Journal Article Critique Assignment

CRIMINAL JUSTICE AND BEHAVIOR, 2016, Vol. 43, No. 10, October 2016, 1310 –1329.

DOI: 10.1177/0093854816640835

© 2016 International Association for Correctional and Forensic Psychology

1310

Can at-Risk Youth be DiveRteD fRom CRime?

a meta-analysis of Restorative Diversion Programs

JENNIFER S. WONg

JESSICA BOUCHARD Simon Fraser University

JASON gRAVEL University of California, Irvine

MARTIN BOUCHARD Simon Fraser University

CARLO MORSELLI Université de Montréal

Existing reviews of the impact of restorative justice programs on juvenile recidivism have reached mixed conclusions. The present meta-analysis identified relevant studies through a systematic search of 20 databases over a 25-year period as well as the ancestry method. Application of inclusion criteria resulted in a set of 21 studies contributing 21 independent effect sizes. Programs were found to be overall effective at reducing recidivism, with a pooled odds ratio of 1.28. Subgroup analyses indicate strong evidence that study and treatment characteristics play a role in evaluation results, such as strength of research design and racial/ethnic mix of program participants. Overall quality of the literature is relatively weak, with the large major- ity of studies derived from non-peer-reviewed sources and a lack of detail presented on treatment characteristics. Limitations with respect to exclusion criteria, sample sizes, and between-study heterogeneity are discussed.

Keywords: meta-analysis; systematic review; restorative justice; youth diversion; mediation

The economic and social ramifications of mass incarceration have sparked many debates regarding how society deals with criminal offenders. Tough on-crime policies of the 1980s and 1990s have led to a meteoric rise in the prison population (e.g., Wakefield & Uggen, 2010). Under these circumstances, community-based correctional programs have emerged as potential alternatives with the promise of alleviating some of the logistical and economic pressures brought about by mass incarceration, yet holding offenders accountable

authoRs’ note: An earlier version of this research was supported by a contract from Public Safety Canada (#201303466) and we gratefully acknowledge the support of our grant managers Lucie Leonard and Cameron McIntosh. Correspondence concerning this article should be addressed to Jennifer S. Wong, School of Criminology, Simon Fraser University, 8888 University Drive, Burnaby, British Columbia, Canada V5A 1S6; e-mail: [email protected].

640835CJBXXX10.1177/0093854816640835Criminal Justice and BehaviorWong et al./Youth RestoRative DiveRsion PRoGRams research-article2016

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1311

for their actions and maintaining public safety (Petersilia, 1999; Redondo, Sanchez-Meca, & garrido, 1999; Roberts & Camasso, 1991). It should come as no surprise that diversion programs, in which eligible offenders are diverted from traditional correctional facilities to serve their sentences in non-institutional settings, have gained traction (Austin & Jacobson, 2013; Jacobson, 2005), especially in light of evidence that they are promising at reducing recidivism (Cullen & gendreau, 2001). It stands to reason that such approaches would be particularly well-suited for juvenile offenders given the particularly damaging effect of incarceration on this population. Through diversion, youth can avoid many of the negative consequences associated with the traditional criminal justice system such as a deviant label and/or an official criminal charge, isolation from peer networks, and removal from educa- tional settings, which may interfere with successful rehabilitation and reintegration (Becker, 1963; Hillian, Reitsma-Street, & Hackler, 2004; Uggen & Wakefield, 2005).

the effeCtiveness of DiveRsionaRY aPPRoaChes at ReDuCinG ReCiDivism

There exist a wide variety of correctional diversion strategies for youths, many of which have been subject to multiple evaluation studies. Participation in diversion programs can be court-ordered or voluntary, and some diversion programs may require an admission of guilt and willingness of a victim or organization to participate in mediation. Successful comple- tion of a program typically results in the termination of formal charges; however, refusal to participate or comply with conditions of the intervention may result in the offender being returned to the traditional justice system for formal processing.

Two recent meta-analyses conducted by Wilson and Hoge (2013) and Schwalbe, gearing, MacKenzie, Brewer, and Ibrahim (2012) summarized findings from evaluations of diver- sion programs. While the first study found that youth diversion programs were more effec- tive than traditional judicial intervention at reducing recidivism, the second found no effects of diversion programs on recidivism. The conflicting findings may be due to the larger sample of studies identified by Wilson and Hoge (45 compared with 28 studies for Schwalbe et al., 2012), which is in part due to the stricter criteria used by Schwalbe and colleagues with regard to methodological quality. Still, despite both studies’ stated aim of evaluating diversion programs—which are defined broadly but fairly similarly in both studies as any program that allows juveniles to avoid imprisonment—only seven evaluation reports were selected by both research teams. Such discrepancies are likely due to the difficulty in clearly defining diversion programs. In this context, the conclusions regarding diversion programs become rather difficult to interpret, as these two meta-analyses, which set out with the same goal, ended with virtually unique sets of included studies and opposing findings.

This article extends from an original study that focused on identifying all evaluations of community-based prevention programs for at-risk youth. The original set of studies con- tained programs exhibiting extreme variability in the approach and type of services pro- vided, and a detailed classification system was developed. Once programs were classified with respect to approach, it became clear that a more narrow focus was needed. Pooling the effects of fairly dissimilar studies is the dilemma that researchers face when seeking to sum- marize the literature on a set of programs falling under an amorphous umbrella such as “diversion programs.” As is evident from the existing meta-analyses of such programs, there is no universally accepted definition of a diversion program. One approach

1312 CRIMINAL JUSTICE AND BEHAVIOR

to addressing this heterogeneity in definition is to limit the type of diversionary program studied. The present study uses this approach, focusing on restorative justice (RJ) diversion programs. Although the use of RJ practices as a response to criminal behavior dates back several decades (Hillian et al., 2004; Latimer, Dowden, & Muise, 2005; Okada, 2011; Zehr, 2002), some have argued that the nature of RJ programs “do not lend themselves to conven- tional methods of evaluation research and performance measurement” (Presser & Van Voorhis, 2002, p. 162), which might explain the lack of high-quality evaluations published in the past.1 The current study seeks to summarize the literature on the effects of restorative diversion programs on recidivism using a comprehensive, rigorous set of criteria that is unique from existing meta-analytic summaries.

RestoRative JustiCe PRoGRams

According to Zehr and Mika (1997), the principle tenets of the RJ approach are that crime is a violation of people and of interpersonal relationships, that these violations create obligations and liabilities, and that RJ attempts to remedy the wrongs caused by criminal behavior. RJ approaches focus primarily on the restoration of harms, rather than on retribu- tion against the offender (Bazemore, 1998), and the RJ response to crime differs substan- tially from the traditional criminal justice system in terms of approach, nature of the proceedings, and the involvement of victims (Cormier, 2002; Kurki, 1999; Zehr & Mika, 1997).

The theoretical basis of RJ is primarily derived from reintegrative shaming (Braithwaite, 1989) as well as procedural justice (Tyler, 1990). Reintegrative shaming allows for the expression of disapproval of the offender’s actions by the community, followed by re-accep- tance of the offender into the community (Braithwaite, 1989). This process avoids stigmati- zation and labeling of the offender, making it more likely for the offender to engage in prosocial behaviors and desist from future crime (Braithwaite, 1989). Procedural justice emphasizes the perception of legitimacy of the criminal justice system, the idea being that individuals who view the justice system process as fair and legitimate are more likely to obey the law (Tyler, 1990). Some evidence suggests that RJ is perceived as more legitimate than is the traditional criminal justice process (e.g., Hayes & Daly, 2003; Kuo, Longmire, & Cuvelier, 2010; McCold, 2003; Sherman, Strang, & Woods, 2000; Umbreit & Coates, 1993).

RJ programs involve the offender taking responsibility for his or her actions, a focus on repairing harm done by requiring direct communication between victim, offender, and asso- ciated third parties, and often a requirement for some form of compensation to victims or reparations via community service. Although the fundamental premise of RJ is clear (repair- ing damage between victim, offender, and the community), the term restorative justice itself involves multiple practices and approaches (Latimer et al., 2005; Presser & Van Voorhis, 2002). Three intervention models have been said to “dominate the practice of restorative justice” (Zehr, 2002, p. 47): victim–offender mediation, family group conferences, and peacemaking circles (Latimer et al., 2005; Zehr, 2002). Although these restorative encoun- ters are similar in that they are all characterized by mediated face-to-face dialogue to dis- cuss the offense and come to an agreement on how to repair damages (Buchholz, 2014), they “differ on the ‘who’ and ‘how’” (Zehr, 2002, p. 49).

In victim–offender mediation, the victim and the offender may begin the process by meeting with a facilitator individually (Zehr, 2002). If both parties agree to meet for

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1313

face-to-face mediation, a trained facilitator will mediate the conversation. In family group conferences, the inclusion and participation of individuals beyond the victim and the offender are encouraged, such as the families of the victim and offender, and important members of the community such as justice officials. In peacemaking circles (also known as sentencing circles), a judge or a respected elder from the community typically facili- tates the discussion; a talking piece represents a unique component to this approach, and its use allows for a balanced discussion among participants (Boyes-Watson, 2005; Zehr, 2002).2

RJ diversion programs are typically considered more victim-sensitive than traditional criminal justice programs, and some research suggests that victim satisfaction is higher with RJ processes (e.g., Umbreit, Coates, & Vos, 2004). Furthermore, RJ programs may be cheaper to implement than traditional criminal justice processing, as diverting offenders from courts and prison can result in substantial savings (gromet & Darley, 2006; Sherman & Strang, 2007; Umbreit et al., 2004). Meta-analyses of the impact of RJ programs on juve- nile recidivism have reached mixed conclusions. On one hand, early reviews of RJ pro- grams found small but significant reductions in recidivism in general populations (Latimer et al., 2005), but also in juveniles specifically (Nugent, Williams, & Umbreit, 2004). On the other hand, more recent reviews of victim–offender mediation programs (Bradshaw, Roseborough, & Umbreit, 2006), RJ conferencing (Livingstone, Macdonald, & Carr, 2013), and programs broadly defined as using RJ principles (Bain, 2012) have failed to identify statistically distinguishable effects on recidivism.

Across all of these prior meta-analyses, concerns over the methodological quality of RJ program evaluations are consistently raised and substantial program/study heterogeneity is evident. Authors typically caution readers to interpret the results carefully, as well as call for additional rigorous evaluation research in this area.

the CuRRent stuDY

The aims of the present meta-analysis are to provide a comprehensive quantitative syn- thesis of restorative diversion programs for youth, consider strategies that are most effective in reducing recidivism, and identify which variables play a role in moderating outcome effects. Achieving these research objectives will further our understanding of whether restorative diversionary programs are effective and, if so, what type of program produces the largest reduction in recidivism among youth. This information is critical for policymak- ers and other decision makers when deciding whether or not to implement this type of approach.

methoD

aPPRoaCh

We conducted a systematic literature review to identify relevant studies for pooling in a meta-analysis. Meta-analysis avoids the typical bias inherent to vote-counting across con- flicting primary studies and nonsystematic narrative reviews. This quantitative literature review technique enables a summary of the state of the existing research by computing a weighted average of the studies’ individual estimates of effect for an overall assessment of treatment impact.

1314 CRIMINAL JUSTICE AND BEHAVIOR

systematic Literature search

In a revised search focusing exclusively on diversion programs, we attempted to identify the entire population of existing studies that met the a priori inclusion criteria. We con- ducted a comprehensive search of 20 bibliographic databases from January 1990 to April 21, 2015 (see Appendix A for a list of the included databases). In addition, we used the ancestry method by reviewing the bibliographies of narrative literature reviews or existing meta-analyses on diversion strategies or other related studies on diversion programs.

Database search terms. Four key constructs were used in the database searches: (a) Youth, (b) Criminal/deviant behavior, (c) Diversion program, and (d) Evaluation. Using the Camp- bell Collaboration’s guidelines (Hammerstrom, Wade, & Jorgenson, 2010), a compre- hensive search strategy using multiple synonyms and defining terms was developed and refined, with a primary goal of identifying all potential candidate program evaluations and a secondary goal of excluding primarily irrelevant studies. Multiple iterations of the search strategy were tested in an effort to include all potentially important search terms; we erred on the side of obtaining liberal (i.e., large) hit lists versus potentially missing studies. The final sets of terms were combined using a Boolean abstract search and multiple truncations (see Appendix B for a list of search terms).

Inclusion/exclusion criteria. We included in the meta-analysis evaluations of RJ diver- sion programs or strategies (i.e., programs were limited to diversionary programs spe- cifically using a restorative approach), reporting on at least one individual-level measure of crime and/or delinquency. Programs were limited to settings in Canada, the United States, Australia, New Zealand, or a Western European country, with delivery in a non- closed community setting (i.e., not delivered primarily in schools during school hours, in youth custody or residential centers, or in hospitals). Studies in the analysis had a minimum treatment sample size of 20 and included at-risk youths (with respect to risks of delinquency involvement or recidivism) predominantly aged 12 to 18 years. The stud- ies used a pretest/post-test design or a treatment/control group design with a control group deemed appropriate for comparison purposes (i.e., random assignment to condi- tions or quasi-experiment with matched control). Studies were published in English or in French between January 1990 and April 2015 (this time frame was selected to avoid concerns of extending the research beyond the last 25 years and including outdated pro- gram approaches that may affect the relevance of the generalizability to programs that exist today). Finally, studies provided sufficient numerical or graphical data to allow for computation of an effect size with respect to treatment impact.

Studies were excluded if the targeted outcomes were limited to substance abuse (tobacco, alcohol, illicit drugs), or if the targeted population were substance users. It appeared clear to the authors that substance abuse programs target a distinct population and a set of issues (e.g., addiction) that require a separate treatment from other types of RJ programs.3 given the large number of evaluations of substance use prevention programs, as well as numerous research syntheses and meta-analyses of the effects of these programs (e.g., Filges, Knudsen, et al., 2015; Filges, Rasmussen, Andersen, & Jorgensen, 2015; Holloway, Bennett, & Farrington, 2006; Lindstrom et al., 2015; Tanner-Smith, Steinka-Fry, Hennessy, Lipsey, & Winters, 2015), we excluded them to focus on less-studied diversion programs.

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1315

Studies were also excluded if the outcome measure was assessed using an entity other than the at-risk youth himself or herself (e.g., parent, sibling, physician, teacher, commu- nity). Our goal was to focus on programs specifically targeting at-risk youths as opposed to larger community-wide crime prevention programs or programs targeting family function- ing. To select a commensurable set of studies, we also excluded programs targeting very specific populations of at-risk youths (e.g., juvenile sex offenders, known gang members, Torres Strait Islander indigenous peoples). Finally, we excluded interventions that included a pharmacological component (e.g., selective serotonin reuptake inhibitors [SSRIs], Vitamin D) or that were part of a standard clinical treatment (e.g., a psychiatrist’s roster of patients), as these programs target very specific types of at-risk youth (e.g., those suffering from clini- cally diagnosed anxiety or depression), and also require medical supervision to implement.

CoDinG of stuDY RePoRts

Data were coded for each study that describes (a) characteristics of the publication (date, publication type, peer-reviewed status), (b) characteristics of the intervention (time period in which the intervention was delivered, geographic location), (c) characteristics of the participants (gender mix, predominant race/ethnicity), and (d) characteristics of the study (sample size, type of research design, type of data [self-report or official report], type of outcome measure [e.g., arrest, police contact, reconviction], length of follow-up period).

the effeCt size

To pool results across the set of included studies, individual study findings are first transformed into an effect size: a standardized outcome format which represents the mag- nitude as well as the direction (i.e., positive vs. negative) of the outcome. given that the large majority of the studies in the current analysis present 2 × 2 outcomes in terms of youth who recidivated or did not recidivate, effect sizes were calculated as log odds ratios (ORs).4 An OR compares two groups in terms of the relative odds of an event—in this case, a measure of crime such as arrest, contact with police, or a new offense. ORs are centered on 1 (rather than zero), with 1 indicating no relationship between treatment and outcome—in other words, recidivism is equally likely to occur in both groups. Studies were coded such that values below 1 indicate a negative effect of the treatment (i.e., the treatment group is more likely to recidivate), and values greater than 1 indicate a benefi- cial effect of treatment (i.e., treatment group participants are less likely to recidivate). ORs were computed using David Wilson’s effect size calculator, available on the Campbell Collaboration’s website.5

anaLYtiC aPPRoaCh

A fundamental concern in meta-analysis is the pooling together of commensurate studies (avoiding an “apples and oranges” comparison; Lipsey & Wilson, 2001, p. 2). We test for the presence of heterogeneity in the effect size distributions using a Cochran’s Q statistic and an I2 test. The Q statistic tests whether differences between study effect sizes are the result of random subject-level sampling error (i.e., whether samples for each of the studies were drawn from the same population; Lipsey & Wilson, 2001). The I2 test ranges from 0%

1316 CRIMINAL JUSTICE AND BEHAVIOR

to 100%, and estimates the percent of total variation across the effect sizes that is due to the true effect of the treatment rather than to sampling variation (Higgins, Thompson, Deeks, & Altman, 2003).

Because between-study heterogeneity emerged as a concern in our analysis (see “Results” section of this article), we opted for a random effects model rather than a fixed effects model. Random effects models (DerSimonian & Laird, 1986) assume that variability across individ- ual study effect sizes is a function of sampling error across studies (noise) as well as variability in the population effects being estimated (Egger & Smith, 2001). This model controls for such variability by using each study’s effect size inverse variance as its weight in the pooled analy- sis. Therefore, more precise effect size estimates have a greater contribution to the pooled mean effect than do less precise effect size estimates. Random effects models are generally more conservative than fixed effects models and have lower statistical power (Lipsey & Wilson, 2001).6 As is typical in the case of between-study heterogeneity, we modeled between- study differences using subgroup and moderator analyses, which enable us to assess whether one or more study-level variables explain some of the observed heterogeneity.

We first conduct a meta-analysis for the entire set of restorative diversion programs. Next, we explore potential sources of variability between studies based on observable study characteristics using moderator analyses. This approach tests whether a portion of the het- erogeneity between studies is associated with particular variables that can be modeled in the analysis. We conducted moderator analyses using four characteristics of the studies: (a) Program delivery year (1990-1999 vs. 2000-2015), (b) Research design (randomized/ strongly matched comparison group vs. weakly matched comparison group), (c) Sample size (<100 students vs. 100+ students), and (d) Sample race/ethnicity (predominantly Caucasian vs. predominantly minority or mix). These variables were selected based on pre- vious meta-analytic research that suggests these factors may be related to study outcomes (e.g., see Farrington & Welsh, 2003; Lipsey & Wilson, 2001; Sweet & Applebaum, 2004; Wilson & Lipsey, 2000). For example, it may be that studies using less rigorous research designs find higher rates of delinquency among study participants than do studies relying on weaker designs, or that programs implemented in more recent years with different cohorts of youth (i.e., millenials) experience different levels of success. Using this approach, the studies were grouped according to each of the moderating variables and a separate meta- analysis was conducted on the results within each subgroup (Deeks, Altman, & Bradburn, 2001). The resulting pooled estimates were then compared to determine whether the group- ing variable is related to differences in pooled effect estimates.

Publication biases and “small study effects” are common concerns when conducting meta-analyses. Published research is more likely to show significant results than is non- published research—a potentially biasing effect as meta-analyses rely primarily on biblio- graphic sources to identify candidate studies (Egger, Dickersin, & Smith, 2001). Related to this is the issue of small study effects; the finding that small studies often show larger treat- ment effects than do large-scale studies (Sterne, gavaghan, & Egger, 2000). We tested for publication bias and small study effects by visually examining a funnel plot in which each study’s effect size is plotted against its standard error. If all of the studies included in the analysis are estimating the same effect, the spread of effect sizes around the average effect should be proportional to their variances. As such, smaller studies would be expected to spread widely around the mean effect, whereas larger samples would demonstrate a

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1317

narrower spread. The funnel plot is accompanied by a Kendall’s Tau adjusted rank correla- tion test to assess the degree of asymmetry in the plot (Begg & Mazumdar, 1994).

Meta-analyses were conducted using the natural log of the ORs (see Lipsey & Wilson, 2001) that were subsequently transformed into ORs, enabling a more straightforward inter- pretation of findings. Analyses were conducted using the metan module for meta-analysis in State SE 14.0.

ResuLts

sYstematiC RevieW

A total of 11,996 hits were obtained through the systematic searches of 20 databases.7 Of these 11,996 hits, all potentially relevant articles based on titles/abstracts were selected for preliminary review, and virtually all were retrieved and reviewed by two team members (the exception being those reports deemed un-retrievable by our inter-library loans depart- ment, such as unpublished conference presentations).

overview of the included studies

After applying the inclusion and exclusion criteria, 21 studies contributing 21 indepen- dent effect sizes met the requirements and were included in the analysis. The effect sizes present data from 5,209 treatment group participants and 13,049 comparison group youth. Table 1 provides the characteristics of the studies included in our analyses.

The date range for publication (start date established a priori) was 1990 to 2013, with 43% of the studies published between 1990 and 1999, 33% published from 2000 to 2006, and 24% from 2007 to 2013. The majority of the studies were technical reports (62%, n = 13); 14% were published in a peer-reviewed academic journal (n = 3), four studies (19%) were theses or dissertations, and one study (5%) was a book chapter. The large majority (71%) of the studies took place in the United States (n = 15), while the remaining 29% were set in Australia, New Zealand, or Europe (n = 6).

With respect to research design, three studies used random assignment to treatment and comparison groups (14%), while five studies (24%) used a quasi-experimental design with a matched comparison group. Thirteen studies (62%) used a quasi-experimental design with a comparison group that was considered appropriate but weakly matched—for example, some differences in risk level were identified at baseline between groups (with one group clearly riskier on one or more variables such as criminal history or age), no baseline com- parisons on demographic or criminal history variables were conducted, or no information on the matching procedure was provided.

Nearly all of the studies used official reports of delinquent or criminal behavior (95%, n = 20); the remaining study used self-reports by participants (n = 1). The types of outcomes measured included either police/court contact or referral (62%, n = 13) or arrest (38%, n = 8). Follow-up periods were categorized as less than 12 months (18%, n = 3), 12 months (53%, n = 9), or more than 12 months (29%, n = 5). Four studies did not report information concerning follow-up period. The shortest follow-up period was 6 months and the longest was up to 3 years.

Treatment group sample sizes varied substantially among studies, with a range from 25 to 917 participants. Sample size was dichotomized for the purposes of analysis, with eight

1318 CRIMINAL JUSTICE AND BEHAVIOR

studies (38%) using a total sample of less than 100 students and 13 (62%) using a sample of 100 participants or more. With respect to gender, 88% of the studies used samples predomi- nantly composed of males (n = 15) and 12% of studies had samples that contained a fairly

Table 1: Characteristics of Studies Included in the Meta-analysis (N = 21 effect Sizes)

Study characteristic N (%)

Publication year (N = 21) 1990-1999 9 (42.9) 2000-2006 7 (33.3) 2007-2015 5 (23.8) Publication type (N = 21) Peer-reviewed (14.3%) Journal article 3 (14.3) Non-peer-reviewed (85.7%) Book chapter 1 (4.8) Thesis or dissertation 4 (19.0) Report 13 (61.9) Location (N = 21) United States or Canadaa 15 (71.4) Australia/New Zealand or Europeb 6 (28.6) Program delivery year (N = 21) 1990-1999 14 (66.7) 2000-2015 7 (33.3) Type of research design (N = 21) Randomized control trial 3 (14.3) Quasi-experiment with matched comparison group 5 (23.8) Quasi-experiment with weakly matched comparison group 13 (61.9) Outcome source (N = 21) Official records 20 (95.2) Self-report 1 (4.8) Outcome (N = 21) Police/court contact or referral 13 (61.9) Arrest 8 (38.1) Follow-up period (n = 17)c

Less than 12 months 3 (17.6) 12 months 9 (52.9) More than 12 months 5 (29.4) Sample size in treatment group (N = 21) Less than 100 8 (38.1) 100+ 13 (61.9) Sample gender mix (n = 17)d

Predominantly male 15 (88.2) Nearly equivalent 2 (11.8) Predominantly female 0 (0.0) Sample race/ethnicity (n = 16)e

Predominantly Caucasian 9 (56.3) Predominantly minority or mix 7 (43.8)

a. No Canadian studies were identified. b. One study took place in Europe and five in Australia/New Zealand. c. Four studies were missing data on follow-up period. d. Four studies were missing data on gender mix. e. Five studies were missing data on race/ethnicity.

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1319

even gender distribution (n = 2). Finally, more than half of the samples were composed of mostly Caucasian youth (56%, n = 9), while 44% (n = 7) of the samples contained primarily minority youth or were fairly evenly distributed with respect to race/ethnicity (five studies were missing data on race/ethnicity).

PooLeD meta-anaLYtiC ResuLts

To begin, all 21 effect sizes were pooled, representing the overall effect of restorative diversion programs for at-risk youth on criminogenic outcomes. The 21 effect sizes are substantially heterogeneous, as indicated by a statistically significant Q statistic of 6694.41 (df = 0, p < .001) and an I2 value indicating that approximately 99.7% of the variation between effect sizes is due to nonrandom factors. As such, the random effects model is appropriate, particularly given that the studies were testing substantially different types of restorative diversion programs, using differing measures, research designs, and follow-up periods.

Figure 1 presents a forest plot of each study’s log odds ratio (LOR) for the treatment/ control group comparison, with standard errors (depicted graphically by the horizontal bars extending from each LOR square), 95% confidence interval (CI), and the study’s relative weight contributing to the overall pooled effect. The LORs ranged from −1.218 to 1.027. Although it is clear from Figure 1 that not all 21 of the studies had beneficial (positive) effects (i.e., six of the effect sizes were negative), the overall picture demonstrates that restorative programs are related to lower rates of criminogenic outcomes for youths. Of the 15 positive LORs, 12 were statistically significant.

The overall pooled random effects estimate was significant, with an LOR of 0.248, 95% CI [0.068, 0.427], z = 2.71, p < .01 (see diamond at the bottom of Figure 1). This is equiva- lent to an OR of 1.28, and suggests that overall restorative diversion programs have a ben- eficial effect in terms of lowered rates of recidivism for youth.

moDeRatoR anaLYses

given the substantial heterogeneity shown among diversion program evaluations, further explorations of heterogeneity were undertaken. To investigate potential moderator effects, four dichotomous indicator variables of general study characteristics were created. Table 2 presents results from the moderator analyses of study characteristics on youth criminogenic outcomes. Subgroup estimates using fixed effects models were compared using the analog to the ANOVA method (Hedges, 1982) to determine whether these four categorical variable groupings provide any additional explanation of the variability in effect sizes (see Lipsey and Wilson, 2001). Table 2 provides measures of QW (within-group heterogeneity) and QB (between-group heterogeneity).

This analysis found significant between-groups heterogeneity for all four of the moderator variables, suggesting that the mean effect sizes produced by all of these dichotomized groups are significantly different from one another. In other words, characteristics of the studies with respect to program delivery year, research design, sample size, and sample race/ethnic- ity were important factors in predicting the magnitude of treatment effect on youth recidi- vism. All of the pooled LORs are positive, with one exception: the effect for the eight studies using a randomized or strongly matched control design (LOR = −0.009, z = 1.35, p = .176, OR = 0.991). This result favored the comparison group but was not statistically significant.

1320 CRIMINAL JUSTICE AND BEHAVIOR

given that between-group heterogeneity was significant for all comparisons, and the results in Table 2 used fixed effects models, we re-conducted analyses implementing ran- dom effects models, which are more appropriate when considering a direct comparison of group results (rather than an assessment of moderator significance; see Table 3). Using this more conservative model, we found that weaker research designs were related to a larger treatment impact (OR = 1.57, z = 2.68, p = .007), while studies using stronger research designs did not have an overall significant pooled effect of restorative treatment (OR = 0.912, z = 0.90, p = .336). More recently implemented programs were less likely to have a significant pooled impact; the effect for these seven programs was non-significant (OR = 1.40, z = 1.24, p = .215). Programs implemented during the 1990s had a significant effect at reducing recidivism (OR = 1.23, z = 2.30, p = .021).

In addition, studies using a sample of less than 100 youths in the treatment group did not have a significant pooled effect (OR = 1.24, z = 1.39, p = .165), nor did programs that treated participants who were predominantly ethnic minorities or a racial/ethnic mix (as opposed to predominantly Caucasian; OR = 0.861, z = 0.53, p = .598). Conversely, studies

Figure 1: effects of restorative programs on criminogenic outcomes (N = 21) Note. Studies are sorted by standard error. CI = confidence interval.

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1321

Table 3: Moderator analysis of Study Characteristics on Criminogenic Outcomes Using Random effects Models

Study characteristic Random effects LOR, z, p-Value, and OR

Program delivery year (N = 21) 1990-1999 (n = 14) LOR = 0.203, z = 2.30, p = .021, OR = 1.23 2000-2015 (n = 7) LOR = 0.333, z = 1.24, p = .215, OR = 1.40 Type of research design (N = 21) Randomized or QE with strongly matched control (n = 8) LOR = −0.092, z = 0.90, p = .336, OR = 0.912 Quasi-experiment with weakly matched control (n = 13) LOR = 0.448, z = 2.68, p = .007, OR = 1.57 Sample size in treatment group (N = 21) Less than 100 (n = 8) LOR = 0.218, z = 1.39, p = .165, OR = 1.24 4,100+ (n = 13) LOR = 0.262, z = 2.34, p = .019, OR = 1.30 Sample race/ethnicity (n = 16)a

Predominantly Caucasian (n = 9) LOR = 0.545, z = 5.00, p < .001, OR = 1.72 Predominantly minority or mix (n = 7) LOR = −0.150, z = 0.53, p = .598, OR = 0.861

Note. LOR = log odds ratio; OR = odds ratio. a. Five studies were missing data on race/ethnicity.

with samples of more than 100 treatment group participants were significant overall (OR = 1.30, z = .34, p = .019), as were studies with samples of predominantly Caucasian youths (OR = 1.72, z = 5.00, p < .001). These results are crucial to understanding the characteristics of diversion program evaluations that are correlated with evidence of treatment success. It appears that whether or not studies find that restorative programs are effective may have much to do with factors of the study design, rather than the programs themselves.

Table 2: Moderator analysis (Fixed effects) of Study Characteristics on Criminogenic Outcomes

Study characteristic Effect size and Q statistics

Program delivery year (N = 21) 1990-1999 (n = 14) LOR = 0.094, z = 20.64, p < .001 2000-2015 (n = 7) LOR = 0.269, z = 33.57, p < .001 Between-studies heterogeneity QB = 358.64 ~χ2(1), p < .001 Within-studies heterogeneity QW = 6,335.77 ~χ2(19), p < .001 Type of research design (N = 21) Randomized or QE with strongly matched control (n = 8) LOR = −0.009, z = 1.35, p = .176 Quasi-experiment with weakly matched control group (n = 13) LOR = 0.207, z = 42.91, p < .001 Between-studies heterogeneity QB = 648.46 ~χ2(1), p < .001 Within-studies heterogeneity QW = 6,045.95 ~χ2(19), p < .001 Sample size in treatment group (N = 21) Less than 100 (n = 8) LOR = 0.359, z = 8.20, p < .001 100+ (n = 13) LOR = 0.135, z = 33.96, p < .001 Between-studies heterogeneity QB = 25.94 ~χ2(1), p < .001 Within-studies heterogeneity QW = 6,688.47 ~χ2(19), p < .001 Sample race/ethnicity (n = 16)a

Predominantly Caucasian (n = 9) LOR = 0.113, z = 24.65, p < .001 Predominantly minority or mix (n = 7) LOR = 0.217, z = 26.61, p < .001 Between-studies heterogeneity QB = 124.67 ~χ2(1), p < .001 Within-studies heterogeneity QW = 6,462.47 ~χ2(14), p < .001

Note. LOR = log odds ratio. a. Five studies were missing data on race/ethnicity.

1322 CRIMINAL JUSTICE AND BEHAVIOR

PubLiCation bias

No evidence of publication bias was found using Begg’s test (adjusted Kendall’s Tau = −18, SD = 33.12, n = 21, z = 0.51, p = .608). Figure 2 presents a funnel plot of each study’s effect size against its standard error, with pseudo 95% confidence limits—in other words, the expected CIs if there was no heterogeneity among the studies. Although the funnel plot appears reasonably symmetric, several studies do fall outside the confidence limits, indicat- ing between-study heterogeneity.

DisCussion

The current study finds that RJ diversion programs are generally effective at reducing juvenile recidivism. The systematic review we conducted identified 21 studies providing 21 independent effect sizes. Of these 21 effect sizes, 15 suggested positive effects of the pro- grams (12 of these were statistically significant) and six suggested a negative effect of the program (five of these effects were statistically significant). Overall, the pooled result was significant and positive (OR = 1.28), suggesting a beneficial treatment impact of restorative programs on juvenile recidivism.

However, the results from evaluations of restorative diversion programs show substantial heterogeneity with respect to outcomes as evidenced by large, significant Q statistics and high I2 values. We examined characteristics of the studies and found strong evidence that study charac- teristics play a role in the results found for the programs evaluated in this study. Studies using stronger research designs did not show evidence that programs were effective at reducing recidi- vism, while studies using weaker research designs showed a significant effect of treatment. It is notable that of the 21 effect sizes used in the analysis, 86% were from non-peer-reviewed sources including book chapters, theses, and technical reports. Overall, the quality of the literature on these programs is for the most part relatively weak. Other authors have expressed similar con- cerns (e.g., Bain, 2012; Bradshaw et al., 2006; Latimer et al., 2005; Nugent et al., 2004).

Figure 2: begg’s funnel plot of study effect sizes versus standard error (N = 21)

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1323

Even after accounting for these characteristics, heterogeneity remained. Such heteroge- neity might be explained by different characteristics of the treatment and/or of the popula- tions targeted. However, investigating the importance of program and/or population characteristics, such as parental involvement, mandatory participation, program provider, program length, and risk level of participants, proved to be difficult—during the coding process, it became clear that such information was not reported consistently across the stud- ies, and missing data were prohibitive.

We were able to identify only one characteristic of treatment that had an influence on the effects reported in evaluation studies: ethnic composition of the sample. Programs serving a predominantly Caucasian sample of at-risk youths resulted in a significant effect of treat- ment, while programs serving samples that were predominantly ethnic minorities or a mix (only one study fell into the latter category) did not have a significant impact. This finding suggests that restorative programs may not adequately address the needs of certain youths, or, at the very least, may be better geared toward the needs of Caucasian adolescents. Conversely, it is possible that this finding has little to do with the program itself, but is a result of differences in baseline probabilities of subsequent contacts with police for differ- ent ethnic and racial groups. given that police contacts or rearrests are used as outcomes to evaluate programs included in this analysis, such a disparity in the effectiveness of RJ diversion programs between Caucasian and minority youths might be an artifact of the long-established finding that minority youths are more likely to be treated more harshly at all levels of the criminal justice system (e.g., Fagan, Slaughter, & Hartsone, 1987; Leiber & Mack, 2003; Rodriguez, 2010; H. Smith, Rodriguez, & Zatz, 2009).

Limitations

Limitations to the current analysis include the use of potentially contentious exclusion criteria. The specification of inclusion and exclusion criteria is critical in meta-analysis. Indeed, prior meta-analyses seeking to summarize the effects of diversion programs have differed dramatically on sets of included studies based on differences in these criteria (see Schwalbe et al., 2012; Wilson & Hoge, 2013). Although we attempted to be as broad as possible with our inclusion criteria with a focus on rigorous evaluations of restorative diver- sion programs for at-risk adolescents, an examination of our initial search results made it clear that some exclusion criteria were appropriate. Some of these criteria may not be obvi- ous choices, for example, the exclusion of programs targeting substance users or studies limited to substance use outcomes. Although substance users could certainly be considered at-risk and substance use outcomes could certainly be considered delinquency, we excluded these primarily due to the large existing literature focusing on these types of programs and the distinct nature of the issues (e.g., addiction) targeted by these programs.

In addition, because we were interested in restorative programs that have some level of generalizability, we excluded programs targeting very specific populations of youths such as sex offenders and gang members, or youths belonging to a very specific minority group. Similarly, we excluded interventions that were medical or clinical in nature as such pro- grams are limited in terms of participants who would be considered eligible, and, in the case of a pharmacological component, may present additional risks.

Although the full set of 21 effect sizes is reasonably large, it is true that for many of the moderator variable groups (e.g., eight studies used a sample size of less than 100 youth), the

1324 CRIMINAL JUSTICE AND BEHAVIOR

sample sizes were small. Most importantly, heterogeneity among the studies included in the analyses was large and was not well-explained through the subgroup and moderator analy- ses. There is no question that these programs and studies are diverse. However, they were selected using a strict set of inclusion and exclusion criteria and represent the best available evaluations of restorative programs for youths.

ConCLusion

RJ is an approach that has been embraced by many and widely implemented as an alter- native to punitive, expensive, and possibly detrimental traditional youth justice programs. Advocates of this approach as well as the theoretical basis for RJ suggest that RJ may lead to increased victim satisfaction, decreased costs, and long-term success in recidivism pre- vention through avoidance of labeling and community willingness for offender reintegra- tion. The results from the present analysis suggest that restorative approaches are a promising way to combat recidivism among youth and should continue to be implemented and evalu- ated. However, moderator analyses complicate any straightforward conclusion of the results, indicating a strong need for more peer-reviewed studies using rigorous research designs that document important program and study components that may play a meaning- ful role (e.g., parent involvement, length of follow-up), as well as programs that target youths of diverse racial/ethnic backgrounds. Although true experimental designs may be challenging in this field given the nature of the RJ process (i.e., victim and offender willing- ness to participate), quasi-experimental designs using matched comparison group controls (e.g., with propensity scores), or time-series designs with multiple pre- and post-test assess- ments are clear avenues for increasing the internal validity of conclusions drawn from pro- gram evaluations. One option for matching is presented by Latimer et al. (2005), who suggest measuring participants’ motivation to participate in the RJ process prior to actual participation, thus allowing for the comparison of treatment and control group offenders of similar motivational levels (e.g., high, moderate, and low). Motivation to engage in and fulfill victim and community reparations is of key importance for offenders to successfully participate in the RJ process. We recommend creating appropriate comparison groups by measuring this construct along with more standard measures of demographics, criminal his- tory, and current offense.

aPPenDix a

List of 20 Databases

ebsco host: 5,229 hits

1. Academic search premier 2. Criminal justice abstracts 3. ERIC 4. Medline 5. PsycARTICLES 6. PsycBOOKS 7. PsycINFO 8. Social sciences full text 9. Social sciences abstracts

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1325

notes

1. Although quantitative measures of recidivism are often a focus in evaluative literature, there is also an extensive lit- erature on qualitative outcomes for diversion programs. Frequently cited outcomes include perceptions of fairness by victim and/or offender, levels of satisfaction, and procedural fairness (Buchholz, 2014). Qualitative analyses offer an alternate per- spective to the concept of “effectiveness” and should be considered alongside recidivistic measures. Recidivism is a crucial measure of effectiveness, however, with clear implications for community safety, community support, and economic viability. As such, recidivism is the key outcome measure of focus in the current study.

Proquest host: 3,273 hits

1. Canadian research index 2. National criminal justice reference service (NCJRS) 3. PAIS international 4. Proquest dissertations and theses full text 5. Social services abstracts 6. Sociological abstracts

ovid: 104 hits

1. Cochrane central register of controlled trials 2. Cochrane database of systematic reviews 3. Database of abstracts of reviews of effects

solo Databases

1. Open access theses and dissertations: 1,091 hits 2. Web of science: 2,299 hits

aPPenDix b

seaRCh teRms

Construct 1: (youth* OR juvenile* OR adolesc* OR teen*) AND Construct 2: (crime* OR criminal* OR devian* OR violen* OR delinquen* OR offend* OR offense* OR recidiv* OR gang OR gangs) AND Construct 3: (“skills training” OR “skills development” OR “academic skills” OR “education* program” OR “treatment program” OR “vocational training” OR “vocational skills” OR “employ- ment program” OR “behavior management” OR “behaviour management” OR “behavioral inter- vention*” OR “behavioural intervention*” OR “behavioral skills” OR “behavioural skills” OR “aggression replacement training” OR “conflict management” OR divert OR diversion OR coun- seling OR mentor* OR rehabilitat* OR mediation OR “police caution*” OR “victim offender con- ferencing” OR “restorative conferencing” OR “family group conferencing” OR “teen court” OR “youth court” OR “juvenile court” OR “student court” OR “peer court” OR “family court” OR “circle sentencing” OR “peacemaking circle*” OR “youth peer panel*” OR “youth panels” OR “neighborhood boards” OR “neighbourhood boards” OR “community diversion boards” OR “reparative boards” OR “prevention program*” OR reconciliation OR restitution OR retribution OR “scared straight” OR “fear based” OR disciplin* OR “boot camp” OR “shock program*” OR “shock incarceration” OR wilderness OR deterrence OR “intensive community program” OR “community service” OR “alternative to imprisonment” OR “alternative to incarceration” OR “alternative to detention”) AND Construct 4: (evaluat* OR effect* OR impact* OR outcome*)

1326 CRIMINAL JUSTICE AND BEHAVIOR

2. Although peacemaking circles are considered to be one of the dominant practices of restorative justice (Latimer et al., 2005; Zehr, 2002), there is little representation of this practice in the current evaluation literature with respect to its effects on recidivism. In our current systematic search, we identified no evaluation studies of this approach that met inclusion criteria.

3. If a sample contained youth who were substance users, they were not necessarily excluded from our set of studies. Rather, we excluded programs that specifically targeted only substance users. We also did not exclude studies measuring substance use outcomes as long as other measures of crime/delinquency were also considered.

4. Three exceptions to the above existed. First, the study Brooks (2013) did not present 2 × 2 data but instead presented a pre-calculated odds ratio. Second, the study by Luke and Lind (2002) presented a continuous outcome measure with mean and standard deviation; this effect size was calculated as a standardized mean difference (Cohen’s d), then converted to an odds ratio using the Cox logit method to ensure compatibility of continuous and dichotomous outcomes (see Sanchez-Meca, Marin-Martinez, & Chacon-Moscoso, 2003). Similarly, the study by Sherman and colleagues (2000) presented a t statistic for the comparison of treatment and control groups; this was used to compute a Cohen’s d, which was converted to an odds ratio using the Cox logit method.

5. Available at http://www.campbellcollaboration.org/resources/effect_size_input.php 6. Mathematically, smaller studies are weighted more heavily than they are in fixed effects models (Higgins & green,

2006), as the inverse variance weights take into account both within-study variability and between-study variability (Sweet & Applebaum, 2004).

7. Note that as this article extends from an original study that focused on identifying all evaluations of community-based

prevention programs for at-risk youth, this set of 11,996 hits encompasses program types other than restorative justice.

RefeRenCes

*Studies marked with an asterisk were included in the meta-analysis. *Allard, T., Stewart, A., Chrzanowski, A., Ogilvie, J., Birks, D., & Little, C. (2010). Police diversion of young offenders and

Indigenous over-representation (Trends & Issues in Crime and Criminal Justice No. 390). Canberra: Australian Institute of Criminology.

Austin, J., & Jacobson, M. (2013). How New York City reduced mass incarceration. New York, NY: Vera Institute of Justice. Bain, K. (2012). Restorative justice and recidivism: A meta-analysis (Unpublished master’s thesis). University of Denver,

CO. Retrieved from http://digitalcommons.du.edu/cgi/viewcontent.cgi?article=1045&;context=etd Bazemore, g. (1998). Restorative justice and earned redemption. American Behavioral Scientist, 41, 768-813. Becker, H. (1963). Outsiders: Studies in the sociology of deviance. New York, NY: Free Press. Begg, C. B., & Mazumdar, M. (1994). Operating characteristics of a rank correlation test for publication bias. Biometrics,

50, 1088-1101. *Bergseth, K. J., & Bouffard, J. A. (2007). The long-term impact of restorative justice programming for juvenile offenders.

Journal of Criminal Justice, 35, 433-451. Boyes-Watson, C. (2005). Seeds of change: Using peacemaking circles to build a village for every child. Child Welfare, 84,

191-208. Bradshaw, W., Roseborough, D., & Umbreit, M. (2006). The effect of victim offender mediation on juvenile offender recidi-

vism: A meta-analysis. Conflict Resolution Quarterly, 24, 87-98. Braithwaite, J. (1989). Crime, shame, and reintegration. Cambridge, UK: Cambridge University. *Brooks, A. (2013). Moving forward: Two approaches to repairing the harm through restorative justice (Unpublished doc-

toral dissertation). American University, Washington, DC. Buchholz, M. (2014). Effectiveness of restorative justice programs in the prevention of juvenile crime (Unpublished master’s

thesis). North Dakota State University of Agriculture and Applied Science, Fargo. *Cannon, J. (2002). A case study of a restorative community justice initiative for juveniles in the southeastern United States

(Unpublished doctoral dissertation). Florida State University, Tallahassee. *Carr, C., & Nelson, P. (2000). Chapter 3: The VORP in Los Angeles County. In A. Evje & R. C. Cushman (Eds.), A summary

of the evaluations of six California victim offender reconciliation programs (pp. 33-43). San Francisco: The Judicial Council of California, Administrative Office of the Courts, Center for Families, the Children, & the Courts.

Cormier, R. (2002). Restorative justice: Directions and principles. Ottawa, Ontario: Public Works and government Services Canada.

*Cosden, M., Casas, M., & Wolfe, M. (1999). Chapter 6: The VORP in Santa Barbara County. In A. Evje & R. C. Cushman (Eds.), A summary of the evaluations of six California victim offender reconciliation programs (pp. 65- 74). San Francisco: The Judicial Council of California, Administrative Office of the Courts, Center for Families, the Children, & the Courts.

Cullen, F., & gendreau, P. (2001). From nothing works to what works: Changing professional ideology in the 21st century. The Prison Journal, 81, 313-338.

*Cunningham, T. (2007). Pre-court diversion in the Northern Territory: Impact on juvenile reoffending (Trends and Issues in Crime and Criminal Justice No. 339). Canberra: Australian Institute of Criminology.

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1327

Deeks, J. J., Altman, D. g., & Bradburn, M. J. (2001). Statistical methods for examining heterogeneity and combining results from several studies in meta-analysis. In M. Egger, g. Smith, & D. g. Altman (Eds.), Systematic reviews in health care: Meta-analysis in context (2nd ed., pp. 285-312). London, England: BMJ Publishing group.

DerSimonian, R., & Laird, N. (1986). Meta-analysis in clinical trials. Controlled Clinical Trials, 7, 177-188. Egger, M., Dickersin, K., & Smith, g. (2001). Problems and limitations in conducting systematic reviews. In M. Egger, g.

Smith, & D. g. Altman (Eds.), Systematic reviews in health care: Meta-analysis in context (2nd ed., pp. 43-68). London, England: BMJ Publishing group.

Egger, M., & Smith, g. D. (2001). Principles and procedures for systematic reviews. In M. Egger, g. Smith, & D. g. Altman (Eds.), Systematic reviews in health care: Meta-analysis in context (2nd ed., pp. 23-42). London, England: BMJ Publishing group.

Fagan, J., Slaughter, E., & Hartstone, E. (1987). Blind justice? The impact of race on the juvenile justice process. Crime & Delinquency, 33, 224-258.

Farrington, D. P., & Welsh, B. C. (2003). Family-based prevention of offending: A meta-analysis. Australian & New Zealand Journal of Criminology, 36, 127-151.

Filges, T., Knudsen, A.-S. D., Svendsen, M. M., Kowalski, K., Benjaminsen, L., & Jorgensen, M. K. (2015). Cognitive- behavioural therapies for young people in outpatient treatment for non-opioid drug use: A systematic review. Campbell Systematic Reviews, 11(3), 1-172.

Filges, T., Rasmussen, P. S., Andersen, D., & Jorgensen, A. K. (2015). Multidimensional family therapy (MDFT) for young people in treatment for non-opioid drug use: A systematic review. Campbell Systematic Reviews, 11(8), 1-125.

gromet, D. M., & Darley, J. M. (2006). Restoration and retribution: How including retributive components affects the accept- ability of restorative justice procedures. Social Justice Research, 19, 395-432.

Hammerstrom, K., Wade, A., & Jorgenson, A. M. K. (2010). Searching for studies: A guide to information retrieval for Campbell Systematic Reviews. Campbell Systematic Reviews, 2010(Suppl. 1), 1-74.

Hayes, H., & Daly, K. (2003). Youth justice conferencing and reoffending. Justice Quarterly, 20, 725-764. Hedges, L. V. (1982). Fitting categorical models to effect sizes from a series of experiments. Journal of Educational Statistics,

7, 119-137. Higgins, J. P. T. & green, S. (Eds.). (2006). Cochrane handbook for systematic reviews of interventions. Chichester, UK:

John Wiley. Higgins, J. P. T., Thompson, S. g., Deeks, J. J., & Altman, D. g. (2003). Measuring inconsistency in meta-analyses. British

Medical Journal, 327, 557-560. Hillian, D., Reitsma-Street, M., & Hackler, J. (2004). Conferencing in the Youth Criminal Justice Act of Canada: Policy

developments in British Columbia. Canadian Journal of Criminology and Criminal Justice, 46, 343-366. *Hitao, g. (1999). Chapter 8: The VORP in Sonoma County. In A. Evje & R. C. Cushman (Eds.), A summary of the evalu-

ations of six California victim offender reconciliation programs (pp. 91-100). San Francisco: The Judicial Council of California, Administrative Office of the Courts, Center for Families, the Children, & the Courts.

Holloway, K. R., Bennett, T. H., & Farrington, D. P. (2006). The effectiveness of drug treatment programs in reducing crimi- nal behavior: A meta-analysis. Psicothema, 18, 620-629.

Jacobson, M. (2005). Downsizing prisons: How to reduce crime and end mass incarceration. New York: New York University Press.

Kuo, S.-Y., Longmire, D., & Cuvelier, S. J. (2010). An empirical assessment of the process of restorative justice. Journal of Criminal Justice, 38, 318-328.

Kurki, L. (1999). Incorporating restorative and community justice into American sentencing and corrections. Washington, DC: U.S. Department of Justice, National Institute of Justice.

Latimer, J., Dowden, C., & Muise, D. (2005). The effectiveness of restorative justice practices: A meta-analysis. The Prison Journal, 85, 127-144.

*Lee, S. (1999). Chapter 7: The VORP in Santa Clara County. In A. Evje & R. C. Cushman (Eds.), A summary of the evalu- ations of six California victim offender reconciliation programs (pp. 75-90). San Francisco: The Judicial Council of California, Administrative Office of the Courts, Center for Families, the Children, & the Courts.

Leiber, M. J., & Mack, K. Y. (2003). The individual and joint effects of race, gender, and family status on juvenile justice decision-making. Journal of Research in Crime & Delinquency, 40, 34-70.

Lindstrom, M., Saidj, M., Kowalski, K., Filges, T., Rasmussen, P., & Klint Jørgensen, A. (2015). Family behavior therapy (FBT) for young people in treatment for illicit non-opioid drug use: A systematic review. Campbell Collaboration Reviews, 11(9), 1-77.

Lipsey, M. W., & Wilson, D. B. (2001). Applied Social Research Methods Series: Vol. 49. Practical meta-analysis. Thousand Oaks, CA: SAgE.

Livingstone, N., Macdonald, g., & Carr, N. (2013). Restorative justice conferencing for reducing recidivism in young offend- ers (aged 7 to 21). Cochrane Database of Systematic Reviews, 2013(2), Article CD008898. doi:10.1002/14651858. CD008898.pub2

*Luke, g., & Lind, B. (2002). Reducing juvenile crime: Conferencing versus court. New South Wales, Australia: Bureau of Crime and Statistics and Research. Retrieved from http://www.bocsar.nsw.gov.au/Documents/CJB/cjb69.pdf

1328 CRIMINAL JUSTICE AND BEHAVIOR

McCold, P. (2003). An experiment in police-based restorative justice: The Bethlehem (PA) project. Police Practice and Research, 4, 379-390.

*McCold, P., & Wachtel, B. (1998). Restorative policing experiment: The Bethlehem Pennsylvania police family group con- ferencing project. Pipersville, PA: Community Service Foundation.

*Miers, D., Maguire, M., goldie, S., Sharpe, K., Hale, C., Netten, A., . . .Newburn, T. (2001, September). An exploratory evaluation of restorative justice schemes: Executive summary (Crime Reduction Research Series Paper No. 9). London, England: Home Office.

*Nugent, W., & Paddock, J. (1996). Evaluating the effects of victim-offender reconciliation program on reoffense. Research on Social Work Practice, 6, 155-178.

Nugent, W., Williams, M., & Umbreit, M. (2004). Participation in victim-offender mediation and the prevalence of subse- quent delinquent behavior: A meta-analysis. Research on Social Work Practice, 14, 408-416.

Okada, D. (2011). Editor’s note. Special issue: Essays celebrating the 35th anniversary of Restorative Justice. Contemporary Justice Review, 14, 365-366.

Petersilia, J. (1999). A decade of experimenting with intermediate sanctions: What have we learned? Corrections Management Quarterly, 3(3), 19-27.

Presser, L., & Van Voorhis, P. (2002). Values and evaluations: Assessing processes and outcomes of restorative justice pro- grams. Crime & Delinquency, 48, 162-188.

Redondo, S., Sanchez-Meca, J., & garrido, V. (1999). The influence of treatment programmes on the recidivism of juveniles and adult offenders: An European meta-analytic review. Psychology, Crime & Law, 5, 251-278.

Roberts, A., & Camasso, M. (1991). The effect of juvenile offender treatment programs on recidivism: A meta-analysis of 46 studies. Notre Dame Journal of Law, Ethics & Public Policy, 5, 421-441.

Rodriguez, N. (2010). The cumulative effect of race and ethnicity in juvenile court outcomes and why preadjudication deten- tion matters. Journal of Research in Crime & Delinquency, 47, 391-413.

Sanchez-Meca, J., Marin-Martinez, F., & Chacon-Moscoso, S. (2003). Effect-size indices for dichotomized outcomes in meta-analysis. Psychological Methods, 8, 448-467.

*Schneider, A. L. (1990). Deterrence and juvenile crime. New York, NY: Springer-Verlag. Schwalbe, C., gearing, R., MacKenzie, M., Brewer, K., & Ibrahim, R. (2012). A meta-analysis of experimental studies of

diversion programs for juvenile offenders. Clinical Psychology Review, 32, 26-33. Sherman, L. W., & Strang, H. (2007). Restorative justice: The evidence. London, England: The Smith Institute. *Sherman, L. W., Strang, H., & Woods, D. J. (2000). Recidivism patterns in the Canberra Reintegrative Shaming Experiments

(RISE). Canberra: Centre for Restorative Justice, Research School of Social Sciences, Australian National University. Smith, H., Rodriguez, N., & Zatz, M. S. (2009). Race, ethnicity, class, and noncompliance with juvenile court supervision.

The Annals of the American Academy of Political and Social Science, 623, 108-120. *Smith, N., & Weatherburn, D. J. (2012). Youth justice conferences versus children’s court: A comparison of re-offending

(Crime and Justice Bulletin, No.160). Sydney, Australia: NSW Bureau of Crime Statistics and Research. Sterne, J. A. C., gavaghan, D., & Egger, M. (2000). Publication and related bias in meta-analysis: Power of statistical tests

and prevalence in the literature. Journal of Clinical Epidemiology, 53, 1119-1129. *Stone, K. (2000). An evaluation of recidivism rates for resolutions Northwest’s victim-offender mediation program

(Unpublished master’s thesis). Portland State University, Portland, OR. *Stone, S., Helms, W., & Edgeworth, P. (1998). Cobb County juvenile court mediation program evaluation. Retrieved from

http://www.cehd.umn.edu/ssw/rjp/resources/Research_Annotated_Bibliography/RAB/VOM/Cobb%20_County_JCM_ Program_Eval.PDF

Sweet, M. A., & Applebaum, M. I. (2004). Is home visiting an effective strategy? A meta-analytic review of home visiting programs for families with young children. Child Development, 75, 1435-1456.

Tanner-Smith, E. E., Steinka-Fry, K. T., Hennessy, E. A., Lipsey, M. W., & Winters, K. C. (2015). Can brief alcohol interven- tions for youth also address concurrent illicit drug use? Results from a meta-analysis. Journal of Youth and Adolescence, 44, 1011-1023.

Tyler, T. (1990). Why people obey the law. New York, NY: Russell Sage Foundation. Uggen, C., & Wakefield, S. (2005). Young adults reentering the community from the criminal justice system: The challenge.

In D. W. Osgood (Ed.), On your own without a net: The transition to adulthood for vulnerable populations (pp.114-144). Chicago, IL: University of Chicago Press.

Umbreit, M. S., & Coates, R. B. (1993). Cross-site analysis of victim-offender mediation in four states. Crime & Delinquency, 39, 565-585.

Umbreit, M. S., Coates, R. B., & Vos, B. (2004). Victim-offender mediation: Three decades of practice and research. Conflict Resolution Quarterly, 22, 279-303.

*Urban, L., & Burge, S. (2006). Victim/offender mediation in St. Louis: An assessment. Justice Research and Policy, 8(2), 89-113.

*URSA Institute. (1993). Final evaluation report: Community involvement in mediation of first and second time juvenile offenders project. San Francisco, CA: Community Board Program of San Francisco.

Wong et al./YOUTH RESTORATIVE DIVERSION PROgRAMS 1329

Wakefield, S., & Uggen, C. (2010). Incarceration and stratification. Annual Review of Sociology, 36, 387-406. *Wiinamaki, L. (1997). Victim-offender reconciliation programs. Juvenile property offender recidivism and severity of reof-

fense in three Tennessee counties (Unpublished doctoral dissertation). University of Tennessee, Knoxville. Wilson, H. A., & Hoge, R. D. (2013). The effect of youth diversion programs on recidivism: A meta-analytic review. Criminal

Justice and Behavior, 40, 497-518. Wilson, S. J., & Lipsey, M. W. (2000). Wilderness challenge programs for delinquent youth: A meta-analysis of outcome

evaluations. Evaluation and Program Planning, 23, 1-12. Zehr, H. (2002). The little book of restorative justice. Intercourse, PA: good Books. Retrieved from http://www.unicef.org/

tdad/littlebookrjpakaf.pdf Zehr, H., & Mika, H. (1997). Fundamental concepts of restorative justice. Contemporary Justice Review: Issues in Criminal,

Social, and Restorative Justice, 1(1), 47-56.

Jennifer s. Wong is an assistant professor of criminology at Simon Fraser University. Her work focuses primarily on crime prevention/intervention programs and policies, with a secondary focus on immigration and crime. She is primarily interested in the development of knowledge that is useful to the public, practitioners, and policymakers about best practices for prevent- ing crime.

Jessica bouchard recently completed her master of arts in criminology at Simon Fraser University. Her research interests include community corrections, alternatives to incarceration, crime prevention and intervention, and program evaluation.

Jason Gravel is a PhD candidate in the Department of Criminology, Law & Society at the University of California, Irvine. His current research focuses on social networks, street gangs, and criminal justice policy. He has published his research in Criminology & Public Policy and the Journal of Criminal Justice.

martin bouchard is an associate professor of criminology at Simon Fraser University. His work focuses on the organization and dynamics of illicit markets and on examining the impact of social networks in various criminal career outcomes. He has also published extensively on street gangs, organized crime, online illicit networks, and methodologies to estimate the size of illicit markets. His current projects focus on the social structure of serious crime in British Columbia, and its implications for understanding the dynamics of violence and illicit markets in the area.

Carlo morselli is a professor at the École de criminologie and the director of the International Centre for Comparative Criminology, Université de Montréal. His expertise lies mainly in the areas of criminal networks, organized crime, and street gang research. He is the author or editor of three books: Contacts, Opportunities, and Criminal Enterprise (2005; University of Toronto Press); Inside Criminal Networks (2009; Springer); and Crime and Networks (2014; Routledge).