NCJ Number
209656
Journal
Criminal Justice and Behavior Volume: 32 Issue: 2 Dated: April 2005 Pages: 223-247
Date Published
April 2005
Length
25 pages
Annotation
The purpose of this study was to determine whether an empirical modeling approach could be applied to the problem of classifying relative levels of recidivism risk in a population of released offenders from the Wisconsin Department of Corrections.
Abstract
A cornerstone of the Wisconsin Department of Corrections’ (WIDOC) approach to risk assessment is the Client Management Classification (CMC) Report (Lauen, 1997) and its predecessor, the Case Classification Staff Deployment (CCSD) project (Baird et al. 1974). This study looked at whether an empirical modeling approach could use offenders’ CCSD data to determine which offenders posed a lower or higher risk for committing new crimes upon release. Examining records for all admissions dating from approximately November 1969 to October 2001, a sample of 5,941 cases was found that contained all the relevant information necessary for modeling. The sample included 5,357 males and 584 females. Each offender was linked to two pieces of data: a CCSD and a crime report in the admission records. The sample was divided into 3 classes: a no-readmission group (n=3,267); a readmitted-without-a-new-conviction group (n=1,718); and a readmitted-with-a-new-conviction group (n=956). To designate the classes, exemplar-based modeling was used to generate a partially refined transitional exemplar library. Through the transitional modeling process, cleaner exemplars were selected for the permanent reference library. Thus, instead of using the data pattern library to model the number of admissions, the data pattern library was used to model risk class, generating a continuum of risk ranging from 1.00 (lower-risk exemplar group) to 2.00 (higher-risk exemplar group). Analyses of the data found that modeled-risk values provided a statistical means to determine the degree of match between a release candidate, with unknown recidivism prospects, and either of two risk level groups with known outcome histories. Issues related to extracting relatively pure classes of exemplars from relatively ambiguous data are detailed and implications for controlling for risk factor patterns are discussed. References, tables, figures, and appendixes