Moderators of Sexual Recidivism as Indicator of Treatment Effectiveness in Persons With Sexual Offense Histories: An Updated Meta-analysis

The present meta-analysis is an update of the meta-analysis by Schmucker and Lösel [Campbell Syst. Rev. 2017; 13: 1–75], which synthesized evidence on sexual recidivism as an indicator of treatment effectiveness in persons with sexual offense histories. The updated meta-analysis includes 37 samples comprising a total of 30,394 individuals with sexual offense histories, which is nearly three times the sample size reported by Schmucker and Lösel (2017: 28 samples, N = 9781). In line with Schmucker and Lösel (2017), the mean treatment effect was small with an odds ratio of 1.54 [95% CI 1.22, 1.95] (p < .001). A moderator analysis suggested three predictors of importance, i.e., risk level, treatment specialization, and author confounding. Greater treatment effectiveness was suggested in high- and medium-compared to low-risk individuals and in specialized compared to non-specialized treatments. Authors affiliated with treatment programs reported larger effectiveness than independent authors. These findings were overall in line with Schmucker and Lösel (2017), though the effects of risk level and treatment specialization were stronger in the current meta-analysis. The findings of the updated meta-analysis reinforce the evidence for the first and second principle of the Risk-Need-Responsivity model. The results may support researchers and decision-makers in interpreting the current evidence on sexual recidivism as an indicator of treatment effectiveness, and, based on that, implement and carry out informative, methodologically sound evaluations of ongoing treatment programs in persons with sexual offense histories.


PRISMA
Figure 1: PRISMA flow diagram.PRISMA flow diagram illustrating the study selection process.Generated using the PRISMA2020 package (Haddaway et al., 2021) in R programming language (Team, 2022).PRISMA flow statement detailing the study selection process.
• Additional studies: N = 2 studies (Buttars et al., 2016;Mews et al., 2017), which were published after the completion of meta-analysis by Schmucker and Lösel (2017), were identified and found to be eligible based on database search and other recent meta-analyses (Gannon et al., 2019;Lösel, 2020).

ICC
Figure 2: Dot plot.Dot plot of the intraclass correlation coefficient (ICC) comparing the ranks of the 28 study-specific effect sizes reported by Schmucker and Lösel (2017) and that collected in the updated meta-analysis.Results indicated an excellent absolute agreement between the two ranks ( (&,!) = .971,p < .001)considering the guideline for interpreting ICC ( > .9excellent) (Koo & Li, 2016).

Moderator coding scheme
Following the coding scheme suggested by Schmucker and Lösel (2017), four types of moderators were collected, i.e., publication characteristics, sample characteristics, treatment characteristics, and individual characteristics.Together, a total of 17 moderators were collected (15 categorical predictors, 9 continuous predictors).In case of categorical moderators, subgroups are reported in brackets.• Descriptive validity: Descriptive validity refers to a report's quality in documenting the relevant details of an evaluation.Descriptive validity was coded following the concept suggested by Lösel & Köferl 1989Lösel & Köferl, 1989.A 4-point-scale (0 = low, 1 = medium, 2 = fair, 3 = excellent) was used to judge descriptive validity in the areas of: treatment concept, treatment realization, study design, presentation of results, overall transparency of report.Continuous variable.
• Design quality: According on the Maryland Scientific Methods Scale Farrington et al., 2002.Categorical and continuous variable.
-Level 1: No control group (excluded from analysis).
-Level 2: No equivalent control group (excluded from analysis).
-Level 3: Incidental assignment of treatment and control group.The assignment strategy gave no serious doubts that assignment resulted in equivalent groups, or sound statistical control of potential differences.The assignment strategy was not related to relevant risk variables.If there was indication of potential group differences, statistical analyses must adequately take care (e.g., regression methods including relevant control variables).
-Level 4: Matching procedures to assign treatment and control group.Systematic strategy to attain equivalence of the treatment and control group, such as theoretically sound matching or propensity score techniques.The variables used for matching must be relevant to differences that actually or potentially arise from treatment assignment for the program under evaluation.
-Level 5: Randomized assignment of treatment and control group.If there is attrition regarding the recidivism data, the study must be downgraded or even excluded depending on its severity.Treatment and control group must remain reasonably well comparable despite (potential) effects of selective attrition.

Figure 4: Forest plot general recidivism. Forest plot
Figure 3: Forest plot violent recidivism.Forest plot illustrating sample-specific odds ratios with 95% confidence intervals (OR [95% CI]) included in the updated meta-analysis with respect to violent recidivism.Square size is proportionate to the precision of the sample-specific effect sizes.Arrows indicate CIs extending beyond the axis limits.The red diamond represents the mean treatment effect on sexual recidivism with its 95% CIs given in brackets and its 95% prediction interval ([95% PI]) depicted as dotted interval around the diamond.

Table 1 :
Sensitivity analyses.Listed are results of the sensitivity analyses of the moderator analyses with respect to sexual recidivism as indicator of treatment effectiveness.For categorical moderators, effect sizes are reported in terms of odds ratio with 95% confidence intervals ( [95% CI]) for each subgroup.For continuous moderators (cont.),effectsizesare reported in terms of regression weights ().An omnibus test was done for each moderator ( ( ) based on a Wald-type test, which is a test estimating whether any of the coefficients in the moderator are significantly different from zero(Viechtbauer, 2021).Significant effects of continuous moderators are underlined (p < .05).Significant subgroup-contrasts of categorical moderators are listed in the following table. 180

Table 2 : Sensitivity analyses subgroup-contrasts.
Listed are contrasts of the categorical moderator subgroups based on general linear hypothesis testing.The Bonferroni correction was applied to counteract the problem of multiple comparisons.Significant subgroup differences are underlined (p < .05).

Table 3 :
Sensitivity analyses heterogeneity.Listed are -values indicating residual heterogeneity after accounting for each moderator based on the moderator analyses with respect to sexual recidivism as an indicator of treatment effectiveness.