Objectives

Current ‘geographical offender profiling’ methods that predict an offender’s base location from information about where he commits his crimes have been limited by being based on aggregate distributions across a number of offenders, restricting their responsiveness to variations between individuals as well as the possibility of axially distorted distributions. The efficacy of five ideographic models (derived only from individual crime series) was therefore tested.

Methods

A dataset of 63 burglary series from the UK was analysed using five different ideographic models to make predictions of the likely location of an offenders home/base: (1) a Gaussian-based density analysis (kernel density estimation); (2) a regression-based analysis; (3) an application of the ‘Circle Hypothesis’; (4) a mixed Gaussian method; and (5) a Minimum Spanning Tree (MST) analysis. These tests were carried out by incorporating the models into a new version of the widely utilised Dragnet geographical profiling system DragNetP. The efficacy of the models was determined using both distance and area measures.

Results

Results were compared between the different models and with previously reported findings employing nomothetic algorithms, Bayesian approaches and human judges. Overall the ideographic models performed better than alternate strategies and human judges. Each model was optimal for some crime series, no one model producing the best results for all series.

Conclusions

Although restricted to one limited sample the current study does show that these offenders vary considerably in the spatial distribution of offence location choice. This points to important differences between offenders in the morphology of their crime location choice. Mathematical models therefore need to take this into account. Such models, which do not draw on any aggregate distributions, will improve geographically based investigative decision support systems.