An Information Theory of Efficient Differential Treatment

Abstract

When are differential treatment policies—such as preferential treatment, affirmative action, and gender equity policies—justified by efficiency concerns? I propose a nonparametric assignment model where a policymaker assigns agents to different treatments or positions to maximize total surplus, based on the agents’ characteristics and noisy information about their types. I provide necessary and sufficient conditions on the agents’ signal structures, which characterize whether surplus maximization requires differential treatment or not, and study how the bias and informativeness of signal structures determine the efficiency implications of differential treatment. I examine implications of this model for inequality, decentralization, and empirical work. (JEL D63, D82, D83, I23, I24, J71)