Using Geographically Weighted Regression to Explore Local Crime Patterns

Abstract

The present research examines a structural model of violent crime in Portland, Oregon, exploring spatial patterns of both crime and its covariates. Using standard structural measures drawn from an opportunity framework, the study provides results from a global ordinary least squares model, assumed to fit for all locations within the study area. Geographically weighted regression (GWR) is then introduced as an alternative to such traditional approaches to modeling crime. The GWR procedure estimates a local model, producing a set of mappable parameter estimates and t-values of significance that vary over space. Several structural measures are found to have relationships with crime that vary significantly with location. Results indicate that a mixed model— with both spatially varying and fixed parameters—may provide the most accurate model of crime. The present study demonstrates the utility of GWR for exploring local processes that drive crime levels and examining misspecification of a global model of urban violence.