What Can We Reasonably Expect to Accomplish in Conducting Actuarial Risk Assessments with Sexual Offenders in Civil Commitment Settings? A Response to Campbell and DeClue: “Maximizing Predictive Accuracy in Sexually Violent Predator Evaluations”

Measuring risk to recidivate in populations of sexual offenders has always been a contentious and complicated process. Whereas predictions of reoffense risk were once offered by clinicians using “unstructured clinical judgment,” the research literature has clearly shown that the accuracy of such methods often fails to meet even chance levels. Consequently, the past 20 years or more has seen the development and refinement of tools designed to actuarially predict risk for recidivism. One such tool, the Static-99, has become the index of choice for most evaluators of sexual recidivism risk, to the extent that it is likely the most widely researched and utilized measure of its type in the world. However, the Static-99 is not without its foibles: It possesses only moderate predictive accuracy, omits such robust predictors as direct measures of sexual deviancy and antisociality, and does not include any dynamic factors. Further, it appears that heterogeneity in the standardization samples presents issues for evaluators and jurists alike in terms of how best to evaluate individual offenders against putative normative groups. In their paper “Maximizing Predictive Accuracy in Sexually Violent Predator Evaluations,” Campbell and DeClue argue that the predictive value of the Static-99 and its related measures is suspect. These arguments center on issues related to cut-off scores, confidence intervals, and assignment of individual offenders to specific normative samples based on factors not included in the Static-99. In this paper, we provide a response to some of the issues raised by Campbell and DeClue, specifically focusing on forcing dichotomous status on continuous variables, a failure to see the big picture in using Static-99 to predict risk in populations of Sexually Violent Predators, and difficulties associated with creating confidence intervals