Crime risk maps

Abstract

In crime analysis, maps for degree of risk are important in decision making of police departments such as where to patrol or how to deploy the location of police force. This paper statistically models spatial crime data and produces crime risk maps. For the modeling and mapping of spatial crime data, we consider the traits of crime occurrences which are spatial dependence and correlation with other crimes. To reflect both of them simultaneously, we use a generalized multivariate conditional autoregressive model. For a real data application, we use counts of vehicle theft, larceny, and burglary at 83 census tracts in San Francisco in 2010. Bayesian approach using Markov chain Monte Carlo is used to estimate the model parameters. From the results, we detect crime hotspots and figure out the advantage of multivariate spatial analysis for crime risk.