Maximizing Predictive Accuracy in Sexually Violent Predator Evaluations

The Static-99 and related procedures (Static-2002, Static-2002R, and Static- 99R) are the most frequently used actuarial instruments for assessing sexual-recidivism risk. We analyze the original 15-year recidivism data for the Static-99 and all of the subsequent “norms” included in datasets from www.static99.org3 and published articles. We use frequency tables to construct traditional 2 X 2 contingency tables for the datasets. We report and analyze corrected values for the Static-2002 dataset that was incorrectly presented in Hanson, Helmus, and Thornton (2010). We use traditional test utilities to seek optimal cutoff scores in order to maximize overall accuracy. We provide summary tables showing optimal cutoff scores for all four instruments. We provide an example that illustrates how an evaluator could use traditional methods for classification and prediction tests to report positive predictive value (PPV), with associated confidence intervals, as recommended by Heilbrun, Douglas, and Yasuhara (2009)