Wednesday 04 Dec 2013: Spatio-temporal dynamics of crime: a statistical modelling approach with an application to crime-rates in Lancashire, UK
Irene Kaimi - Plymouth University
Plymouth University, Fitzroy 210 14:00-15:00
Spatio-temporal data are often aggregated into small areas. Motivated by a specific problem concerning the spatio-temporal distribution of crime in Lancashire, UK, we describe a flexible spatio-temporal log-Gaussian Cox process model appropriately adjusted for count data. The model includes a multiplicative decomposition of the spatio-temporal intensity function into separable spatial and temporal terms, through which the effects of spatial and temporal covariates can be estimated, and a stochastic component that allows for non-separable spatio-temporal dependence. In contrast to widely used Markov random field models for spatial count data, we derive the joint distribution of the counts from an underlying spatially and temporally continuous process, rather than tying the model to a fixed set of geographical units.