Predictive Crime Mapping: Arbitrary Grids or Street Networks?

Objectives Decades of empirical research demonstrate that crime is concentrated at a range of spatial scales, including street segments. Further, the degree of clustering at particular geographic units remains noticeably stable and consistent; a finding that Weisburd (Criminology 53:133–157, 2015) has recently termed the ‘law of crime concentration at places’. Such findings suggest that the future locations of crime should—to some extent at least—be predictable. To date, methods of forecasting where crime is most likely to next occur have focused either on area-level or grid-based predictions. No studies of which we are aware have developed and tested the accuracy of methods for predicting the future risk of crime at the street segment level. This is surprising given that it is at this level of place that many crimes are committed and policing resources are deployed. Methods Using data for property crimes for a large UK metropolitan police force area, we introduce and calibrate a network-based version of prospective crime mapping [e.g. Bowers et al. (Br J Criminol 44:641–658, 2004)], and compare its performance against grid-based alternatives. We also examine how measures of predictive accuracy can be translated to the network context, and show how differences in performance between the two cases can be quantified and tested. Results Findings demonstrate that the calibrated network-based model substantially outperforms a grid-based alternative in terms of predictive accuracy, with, for example, approximately 20 % more crime identified at a coverage level of 5 %. The improvement in accuracy is highly statistically significant at all coverage levels tested (from 1 to 10 %). Conclusions This study suggests that, for property crime at least, network-based methods of crime forecasting are likely to outperform grid-based alternatives, and hence should be & Toby Davies [email protected] 1 SpaceTimeLab, Department of Civil, Environmental and Geomatic Engineering, University College London, Gower Street, London WC1E 6BT, UK 2 Department of Security and Crime Science, University College London, 35 Tavistock Square, London WC1H 9EZ, UK 123 J Quant Criminol DOI 10.1007/s10940-016-9321-x