Practice Profile: Treatment for Adults Who Have Committed Sex Offenses

The 2006 meta-analysis by Aos, Miller, and Drake updated and extended an earlier 2001 review by Aos and colleagues. The overall goal of the review was to provide policymakers at Washington State Institute for Public Policy with a comprehensive assessment of adult corrections programs and policies that have the ability to affect crime rates. This meta-analysis concentrated exclusively on adult corrections programs.

A comprehensive search procedure was used to identify eligible studies. Studies were eligible to be included if they 1) were published in English between 1970 and 2005, 2) were published in any format (peer-reviewed journals, government reports, or other unpublished results), 3) had a randomly assigned or well-matched comparison group, 4) had intent-to-treat groups that included both complete and program dropouts, or sufficient information was available that the combined effects could be tallied, 5) provided sufficient information to code effect sizes, and 6) had at least a 6-month follow-up period and included a measure of criminal recidivism as an outcome.

The search resulted in the inclusion of five evaluations of sex offender programs. The five studies included close to 900 participants. Two studies were published in a journal. The other studies were government reports or unpublished evaluations. No information was provided on the age, gender, or racial/ethnic breakdown of the studies’ samples, nor on the location of the programs.

The mean difference effect size was calculated for each program. Adjustments were made to the effect sizes for small sample sizes, evaluations of “non–real world” programs, and for the quality of the research design. The quality of each study was rated using the University of Maryland’s five-point scale; only studies that received a rating of 3 or higher on the scale were included in the analysis (a rating of 3 means a study used a quasi-experimental design with somewhat dissimilar treatment and comparison groups but there were reasonable controls for differences). Once effect sizes were calculated for each program effect, the individual measures were added together to produce a weighted average effect size for a program or practice area. The inverse variance weight was calculated for each program effect, and those weights were used to compute the average. The fixed-effects model was used for the analysis.

Lösel and Schmucker (2005) conducted a comprehensive, review of treatment programs. Studies were included in their meta-analysis if they met the following criteria: 1) participants in each study had to have been convicted of a sex offense or must have been committing acts of illegal sexual behavior that would result in a conviction if prosecuted, 2) the intervention program had to be aimed at reducing recidivism, 3) interventions had to include therapeutic measures and did not need to be specifically designed for offenses of a sexual nature, 4) studies could not have offered only purely deterrent or purely punishing treatment programs, 5) recidivism had to be included as a dependent variable, 6) each study had to report the same recidivism outcomes for the comparison group who were not receiving the same treatment, 7) the studies had to have a sample size of at least 10 with a minimum of 5 persons placed in each group, 8) data had to be reported that allowed for the calculation of effect sizes, 9) studies had to be either in Dutch, English, French, German, or Swedish, and, finally, 10) studies were not limited by report date and all studies were included until 2003.

The authors conducted their literatures search by compiling references from previous meta-analysis reviews and individual studies, using electronic databases (such as PsycINFO and MEDLINE), hand-searching journals directly related to the topic, contacting researchers in the sex-offender treatment field for additional suggestions or for unpublished reports, and using a basic Internet search of relevant institutions and corrections departments for pertinent materials. This method resulted in the identification of 2,039 possible citations. Further analysis resulted in a final 66 reports meeting the authors criteria. Several of the reports contained more than one eligible study, so the authors used each eligible individual study as a separate unit of analysis. Other studies presented results of different subgroups, such as type of offense. The authors chose to use those subgroups as individual units of analysis as well. This resulted in 80 eligible comparisons from a total of 69 studies.

An adapted version of the Maryland Scale of Scientific Rigor was used to evaluate overall methodological quality of the 69 individual studies. Only studies that received a rating of 2 or higher on the scale were included in the analysis. A large majority of the studies (69.5 percent) included in the meta-analysis came from North America (Canada and the United States). About half of the studies were published in peer-reviewed journals, and the other half were book chapters or unpublished studies. Treatment material differed greatly, with half using cognitive–behavioral therapy (CBT) programs and 14 comparisons using physical therapy programs (8 of these were surgical castration). Sample size varied from 15 to 2,557, with about one third of comparisons containing fewer than 50 individuals. Almost all of the studies had a follow-up period longer than 5 years.

Participants in the studies were mostly adults (56.3 percent adult versus 17.5 percent juveniles, while 10 percent included mixed age groups, and 25 percent had unclear age groups). Participation in the treatment program was usually voluntary (46.3 percent compared to 20 percent nonvoluntary, while 10 percent had mixed participation, and 23.8 percent had unclear participation). Child molestation (73.7 percent), rape (55.0 percent), and incest (47.5 percent) were the most common offense types (individual comparisons may cover multiple categories of offenses).

The meta-analysis used odds ratios (OR) to measure effect sizes. For studies that did not report statistics that could be easily transformed into OR, the authors used standard procedures to calculate Cohen’s d and then used this statistics to calculate OR.

Hanson and colleagues (2009) conducted a meta-analysis on whether principles associated with effective interventions for those convicted of general offenses (risk–need–responsivity) would also apply to psychological treatments of people convicted of sex offenses. The basis for this research was formed from earlier data that shows that the most effective types of programs for the general group are those that follow the risk–need–responsivity (RNR) model. Treatments using the RNR methodology follow 3 main tenants: 1) treatment is more likely to be effective for those who are at a moderate or higher risk of offending, 2) target the characteristics that are most closely related to the offending behaviors, also known as criminogenic needs, and 3) match the treatment with the their learning styles and abilities.

The authors selected studies by conducting searches of databases, such as PsychINFO, Web of Science, Digital Dissertations, and the National Criminal Justice Reference Service. They found additional materials using the reference lists of collected articles, program materials from conferences, and a general review of articles on the topic. To be included in the meta-analysis, studies had to examine treatment effectiveness by comparing recidivism rates using a sex-offender population with a matching comparison group of sex offenders. The authors defined sex offenders as “offenders with sexually motivated offenses against an identifiable victim” (Hanson et al. 2009, 868). Participants in the control/comparison group could have received an alternate treatment, less treatment, or no treatment. For the studies to meet the need principle of the RNR model, at least 51 percent of the treatment had to target criminogenic needs, such as antisocial lifestyle, impulsivity, or negative peer associations. Services in the treatment program met the responsivity aspect of RNR when the treatment was provided in such a way as to match the needs and learning style of the client.

The number of eligible studies was narrowed down using the Collaborative Outcome Data Committee (CODC) guidelines, which help in determining the extent to which a study’s features indicate possible bias when estimating treatment effect. Only studies categorized as weak, good, and strong were included. This resulted in a total of 23 studies included in the analysis. Of these, 14 were published and 9 were unpublished. Most of the studies (74 percent) were from North America (Canada and the United States). Of the 14 studies, 10 included adult males while 4 studies focused exclusively on juveniles. In addition, 3 studies indicated the inclusion of some females in the sample). The sample sizes ranged from 16 to 2,557. In 10 of the studies the treatment programs were offered in institutions, and in 11 studies they were offered in the community (in two studies the treatment was offered in both settings).

Statistics were calculated using both a fixed effect and random effect models and by calculating odds ratio.