Statistical Discrimination Against Underrepresented Groups

When employers make hiring decisions, they have to predict a job candidate's performance on the basis of observable attributes. Demographic characteristics, such as gender and race, affect these assessments even when they are not predictive of performance. In this paper, we propose that a simple cognitive mechanism can lead people to form false beliefs about performance differences after receiving true information. Specifically, we suggest that people who learn about the demographic characteristics of top performers fail to adjust for the prevalence of people with those demographics in the pool from which the top performers emerge. This process systematically generates statistical discrimination against minority groups. Across two preregistered experiments in which participants make incentivized hiring decisions, we find that people who receive demographic information about the top performers fail to adjust for the demographic composition of the pool they receive information about. Study 1 (n = 3,002) uses a pool composition unbalanced toward male workers, reflective of some high-profile industries. Receiving information about the top performers' gender leads participants to infer gender differences where none exist. Study 2 (n = 2,000) shows the effect also occurs in a sample representative of the US population, where there are inherently fewer Black or Asian than White candidates. Here, participants infer performance differences across race that are opposite to actual performance differences.