Reasonable Doubt: Experimental Detection of Job‐Level Employment Discrimination

This paper develops methods for detecting discrimination by individual employers using correspondence experiments that send fictitious resumes to real job openings. We establish identification of higher moments the distribution job‐level callback rates as a function number sent each and propose shape‐constrained estimators these moments. Applying our three experimental data sets, we find striking heterogeneity in extent which probabilities differ race or sex. Estimates reveal while most jobs barely discriminate, few discriminate heavily. These moment estimates are then used bound share posterior probability is engaged discrimination. In recent experiment manipulating racially distinctive names, at least 85% contact both two white applications neither black To assess potential value regulators, consider accuracy decision rules investigating suspicious behavior various designs under simple two‐type model rationalizes data. Though estimate only 17% on basis race, an sending 10 would enable detection 7–10% discriminatory yielding Type I error below 0.2%. A minimax rule acknowledging partial yields slightly fewer investigations than Bayes based model. findings suggest illegal labor market can be reliably monitored with relatively small modifications existing designs.