Combining discrimination diagnostics to identify sources of statistical discrimination

Abstract

Statistical discrimination is usually flagged by economists as a potential source of treatment disparities. The literature, however, lacks reduced-form tests that provide information about the relative importance of statistical discrimination in explaining aggregate patterns. This article explores whether combining three different diagnostics of aggregate discrimination – those being, unconditional treatment disparities, benchmark tests, and outcome tests – can provide insights into sources of statistical discrimination. We discuss the difficulties concomitant with this exercise and argue, using an identification result that relies on restrictive (and presumably implausible) assumptions, that the answer is most likely negative.