Policy evaluations of police traffic stops have increasingly used variation in the timing of sunset, “Veil of Darkness” (VOD) tests, based on the assumption that police are better able to detect race during daylight. Here, we propose that African-American motorists adjust their driving behavior in response to increased scrutiny by police during daylight when their race is more easily observed, potentially biasing estimates of discrimination. Using nationally representative data, we demonstrate that African-Americans are less likely to have fatal motor vehicle accidents when driving in daylight. These effects are largest in states with historically high levels of structural racism and that rank highest in terms of police shootings of unarmed African-Americans. We find no such daylight effects on fatal accidents across any other observed motorist or vehicle attributes. Using police traffic stop data, we also find that the speed distribution of stopped African-American motorists in Massachusetts and Tennessee is shifted towards slower speeds in daylight with no evidence of such shifts for whites or over other observables. Finally, we develop and calibrate a model of police stop and motorist speeding behavior and use this model to demonstrate the effects of changes in driving speed on the VOD test. Theoretically, we show that the VOD test statistic can be reversed by motorist responses to the presence of police prejudice, and our calibration model demonstrates that substantial bias is introduced into the VOD test statistic by these responses.
Second version, August 30, 2019
K14: Criminal Law
K42: Illegal Behavior and the Enforcement of Law
H11: Structure, Scope, and Performance of Government
I31: General Welfare
J70: Labor Discrimination: General