Crime Hotspots In Belgium: Spatial He terogeneity In The Multivariate Regression Model?

Place attractiveness is an important property to consider when modelling the spatial behaviour of a serial offender. Small spatial units are privileged in the recent literature to study the crime concentration-urban backcloth relationship, with the drawback of focusing on a single city. However, the urban backcloth is also shaped by the inter-cities and city-hinterland relationships.
Our objective is to model the concentration of sexual offences for entire Belgium, at the very precise scale of the statistical sector in order to take into account the role of the urban hierarchy. Methods: We analyse the ViCLAS data for Belgium between 2004 and 2011. Specifically, we study the relationship between the concentration of initial contact scenes and socio-economic characteristics and facilities of the statistical sector, using two spatial models. First, we computed a spatial lag zero-inflated model (ZINB) in order to evaluate the global predictive capacities of the model. Then, a Poisson geographically-weighted regression (GWR) was run to study the spatial heterogeneity for each of the explanatory variables. Results: The zero-inflated negative binomial model with spatial lag recovers the neighbourhood environmental quality, the concentration of economic and cultural activities, and drinking places as major significant factors (Mc Fadden’s pseudo-r² of 0.2). The Poisson geographically-weighted regression underestimates the crime concentrations but highlights the spatial heterogeneity