The long-term impact of restorative justice programming for juvenile offenders

Abstract

While extant research generally supports restorative justice as an alternative to traditional juvenile court processing, much of this research is limited to short-term follow-up periods examining only prevalence of reoffense. In addition, recent meta-analyses point to several study design characteristics, the impacts of which are not well understood. This study compared long-term outcomes of youth referred to restorative justice and traditional juvenile court processing using multiple outcome measures. Specifically, the authors examined the impact of restorative justice referral on prevalence of reoffense, number of later official contacts, and seriousness of later offending behavior over several follow-up periods up to four years post-referral.

Introduction

Juvenile justice processing over the past twenty-five years has consisted of two distinct trends. On the one hand is the movement for increasingly harsher penalties for serious and violent juvenile offenders, marked primarily by the increasing use of waiver and legislative changes that allow for or mandate adult court processing for younger offenders (Sontheimer, 2001). On the other hand, there has been an expansion of various rehabilitative and/or restorative approaches, including teen or peer courts (Butts & Buck, 2000), juvenile drug courts (Butts & Roman, 2004), restorative justice (Braithwaite, 2002), and other diversionary programs to deal primarily with less serious youthful offenders. While each trend is grounded in unique theoretical orientations, they both provide alternatives to formal juvenile justice system processing, thus lessening case load burdens on juvenile courts and reducing their potentially criminogenic effects. This study examined the effectiveness of restorative justice as an alternative to traditional juvenile court processing, with an emphasis on long-term recidivism outcomes including prevalence of new offending, the number of new offenses, and the seriousness of later offending behavior. The sample in this evaluation also included a number of juvenile offenders not often included in restorative justice (RJ) programs, specifically those with prior offending histories and those with ‘persons crimes’ (i.e., violent offenses).

Restorative justice approaches to minor delinquency or criminal violations have gained popularity in the U.S. and elsewhere since the 1970s and are increasingly employed as responses to serious delinquency or adult criminal behaviors (Bazemore & Umbreit, 2001). A range of strategies are generally included under the restorative justice umbrella, such as victim-offender mediation, community reparative boards, family group conferencing, and circle sentencing (Bazemore & Umbreit, 2001), although some also include programs utilizing community service or restitution components (e.g., Bonta, Wallace-Capretta, & Rooney, 1998). The overarching purpose of restorative justice programming is restoration of both victims and offenders, as well as the reparation of harm done to the wider community, whose fabric has been negatively impacted by the crime (Smith, 2001). Restorative justice, and various forms of restorative conferencing in particular, involves a series of strategies that attempt to bring together those most affected by a criminal incident (offenders, victims, and community members) in a non-adversarial process to promote offender accountability and repair harms resulting from crime (Bazemore & Umbreit, 2001). While administrative and procedural differences exist, the four basic models of restorative processing (i.e., victim-offender mediation, family group conferencing, circle sentencing, reparative board) share common features including a community-based sanctioning focus, non-adversarial and informal processes, and decision-making by consensus (Bazemore & Umbreit, 2001).

Restorative justice (RJ) advocates often distinguish restorative programs from traditional programs based on the dichotomy between ‘retribution’ (i.e., ‘an eye for an eye’ philosophy) and ‘restoration’ (i.e., repairing the harms associated with crime) (Bazemore, 1998). In addition, the restorative justice response to crime is often contrasted with traditional system processing in terms of differences in the definition of crime, the nature of the proceedings, the primary focus of each approach, and divergent roles afforded to victims (Bazemore, 2000, Bonta et al., 1998, Cormier, 2002, Kurki, 1999, Pranis, 1998, Smith, 2001, Zehr and Mika, 1997). For instance, while traditional justice approaches define crime as an offense against the state, restorative approaches define crime in terms of harm to victims or communities (Cormier, 2002) or a violation of relationships (Zehr & Mika, 1997). Similarly, victims play a limited or passive role in traditional criminal processing, while in restorative approaches they are given a central role and encouraged to actively participate, for instance through in-person meetings with offenders. During these meetings victims are given the opportunity to express their feelings, ask questions of the offender, and articulate the impact of the criminal event. The basic ideas underlying RJ processing focus on attempts to promote offender accountability and change and to meet the needs of victims (e.g., need to be heard and have a say in the outcome of their victimization). This is accomplished by bringing together those most affected by a crime to discuss the event and its repercussions and to develop a plan to repair harms. Advocates for RJ approaches claim that these features make RJ processing superior to traditional court processing in meeting the needs of victims, strengthening the community, and potentially reducing recidivism by offenders.

While the potential for reduced recidivism is only one purported benefit of RJ programs, reoffending is a key concern for policymakers considering restorative justice as an alternative to formal court processing. Numerous evaluations of RJ programs have demonstrated high levels of victim and offender satisfaction and compliance with restorative agreements (for excellent reviews see Braithwaite, 2002, Latimer and Kleinknecht, 2000), however, evidence regarding the impact on recidivism is less consistent. Some of the existing research demonstrated reductions in reoffending (Bonta et al., 1998, Hayes and Daly, 2004, Luke and Lind, 2002, Maxwell and Morris, 2001, McGarrell, 2001, Rodriguez, 2005), while other evaluations failed to find significant reductions in recidivism (McCold and Wachtel, 1998, Niemeyer and Shichor, 1996, Roy, 1993, Umbreit, 1994).

An early meta-analysis of restorative justice programming conducted by Bonta et al. (1998) found an average reduction of 8 percent in the reoffending rate of those who participated in programs involving restorative features (e.g., restitution, community service) relative to those who did not. Other recent meta-analyses examined either the effectiveness of justice processes incorporating some restorative features (Bonta, Wallace-Capretta, Rooney, & McAnoy, 2002), a variety of restorative justice programs (Bradshaw and Roseborough, 2005, Latimer et al., 2001, Latimer et al., 2005), or victim-offender mediation programs for juvenile offenders in particular (Nugent et al., 2003, Nugent et al., 2004). Each of these meta-analyses found positive effects for RJ-style programming, with effect sizes including .03 (Bonta et al., 2002), .07 (Latimer et al., 2005), .26 (Bradshaw & Roseborough, 2005), and as high as .30 among studies with stronger methodological characteristics (Nugent et al., 2004).

While this group of meta-analyses generally agreed on the positive effects of RJ processing, studies differed in regards to the impact of various methodological characteristics on the magnitude of effect sizes uncovered. Specifically, there were a number of methodological shortcomings which persisted in the evaluation literature on restorative justice, including varying definitions of reoffense, the length of the follow-up time period studied, and various analytic strategies for comparing RJ-involved juveniles to those receiving other forms of processing. Each of these factors had been demonstrated to impact the magnitude of effect sizes reported in one or more of the meta-analyses cited above.

Restorative justice programs are generally perceived as diversionary programs, and as such, studies of their effectiveness typically compare RJ participants (who are generally first time offenders without prior offending histories) to participants from other diversion programs (e.g., Roy, 1993) or offenders in the traditional court process (Bonta et al., 1998, Luke and Lind, 2002, McCold and Wachtel, 1998, McGarrell, 2001, Rodriguez, 2005). Some studies had included comparison samples drawn from “true” diversionary programs (e.g., all first-time offenders) (McCold and Wachtel, 1998, McGarrell, 2001), while others had made comparisons of RJ participants to offenders with prior records (e.g., Bonta et al., 1998, Umbreit, 1994). As a result, some of the existing RJ research utilized biased samples (comparing offenders in RJ without offending histories, to those in comparison samples who may have had such histories) which might have produced favorable, but less credible conclusions regarding the impact of RJ. It is important then for evaluations of the impact of RJ programs to utilize appropriate (i.e., matched) comparison samples so that conclusions regarding the effectiveness of RJ interventions can be drawn from differences in recidivism between ‘treatment’ and ‘comparison’ groups that are as similar as possible. When differences do arise, it is also important for evaluations to make use of appropriate multivariate statistical procedures to control for any initial group differences in the groups' underlying propensities to recidivate.

Another important issue, not adequately dealt with in some existing RJ evaluations, relates to self-selection biases introduced by the voluntary nature of restorative programs. In most cases, once a juvenile offender is deemed ‘appropriate’ for participation in such an intervention, the juvenile is then afforded the opportunity to volunteer for actual participation. As a result, the potential for self-selection bias is introduced if only those most ‘amenable to treatment’ decide to follow through with RJ participation. For this reason, it is crucial for evaluations to examine differences in recidivism rates between those who were assigned to RJ interventions (even if they did not receive them) and those who were assigned to the comparison group. Several studies of RJ effectiveness employed such “intention to treat” designs (e.g., McGarrell, 2001, Sherman et al., 2000), while some analyzed groups based on the type of treatment actually delivered (e.g., Hayes, 2005), and still others based on assignment-treatment status (e.g., McCold & Wachtel, 1998). This distinction is important, as differences in offender motivation may be confounded with treatment effects when groups are analyzed based on whether they chose to actually participate in their assigned intervention. Analyses of “intention to treat” (ITT) or treatment as assigned can provide more credible tests of the impact of restorative justice (Sherman & Strang, 2004) by maintaining the initial group assignment independent of what was actually received. Thus the ITT approach is more suited to disaggregating treatment from motivation effects than are other analytic approaches.

While experimental designs are generally considered the most effective way to control for initial group differences in motivation and offending propensity, only a few RJ evaluations were able to carry out such designs. Even the few RJ evaluations which utilized these powerful research designs were compromised by several factors, including the voluntary nature of restorative processing and possible differential attrition rates. Random assignment to RJ processing often occurs after cases are screened for various case and offender characteristics (e.g., admission of the offense); however, whether the offender then actually receives restorative processing depends on other offender (e.g., remorse, willingness to meet with victim, later offending) and victim (e.g., willingness to meet offender, beliefs regarding restorative processing as adequate resolution) factors. As such, even studies intending to conduct random assignment to restorative versus traditional processing can suffer from self-selection biases. In fact, in one study (McCold & Wachtel, 1998), only 42 percent of cases randomly assigned to restorative conferencing (versus court processing) after initial eligibility screening actually resulted in the juvenile's participation in a restorative conference.

Yet another limitation existed in the extant literature on RJ programming which relied on quasi-experimental designs in that a number of these studies compared those who completed the restorative justice intervention with those who completed traditional court processing (Rodriguez, 2005), those who were assigned to but did not complete the restorative intervention (Niemeyer & Shichor, 1996), and offenders in other diversionary programs (Roy, 1993). Again, when analyses include as the ‘treated’ group only those juvenile offenders who complete a restorative intervention, the risk arises that desired program effects may be confounded with motivation effects. Specifically, those who successfully complete any program (not just an RJ program) may be less likely to recidivate, not because of program impacts, but because they were less “serious” offenders or were generally more amenable to treatment than those who failed to complete the intervention. In other words, “successes succeed and failures fail.” This problem may remain when comparing restorative justice completers to traditionally processed completers (e.g., Rodriguez, 2005) if restorative processing itself is seen as more difficult or cumbersome (e.g., emotionally charged) for offenders to complete, thus making the average RJ participant more likely to drop out of his/her program than is the average juvenile assigned to traditional processing (because only the most motivated RJ participants will remain in that sample).

While meta-analyses by Bradshaw and Roseborough (2005) and Latimer et al. (2005) found no differences in effect size by quality of research design (e.g., randomized designs, or the nature of the comparison group), Nugent et al., 2003, Nugent et al., 2004 found that variation in effect sizes was largely explained by a twelve-item “group formation methodology” (GFM) scale which included items related to random assignment, matching of clients on key demographic and history variables, and group placement in an unbiased manner, with studies in the top quintile on GFM score finding reductions in reoffense as large as 30 percent (Nugent et al., 2004). In addition, Bradshaw and Roseborough (2005) found stronger effect sizes among studies employing a comparison group of restorative justice failures than among studies comparing restorative justice participants to those in an alternative treatment. The extent to which such variability in study design characteristics (including randomization, method of analyses, and statistical controls for group differences) influence conclusions regarding the effectiveness of RJ programming is an important consideration. As such, additional research designed to overcome some of these lingering shortcomings is needed to further, credibly demonstrate the impact of RJ interventions on recidivism.

Yet another methodological shortcoming in the existing literature on restorative justice programming involves how recidivism is defined in various evaluations. Most RJ studies examined changes in the likelihood (prevalence) of reoffending, while fewer studies addressed new offending rates (Sherman et al., 2000) or the severity of later offending (Nugent & Paddock, 1995). While some studies utilized a broad definition of reoffense, including new officially-recorded contact with the police or the filing of new petitions to the juvenile court (Hayes and Daly, 2004, Niemeyer and Shichor, 1996, Nugent and Paddock, 1995, Rodriguez, 2005), others employed a more narrow definition of reoffense, including only new convictions or new RJ conference assignments (Luke & Lind, 2002).

These varying definitions are important because, while the meta-analysis by Bradshaw and Roseborough (2005) found no differences in effect size based on how reoffense was defined, Nugent et al., 2003, Nugent et al., 2004 again found greater variation in effect sizes in studies employing a broad reoffense definition. In addition, Nugent et al. (2003) found that studies using broad reoffense definitions also tended to have poorer group formation methods and nonequivalent groups, leading the authors to conclude that the impact of victim-offender mediation on these more broadly defined recidivism measures was unclear.

While the narrow definition of reoffense is advocated as a more conservative indicator of program impact leading to fewer ‘false positive’ errors (Nugent et al., 2003), broader reoffense measures may more accurately represent the actual behavior of the offender and may also be less influenced by juvenile justice system factors (e.g., various processing decisions within the juvenile court system). Therefore, it is important to further examine the impact of RJ processing relative to traditional processing on broadly defined reoffense measures in studies with careful comparison selection methods.

Finally, the extent to which the impact of restorative justice processing changes over time is open to debate, as some research found that program effects disappear or diminish over time (McCold and Wachtel, 1998, McGarrell, 2001), and other research found that effects that maintain over longer periods (e.g., twenty-seven to thirty-nine months) and were not evident using shorter follow-up periods (Luke & Lind, 2002). While the meta-analysis by Bradshaw and Roseborough (2005) found no indication that the length of follow-up period impacted the effect sizes reported among nineteen studies with follow-ups ranging from nine to forty-eight months, the meta-analysis presented by Nugent et al., 2003, Nugent et al., 2004 found that effect sizes generally declined over time. For instance, they reported an estimated reduction in offending of 27 percent at six months, but only 9 percent at thirty months (among studies with strong methods employing a narrow definition of reoffense) (Nugent et al., 2003). Further examination of the long-term impacts of restorative processing are thus important as extant research on the durability of RJ program effects is mixed.

In summary, recent meta-analyses provided support for restorative justice programming as an alternative to traditional justice processing. These same meta-analyses also suggested that the impact of several important study design characteristics, such as comparison/control group formation, analytic strategy, breadth of reoffending definitions, and follow-up time period on the effectiveness of restorative processing are still not completely understood. In addition, while several studies examined likelihood (prevalence) of later offending as a key outcome criterion, less is known about the effect of RJ programming on other reoffense measures, including the number of later offenses and the seriousness of later offending.