It has long been recognized that a small proportion of individuals are responsible for a disproportionate number of crimes. Wolfgang, Figlio, and Sellin (1972) reported that 6% of their Philadelphia cohort accounted for 50% of the criminal acts to age 17. More recently, Welsh, Loeber, Stevens, Stouthamer-Loeber, Cohen, and Farrington (2008) found that 10.2% of their sample from the Pittsburgh Youth Study (PYS) accounted for 50.1% of all self-reported offences. This small collection of high-rate habitual offenders often begin their criminal activity at an early age and continue into adulthood, commit serious and violent crimes, impose considerable financial costs on society, and pose a significant challenge to the criminal justice system (Cohen, Piquero, and Jennings 2010; Piquero, Farrington, and Blumstein 2003). Therefore, targeting early intervention and prevention programs for children and youth likely to become chronic offenders is an important task for our society that makes good economic sense and holds the greatest promise for crime reduction (Alltucker, Bullis, Close, and Yovanoff 2006; Aos, Phipps, Barnoski, and Lieb 2001; Craig, Schumann, Petrunka, Khan, and Peters 2011).
Using data from the Second Philadelphia Birth Cohort (Tracy, Wolfgang, and Figlio 1990), Cohen et al. (2010) estimated that the financial burden of the small group of high-rate chronic offenders identified in their study (3.1% of the sample) was nearly half the total cost of offending for the entire sample. Moreover, emerging evidence suggests that the greatest gains from crime prevention efforts come from targeting those individuals with the highest risk factors (Dodge and McCourt 2010; Welsh and Farrington 2007). This was the conclusion reached by Foster, Jones, and the Conduct Problems Prevention Research Group (2006) when they evaluated the cost-effectiveness of the Fast Track prevention program for at-risk children in first through tenth grade. Using incremental cost-effectiveness ratios (ICERs), an index of the costs of the program relative to the outcomes, they determined that, for the group at lowest risk at intake, the ICER was negative and the effectiveness probability was 6%, indicating that the program was neither cost-effective nor likely to be effective with this group. By contrast, for the group at highest risk at intake (defined as those above the 90th percentile on the screening measures), the estimated ICER was found to be $752,103 (less than the $1 million threshold) and the effectiveness probability was 99%, indicating both cost-effectiveness and a high likelihood of being effective with this group.
Across Canada, many child- and youth-services agencies engage in early intervention and prevention programs for crime prevention. A challenge for any targeted (i.e., indicated or selected) prevention or early intervention program is to identify those individuals most at risk of a maladaptive outcome, such as a life of crime (Cohen et al. 2010; LeBlanc 1998; Lochman 2006). Considerable efforts have been expended over the past decades to identify factors that are most strongly associated with the onset and maintenance of criminal behaviour (Leschied, Chiodo, Nowicki, and Rodger 2008; van Domburgh, Vermeiren, Blokland, and Doreleijers 2009). These efforts have informed the development of myriad programs for young people aimed at preventing or forestalling the onset of antisocial activity by strengthening protective factors and reducing the impact of risk factors; some of these programs have been shown to be effective (Farrington 2007). However, further work needs to be done to identify individuals who show risk factors associated with the most serious, protracted, and highest-rate criminal careers.
Our framework for understanding the nature and course of offending comes from the risk factor research (RFR) and developmental and life course (DLC) paradigms advanced by Farrington (2003, 2005a), which are concerned with identifying and investigating linkages between past events (i.e., risk factors) and future outcomes. Consistent with these approaches, longitudinal studies that track criminal activity over time, ideally across major developmental periods such as adolescence and adulthood, and are able to identify, prospectively or retrospectively, early risk factors associated with serious (i.e., high-rate, persistent) offending could aid the development of targeted intervention and prevention strategies. Longitudinal studies have the advantage over other methodologies such as cross-sectional research of tracking within-individual developmental pathways and of identifying how life events are associated with change and continuity across the life course. Moreover, recent advances in person-centred statistical analyses have enabled longitudinal researchers to examine within-individual change over time as well as to identify distinct patterns of within-sample offending behaviour (i.e., age-crime trajectories).
One of these statistical techniques is group-based trajectory analysis (Nagin 2005). Group-based trajectory analysis is a specialized application of finite mixture modelling (McLachlan and Peel 2000) that allows the researcher to identify clusters of individuals whose pattern of offending is statistically similar as it unfolds over time. Furthermore, consistent with the classify/analyse paradigm (Piquero 2008), once individuals are sorted into discrete trajectory groups, regression analysis (or other statistical approaches) may be applied to identify the best set of developmental predictors to differentiate among the groups (Nagin and Odgers 2010a). Childhood and adolescent variables reflecting various life domains (e.g., individual, family, peer, school, and neighbourhood) are recorded and are then subjected to the analysis. In this regard, group-based trajectory analysis may be well suited to identify risk factors that could inform the development of targeted early intervention and prevention programs (Chung, Hill, Hawkins, Gilchrist, and Nagin 2002; Cohen et al. 2010; Piquero, Paternoster, and Brame 2011; Wiesner and Capaldi 2003). (2) Adopting this approach reflects the influence of both the DLC and the risk factor research (RFR) paradigms. However, the approach has been called into question, as some evidence suggests that trajectory analysis may not be useful for this purpose (Bersani, Nieuwbeerta, and Laub 2009; Sampson and Laub 2003; Skardhamar 2010). The next section briefly reviews the literature on group-based trajectory analysis.
Findings from group-based trajectory analyses
The criminology field has widely embraced group-based trajectory analysis since its advent about 20 years ago (Nagin and Odgers 2010b). Piquero (2008) identified over 80 studies that have used this statistical technique. As a review of all these studies is beyond the scope of this paper, selected findings will be highlighted, with a particular emphasis on risk factors associated with the most serious offence trajectories. Across studies, the number of trajectory groups yielded varies from as few as two (Yessine and Bonta 2009) to as many as eight (Thornberry, Bushway, Krohn, and Lizotte 2004), though four to six is typical (Piquero 2008). Reasons for differences in the number of trajectory groups include sample characteristics, methodological design, the number of time points for assessment, and outcome variable definition.
Comparisons across trajectory groups on offending-related variables indicate that groups differ in terms of the average age of onset, length of criminal trajectory, peak age of offending, and number of offences committed. Moreover, studies with community samples commonly identify a non-offender group, which often comprises the majority of individuals in the sample (e.g., Piquero, Farrington, and Blumstein 2007). Studies with offender samples typically identify a low-rate (e.g., near-zero) trajectory group, which often comprises the largest group in the sample. For example, Bersani et al. (2009) found that 70% of their offender sample fell into the lowest-rate group (referred to as sporadic offenders).
In addition to identifying a non-offender or low-rate group, the other end of the trajectory group spectrum reports a high-rate trajectory group who show persistence in their offending. These high-rate chronics (as they are typically labelled) generally constitute about 10% of the sample, irrespective of sample characteristics. There is now a growing body of literature that has examined the risk factors (3) of these trajectory groups. As noted, this literature might prove useful in identifying early predictors of high-rate persistent offenders, for whom targeted early intervention and prevention programs might be developed, as well as the particular risk and protective factors that could be targeted by the intervention.
Twenty studies were identified that examined both trajectory groups and risk factors of trajectory-group membership. With regard to the relations between risk factors and trajectory groups, some studies have reported dose effects such that high-rate groups evince the most risk factors, low-rate groups show the most favourable backgrounds, including the most protective factors, and moderate-rate groups fall somewhere in between (e.g., Fergusson, Horwood, and Nagin 2000; Maldonado-Molina, Piquero, Jennings, Bird, and Canino 2009; Sampson and Laub 2003; van Domburgh, Loeber, Bezemer, Stallings, and Stouthamer-Loeber 2009). Other studies did not report dose effects but rather reported on differences for discrete variables such that specific risk factors were associated with particular trajectory groups (e.g., Ward, Day, Bevc, Sun, Rosenthal, and Duchesne 2010; Wiesner and Windle 2004). Table 1 summarizes the specific risk factors associated with serious offence trajectories. The risk factors identified were those that were unique to the high-rate chronic group and that distinguished that group...