Endogenous learning, persistent employer biases, and discrimination

I present a new discrimination model of the labor market in which employers are initially uncertain about the productivity of worker groups and endogenously learn about it through their hiring. Previous hiring experiences of an employer shape their subsequent decisions to hire from a group again and learn more about its productivity, leading to differential learning across employers and biased beliefs about the group's productivity. Given a market-clearing wage, optimal hiring follows a cutoff rule in posterior beliefs: employers with sufficiently negative experiences with workers from a group stop hiring from the group, preserving negative biases and leading to a negativelyskewed distribution of beliefs about their productivity. When employers have noisier initial information on the productivity of one worker group, discrimination against that group can arise and persist without productivity differentials or prior employer biases, with market competition, and with or without worker signaling or investment decisions. The model generates steady state predictions analogous to the Becker (1957) tastebased model with beliefs replacing preferences, but is set within a statistical framework, explaining apparent prejudice as the result of \incorrect" statistical discrimination. The model also generates additional predictions and policy implications that contrast with previous models.