An economic theory of differential treatment

Abstract

I study differential treatment policies, such as affirmative action in school assignment, university admissions, labor market hiring, and promotions within organizations. I consider an assignment model where a policy-maker assigns agents to different positions or treatments, to maximize surplus. The agents have unobserved types (abilities, productivities, etc.), and observable noisy signals of their types, as well as observable characteristics (socio-economic status, gender, race, etc.). The optimal policy assigns agents monotonically with respect to an index defined by their expected incremental gains from different treatments. This policy treats agents differently whenever the signal distributions vary across characteristics—e.g. if signals are biased predictors of types, or if signals are unbiased but their informativeness varies across characteristics. Therefore differential treatment arises as a feature of the surplus maximizing policy, and the model provides a rationale for affirmative action based on efficiency concerns, rather than fairness.