Forecasting Dangerous Inmate Misconduct: An Application of Ensemble Statistical Procedures

• NCJ Number: 214958
• Journal: Journal of Quantitative Criminology Volume: 22 Issue: 2 Dated: June 2006 Pages: 131-145
• Author(s): Richard A. Berk; Brian Kriegler; Jong-Ho Baek
• Date Published: June 2006
• Length: 15 pages
• Grant Number(s): SES-0437169
• Sponsoring Agency: National Science Foundation
• Publication Format: Article
• Publication Type: Test/Measurement
• Language: English
• Country: United States of America

Annotation

This study used data from the California Department of Corrections to examine how "ensemble" statistical procedures (procedures that use a variety of factors related to various behaviors and offender characteristics) c be applied to identify inmates likely to engage in very serious misconduct (a major felony) while incarcerated.

Abstract

The classification procedure developed identified inmates at high risk for serious misconduct as young offenders with long criminal records, active participation in street and prison gangs, and sentenced to long prison terms. Sentence length was most predictive of serious misconduct when it was between 5 and 10 years. Criminal record had its greatest predictive impact for inmates first arrested at a very young age. Very young inmates with long criminal histories who have been active in gang activities and are serving sentences of more than 10 years will almost certainly engage in serious misconduct while in prison. This might include drug trafficking, assault, rape, and attempted murder. Although less than 3 percent of the inmates studied over 2 years were reported for very serious misconduct, these inmates were correctly identified for such behavior about half the time. Data were drawn from a completed randomized trial testing of the California Department of Corrections inmate classification system. Specifically, this study focused on the 9,662 male inmates assigned to placement under the revised classification system that included an updated list of risk factors and new set of weights for the items that were to be combined into each inmate's classification score. In addition, this sample was assessed with "random forests," a promising approach that constructs an ensemble of classification "trees." Each "tree" is built from a sample of the data, and a random sample of predictors is examined. Classification is determined by a majority vote for each case over the ensemble of classification "trees." 5 tables, 3 figures, and 23 references