Measurement issues are important because when policy and practice are evidence-based, the accuracy of that evidence is paramount. The "dark figure" of unrecorded crime that does not appear in police data has long been recognized as important. The need to estimate it was influential in the development of crime victim surveys (Hindelang, Gottfredson, and Garofalo 1978; Sparks, Genn, and Dodd 1977). Yet large scale representative surveys are rarely conducted on a local or regional level, are often dated in operational terms, and seldom contain geo-coded, time-series or time-specific information on crime events. This means recorded crimes remain a key information source for many analytic purposes. Such data can produce informative analysis, particularly where they offer a representative sample. However, when recorded crimes are not representative, the analysis is potentially misleading. The present study identifies a tool for dealing with such an instance, and we name that tool the Recorded Repeats Adjustment Calculator (RRAC). The hope is that this will lead to more informed predictive-policing and crime-prevention efforts.
The broader context of this study is the concentration of crime on the same targets--persons or places, including households, businesses or other facilities, cars, or other targets however defined. It is well established that repeat victimization occurs far more frequently than would be expected if crime were randomly distributed. (2) Yet such repeat victimization is often disproportionately undercounted in recorded crime data, including that relating to residential burglary, which is the focus of the empirical evidence presented here (Genn 1988; Ratcliffe and McCullagh 1998; Bichler 2004; Farrell and Pease 2003). The result is that research analysts, and thereafter police managers and policy makers, may be likely to think that repeat victimization is not a problem in their patch. This, in turn, may lead to a tendency to disregard crime prevention efforts focused upon the prevention of repeat victimization, even though recent systematic reviews suggests they succeed more often than not (Farrell and Pease 2006; Grove and Farrell 2012). The bulk of previous prevention effort has focused upon residential burglary, and so data relating to that crime type is used here.
There is a growing literature on repeat victimization, and the interested reader is referred to a recent annotated bibliography which includes a gross of studies (see Grove and Farrell 2011). Those studies cover the history of this area of research, the various forms of crime concentra- tion that involve forms of repeat victimization, theories and theory testing relating to the causes of repeat victimization, the impact of recurrent crime upon victims, victimization over the life course, the development of related strategy and policy, prevention efforts that have been undertaken, and details of existing, key guides and tools for policing- and crime-prevention practitioners (Grove and Farrell 2011).
The present study outlines a methodology to develop a truer estimate of the rate of repeat victimization than that observed in police data. It thereby aspires to empower researchers and analysts with information to better inform crime prevention goals.
This study analyses patterns in break and enter crimes recorded by police over a 5 year period in Metro Vancouver. (3) Break and enter is hereafter referred to as burglary. The data set is part of that known as the Police Information Recording System, or PIRS. It consists of 23,659 burglaries at single-family dwellings in Metro Vancouver during the 5 year period ending 31 June 2006. Single-family dwellings were selected to avoid the problems inherent in identifying individual addresses within large, multi-occupancy apartment buildings. (4)
The frequency of recorded repeats
For any given address that was burgled, repeats were identified that occurred within the next 12 months. This meant that, for our 5 years of data, a 'first' burglary could occur at any time in the first four years, so that the subsequent 12 month window could always be assessed. Skogan (1981) identified the importance of the "time window" issue for the study of repeats, the rolling-window measure used here was introduced by Tilley (1993), and variations in usage have emerged (e.g., Anderson, Chenery, and Pease 1995; Farrell, Sousa, and Weisel 2002; Johnson, Bernasco, Bowers, Elffers, Ratcliffe, Rengert, and Townsley 2007; Short, d'Orsogna, Brantingham, and Tita 2009). A limitation of the approach used here is that two events occurring more than 12 months apart at the same address are not linked. This means that date for households with long-term burglary problems are somewhat truncated in our analysis.
The frequency distribution of 21,329 burglaries at 19,008 dwellings is shown in Table 1. For presentational simplicity, those addresses reporting more than five burglaries within 12 months are grouped together (so the total number of burglaries is not a multiple of five because some households reported more).
In what follows, we estimate the frequency distribution of actual burglaries rather than solely those that are recorded. Two assumptions are made...