Estimating the probability of local crime clusters: The impact of immediate spatial neighbors

PurposeTo investigate the importance of immediate spatial neighbors when investigating local crime patterns.MethodsLocal indicators of spatial association are used to identify local crime clusters. The classification scheme of these local crime clusters is then modeled in a multinomial logistic regression.ResultsThe results show that immediate spatial neighbors are important for understanding local crime patterns. Though (positive) spatial autocorrelation has long been known to be present with crime data, this analysis suggests that negative spatial autocorrelation (if present) has a significantly different implication. Generally speaking, when predicting a local crime cluster type, the immediate spatial neighbors are more important for correct prediction. As such, a low local crime area that is surrounded by high crime areas presents itself as a high crime area in the regression results.ConclusionsTherefore, efforts to understand the criminal nature of an area must not consider that area in isolation.

► We identify local crime clusters using a local indicator of spatial association, local Moran's I. ► The local crime clusters are then model in a multinomial logistic regression to identify their predictor variables. ► Knowing the type of immediate spatial neighbours is critical when identifying local crime clusters. ► Local crime areas relate to their spatial neighbours: low crime areas with high crime neighbours present as high crime areas. ► Efforts to understand the criminal nature of an area must not consider that area in isolation.