This is the 8th in this year’s series of posts by PhD students on the job market.
Employers have been shown to discriminate against various groups. However, we have little understanding of whether and how jobseekers anticipate and react to discrimination while searching for work. Can jobseekers sidestep discrimination, for instance by strategically changing how they present themselves? Or do jobseekers’ responses to expected discrimination amplify the effects of employer discrimination by discouraging applications, or perhaps by making individuals too nervous to do well in interviews? In the latter case, even firms interested in hiring diverse candidates can have difficulty finding suitable candidates. To study the effects of expected discrimination on labor suppliers, in my job market paper, we partnered with a firm to set up an experiment inviting 2,200 jobseekers to apply for real sales jobs in Rio de Janeiro, Brazil.
Employers in Rio de Janeiro may discriminate based on many stigmas (marks associated with negative stereotypes), and our study focuses on the stigma of living in a favela (urban slum). There are many reasons why favela residents are stigmatized: these neighborhoods have a history of poverty since they grew without urban planning or public services and were initially populated by marginalized groups. Further, criminal organizations hold the monopoly of violence in favelas, giving rise to stereotypes related to gang affiliation or drug use. More generally, there are cultural stereotypes around favela residents; the derogatory term “favelado” is widely used. Because there are limited employment opportunities in favelas, jobseekers typically search outside them.
Expected vs. Actual Discrimination
To measure how accurately jobseekers perceive the discrimination they might suffer, we needed an objective measure of discrimination to establish some "ground truth", so we ran an audit study. In our door-to-door sample recruitment survey (in favelas), we asked people to guess what we would find. The surveyor explained that we were sending (fake) résumés to 700 sales jobs in Rio, with addresses either from a favela or an adjacent non-favela neighborhood. Then, the jobseeker could see a sample résumé and guess (to get a money prize) the percentage of those résumés that would get a positive response (a “callback”).
Our audit study found callback rates of 19.3% and 19.6% for favela and non-favela addresses, so we cannot reject the null of no discrimination. Jobseekers typically overestimate the non-favela callback rate by a lot and are somewhat more accurate – but still too optimistic – when predicting callback rates for a favela. The top panel in Figure 1 shows the distributions of guesses for each neighborhood, and the bottom panel shows the distribution of the implied discrimination rates, i.e., the expected percent drop in callback caused by a favela address. The median jobseeker expects a 50% discrimination rate; we call people who expect 50% discrimination or more the “high” expected discrimination group.
Randomizing Expected Discrimination
We implemented two experiments that randomized expected stigma visibility (and a third experiment that revealed the audit study results, which we discuss further in the paper). The basic idea in these experiments is that when favela jobseekers believe that an employer does not know their neighborhood names (which are always included in Rio addresses), they should not expect anti-favela discrimination.
Pre-callback Experiment. A few days after participating in the door-to-door survey, an HR firm – which we controlled as researchers, but kept separate from the field survey – invited jobseekers to apply for full-time sales jobs with our partner. The text message invite briefly described the job and told each jobseeker they were selected to participate in one of the firm’s new streamlined processes: for that application stage, they only needed to provide information on their “education and any courses or previous experiences”. We randomized the sentence right after that: “Your home address is [ALSO/NOT] required”.
Interview Experiment. When an applicant arrived at the interview office, the receptionist would ask to confirm name, date of birth, and address. The receptionist then told the jobseeker that, to keep the process objective, “the interviewer will only know your [NAME/NAME AND ADDRESS]”. Those two words, “and address”, were the only difference in our experimental conditions. Importantly, though, the interviewer only knew the candidates’ names at the moment of the interviews and learned they were favela residents only at the end of the study when we debriefed them.
Our experiment randomizing whether address was ALSO or NOT required to apply (N=1,303) yielded null average effects. We saw very similar application and interview show-up rates (about 40% and 20% of the whole sample, respectively), whether or not address information was required for applying.
In contrast, we find some effects of stigma visibility on interview performance. We conducted 422 interviews, and our main outcomes are indexes of interview performance. We construct one index using the three pre-registered dimensions the interviewer would rate candidates on (overall performance, nervousness, and professional behavior) and one index aggregating the same dimensions but using candidates’ self-assessments. In the specifications without controls, expecting that the interviewer would only know "name-only" increases the self-assessed performance index by 0.17 SDs (p<0.01), and the interviewer-assessed index by 0.09 SDs (p=0.28) -- see Figure 2 below. We cannot reject that both effects are the same (p=0.34).
Since our research goal is to understand the effects of expected discrimination, we pre-specified heterogeneity analyses by our baseline measure of that variable. Consistent with our hypothesis, we see that expecting to have a hidden stigma increases both kinds of performance by about 0.2 SDs in the group that expected high discrimination (see estimates with saturated interactions in Figure 2) – evidence that anticipated discrimination hurts interview performance. For comparison, jobseekers with some college had an aggregated performance index 0.56 SDs larger than those without any college, suggesting that an effect of 0.2 SDs is neither irrelevant nor implausibly large.
Correlated Stigmas: Race and Address
In Rio de Janeiro, the white population is a majority outside the favela but only about one-third of the favela population. Most jobseekers in our survey believed that racism was one of the main reasons why employers discriminated against favela jobseekers. In our data, we also see that race plays a big role: in both experiments, white jobseekers responded more to believing their addresses were hidden, showing up for interviews over 50% more often when they are told they don’t need to declare an address to apply, and increasing their interview performance index by 0.3 SDs when they are told the interviewer knows only their name. Non-whites are not affected in the pre-callback experiment, helping to explain our null average effects on application rates. In summary, since white jobseekers are more able to pass as non-favela residents, or because non-white jobseekers expected discrimination either way, address visibility tends to be marginal only for whites.
What Do We Learn?
Our results give some merit to the theory that anticipated discrimination amplifies the negative consequences of discrimination. We see that jobseekers have heterogeneous beliefs about discrimination levels, as most predict substantial discrimination in our audit study, and those who are the most pessimistic about discrimination also do worse in interviews when they expect to have visible addresses.
We suggest two main policy implications. First, if a firm wants to give a candidate their best shot, they should consider “blind” interviews, but they should do so in a way that hides all relevant stigmas – and note that could be a reason why blind orchestra auditions succeeded in increasing gender representation (Goldin and Rouse, 2000). Second, when it comes to disseminating information, focusing on market-level discrimination (e.g., discrimination rates, rather than, say, policies raising discrimination awareness or reporting on the most shocking cases) could help aligning beliefs.