Abstract
Clinicians have at their disposal a variety of instruments with which to evaluate the risk presented by sex offenders. These measures yield similar predictive potency, and combining them does not appear to enhance prediction. The present study examined whether classification and regression tree analysis could identify new combinations of predictors that would improve predictive validity. Items from seven actuarial instruments were used to construct classification trees. Overall, classification trees achieved slightly higher predictive accuracy than did actuarial instruments. In addition, these analyses highlight the heterogeneity among sex offenders. Despite this improvement, one should consider the incorporation of other predictors into the instruments—including dynamic factors, protective factors, and measures with strong theoretical justification.