WHAT THE LITERATURE SEEMS TO SUGGEST ABOUT RESTORATIVE JUSTICE EFFECTIVENESS: AN OVERVIEW OF A META-ANALYSIS

The effects of crime are deleterious to victims, offenders, and communities, and the current approach to justice seems both ineffective and inefficient. Restorative justice proponents have posited that their humanistic approach to justice through dialogue and negotiation may be more effective than traditional retributive justice. The purpose of this study was to conduct a meta-analysis to determine if such conjecture was supported or rejected. Evaluations of victim-offender mediation (VOM) and family group conferencing (FGC) programs, the two most common forms of restorative justice, were analyzed to determine what factors, if any, explained variability across studies in outcomes. Four outcomes were analyzed: victim fearfulness of revictimization, offender recidivism rates, victim/offender satisfaction rates for both the justice process and the justice outcome, and restitution negotiation and completion rates. If variability across studies in outcomes existed, then potential explanatory variables were analyzed to determine if they significantly contributed to the variability.

Inclusion Criteria and Sample

Four inclusion criteria were developed to find appropriate studies to be used in the meta-analysis. A study had to have: (1) used either juvenile and/or adult samples; (2) implied the use of victim-offender mediation or family group conferencing (other forms of restorative justice, such as peace circles, were not evaluated); (3) indicated that participants experienced face-to-face meetings; and (4) focused on criminal rather than civil offenses. Searches for peer-reviewed articles and program evaluations were conducted through social service and criminal justice database catalogs, on the Internet, through a bibliographic review of existing studies, and by contacting researchers and institutions in an attempt to find unpublished, overlooked, or new research. A thorough review of the literature found a total of 40 studies that were suitable for analysis.

Variables

Several variables were identified as potential predictors of magnitude of the effect sizes found in the meta-analysis. First, the type of justice approach that victims and offenders participated in was determined to be an important factor: traditional justice (comparison groups), victim-offender mediation, or family group conferencing. Participants of family group conferencing (FGC), for example, may have different results than participants of victim-offender mediation (VOM) because VOM offenders may not experience the same level of discomfort as FGC offenders. Participants of FGC appear before more people in the community than participants of VOM. Second, adolescents are generally seen as being more amenable to treatment, so the age of the offender may be an important predictor of the magnitude of effect size. Next, because better-designed inquiry enables researchers to draw more accurate inferences, whether or not a study was rigorous in terms of the creation of comparison groups that were initially statistically equivalent was determined to be a potentially important predictor. Another variable that may be important in predicting the magnitude of effect size was the length of time offenders were followed in order to assess recidivism. Finally, several countries have policies encouraging the use of restorative justice approaches. Studies conducted in countries outside the United States may be more likely to have larger effect sizes.

Data Analysis

Hierarchical Generalized Linear Models (HGLM) with a logit link function was used to analyze data. All outcomes analyzed in the meta-analysis were dichotomous. The HGLM binomial model is an appropriate model to use for such data (Bryk, Raudenbush, and Congdon, 1996).

Results

Almost twenty percent of the sample was found on the Internet. The majority of the studies analyzed were program evaluations (65 percent), thirty percent were journal articles and only two (5 percent) were heses/dissertations. A common complaint of meta-analyses is the increased risk of the “file drawer problem” whereby the sample contains mostly articles from peer-reviewed journals. Thus, the sample is often not representative of the research that has been conducted as evidenced by the exclusion of program evaluations. Based on the current sample composition, the risk of this potential problem was reduced.

There were three significant findings for this meta-analysis. Other findings are important to note and this is followed by a delineation of the study limitations. Many times, certain variable combinations were excluded from analysis because to include "sparse" data could have led to erroneous results. Williams-Hayes (2002) reported that: "A preliminary analysis of the data set indicated that there were few effect sizes in table 'cells' for studies with specific independent variable combinations. In fact, for each dependent variable, preliminary analyses revealed few effect sizes in certain table cells with specific independent variable combinations. Agresti (1996) refers to this situation as 'sparse' data and shows how sparse data can lead to biased estimates of regression coefficients and odds ratios" (p. 54).

Fearfulness of Revictimization

After controlling for all explanatory variables, victims participating in both VOM (=-1.07, t[18]=-3.55, p=.003 FGC (=-1.32, t[18]=-4.10, p=.001) reported feeling less fearful of revictimization than victims in comparison groups. There was not a significant difference between the two forms of restorative justice on this outcome.

Recidivism

After controlling for all included explanatory variables, offenders who participated in FGC were not significantly less likely to have been reported to reoffend then were those in the comparison group (= -.19, t[52]=-0.58, p=.563), and neither were offenders who participated in VOM (=-0.27, t[52]=-1.52, p=.136). The length of time that offenders were followed to assess recidivism rates was, however, significantly associated (=0.02, t[52]=2.84, p=.007). The longer these offenders were followed for reoffense, the more likely they were to reoffend.

Participant Satisfaction

There were few effect sizes for FGC studies, so VOM and FGC was collapsed into a single category. Both offenders and victims in restorative justice were more likely to report feeling satisfied with the justice process than participants in comparison groups (=0.97, t[31]=4.28, p<.001 and =1.41, t[34]=5.27, p<.001, respectively). The methodological rigor of a study was also significantly associated with reports of greater satisfaction with the justice process (=0.63, t[34]=2.05, p=.048). Offenders and victims participating in restorative justice were not significantly more likely to report feeling satisfied with the justice outcome than participants in comparison groups.
Restitution Negotiation and Completion

The two forms of restorative justice were collapsed into one category because there were too few effect sizes for FGC studies. Participants in restorative justice were more likely to report higher rates of negotiating restitution contracts than participants in comparison groups (=3.38, t[15]=3.57, p<.003). Comparison group participants were significantly more likely to report smaller rates of restitution completion than restorative justice participants (=-1.99, t[16]=-2.48, p=.025).

Limitations and Conclusion

These tentative results should be taken with caution as there were potential limitations to the meta-analysis and to the designs of previous research used for this study. First, results were analyzed using a small sample (effect sizes ranged from 12-57) and several variables had to be omitted from analyses because there were too few effect sizes available. At other times significant variables were excluded because they were not available in the original research. There were also limitations of original research. For instance, seldom were participants randomly assigned to treatment and comparison groups. Findings were often from studies with non-equivalent post-test only groups. Finally, oftentimes categories were collapsed into a dichotomous variable if there were not enough studies that reported ordinal rankings (e.g., satisfied or dissatisfied rather than levels of satisfaction). Subtle, but important differences in levels of satisfaction, for example, were possibly lost. Finally, one of the cornerstones of restorative justice is the ability for potential participants to give consent or decline services. Self-selection bias is a potential threat.

Much of this research corroborates previous research. Findings on reoffense rates have been more mixed with some research demonstrating positive effects of restorative justice on recidivism and/or seriousness of reoffense (e.g., Nugent and Paddock, 1995). The following are recommendations for future research: replication, the need to diversify study samples, evaluation of programs utilizing random assignment, and the use of more comparison groups.

by Mona M. Williams-Hayes (University of Tennessee, Knoxville, TN, USA).

Suggested Readings

Agresti, A. (1996), An Introduction to Categorical Data Analysis, John Wiley: New York.

Bryk, A.; Raudenbush, S. and Congdon, R. (1996), Hierarchical Linear and Nonlinear Modeling with the HLM/2L and HLM/3L Programs, Scientific Software International: Chicago, IL.

Nugent, W.R. and Paddock, J. (1995), The Effect of Victim-Offender Mediation on Severity of Reoffense, Mediation Quarterly, Vol. 12, pp. 353-67.

Williams-Hayes, M.M. (2002), The Effectiveness of Restorative Justice: A Meta-Analysis, dissertation, The University of Tennessee: Knoxville, TN.