The effectiveness of calibrated versus default distance decay functions for geographic profiling: a preliminary examination of crime type

Abstract

This study examines whether a distance decay function calibrated for a particular crime type results in more accurate geographic profiles compared to default functions that are not calibrated for one specific crime type. Decay functions were calibrated for three different types of serial crime (residential burglary, theft, and auto theft) collected from the same geographic region (Glendale, AZ, USA). The two default functions used for comparison purposes (truncated negative exponential and negative exponential) came from CrimeStat (v. 3.1), a computerized geographic profiling system. The hypothesis that calibrated functions would possess more predictive power than default functions was not supported. Potential explanations for these findings are provided and implications are discussed.

Keywords:

• geographic profiling
• distance decay
• journey-to-crime
• serial crime
• offender spatial behaviour

Acknowledgements

Special thanks to Bryan Hill of the Glendale Police Department for providing the data necessary to carry out this study.

Notes

1. Although many, if not the majority of, serial offenders commit multiple types of crime (e.g. Leitner & Kent, 2009), almost all of the GP research to date has used a single crime type in their analyses (e.g. Bennell, Emeno, Snook, Taylor, & Goodwill, 2009; Canter et al., 2000; Paulsen, 2006; Rossmo, 2000; Snook et al., 2005). Thus, the current study will focus on single crime types as an initial step in assessing calibrated versus default functions.

2. While the size of the sub-samples prevented us from running formal inferential tests, a visual examination of the means for both error distance and hit percentage indicated that the profiles generated in this study were more accurate for marauding offenders compared to commuting offenders. This trend was consistent across all three data files and for each of the three functions examined. Please contact the first author if you are interested in seeing these results.

3. Interestingly, there has also been a lot of debate about the relative merits of these two measures, especially error distance. While some argue that error distance is an appropriate measure of GP accuracy (Levine & Associates, 2007; Snook et al., 2005), others disagree, typically pointing out the fact that measures such as error distance do not reflect how GP systems are often used (i.e. error distance focuses on one single location or point of prediction, rather than on an area that can be searched, as is produced by GP systems; Gorr, 2004; Rossmo, 2005).