Policing Bias with Naive Bayesian Officers

Joint with Andrew T. Little.

We analyze a formal model of policing where differential treatment of groups and inaccurate beliefs about the relative criminality of those groups are joint and mutually reinforcing outcomes. In the model, a police officer forms beliefs about the level of criminality in two different groups. These beliefs are based on crime he observes, and so respond to how he allocates his time between policing and socializing with members of those groups outside work. He is a “naive Bayesian” in the sense that he estimates each group’s rate of criminality based on the frequency of his interactions with those who commit crimes, but does not adjust for the fact that he is more likely to encounter criminal behavior while working. We show how the police officer’s beliefs about criminality can be distorted by his own choices about how he interacts with members of different policed groups. In particular, he will overestimate the relative criminality of whichever group he socializes with less, which leads him to police that group more, which in turn reinforces his inaccurate belief that this group is more criminal. A contribution of our model is to provide a fully endogenous explanation for disparities in police surveillance that result from neither taste-based discrimination nor statistical discrimination by police officers. Instead, it highlights the potentially important role of social segregation and police officer social networks in explaining those disparities.

Selected Presentations: Conference on Political Economy and Public Law (June 2019, Princeton University)