This paper examines how the Akaike information criterion can be used to evaluate candidates for models of offender behavior, and how multimodel interference can be used to combine estimates obtained from different models.
In this paper, the author briefly reviews his research team’s geographic profiling method, which is based on the Akaike information criterion; it introduces the relevant mathematical theory of model selection and multimodel interference, then illustrates those ideas by applying them to a series of convenience store robberies in Baltimore County, Maryland. The author discusses the primary question of geographic profiling, given the locations of a series of crimes committed by a single serial criminal, to estimate the location of that offender’s anchor point. The author also reviews several other approaches which have been developed to solve the problem, including a method using Bayesian interference to produce estimates of the offender’s anchor point but suggests that the Bayesian approach can only be as accurate as the underlying models of offender behavior. The research described in this paper pursued the use of model selection and multimodel interference based on the Akaike information criterion as a basis for comparing and evaluating models for offender behavior. The author concludes that the technique presented can be used to compare and evaluate models of offender behavior against one another, and can also be used to draw inferences from multiple models in a reasonable fashion for data taken from a real crime series. The author also notes that although the methods presented allow the comparison and evaluation of models, they do not specify a way to craft or create those models.