2 49 PREDICTIVE PROPERTIES DYNAMIC INSTRUMENTS (2013), the variance of studies with the largest weights were adjusted so their weights would not exceed the next largest weight by factor 2. 2.3.5.4 Moderator analyses Moderator analyses were performed, after excluding studies that were considered outliers, to assess the influence of method variations and sample characteristics. To be qualified as a moderator, a sufficient number of studies with information on that moderator needed to be available (at least three studies in each category for categorical moderators, at least 10 studies for continuous moderators), and they should be meaningfully related to theoretical models or prior research findings. The following categorical moderators were analyzed by means of a fixed-effect test of the Qbetween, measuring the variability accounted for the levels of the moderator; publication (published study/unpublished study), development (instrument developed with study sample/replication in new sample), design (prospective study/retrospective study), blind (scored blind to recidivism outcome/unclear if scored blind to recidivism outcome), location type (samples from institutions/community/mixed samples), setting (corrections in prison or community/treatment in prison or community/treatment in psychiatric setting/mixed group), and offender type (child molesters and child porn/ rapists/mixed group). The following continuous moderators were analyzed by means of meta-regression: sample size, length of follow-up, and year of publication. If multiple publications were based on (partly) overlapping samples, the first publication date was used for the moderator analyses. Appendix B, available in the online supplemental materials, provides a detailed description of the included samples. 2.4 RESULTS 2.4.1 OVERALL PREDICTIVE PROPERTIES The average follow-up time for studies predicting sexual recidivism was 67.1 month (range = 18–240), for violent (including sexual) recidivism 53.7 months (range = 36–240) and for any recidivism (sexual, violent or any) 75.5 months (range = 9–240). Observed recidivism rates were 16.9% for sexual recidivism (23 samples, n = 5,877), 11.8% for violent (including sexual) recidivism (14 samples, n = 10,368), and 33.3% for any recidivism (10 samples, n = 3,405). Table 2.1 summarizes the effect sizes of the predictive properties of dynamic risk assessment instruments developed for adult male sex offenders on the three types of recidivism. Effect sizes of the individual samples are reported in Appendix B1 (available in the online supplemental materials). 2.4.1.1 Predictive properties for sexual recidivism After excluding one outlier (Allan & Dawson, 2002), the fixed-effect weighted Cohen’s
RkJQdWJsaXNoZXIy MjY0ODMw