NCJ Number
196859
Journal
Criminal Justice and Behavior Volume: 29 Issue: 5 Dated: October 2002 Pages: 538-568
Editor(s)
Kirk Heilbrun
Date Published
October 2002
Length
31 pages
Annotation
This article discusses the operationalization of a multiple models approach in predicting recidivism risk and its implications in judicial decision-making. It examines whether a newly developed actuarial approach for constructing risk assessment tools, the multiple models approach, could be used to develop a more accurate tool for predicting the future criminal behavior of offenders.
Abstract
Using a large recidivism data set, this study developed and validated a multiple models tool for predicting recidivism risk. A series of classification-tree-based actuarial models for predicting criminal recidivism were developed and showed how combining the predictions from these models enhances the ability to classify cases into groups that vary along a spectrum of risk. The study examined the robustness of the multiple models approach by testing it on a series of cross-validation samples. Using the large recidivism data set found that combining the predictions of multiple classification-tree models enhanced the ability to predict criminal recidivism over the traditional single-model approach. However, judges seldom use actuarial prediction tools. Instead, they rely on perceptual shorthand or intuition to predict the future behavior of offenders. Careful consideration must be given to the procedural question of how judges are to incorporate actuarial tools into their assessments of future criminal behavior. Should the actuarial tool supplant judicial estimates of risk or be used to support judicial risk assessments? As long as judges remain concerned with protecting the public from future criminal harm, assessing recidivism risk will remain a core judicial activity. As long as risk assessment remains a primary judicial activity, efforts to incorporate actuarial tools into the judicial risk assessment process must be pursued. References