Developing a combined risk model for the prediction of temporally clustered offences

Having the means to estimate when and where future offences are likely to occur is of immense value to crime practitioners and partner agencies, hence a number of crime models have been developed to this end. Some of these models, including ProMap and SEPP, incorporate elements of a criminological theory known as ‘repeat/ near-repeat victimisation’. This theory is based on the idea that offenders typically operate within the confines of residentially-anchored routine activity spaces, thus rendering them well-placed to return to previously targeted locations over time. Therefore, associated offences are likely to present as spatially-clustered, yet temporally extended crime series, which provides a window of opportunity for operational intervention. Although ‘repeat/ near-repeat victimisation’ theory has informed the modelling of some high-level crime recording categories, including residential burglary, this thesis presents empirical evidence that failure to disaggregate beyond official crime classifications risks neglecting heterogeneity of offence characteristics within these. A potential implication of this is that the spatio-temporal parameters on which some prevailing crime modelling techniques are based might not apply to all offences, meaning that any related decision-making could be misinformed.