Robust spatial analysis of rare crimes: An information-theoretic approach

In this paper we describe a semiparametric information-theoretic framework for modeling the determinants of rare events aggregated at intracity areal units while allowing for various forms of error correlation structures. The approach is applied to an examination of the effects of socioeconomic and demographic macrocharacteristics of communities on the amount of violence they experience. We investigate and find evidence of some instability in these processes across types of violence and level of areal aggregation. However, we also find evidence of a stable predictor - resource deprivation - for all the types of violence analyzed and at both levels of areal aggregation considered. In addition, we find evidence of a spillover effect of a community's resource deprivation on the level of violence its neighboring areas can expect. We discuss our findings in light of their substantive, methodological, and practical implications.