Spatial analysis with preference specification of latent decision makers for criminal event prediction

Spatial analysis looks for statistically significant patterns in observed events that occur at specified locations. Most examples of spatial analysis consider aggregate characteristics over a number of coarsely defined regions rather than point processes. However, criminal events are point processes and should be modeled as such. In this paper, we combine recent advances in discrete choice theory and data mining to develop point process models for spatial analysis. We use this new methodology to analyze and predict the spatial behavior of criminals, and more generally, latent decision makers. The paper compares the performance of this methodology to more traditional hot spot methods of crime analysis.