Andrew Little, UC Berkeley

"A Behavioral Theory of Discrimination in Policing" (with Ryan Hubert)

Racial disparities in policing are well documented. In addition to officer animus towards some groups (``taste-based discrimination''), these could be driven by officers' beliefs that some groups commit crimes at a higher rate (``statistical discrimination''). But where do these beliefs come from, and what if they are incorrect? We analyze of formal model where officers form these beliefs using crime statistics, but make a common inferential mistake when doing so: they do not fully adjust for the fact that they will detect more crime in communities that they police more heavily. This creates a feedback loop where officers (incorrectly) believe there is relatively more crime in communities that are policed more heavily, which leads to continued over-policing. This inferential mistake amplifies whatever disparities would otherwise exist due to taste-based or statistical discrimination. We first analyze this dynamic in a model with a single, representative officer, and then extend to the case of multiple officers. Since crime data are generated by the decisions of all, discrimination driven by false beliefs becomes contagious. As a result, inferential mistakes can exacerbate discrimination even among officers with no animus who sincerely believe disparities are driven by real group differences.