Abstract
A mature body of evidence exists on the utility of clinical versus statistical-prediction of criminal behavior. Less is known about the predictive validity of either model in terms of spatial distribution of crime. How accurately do experts—i.e. frontline-officers—predict which geographic areas to target with preventative tactics? Research on crime and place shows distinctively that incidents concentrate spatially within “recidivist” hotspots. Statistical predictions based on this spatial persistence characteristic of hot spots can serve as a comparator for the prediction capacity based on professional judgment. We turn to the Police Service of Northern Ireland, where we compared “Waymarkers,” maps drawn as a predictive tool for deployment purposes, against recidivist street segments based on statistical analyses. While statistical “hot spots” and “harmspots” accurately predict future incidents, Waymarkers are nearly always misplaced. Professional judgement inaccurately locates future spatial crime concentrations, and is more suited to identifying appropriate responses. We conclude that preventative policing ought to be based on statistical forecasting, rather than on professional judgment.
Acknowledgements
We wish to thank the Police Service of Northern Ireland and in particular the information and communication service. We would particularly like to express our gratitude to Chris Noble, Maura Muldoon and Keavy Sharkey. We would also like to thank Lawrence Sherman, Heather Strang and Sir Denis O’Connor for their insightful comments on earlier drafts of this paper.
Disclosure Statement
No potential conflict of interest was reported by the authors.
Notes on Contributors
Elizabeth Macbeth is a Business Improvement Coordinator for the Police Service of Northern Ireland. She has worked in policing since 2003, primarily as a Crime Analyst, in Scotland and Northern Ireland.
Barak Ariel is a lecturer in Experimental Criminology at the Institute of Criminology, Cambridge University and associate professor at Hebrew University.
Notes
1 We note, that there is no full agreement in the technical literature that street segments are superior over other mapping techniques. Those who argue in favor of the street segments approach suggest that a key limitations of KDE mapping is its potential “smoothing effect,” which creates hot spots in locations where no crime actually occurred, or in locations where it is not possible for crime to have been committed, due to topological features (Chainey & Ratcliffe, Citation2005c). However, not everyone agrees with this argument (Tompson, Partridge, & Shepherd, Citation2009), commenting that using a smaller search radius (search bandwidth) of the kernel would reduce the smoothing effect. Similarly, some scholars prefer using street segment over KDE is due to the benefit of data aggregation on the street segment. Specifically, geocoded crime incidents can be easily allocated to any street segment, making all other statistical analysis easy and pragmatic. On the other hand, KDE does not need this aggregation/allocation procedure and shows an actual distribution of crime regardless of street segment structure, which “works” well for certain visual purposes. Wheeler (Citation2015) has made a similar observation, suggesting that “aggregation is unlikely to hinder the goals of the particular research design, and how heterogeneity of measures in smaller units of analysis is not a sufficient motivation to examine small geographic units” (p. 1). From a practical perspective, however, there is a more robust line of research that shows how focusing on micro-units is superior to these aggregation approaches (see, among others, Weisburd et al., Citation2016). There is another reason why using street segment needs caution. Every crime occurs at address and is not equally distributed on the street segment. However, using street segment as a geographic unit of analysis would “smudge” the actual location of crime into street segment. Therefore, from the prediction stand point, prediction should be made on the address level. For a more elaborate discussion on single address predictions, see Bottoms, Citation2014; Buerger, Cohn, & Petrosino, Citation1995; Gorman, Gruenewald, & Waller, Citation2013, Sherman et al., Citation1989). Still, we are in the opinion that street segments provide the most efficient method of predicting crime, preventing crime and testing new methods of dealing with crime and delinquency, given the maximin rule.
2 As stipulated by the Crown Prosecution Services (CPS; cps.gov.uk/legal/s_to_u/sentencing_manual/), “The CPS Sentencing Manual has been designed as a source of information for advocates to assist them with trial preparation, in particular when addressing the court at the sentencing hearing. It consists of templates, grouped by subject headings, based on the chapter headings in Archbold, and provides sentencing guidance on the most commonly encountered offenses. It is intended to complement established texts on sentencing, such as Current Sentencing Practice.”
3 Gorr and Lee (Citation2015) have recently shown that employing a combination of KDE and non-statistical rules performs better than GIS-based KDE hot spots. However, these findings might allude more to a criticism against KDE rather than a broader distinction between statistical/actuarial versus clinical predictions. Their findings could also be a deviation from the overall pattern given their focus on temporary hot spots. More research is therefore needed, comparing the predictive validity of different methodological techniques and clinical predictions.