Abstract

Statistical discrimination is typically defined as the use of statistical generalizations about ascriptive identities, such as race or gender, as a proxy for individual attributes that are unknown and difficult to observe. While moral and political philosophers have questioned whether it is fair, they usually concede that it is rational. This paper examines the invention of the concept of statistical discrimination in neoclassical economic models of the labor market in the 1970s and exposes the problematic presuppositions behind the claim that it constitutes a distinctly rational and nonprejudiced form of discrimination. I show that these models followed standard disciplinary fictions of economics without considering their accuracy or plausibility in the discrimination case, and the empirical literature has accepted these assumptions without verifying them. The result has been a conceptual blind spot around the possibility of prejudiced but belief-driven discrimination, as well as the popularization of a problematic notion of rationality whose normative effect is to condone profit maximization. I go on to show how findings of statistical discrimination in the empirical literature have been used to push for policy prescriptions that exonerate those who discriminate as a side effect of the pursuit of profit, while putting the burdens of discrimination on those who are discriminated against.