat job seekers differently on the basis of race
and criminal record, even relative to otherwise equally qualified applicants. These
findings are consistent with the hypothesis that employer discrimination along the lines of
race, ethnicity, and criminal conviction status remains a salient source of inequality in
contemporary urban labor markets.
Upon the completion of our data collection, there are numerous additional questions we
plan to address. First, an additional two testers pairs are being used to investigate
whether improving the educational credentials of former inmates can in part reduce the
negative effects of criminal stigma. In these pairs (one white pair and one black pair), the
applicant posing as the ex-offender presents evidence of an Associates Degree. We will
compare the outcomes of these teams to the original tester teams in order to assess
whether (and how much) an advanced educational credential can improve the relative
outcomes of ex-offenders.
Second, the audit data are rich with qualitative information that will allow us to
contextualize the differential treatment we observe. Testers often write pages of narrative
following their visits to employers, recounting in great detail the characteristics of the
employer, the content of their interaction, and their impressions of the visit. These
narrative descriptions (in addition to information from the 4 pages of close-ended
questions testers complete following each audit) can be used to better understand how

2

employers gather information about entry-level job applicants, and what sorts of microlevel interactions work to produce (or reflect) discrimination.
And finally, the information about employers coded by testers after each interview can be
used to analyze the audit outcomes in a multivariate framework. In these analyses, we
can assess the effects of race of employer, occupation, industry, firm size, the use of tests,
and a multitude of other job/employer characteristics. We can also better control for
possible tester, period, and job source fixed effects. These analyses will move us toward
a better understandings of the contexts in which discrimination is most likely to occur.
The continuing significance of race in contemporary labor markets is hotly contested
among academics and policy makers. Unfortunately, little hard evidence is available to
adjudicate among competing claims. The present study represents one attempt to move
beyond rhetoric by providing solid empirical measurement of this important social
process. Our preliminary evidence suggests that direct discrimination does indeed remain
a significant barrier to employment for ex-offenders and minority men. In our upcoming
analyses, we hope to better explain where and how this discrimination comes into play.

3

Table 1. Percentage of positive responses (callbacks or job offers) received by white, Latino and
black testers (no criminal record).
Positive
Diff. From
Responses (%)
Whites (s.e.)
N
Whites
24.3
239
Hispanics
21.3
2.9 (2.0)
239
Blacks
16.3
7.9 (2.6)
239

Table 2. Percentage of positive responses (callbacks or job offers) received by white ex-offenders
relative to Latino and black non-offenders.
Positive
Diff. From
Responses (%)
Whites (s.e.)
N
240
White ex-offender
12.9
240
Hispanic non-offender
13.3
.4 (2.4)
240
Black non-offender
9.2
-3.7 (2.3)

Table 3. Percentage of positive responses received ex-offenders and non-offenders, by race.
Whites
Blacks
Non-offenders
19.3%
20.1%
Ex-offenders
13.2
8.9
Difference (s.e.)
6.1 (2.5)
11.2 (3.0)
N
197
179

4