Modeling Local-Scale Violent Crime Rate: A Comparison of Eigenvector Spatial Filtering Models and Conventional Spatial Regression Models

Environmental factors have both direct and indirect impacts on crime behavior decision making. This study aimed to examine to what degree the occurrences of violent crimes can be affected by social and built environment over space. Although a few studies have attempted to model crime rate using spatial regression models, there is a lack of comparison of spatial regression models. Particularly, the eigenvector spatial filtering type of spatial regression models has reportedly been effective in urban and regional studies, but it has not been widely applied to crime data. In this study, we aimed to examine whether the spatial filtering type of spatial regression models outperforms conventional types of spatial regression models in modeling violent crime rates over space. Moreover, we aimed to investigate the impacts of land use mix and street connectivity on the occurrences of violent crimes as the routine activity theory explained. In empirical studies, two types of spatial regression models (i.e., spatial error model and eigenvector spatial filtering model) were selected and estimated successfully to model local-scale violent crime rates across New York City. The eigenvector spatial filtering models outperform the spatial error models as well as the nonspatial models. Model estimation results show that occurrences of violent crimes (i.e., assaults and robberies) can be well determined by socioeconomic and built environment factors and thereby environmental factors can affect the occurrences of violent crimes. The contributions of socioeconomic and built environment factors to violent crime can offer insights on urban planning and policymaking toward violent crime prevention. Particularly, this study offers new evidence on the routine activity theory that increasing land use mix and street connectivity can enhance street activity, thereby reducing occurrences of violent crimes. Policymakers and urban planners should continue to enhance street activity through increasing land use mix and street connectivity. In addition, eigenvector spatial filtering models are advocated for use in crime or other applications in urban and regional studies