Evaluation of risk assessment instruments for sex offenders

In order to evaluate risk assessment instruments for sex offenders in Germany, we compared the predictive validity of the Static99, HCR-20, SVR-20, and PCL-R scales for 134 sex offenders. The mean follow-up time was 9 years (range 1-340 months), using the first entry into the National Register of Criminal Convictions as endpoint variable. For the estimate of predictive power, the area under the curve (AUC) of receiver operating characteristic (ROC) analysis was calculated. The AUC plots accurately identified violent or sexual recidivists and "false positives" at all scale levels. Comparing the predictive validity of these four instruments, the results favored Static-99. As for the limited sample size, differences between the assessment instruments were, however, not statistically significant. The ROC analysis for Static-99 showed that including treatment dropouts does not improve predictive accuracy (including dropouts: AUC 0.710; excluding dropouts: AUC 0.721). Kaplan-Meier survival analyses yielded highly a significant correlation to recidivism time point for two Static-99 and SVR-20 risk categories. Higher-risk categories were related to earlier recidivism. However, relying on the Static-99 and SVR-20 alone showed false positive results: for up to two out of three sex offenders, they predicted recidivism which did not occur.