NCJ Number
240942
Journal
Criminal Justice and Behavior Volume: 39 Issue: 12 Dated: December 2012 Pages: 1647-1667
Date Published
December 2012
Length
21 pages
Annotation
This study examined whether classification and regression tree analysis could identify new combinations of predictors that would improve predictive validity.
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 instrumentsincluding dynamic factors, protective factors, and measures with strong theoretical justification. Abstract published by arrangement with Sage Journals.