Boris Babic (California Institute of Technology)
The Stanford Open-Policing Project, launched in 2017, currently contains information on more than 60 million police stops across 20 US states conducted between 2011-2015 (Pierson et al., 2017). The dataset was initially compiled through a series of public records requests and for each stop, we have information about the infraction, the driver’s age, gender, and race, whether a search of their vehicle was conducted, and whether contraband was found. With such a rich dataset we should be able to gain important insights about the pervasiveness of racial profiling in the United States.
This raises an important methodological question, however; one that philosophers ought to consider: how should we test for racial profiling? In other words, what sorts of patterns in this (or any other) data would be indicative of wrongful police discrimination? To answer this question we need a test for discrimination – or at least some device that allows us to make inferences as to whether, and the extent to which, it is happening. Ideally, we want this test (or other device) to be grounded in ethical theory but we also need a test that can be operationalized and applied in empirical research. In this project, I try to offer one such test. It’s really an advertisement for a sketch of an outline of a test, to paraphrase Yablo. I also apply this test in three different ways to approximately 300,000 police stops in Connecticut. I show that all three ways suggest substantial racial profiling against blacks and hispanics in Connecticut and I discuss the limitations of each.
This project proceeds as follows. First, I briefly describe the particular subset of data I will rely on, and I offer a brief overview of how lawyers, economists, and criminologists have defined discrimination. Then, I give a conception of discrimination of my own, which relies on considering how individuals of different races are viewed by government institutions. Impermissible discrimination exists to the extent we can conclude that state actors are deemed to hold people of different races on different moral footing. I then show that a sufficient condition for finding evidence of discrimination under this principle is to find evidence that state actors have different epistemic risk profiles with respect to people of different races. I carefully define the expression `epistemic risk profile'. Finally, I show that a sufficient condition for finding a difference in epistemic risk profiles in the specific context of police stops is to show that officers have different search thresholds for people of different races. Finally, I apply three different tests aimed at estimating the search thresholds. First, I evaluate the marginal effect of race in a hierarchical Bayesian logistic regression model with the probability of search as the dependent variable. Next, I consider and apply the famous outcome test proposed by Becker (1957) and developed by Knowles et al. (2001). And third, I apply the direct threshold test proposed by Simoiu et al. (2017).
