Proefschrift

2 55 PREDICTIVE PROPERTIES DYNAMIC INSTRUMENTS Table 2.3. Effect sizes for research question 3: Predictive validity of change scores derived from dynamic risk assessment instruments (van den Berg et al., 2018, p.185) Recidivism Type Fixed-Effect Random-Effect Q I2 (%) N K L Cohen’s d 95% CI Cohen’s d 95% CI Sexual 0.26 [.10, .42] 0.27 [.10, .44] 5.8 13.8 2,043 9 6 Violent 0.14 [.05, .24] 0.15 [-.00, .29] 10.54 52.6 6,234 8 6 Any 0.10 [-.01, .22] 0.10 [-.01, .22] 1.04 <0.00 1,466 6 4 K = Number of studies; L = Number of unique samples 2.4.4.1 Predictive validity of controlled change scores for sexual recidivism For the analysis of sexual recidivism, the fixed-effect weights of the individual studies (the inverse of the variance) varied between 976.56 and 19.75, with a median value of 257.56. The study producing the largest weight had a sample size of 481 (recidivism rate was 3.95%) and had more than twice the weight of the next largest study (349.41) and 49 times more weight than the smallest study weight. To reduce the influence of these studies, the variance was artificially increased to ensure that these study weights would not exceed the next largest study weight by more than twice its size. The fixed-effect weighted hazard ratio on sexual recidivism for the corrected change score on the dynamic risk assessment instruments was 0.91, 95% CI [0.87, 0.95], N = 1,980 based on 6 unique samples. Thus, changes on the dynamic factors added to the prediction of sexual recidivism above and beyond static and initial dynamic risk scores. The Q indicated that the variability across studies was not significant (Q = 2.4, p = .79). The I2 (<0.00%) indicated that the variability among studies that could not be explained by chance was low. Not enough information was available to run moderator analyses. 2.4.4.2 Predictive validity of controlled change scores for violent (including sexual) recidivism For the analysis of violent (including sexual) recidivism the fixed-effect weights of the individual studies (the inverse of the variance) varied between 1,736.11 and 198.37 with a median value of 625.00. The study producing the largest weight came from an effect with sample size of 414 (recidivism rate of 8.21%) and had more than twice the weight of the next largest study (625.00) and more than 8 times more weight than the smallest study weight. To reduce the influence of this study, its variance was artificially increased so that its study weight did not exceed the next largest study weight by factor 2. The fixed-effect weighted hazard ratio on violent (including sexual) recidivism for the corrected change score on the dynamic risk assessment instruments was 0.93, 95% CI [0.90, 0.97], N = 4,168, based on 5 unique samples. Thus, changes on dynamic risk factors added to the prediction of violent (including sexual) recidivism above and beyond static and initial dynamic scores. The Q indicated that the variability across

RkJQdWJsaXNoZXIy MjY0ODMw