Non-Bayesian Statistical Discrimination

Abstract

Models of statistical discrimination typically assume that employers make rational inference from (education) signals. However, there is a large amount of evidence showing that most people do not update their beliefs rationally. We use a model and two experiments to show that employers who are conservative, in the sense of signal neglect, discriminate more against disadvantaged groups than Bayesian employers. We find that such non-Bayesian statistical discrimination deters high-ability workers from disadvantaged groups from pursuing education, further exacerbating initial group inequalities. Excess discrimination caused by employer conservatism is especially important when signals are very informative. Out of the overall hiring gap in our data, around 40% can be attributed to Bayesian statistical discrimination, a further 40% is due to non-Bayesian statistical discrimination, and the remaining 20% is unexplained or potentially taste-based. This paper was accepted by Marie Claire Villeval, behavioral economics and decision analysis. Funding: F. Mengel thanks the European Research Council for financial support [Starting Grant 805017]. P. Campos-Mercade acknowledges funding from the Danish National Research Foundation [Grant DNRF134 (CEBI)]. Supplemental Material: The online appendix and data are available at https://doi.org/10.1287/mnsc.2023.4824 .