The effects of sexual offender treatment on recidivism: an international meta-analysis of sound quality evaluations

Abstract

Objectives

Sound evaluations of sexual offender treatment are essential for an evidence-based crime policy. However, previous reviews substantially varied in their mean effects and were often based on methodologically weak primary studies. Therefore, the present study contains an update of our meta-analysis in the first issue of this journal (Lösel and Schmucker Journal of Experimental Criminology, 1, 117–146, 2005). It includes more recent primary research and is restricted to comparisons with equivalent treatment and control groups and official measures of recidivism as outcome criteria.

Methods

Applying a detailed search procedure which yielded more than 3000 published and unpublished documents, we identified 29 eligible comparisons containing a total of 4,939 treated and 5,448 untreated sexual offenders. The study effects were integrated using a random effects model and further analyzed with regard to treatment, offender, and methodological characteristics to identify moderator variables.

Results

All eligible comparisons evaluated psychosocial treatment (mainly cognitive behavioral programs). None of the comparisons evaluating organic treatments fulfilled the eligibility criteria. The mean effect size for sexual recidivism was smaller than in our previous meta-analysis but still statistically significant (OR = 1.41, p < .01). This equates to a difference in recidivism of 3.6 percentage points (10.1 % in treated vs. 13.7 % in untreated offenders) and a relative reduction in recidivism of 26.3 %. The significant overall effect was robust against outliers, but contained much heterogeneity. Methodological quality did not significantly influence effect sizes, but there were only a few randomized designs present. Cognitive-behavioral and multi-systemic treatment as well as studies with small samples, medium- to high-risk offenders, more individualized treatment, and good descriptive validity revealed better effects. In contrast to treatment in the community, treatment in prisons did not reveal a significant mean effect, but there were some prison studies with rather positive outcomes.

Conclusions

Although our findings are promising, the evidence basis for sex offender treatment is not yet satisfactory. More randomized trials and high-quality quasi-experiments are needed, particularly outside North America. In addition, there is a clear need of more differentiated process and outcome evaluations that address the questions of what works with whom, in what contexts, under what conditions, with regard to what outcomes, and also why.

Appendix

Abbreviations: SOTP = Sexual offender treatment program; TG = Treatment group, CG = Control/comparison group

Relevant studies that appeared more recently

The following is a commented list of more recent evaluations of sexual offender treatment programs (SOTPs). Some of these studies might have been eligible for the current meta-analysis, but we did not fully code them because we know of two rather large evaluations in progress that require a further update of our meta-analysis in the near future. In the meanwhile, the following list serves to check whether recent studies are in accordance with the findings of the present meta-analysis:

Abracen et al. (2011)

This study compared a TG of 64 sex offenders from an institutional SOTP in Ontario with a CG of 55 untreated offenders from the region’s correctional service. The groups were at high risk/need and matched for age at index offense, offender type, psychopathy scores, and risk of recidivism. The treatment was both individual- and group-based, applied a cognitive–behavioral relapse prevention concept and incorporated the Good Lives Model (e.g., Ward and Maruna 2007). Outcome was measured by official reoffending with mean follow-up periods of 9.4 (TG) and 11.2 (CG) years. The base rate of sexual reoffending was low (ca. 10 %) in both groups. Although the comparison between actual and predicted reoffending was somewhat more favorable in the TG, there was no significant difference in recidivism between TG and CG. It has to be noted that only a handful of offenders reoffended, both groups received other programs beyond SOTP, and there were more offenders with higher risk scores and mental health problems in the TG than in the CG.

Comment: This study has the strength of evaluating SOTP in daily practice and with a rather long follow-up period. The findings of overall low rates of sexual recidivism are consistent with our meta-analysis. They also agree with the non-significant effect we observed for custodial treatment.

Grady et al. (2013)

This study mainly investigated whether volunteering for treatment has an impact on reoffending. However, the analyses included actual treatment participation as a variable in a Cox regression model that also controlled for a number of variables (e.g., Static-99 risk score, volunteering for treatment, type of sexual offense, pedophilia diagnosis) and thus also allowed estimation of a treatment effect. A total of 161 sex offenders volunteered and participated in a mainly cognitive–behavioral, group-based institutional program in North Carolina. Non-participants had either volunteered for treatment but were not selected (n = 282) or were eligible but had not volunteered (n = 443). The study participants had a moderate–low recidivism risk. Official recidivism was assessed after 5 years for all study participants. The results indicate a significant reduction in sexual, violent, and non-violent reoffending in the TG.

Comment: The study is not designed to evaluate a treatment effect in the first place and, thus, selection bias is not controlled for, although the analyses applied incorporate a number of relevant control variables. The results differ somewhat from our meta-analytic findings as there is a significant effect in a custodial setting in a rather low-risk group.

Letourneau et al. (2013)

This study is a further U.S.-based evaluation of MST for young sex offenders; for others, see Borduin et al. (1990, 2009) in our results. It is asked whether positive results in efficacy trials could be replicated and sustained after 2 years in an implementation in a community mental health center. In a blockwise RCT on juvenile sex offenders (mean age 14.7 years), a TG of 66 young male offenders was compared with a CG of 58 offenders who received ‘treatment as usual’, that is, mainly group-based CBT interventions. The study reports on a 2-year follow-up for a number of outcomes including official recidivism (re-arrests), but differences in sexual reoffending could not be analyzed because of a very low base rate. There was also no significant decrease in re-arrests when analyses were controlled for baseline status.

Comment: The randomized design is a clear strength of this study. However, as the study does not provide enough ‘hard’ recidivism data it would not influence our results.

Olver et al. (2012)

This study compared a TG of 625 incarcerated sex offenders in Canadian institutions with a CG of 107 sex offenders who did not receive the respective treatment. All programs were based on the Canadian standards of the Risk–Need–Responsivity Model. There were some pre-treatment differences between TG and CG (e.g., less singlehood, more unrelated victims, lower risk scores and higher age at release). The authors used a brief actuarial risk scale to assess and control for group differences. A Cox regression controlling for risk found a significant effect on violent reoffending but only a smaller and not significant effect on sexual recidivism. In further analyses, treated and untreated offenders were stratified for risk level. These showed that only for the high-risk group was there a significant treatment effect on sexual recidivism. In addition, in the TG, the time to new sexual offenses was longer for treated offenders and the offenses committed were somewhat less harmful.

Comment: This is a relatively large study with particular strengths in risk-oriented analyses and differentiated outcome measurement. The overall nonsignificant effect is consistent with our above findings on custodial treatment and large sample sizes. The significant effects for offenders at higher risk are also in accordance with our results.

Smallbone and McHugh (2010)

This study evaluates prison-based treatment in Queensland, Australia. The Queensland prison system offers different treatments according to the risk (medium vs. high) and cultural background of sexual offenders. In total, 158 sexual offenders had attended a treatment program and were compared to 251 untreated sexual offenders with regard to official recidivism (police records) after an average of 29 months. The two groups differed on a number of variables (including risk measures). Treated offenders mostly had moderate–low risk while untreated offenders were at higher risk according to Static-99. Analyses controlling for risk only found a small and nonsignificant treatment effect for sexual recidivism and a marginally significant effect for any recidivism.

Comment: The results are basically in line with our findings in that a prison-based treatment of mainly low to moderate risk sexual offenders showed only a weak effect.

Smid et al. (2014)

This study from the Netherlands applied a quasi-experimental design to evaluate inpatient treatment for high-intensity sex offenders. The sample consisted of 25 % of all convicted Dutch sex offenders that were not referred to any kind of treatment between 1996 and 2002 (CG; n = 176) and all convicted Dutch sex offenders of the same time period who received an inpatient sex offender treatment (TG; n = 90). The treatment took place in special institutions that contain elements of social-therapeutic prisons and forensic hospitals. The concept is based on the Risk–Need–Responsivity model. The Static-99 was applied to control for nonequivalence in risk between the TG and CG. Outcome was measured by official data on sexual reconviction. The mean follow-up period was 12.33 years. In total, 15 % had a sexual and 38 % a violent (including sexual) offense. There were some differences in demographic and offense characteristics between the TG and CG. The results showed no overall significant treatment effect on sexual recidivism when regressions controlled for risk level, age and ethnicity. However, there was a marginally significant treatment effect for high-risk offenders. The latter was stronger for violent recidivism in general and untreated sex offenders at higher risk recidivated more frequently and faster.

Comment: This study has various strengths: It evaluates a complex institutional treatment facility outside North America, uses a long follow-up period, applies a risk-related analysis and investigates survival time curves. Although the Cox regressions may not fully control for baseline differences between TG and CG, the findings are in accordance with our results: the mean recidivism rates were in a similar range, treatment in a custodial setting had no significant effect on sexual recidivism, and the outcome was more favorable for high-risk offenders.

Worling et al. (2010)

This is an update of the Worling and Curwen (2000) study included in our meta-analysis. It is less detailed with regard to the subgroups studied and only compares treatment completers versus a comparison group comprised of non-treated sex offenders as well as treatment dropouts. Thus, we decided to retain the “older” study with a 10-year follow-up but more differentiated reporting of subgroups that allowed a more sensible intent-to-treat estimate of treatment effects. The update that recurs on a 20-year follow-up shows that the results are virtually unchanged and there were only a few additional offenders who recidivated in the 10 years after the first report.

Comment: While the very long follow-up period is a clear strength, the report does not meet stricter methodological criteria. It corroborates the results from the shorter follow-up that met the inclusion criteria of our meta-analysis.