In Chapter 3, the concept of ‘risk’ was explored in a static sense, in that no account is taken of the temporal aspect of crime: the results simply concern aggregated patterns over an extended time period. While this is valuable, empirical research has demonstrated consistently that considering the spatial component of crime in isolation affords only a partial understanding of patterns of offending (Johnson & Bowers, 2004a,b). Perhaps most importantly, patterns revealed by considering the
time-course of victimisation suggest that it is possible, to an extent, to predict dy- namically the locations at which crimes might occur over short time-scales and, more generally, to model the evolution of patterns of crime (Bowers et al., 2004).
The patterns revealed when time is considered in conjunction with space are, in general, variations on the concept of space-time clustering, whereby incidents oc- curring close in space tend also to occur close in time. This is the same general phenomenon as was investigated in Chapter 2, in which ‘event networks’ were used to characterise different forms of clustering. In that chapter, clustering was con- sidered in relatively abstract terms, in the sense that it was regarded simply as a phenomenon to be characterised and described without detailed recourse to its rela- tionship with theory. When such patterns are considered in more specific contexts, however, their nature and the possible reasons for their formation can be described in more concrete terms.
Urban crime, and burglary in particular, is one example for which this is the case. In this setting, clustering is most obviously manifested through the phenomena of repeat and near-repeat victimisation (see Pease, 1998; Morgan, 2001). The mean- ing of these concepts can be defined in several ways (repeated offending against the same individual, or group of individuals, for example) but, when framed in spatio-temporal terms, they refer to cases where an offence at a particular location is followed soon after by another at the same location, or elsewhere in the near vicin- ity. The prevalence of this phenomena has been demonstrated for numerous crimes (Pease, 1998; Grubesic & Mack, 2008), and appears to be ubiquitous for burglary in particular (Johnson et al., 2007).
A number of hypotheses have been advanced as possible explanations for why of- fences should occur in this way, incorporating such notions as risk heterogeneity and event dependence (described in detail in the following section). These hypothe- ses are, in turn, grounded in more general environmental theories of crime, such
as routine activity theory, pattern theory and the optimal forager principle. It is immediately apparent that many of these are the same as those invoked when mo- tivating the study of the street network in the previous chapter: ‘awareness space’, of which the street network is a primary determinant, is a central concept in both cases. The existence of this common theoretical foundation suggests, therefore, that network effects might be expected to also play a role in space-time clustering.
In some respects, the manner in which the influence of the network might be exerted is merely an extension of the static case: to the extent that it is described by risk heterogeneity, space-time clustering can simply be seen as a secondary consequence of inherent potential for crime on some streets. When other aspects are taken into account, however, a change of perspective is required. Whereas static patterns can be considered globally - in the sense that only the absolute location of an incident is important - understanding repeat victimisation necessarily implies a local perspec- tive. This is because the conceptual meaning of any (near-)repeat offence is defined by its location relative to the initial incident: it is precisely that link which defines it as a repeat. If the victimisation is thought of as the outcome of a choice process - either explicitly, as is the case under a same-offender hypothesis, or otherwise - then the choice is made, in some sense, from the perspective of the initial incident.
This change of perspective represents a technical challenge for the analysis and modelling of street network effects. Study of (near-)repeat victimisation concerns two primary issues: whether a given incident is followed by a (near-)repeat and, if so, the location of the follow-up incident. For the second of these, the fact that location is defined in relative terms suggests that streets should be considered in terms of their characteristics relative to the initially-targeted street. Few existing network metrics provide this form of comparison, and to do so therefore requires the development of new methods for network measurement. A possible candidate, referred to here as ‘commonality’, will be defined and used in the analysis.
The purpose of this chapter is, therefore, to explore the relationship between street networks and repeat victimisation, thereby extending the approach of Chapter 3 to apply to dynamic patterns. As with that chapter, the example of burglary will be used.
The existence of such a relationship would have significant implications both for the modelling of crime and for practical interventions. While the concept of (near-)repeat victimisation per se is relatively well-understood, there has been relatively little research concerning whether there is heterogeneity of risk among possible targets for follow-up incidents. Accordingly, risk elevation is typically modelled as being isotropic, simply acting uniformly in all directions around an initial incident. If the street network can be found to influence the location of secondary victimisation, this would provide a means of inferring directionality in the elevation of risk, thereby allowing preventative efforts to be directed more precisely.
The chapter will begin by presenting the theoretical background concerning space- time clustering in urban crime, particularly in the context of repeat and near-repeat victimisation. This will then be used to motivate the study of network effects, before the hypothesised effects are tested using burglary data from Birmingham, UK. In the first instance, it will be shown how an existing test of space-time clustering can be adapted to the network setting, demonstrating that clustering is indeed still present when space is considered in these terms. The focus will then move to testing for di- rectionality in patterns of near-repeat offending. The concept of ‘commonality’ will be introduced and measured for the network of Birmingham, before incorporating it in a discrete choice model of target choice in near-repeat offending.