Association for Public Policy Analysis & Management

With the rise of violent crime in the 1990s, the New York Police Department (NYPD) turned to proactive policing practices. Perhaps the best known of these practices is Stop, Question, and Frisk. This practice, commonly known as Stop and Frisk, has had a disproportionate impact on the black and Hispanic communities of New York. More than 80 percent of all stops involve black and Hispanic pedestrians. Some academics have argued that this racial distribution of stops does not reflect bias. They distinguish between statistical discrimination and racial prejudice by testing whether the hit-rates, the rates at which pedestrians are arrested, are equal for each racial group, given that they were stopped. They show that black and Hispanic pedestrians are arrested at a rate similar to that of white pedestrians, indicating statistical discrimination. This paper challenges that conclusion and finds evidence of differential police behavior by race that may suggest racial bias.

The dependent variable is of post-stop outcomes: nothing, summons, arrest. A multinomial logit model, with coefficient interpretations as relative risk ratios, and 3.9 million stops reported by the NYPD from 2006 to 2012, are used for the estimation. Heterogeneity across types of stops and subsequent police activity masks lower arrest rates of black and Hispanic pedestrians. Specifically, three sets of regressions are estimated. In each one of the sets the specifications are the same multinomial logit on the full subsample, a restricted subsample of only males, only youth, only young adults, and of only stops in which the police did not wear a uniform. The first set of regressions is estimated on the full sample. The second set of regressions is estimated on a subsample of only stops in which the pedestrian was frisked. The third set of regressions is estimated on a subsample of stops for which reporting was mandatory (60 percent of the stops).

Black and Hispanic pedestrians are less likely to be issued a summons. When the sample is restricted to stops that involved a frisk or stops that required reporting, black and Hispanic pedestrians are arrested at a lower rate that is both statistically and substantively significant. The difference between white and black young adults is 30.8 percent (P<0.001) and between Hispanic young adults is 18.7 percent (P<0.001). Assuming decreasing marginal returns to stops suggests that the NYPD is over-stopping people of color aged 25 to 35 in New York City. Another consistent result across the specifications is the lower rate at which black and Hispanic pedestrians are arrested when the police officer is not in uniform. Perhaps the most concerning result is that when the subsample is limited to only the stops that required reporting, the hit-rate analysis shows that black and Hispanic pedestrians are arrested in a both statistically and substantively significant lower rate, 16.5 percent and 10.4 percent (P<0.001), than white pedestrians. These results supports the conclusion that Stop and Frisk is an uneven policing practice and that racial disparities are not immediately explained as a consequence of statistical discrimination.