Simultaneous Sensitivity Analyses: An Application to Racial Bias in Policing

• In Event: Quantitative Study of Race & Gender

Abstract

Inferences about racial bias in police use of force depend on assumptions of no such bias in police behavior preceding the potential use of force — namely, no racial bias in whom officers encounter and in whom officers stop. Existing sensitivity analyses typically assess violations of only one of these assumptions in isolation. We instead derive a simultaneous sensitivity analysis, which we apply to an original matched design from the New York Police Department’s Stop, Question and Frisk (2003 – 2013) data. In so doing, we first establish the importance of a multidimensional sensitivity analysis by showing that increasingly severe violations of each assumption can have countervailing consequences for inferences about racial bias in police use of force. We then show that set restrictions on racial bias in stops often depend on assumptions about racial bias in encounters. This dependence between assumptions underscores the importance of not simply assessing sensitivity to multiple assumptions, but doing so simultaneously. Overall, our simultaneous sensitivity analysis detects racial bias in police use of force and shows that it is moderately robust to increasingly severe racial bias in behavior prior to officers’ potential use of force.

Authors

• Thomas Leavitt, City University of New York (CUNY)

• Jacob Bowers, University of Illinois, Urbana-Champaign

• Luke Miratrix