SPATIAL PATTERNS IN ARRESTS BY RACE IN NORTHERN VIRGINIA AND WASHINGTON, D.C.

Geographical analysis, and spatial statistics more specifically, have been widely used in understanding spatial patterns in crime, as well as in public health data. The spatial scan statistic and a spatial autoregressive model are used here on arrest data in northern Virginia and Washington, D.C. to identify clusters in arrests and check for spatial autocorrelation of the arrests. The spatial scan statistic identifies a number of clusters, with the largest encompassing most of southeast D.C. Adjusting the scan statistic for the race of the individuals arrested, there is a change in the cluster locations. Areas with a large Black population have fewer clusters after the adjustment while new clusters appear in areas with a small Black population but comparatively high number of Black people arrested, suggesting that racial bias among police or community members is affecting arrests. There are also a number of clusters that occur due to a courthouse, jail, or mall in the Census tract, where the high number of arrests is not well predicted by the explanatory variables used for modeling. There is some evidence of spatial autocorrelation in the data, but the variation in the full dataset is largely explained by covariates in the model representing demographic characteristics of each Census tract. Looking at different subsets of the data, there is evidence of spatial autocorrelation beyond that accounted for by the covariates in the arrests in Washington, D.C., as well as in overall arrests of Black or Latino people when arrests are modeled separately by race or ethnicity.