3.7.1: Labor Market Discrimination

Consider the study introduced earlier (Section 2.5.1) which used virtual resumes to demonstrate that resumes designed to be perceived as coming from African American applicants received fewer callbacks from potential employers than resumes designed to be perceived as coming from white applicants (Bertrand and Mullainathan 2004). This study demonstrated bias in the way that employers evaluated resumes, even when the race of the applicant was not explicitly revealed to the evaluator.

This study was based on a designed experiment. The researchers controlled the perceived race conveyed by the names on the resumes, as opposed observing how employers reacted to real resumes that happen to have been sent to them. Conducting the study in this way had the advantage that the researchers were able to carefully control the level of the experience on the resumes, as well as other factors that could eliminate many confounding issues from the study. Observational studies of this same topic have problems with confounding in that the experience of the applicants, the appearance of the resume, and other potentially important factors are not controlled. In such cases it would be impossible to conclude that the behavior of the employers was due solely to the perceived race of the applicant, and not one of these other factors.

The authors of this study did a good job of pointing out possible remaining confounding problems. One potential problem was that employers may not be evaluating the race of the applicant based on the name but may really be evaluating the social background of the applicant. This assumes that the perceived African American names indicate individuals from disadvantaged backgrounds. In this case the employer would be discriminating based on the perceived social background of the applicants instead of race. The researchers performed a more nuanced analysis of the data and argue against this confounding problem. Another potential confounding issue may be that the perceived African American names are less well known and may appear “odd” to human resource managers. In this case the discrimination would be against odd sounding names. Again, the researchers pointed out that they used more common names and provided additional statistical analysis to support the idea that this confounding problem should not be an issue.

The authors pointed out some other potential problems with the study that are obliquely related to confounding problems. The first was that a callback from an employer does not necessarily mean that the applicant was being seriously considered for the job or that they would be hired once their race was revealed. Secondly, not all African American applicants have names that would be perceived as being African American, and hence once again the study did not reveal what would happen when these applicants' race was revealed at some point in the application process. Combining these issues, the study seemed to address whether these resumes would be rejected in their first evaluation based on the perceived race of the name.

The authors emphasized the techniques that they went through to eliminate as many possible confounding issues as possible. For example, the resumes were designed to be both believable and to vary qualifications as little as possible between the perceived African American and white named resumes. The main problem with this study, which the authors emphasized throughout, was that the perception of race, and the exhibition of a corresponding bias in the employment process is a very sophisticated system that is difficult to summarize with a simple experiment. Therefore, it was very difficult to pinpoint exactly why the differences in the callback rates was observed. Overall, this study certainly provided evidence of callback bias for potential employment of the basis of the perceived race of the applicant, though the issues involved may be more complicated than could be addressed by this study.