Public perceptions of police behavior during traffic stops: logistic regression and machine learning approaches compared

Abstract

Prior research on citizen perceptions of police has taken a wide-angle lens approach to the topic, with only a few studies investigating public perceptions of particular types of citizen–police encounters. In the current study, we make use of archival data on police traffic stops drawn from four waves of the BJS police–public contact surveys (PPCS) conducted in 2005, 2008, 2011, and again in 2015. In addition to employing conventional logistic regression, we make use of random forest classification to analyze survey data from a machine learning perspective. We use conventional logistic regression as a tool of explanation and random forest classification as a tool of prediction. We compare the findings generated by these two distinct analytical approaches. Substantive findings are quite similar for the explanatory and forecasting approaches. Driver’s belief that a traffic stop is legitimate is a major factor in how he or she evaluates police behavior in traffic stops, and whether the police use or threaten force during traffic stops may be the second most important factor. We draw out the implications of our work for our understanding of traffic stop dynamics, for the theory of procedural justice, for the theory of negativity bias, and for the enhanced use of machine learning in criminal justice.