TESTING ECONOMIC MODELS OF DISCRIMINATION IN CRIMINAL JUSTICE

Abstract

In this paper I derive several straightforward restrictions imposed by the Becker model of discrimination in the highway search and pre-trial release contexts. I explain how these restrictions may be tested using real-world data at the decision level (e.g., whether to search or whether to release a defendant). I then apply one of these restrictions to Florida data used in Anwar & Fang's (2006) influential study and more recent data from Harris County, Texas, provided by the Stanford Open Policing Project. The Florida data pass the restriction, but the Harris County data do not, with obvious implications for the appropriateness of the Becker model in each context. Further, data from both locations powerfully reject the prediction, from Knowles, Persico & Todd's (2001, KPT) two-sided model, that drivers will carry contraband at identical rates. Next I apply the Becker model restrictions to published estimates from Arnold, Dobbie & Yang's (2018) influential study of racial discrimination and pre-trial release. Their published estimates starkly violate the Becker model's restrictions, regardless of whether these are viewed as flowing from animus or inaccurate stereotyping. It is unclear whether the culprit is econometric assumptions, a failure of the Becker model, or both. These findings suggest the importance of specifi cation testing when we attempt to measure racial discrimination. They also suggest the need to consider alternatives to the workhorse Becker model, although doing so is beyond the scope of this paper.