By David Abrams and Priyanka Goonetilleke

Like most locally governed activities, police stops are evaluated, if at all, using data from a single municipality. This paper aggregates data on over 8 million pedestrian and vehicle police stops from 16 U.S. cities between 2019 and 2023 to better understand how the widely used police tactic varies by place, time, and race. We find immense variation in the implementation of these policies across cities, something that has not been previously highlighted in the literature. Stop rates vary across cities by almost two orders of magnitude, well in excess of inputs like funding and size of police force vary far less, or outcomes like different measures of crime. Contraband discovery rates (hit rates) for pedestrian stops are consistently low, with gun hit rates never exceeding 10% of frisks in any city. Consistent with optimizing models of policing, hit rates rise over time as frisk rates fall substantially. However, this result is concentrated in gun contraband, and for other types - mostly drugs - hit rates do not vary much even with vast declines in frisks, suggesting police are not optimizing for other types of contraband. This supports the notion that frisks are consistent with the law, which requires they be based on suspicion that an individual is armed and dangerous. Consistent with prior work, there are substantial racial disparities in all examined cities, with Blacks frisked as much as 10 times more frequently than Whites in some locations. Taken together, these facts point to a standard practice that seemingly has no standards, given the massive variation across the country. This suggests there are potentially large gains in efficiency and fairness from sharing best practices and nationwide data collection on police stops.

U of Penn, Inst for Law & Econ Research Paper No. 26-01,