Abstract: 

High profile cases have raised public awareness of racial bias in the outcomes of police interactions in the United States. In response, a growing number of police departments are joining the nationwide movement to publish data associated with traffic stops in an effort to encourage understanding while promoting accountability for individual officers within and across police districts.

Police interaction disparities are often studied along a single dimension (for example, race) without accounting for differences in other driver characteristics (age, driving history), the type of moving violation observed, officer characteristics (years of service, rank, prior complaints, awards), vehicle characteristics (make, model, year, condition, history of infractions), geographic location of the traffic stop, driving conditions (weather, school zone proximity), time of day, time of the month, and other factors. Unsupervised learning techniques such as clustering could be used to create traffic stop segments, or subgroups of traffic stops that have similar attributes. These segments facilitate comparison of similar encounters with respect to any number of outcomes, including citation issuance, whether the driver was arrested, whether a search was conducted, etc.

Segmentation analyses help elucidate the nuanced relationship between race and traffic stop outcomes when evaluated in the context of other factors. In this presentation, we will describe the proposed framework for studying racial bias in police interventions using segmentation analysis. We will demonstrate how these techniques were used to find examples of racial bias in the Illinois Traffic Stop Study data that were masked in bivariate studies of race and traffic stop outcomes.

Bio: 

Coming Soon