- Top of page
- Analytic strategy and results
- Supporting Information
Research consistently indicates that a small proportion of offenders are responsible for a majority of crimes (Farrington, 2003; Wolfgang, Figlio, & Sellin, 1972). As such, substantial resources are devoted to policing prolific offenders (Innes, Fielding, & Cope, 2005). Behavioural crime linkage (hereafter, crime linkage) – an analytical technique concerned with the identification of a series of offences committed by a single perpetrator out a pool of unsolved crimes (Grubin, Kelly, & Ayis, 1997; Woodhams, Hollin, & Bull, 2007) – can play a significant role in the detection of prolific offenders.
Ideally, crime linkage could always be achieved using physical or other identifying evidence (e.g., fingerprints, DNA, witness statements), but such evidence is often unavailable (Davies, 1991). As a result, crime linkage is carried out using other categories of crime scene evidence, such as the offender's modus operandi (MO), loosely defined as the constellation of the observable traces of the actions performed by the perpetrator during the commission of the offence. Data relative to MO are analysed to uncover patterns of similarity and differentiation between crimes of the same type, such as burglary or theft from a motor vehicle (Canter, 2000). If successful, crime linkage allows for the centralization and rational allocation of investigative resources (Godwin, 2001), facilitating the detection of prolific offenders. The aim of crime linkage research is, ultimately, to develop decision-making support systems for use by crime analysts and investigators (Bennell, Snook, Macdonald, House, & Taylor, 2012; Woodhams & Toye, 2007).
Underpinning crime linkage are two explicit hypotheses (Woodhams et al., 2007). The first, the consistency hypothesis, states that offenders will behave somewhat consistently across a series of offences of a similar type. The second, the specificity hypothesis, states that offenders can be differentiated on the basis of their offence behaviour. For crime linkage to be possible, some degree of similarity must be observed between the MOs displayed by the same offender at multiple crime scenes, and some degree of difference must be observed between the MOs displayed by different offenders across crime scenes (Canter, 1995).
While there has been much work on crime linkage, arguably, the explicit testing of these hypotheses has not received the attention it deserves using appropriate statistical methods (see also, Salo et al., 2013). In this article, we set out to test directly the consistency and specificity hypotheses using data for a sample of burglars who committed prolific residential burglary in the United Kingdom. First, we discuss why we might expect behavioural consistency and specificity (or lack thereof) in the crimes of prolific burglars, and then review previous crime linkage research, focusing in particular on that concerned with residential burglary. We then discuss a metric for measuring consistency and a method for establishing its statistical significance not used hitherto. After presenting a set of analyses for our sample of burglars who committed residential burglary, we conclude with a discussion of the implications of our findings for future crime linkage research on residential burglary.
Sources of consistency and specificity in burglars who commit residential burglary
Studies of convicted burglars have revealed clear targeting and offence commission strategies (Cromwell, Olson, & Avary, 1991; Hearnden & Magill, 2004; Palmer, Holmes, & Hollin, 2002; Rengert & Wasilchick, 1989; Wright & Decker, 1994), which leads to the expectation that burglars will display some notable degree of consistency in their MO. The evidence regarding possible sources of specificity is more ambiguous. To the extent that a majority of burglars who commit residential burglary use the same criteria to evaluate and search for targets (Wright & Decker, 1994), their behavioural consistency would lack specificity; that is, burglars could not be differentiated on the basis of MO, since they would all go after similar targets in the same way. The fact that burglary skills are transmitted among offenders (Bennett & Wright, 1984; Maguire, 1982; Rengert & Wasilchick, 1985) could be another factor of uniformity, and therefore low specificity. However, studies have also indicated that offenders differ on the basis of their sensitivity to situational characteristics and the amount of risk they are willing to take, as a factor of age and related level of expertise (Coupe & Blake, 2006; Nee & Meenaghan, 2006) or relative to their degree of sophistication (Jacobson, Maitland, & Hough, 2003; Rengert and Wasilchick, 1985). Burglars have also been shown to differ in their ability to articulate the perceptions and reasoning behind their decision-making process, some appearing more deliberate than others in their assessment of environmental cues and their cost–benefit analysis (Bennett & Wright, 1984; Tunnell, 1992; Wright & Decker, 1994). Such differentiation in burglars' thinking styles and decision-making abilities could be another source of inter-individual variation in their mode of operation.
Importantly for crime linkage, a number of factors could lead to the alteration of MOs over time, both across the population of burglars and within individual offenders. New MOs emerge as new tools are made available, knowledge propagates, or new opportunity-blocking measures are encountered (Ekblom & Tilley, 2000). Prolific burglars are likely to rely on cognitive scripts that have proven successful in the past (Nee & Meenaghan, 2006; Wright & Decker, 1994), but they are also subject to learning processes (Douglas & Munn, 1992; Turvey, 2002; Wright, Logie, & Decker, 1995), notably in the early stages of their career. Criminal career research indicates that offender versatility is the rule rather than the exception (Farrington, 1994; Gottfredson & Hirschi, 1990; but see, Johnson, Summers, & Pease, 2009), suggesting that learning processes relevant to MO could occur intra-individually across crime types, though this does not rule out short-term specialization in MO within crime type (Hochstetler, 2002). It has also been proposed that criminals' MOs can evolve or ‘de-evolve’ over time, as a result of increased confidence or deteriorating mental states brought on by drug use (Turvey, 2002).
In short, while there are grounds to expect that burglars can be differentiated on the basis of traditional MO behaviours, it is also possible that a large measure of similarity prevails across individuals, to the extent that several features have been commonly associated with a target's vulnerability to burglary (Budd, 2001; Hakim, Rengert, & Shachmurove, 2001), and that these features could, in turn, characterize a non-trivial number of offences. If this were the case, then while offenders may be consistent with respect to their own MO, that MO may simply reflect a profile that is common across offenders, thereby offering little potential for crime linkage.
Testing consistency in serial offenders
Hypothesis testing requires the use of specific methodological procedures and statistical tests. Different approaches may provide different insights. In the past, researchers working on crime linkage have obtained data for a sample of offenders who committed two or more detected crimes and included in their analysis a maximum of three crimes per offender (e.g., Bennell & Canter, 2002; Bennell & Jones, 2005; Bennell, Jones, & Melnyk, 2009; Davies, Tonkin, Bull, & Bond, 2012; Markson, Woodhams, & Bond, 2010; Santtila, Junkkila, & Sandnabba, 2005; Tonkin et al., 2012; Woodhams & Toye, 2007; see, however, Harbers, Deslauriers-Varin, Beauregard, & van der Kemp, 2012; Salo et al., 2013; Woodhams, Hollin, & Bull, 2008), either to estimate the degree of consistency and specificity exhibited by the offenders, or to classify events as belonging to one series or another. The rationale behind the sampling strategy adopted is that randomly sampling two offences (occasionally three) for each offender should ensure that the findings cannot be distorted by prolific criminals, who may display greater or lesser behavioural consistency, as well as allowing more offenders to be considered ‘serial’ (having committed two or more offences), thus enabling researchers to include a larger number of offenders – though not necessarily offences – in the sample (Bennell & Canter, 2002; see Woodhams & Labuschagne, 2011, for a discussion of issues with this approach).
Such tests focus on how consistent a group of offenders are in the aggregate for a particular type of MO, but not on how consistent individual offenders are across a series of offences. Where tests such as logistic regression are used to test hypotheses, one assumption is that patterns will be relatively homogeneous across offenders (i.e., fixed, rather than random, coefficients are estimated), such that a sufficiently large number of offenders will be consistent and equally distinctive for particular behaviours. This is not the same as examining whether offenders display any evidence of consistency or specificity for any aspect of their MO, which is the aim of this study.
In the article, we use a different approach and estimate how consistent prolific offenders are across the entire series of their detected offences, and how typical it is to observe consistency for particular behaviours in our sample. These are different questions from those that have been addressed in the crime linkage literature to date. These questions test the consistency and specificity hypotheses more directly than has been the case in much of the previous research.
For what behaviours do we anticipate consistency?
Having established how what we aim to achieve differs from prior research, we now consider additional hypotheses. Ecological theories of crime (Brantingham & Brantingham, 1993; Cohen & Felson, 1979) explain how criminal opportunities emerge at the intersection of an offender's routine activity space and of the environmental backcloth. The organization of the backcloth determines the location of the offender's activity nodes and the routes between them, which in turn shapes their awareness of the criminal opportunities afforded by the environment. In line with this, research consistently indicates that offenders commit crimes predominantly within this awareness space (e.g., Rengert & Wasilchick, 1985; Wiles & Costello, 2000), and hence whatever the methodology adopted, offenders would be expected to display consistency with respect to where they commit offences. Insofar as offenders will live at different locations and have different routine activity nodes, they should also display some degree of distinctiveness for where they offend.
In line with this, in the existing crime linkage literature and using the methods described above, spatial (inter-crime distance) measures of behavioural similarity have been found to be particularly good at distinguishing between pairs of linked or unlinked offences, within and across crime types, and such findings have been replicated across different study areas, which suggests the external validity of this finding (Bennell & Canter, 2002; Bennell & Jones, 2005; Goodwill & Alison, 2006; Markson et al., 2010; Tonkin, Santtila, & Bull, 2011; Tonkin, Woodhams, Bull, Bond, & Palmer, 2011). In contrast, what might be termed MO behaviours (e.g., target selection, mode of entry, property stolen in cases of burglary, acts committed in cases of sexual assault) have been found to perform less well as linkage variables (with some exceptions; see, notably, Woodhams & Toye, 2007; Tonkin et al., 2012), especially when considered independently of other characteristics. However, when interpreting such results with respect to the core assumption of crime linkage, it is important to remember how the indices of consistency typically used are derived and the limitations associated with this approach.
Returning to theoretical issues, we should note that any categorization of a burglar's offence behaviour into ‘domains’ (e.g., traditional MO vs. geospatial domains) is somewhat artificial, inasmuch as domains cannot be conceptualized as strictly independent from each other. Furthermore, studies of decision-making stress the role of environmental cues in determining burglars' targeting and search strategies (Nee & Taylor, 1988). From an ecological perspective, motivation to offend is a situational mechanism (Wikström, 2006). For prolific burglars, search and targeting decisions can seem automatic to the point of unintentionality, in that a stable set of cues is perceived and elicits scripted responses or ‘obligatory processes’ (Nee & Meenaghan, 2006). In addition, prior offending leads to awareness of opportunities for future burglaries (Bennett & Wright, 1984). In the light of all this, we may expect some degree of codetermination, or – if the term is too strong – of mutual constraint between physical environment, geospatial behaviour, planning, target selection, and extant MO behaviours (e.g., the type of target may constrain the method of entry). Put differently, backcloth configuration could limit the potential for variability in MO behaviours, and this ‘ecological constraint’, if it affects any one MO behaviour, might constrain others by logical association.
As discussed above, the two core assumptions that must be met for crime linkage to be successful are that offenders must display consistency in the way that they commit their crimes, and that they are distinctive (specific) in the way that they commit their offences, such that all offenders are not consistent in exactly the same way. We suggest that to test these core assumptions directly, and to allow a better test of the underlying theory, it is necessary to examine patterns observed across prolific offenders' entire series of detected crimes, rather than pairs of their offences, as is the case in a majority of crime linkage studies.
In the existing literature, patterns of consistency are typically examined in the aggregate and in such studies, estimated coefficients are fixed for the entire sample.1 In this way, it is essentially (arithmetically if not theoretically) assumed that the degree of consistency observed will be relatively homogeneous across the sample. However, this is not necessarily the case and some offenders may be more consistent than are others. In line with this, Townsley and Sidebottom (2010) show that while most residential burglaries are committed close to an offender's residence, offenders vary quite considerably (see also, e.g., Markson et al., 2010) in the extent to which their journey to crime conforms to a pattern of distance decay – a finding that illustrates how patterns observed in the aggregate may not apply at the level of the individual – the problem of the ecological fallacy. Thus, to estimate the extent to which consistency and specificity exist across offenders, or for particular aspects of an offender's MO, we use a more disaggregated form of analysis than has typically been employed hitherto, and test the following hypotheses:
Hypothesis 1: Offenders will display consistency in the way (area of offence, type of home targeted, method and point of entry, time of day, occupancy of home) they commit offences across a series of crimes – the consistency hypothesis.
Although findings from previous linkage studies lead us to expect that offenders are more likely to display consistency with regard to temporal and geospatial behaviour (time of day, area of offence), rather than more ‘behavioural’ MOs (type of target, method and point of entry, occupancy), we restate the consistency hypothesis in full, as we believe it is worth putting wholly to the test in the light of the new methodological approach adopted here. This said, we do expect that some categories of MOs (notably, geospatial) will display greater consistency than others.
Hypothesis 2: Patterns of consistency for particular MO characteristics will differ between individuals, and individuals will exhibit preferences that differ from the aggregate patterns observed across all offenders – the specificity hypothesis.
Hypothesis 3: Offenders who are more consistent in where they commit their offences will also be more consistent in other types of MOs, which are plausibly constrained by the choice of offence area (e.g., type of home targeted; mode of entry) – the ecological constraint hypothesis. In turn, some types of MOs may also constrain each other (e.g., type of target and mode of entry).
- Top of page
- Analytic strategy and results
- Supporting Information
Most of the research concerned with offender consistency has focused on the utility of using details of an offender's MO to link pairs of crimes together – usually one pair of crimes for each offender sampled. Relatively little research has directly examined the extent to which prolific offenders are consistent across an entire series of detected offences using statistical methods, including tests that estimate the significance of obtained values, which is necessary for hypothesis testing. Consequently, previous research has not examined how typical it is for offenders to be consistent for a particular MO, and whether any consistency exhibited is distinct to them. The aim of this study was to examine these issues for a sample of burglars who committed residential burglary and consequently test theories of offender behaviour and decision-making.
Our findings suggest that when we examine each aspect of an offender's MO without reference to general patterns of offending, offenders appear to display consistency in the way in which they offend (Hypothesis 1), though they do so for some attributes more than others, an outcome in line with previous research in burglary linkage (e.g., Bennell & Canter, 2002; Bennell & Jones, 2005; Markson et al., 2010; Tonkin, Santtila et al., 2011; Tonkin, Woodhams et al., 2011). However, when we compare the extent to which they are consistent with what would be expected, assuming that all offenders commit offences in much the same way, it appears that offenders are more specific for different aspects of their MO, than we might otherwise have concluded. They exhibited the most specificity for the geographical area targeted, with more offenders (78%) demonstrating specificity for this aspect of their MO than any other. Although fewer offenders (37%) exhibited specificity for the type of home targeted, on average they demonstrated nearly twice as much consistency for this aspect of their MO than would be expected. These findings are important from a theoretical perspective, providing support for the theories of offender decision-making discussed in the introduction, and hence for the underlying principles of crime linkage.
However, it is also notable that in the case of spatial decision-making, offenders who had committed more offences appear to be less consistent – across their entire detected crime series – in terms of where they committed offences, than were those who committed fewer. Such a finding suggests that while geospatial information may be the most useful aspect of an offender's MO for crime linkage, there may eventually be diminishing returns, an aspect which the empirical literature has not investigated explicitly up to this point. This finding is not incompatible with the theories discussed in the introduction and may in fact be explained by contemporary theories of offender spatial decision-making. For example, Johnson and colleagues (e.g., Johnson & Bowers, 2004; Johnson et al., 2009; Summers, Johnson, & Rengert, 2010) discuss offender foraging strategies and provide evidence, both quantitative and qualitative, to suggest that prolific offenders often commit offences in one area over a period of days but then quickly relocate their activity in the fear that it would otherwise attract police attention and increase their likelihood of apprehension. Such a strategy could easily lead to a more dispersed geographical pattern of activity than would be expected for less active offenders who would not be so concerned that their activity would attract police attention in a particular geographical area. Further research would be necessary to test this hypothesis more directly, and to establish more precisely which categories of behaviours may gain or lose in consistency as series length increases, since different mechanisms are likely to be implicated in different aspects of offence behaviour (e.g., learning processes; see Sorochinski & Salfati, 2010, for a discussion in the context of serial homicide; or ecological constraint; see the discussion below).
Our findings also show that for some aspects of their MO (e.g., the point of entry used, whether a home was occupied), a large proportion of the sample (around 75%) exhibited insufficient specificity in their pattern of offending for them to be differentiated from their counterparts on that aspect of their MO (Hypothesis 2). On this basis, these aspects of their MO would – at least in isolation – seem to offer less insight for crime linkage analysis, particularly across an offender's series of offences. This suggests that some aspects of an offender's MO either conform to a common preference, or are perhaps more influenced by situational factors than by stable scripted preferences.
With respect to this latter point, it is notable that the two aspects of their MO that offenders exhibited the most specificity for were, perhaps contrary to expectations, aspects which are arguably less ecologically constrained (Hypothesis 3). That is, while offenders can select an area in which to offend and (to a lesser extent perhaps) particular types of homes to target, their choice of point and method of entry will in part be determined by the presenting opportunities in the geographical area selected. Likewise, assuming that offenders wish to avoid homes that are occupied at the time of the offence, this aspect of their MO will in large part be determined by the area in which they decide to offend, as it is the routine activities of the residents that determine when homes are occupied and when they are not. The issue of ecological and mutual MO-constraints is perhaps one that deserves more systematic attention than it has been afforded up to now, and one which might, in particular, benefit from cross-contextual, comparative research designs looking at entire detected crime series.
As with any research there are, of course, limitations to the findings reported. First, not all crime is detected by the police and hence our findings are based on only a sample of those committed and solved, and this may not be representative. Second, the offences analysed included TICs and it is possible that some of these offences were not actually committed by the offenders to whom they were attributed. While we cannot rule this out, this seems unlikely for the following reasons.4 As discussed, Home Office counting rules require that adequate evidence (see above) is available to corroborate any offences that offenders nominate as TICs. Furthermore, interviews with burglars in the study area suggest that they nominate offences to be TIC only if they have committed them and believe that the police are likely to detect evidence that could be used to prosecute them (for details of the study, see Summers et al., 2010). These issues are, of course, true of all research that uses crimes detected by the police.
Third, for obvious reasons, we analysed only those crimes that were cleared to prolific offenders, and it may be that those who are less active exhibit more consistency than do those considered here. Second, crimes that were known to involve more than one offender were not examined and it is possible that patterns for co-offending differ from those reported here. Fourth, the study focuses on a single crime type, in a defined geographical area, and hence we cannot make claims as to the relevance of the findings for other crime types in other contexts (see Tonkin, Santtila et al., 2011; Tonkin, Woodhams et al., 2011; for a cross-national study of crime linkage, showing varying results in each country), though the study does introduce new questions of relevance for crime linkage, which are not conceptually or practically limited to residential burglary in our study area. Fifth, the customary limitations associated with the use of police data apply here. Police data are not collected for the purpose of research; therefore, they afford little information as to the accuracy of the information recorded (e.g., time of offence) or as to coding reliability (see Farrington & Lambert, 1997; for a discussion of police data reliability in the context of offender profiling). However, we concur with other crime linkage researchers that in natural conditions any crime linkage system would be reliant on police data; hence, studies in this domain should make use of the self-same police data, flaws and all, if they aim to be of practical relevance (e.g., Bennell & Canter, 2002).
To conclude, our findings suggest that offenders exhibit consistency in the way they commit offences, but that for some aspects of their MO they often fail to display the specificity that would be necessary for accurate crime linkage analysis. While none of this contradicts most previous research in residential burglary linkage specifically, or crime linkage generally, there is inherent value in replicating these findings through an analytical methodology which allows for the conceptual separation between, and more direct tests of, the consistency and specificity hypotheses, and their respective theoretical underpinnings. More importantly, our results demonstrate that patterns are not homogeneous across offenders. Looked at through the prism of their entire series, some offenders exhibit more specificity than do others. Around one quarter of the offenders in the sample exhibited significant specificity, even for those types of MO for which offenders are, in the aggregate, the least specific, committing offences across their series in ways that differed from the rest of the sample of offenders. Crime linkage analysis offers the most potential for these offenders, and hence identifying which crimes are likely to be the work of offenders who display more specificity a priori constitutes a major challenge for future research of this kind. Looking further, more fundamental questions are raised regarding the explanation behind these differences among the offender population.