- Full text keyword search
- What’s the Real Rate of Sex-Crime Recidivism?
- Sex Offender Risk, Recidivism, and Policy
- Recidivism Rates Among Child Molesters and Rapists: A Methodological Analysis
Table 1 shows the number of offenders in the four risk bands for the yet unmatched sample. The untreated offenders were underrepresented in the lowest risk band and overrepresented in the two high-risk bands.
- videos on sex offenders from msnbc.
- what shows onf fingerprint background check.
- Related eJournals!
Table 1. The matching was based on the sum scores ranging from 0 to 9. Because the number of individuals in our treatment group exceeded the number of untreated participants, we chose weighting over a matching procedure to preserve the sample size for an overview on matching methods, see Stuart, This also results in a fully matched sample regarding the matching variable but avoids dropping cases when there are not enough matches and having to choose individual cases over others when there are more matches available, respectively.
Nonetheless, we had to exclude one treated participant with a score of 9 because none of the untreated offenders had an equivalent score. The weights were calculated separately for controls within each risk score group. Each case in the CG received a frequency weight reflecting the number of treated individuals with the same Static score divided by the number of untreated individuals with the same risk score.
For example, in the initial sample, there were 23 individuals in the TG and 25 individuals in the CG with a Static sum score of 5. Thus, each of the 25 individuals in the CG received a weight of 0. Individuals in the TG received a weight of 1. This resulted in a total number of cases and an equal number in each group in the weighted sample. Propensity score methods have gained much popularity in evaluation research as they allow to control for a large number of possible confounders. Originally proposed by Rosenbaum and Rubin , the propensity score is defined as the probability of being assigned to a treatment group, given a certain set of pretreatment covariates.
In randomized experiments, the true propensity score is predetermined by the study design, with each participant having a propensity score of 0. In quasi-experiments, propensity scores can be calculated via logistic regression given a set of pretreatment characteristics confounders. In the regression model, treatment assignment is the dependent variable dummy coded 0, 1 , and potential confounders are the predictors Austin, For balancing the covariates in the TG and the CG, different techniques, such as matching or weighting can be applied Austin, Matching on the propensity score requires a large number of participants especially in the CG.
The first step in specifying the propensity score model is the selection of relevant covariates. In general, relevant covariates are variables that affect the treatment assignment as well as the outcome. As there is a lack of empirical evidence concerning the treatment selection process, all outcome-related variables potential confounders; see Austin, were considered as covariates in our model.
Four variables could not be included due to missing or unreliable information, especially in the CG diagnosis of a personality disorder, diagnosis of paraphilia, having committed an offense under the influence of drugs, and psychopathy.
Finally, a total number of 37 variables see Table A1 in the Supplemental Material were selected as covariates and ordinal and categorical variables were dummy coded for subsequent analyses. Only Thus, in a complete case analysis, one third of the original sample would have been lost for further analyses.
As the overall proportion of missing data was only 2. Propensity scores were then calculated in SPSS via logistic regression as explained above. Participants with propensity scores outside the area of common support i. Referring to Xu et al. As shown in Figure 1 , the distributions of propensity scores showed a substantial overlap between the TG and the CG in the original sample area of common support. Nonetheless, they were different, indicating imbalance in the measured covariates. In the unweighted sample, 33 In the weighted sample, balance has improved substantially with only two 3.
Full text keyword search
Figure 1. Propensity score distributions in the original sample and the weighted sample by treatment assignment. The pattern of findings was somewhat different for the various outcome criteria. Figure 2. Recidivism rates in the treatment group and the control group matched on the Static sum score. The balance diagnostics indicated that all relevant confounders were sufficiently balanced in the TG and the CG after weighting.itlauto.com/wp-includes/apps/3835-localiser-telephone.php
What’s the Real Rate of Sex-Crime Recidivism?
No difference was statistically significant all p s larger than. Figure 3. Recidivism rates in the treatment group and the control group matched via propensity score weighting. The aim of this study was a comparison of two widely used matching methods in the evaluation of sex offender treatment. This was triggered by a recent politically very influential study in Britain that used PSM as the best design under given practical circumstances Mews et al. In the meanwhile, the U. MoJ decided to prioritize PSM as the most feasible and adequate design for evaluations of offending behavior programs.
Within this context, we first discuss the specific findings in our study and then address broader issues of sex offender treatment, research, and policy. In our study, the more traditional risk-based matching and PSM showed mainly similar, but also partially different, results on recidivism. In both analyses, there were no significant effects of treatment in prison on the rate of sexual reoffending.
Sex Offender Risk, Recidivism, and Policy
Some of the latter studies had smaller samples and applied matching methods that may have impaired equivalence of TG and CG. However, there were also studies with relatively large samples and sophisticated PSM methods e. Some results were nonetheless encouraging as they showed less harmful or delayed sexual reoffending of treated offenders Olver et al.
The latter may lead to less repeat recidivism according to the age—crime curve. In our ongoing project, we are gathering such differentiated data and will apply a more complex metric harm index on a much larger sample for which we just received recidivism data. Although there is widespread agreement among researchers about the need of more differentiated outcome measures, they need to acknowledge political reality.
As the British evaluation of Mews et al. We cannot exclude the possibility of similar events in the future when the new programs will be thoroughly evaluated.
Recidivism Rates Among Child Molesters and Rapists: A Methodological Analysis
Our above findings did not reveal significant prison-based treatment effects on sexual reoffending, but the negative trend in the Static matching analysis requires further consideration. The difference of 1. In our smaller study, the difference may have been due to chance, but it is not negligible when the low base rate in the CG is taken into account.
However, the absolute numbers of recidivism in the TG and CG were only 25 versus Accordingly, very few cases influenced the whole picture, what is similar in other studies on sexual offender treatment. Researchers are familiar with these issues, but the recent experience in Britain suggests that we should inform policy makers, practice and broader audiences about these details. Perhaps plain language documents could help to avoid potential misinterpretations of small and nonsignificant differences between TGs and CGs and underline the need for large and sound experiments.
Our study also showed a somewhat different trend on sexual reoffending in the PSM analysis. In contrast to Mews et al. This small and nonsignificant difference should not trigger far-reaching speculation. However, the different trend in comparison with the Static analysis suggests that the more sophisticated PSM approach does not necessarily lead to negative results. Therefore, we emphasize the urgent need of replications not only across different studies Farrington et al. Policy and practice should be more informed about the fact that evidence comes by replication, and single results need to be embedded in a broader framework.
Another important message from our study is the consistency between both analyses on the other criteria of recidivism. Both analyses on violent recidivism showed also lower rates in the TG. The differences were 2. With regard to the criterion of severe recidivism, there was also consistency between the two analyses. The Static matching showed a nonsignificant difference of 2. Of course, we do not wish to over-interpret nonsignificant trends, but evaluations of sex offender treatment should be realistic about small absolute differences and low statistical power in clinical studies. Over decades, thresholds of statistical significance have been discussed controversially e.
Basically, statistical significance is most appropriate in testing a specific theoretical hypothesis, and in these cases, a one-sided approach is often appropriate. However, in applied fields such as sex offender treatment, significance testing became a rarely reflected routine that did not always fit to realistic outcome expectations and questions of practical significance e. Therefore, more homogeneous results of meta-analytic integrations are very valuable. Comparing our results with other studies is limited by differences such as treatment parameters e. Furthermore, methodological limitations of our study have to be taken into account when interpreting the results.
One limitation relates to the variables that we could use for the PSM. These were assessed by practitioners in the daily practice in prisons. As a reliability study on interrater agreement on the basis of prison files showed, not all of these variables seem to be sufficiently reliable and as a consequence more valid; Haas We do not exactly know how individual expertise may have influenced assessments beyond the respective instructions.
This is a general matter of evaluations of routine practice in contrast to closely monitored model projects , but there is space for improvement by principles of implementation science e. Another limitation is our inclusion of three types of treatment to increase sample size. These approaches varied in content and intensity, but there was also overlap and their recidivism outcomes were similar when the risk level was controlled.
It needs also to be mentioned that the offenders in our sample were released in at the latest. Although a long follow-up period is particularly important in sex offender treatment evaluations, this implies that the treatment content may have improved in the meanwhile. We have not yet data on this issue, but we will test it in our currently enlarged data set.