Michigan Risk Prediction - A Replication Study - Final Report

• NCJ Number: 79872
• Author(s): T H Murphy
• Date Published: 1981
• Length: 37 pages
• Grant Number(s): AP-0
• Sponsoring Agency: National Institute of Corrections
• Corporate Author: Michigan Dept of Corrections
• Sale Source: National Institute of Justice/
• Publication Type: Statistics
• Language: English
• Country: United States of America

Annotation

Based on a random sample selected from all inmates in Michigan paroled in 1974, this study used statistical methods applied in a 1971 project to predict recidivism and reached similar conclusions, such as juvenile arrest and commitment being significant indicators of future criminal activity.

Abstract

The original risk study conducted on 2,200 inmates randomly chosen from all those paroled in Michigan during 1971 collected data on 360 variables per individual and developed the risk groups through the Automatic Interaction Detection Program developed at the University of Michigan. This analysis identified the following groups: high risk, reported felonies while juvenile and major misconduct or reported felony while juvenile and no major misconduct and age of first arrest before 15 years; middle risk, reported felony while juvenile and no major misconduct and first arrest after age 15 or no reported felony while juvenile and drug problem at the time of offense; and low risk, no reported felony while juvenile and no drug problem at the time of offense. Recidivism rates were 37.9 percent for the high group, 27 percent for the middle, and 15.1 percent for the low risk group. A discussion of both studies' methodology emphasizes that large samples were used, selection was random, a 3-year follow-up period was adopted, and criminal behavior rather than judicial or administrative decisions which can affect parole was measured. Procedures developed to establish reliability of the coding are also detailed. The replication study was based on a random sample of 30 percent of 1,200 individuals selected from all inmates paroled in 1974. Data were collected on 90 variables per individual. The analysis created a computer program to duplicate the subgroups of the first study, addressed the replication issues, and examined the impact of selected control variables on parole outcome. Comparisons showed the two samples to be similar with respect to most variables examined, and no differences were found regarding those variables which were predictive of future criminal activity, such as age of first arrest and prior prison commitment. Significant differences were discovered between years in the very low and middle risk groups for violence felony behavior, possibly because of the 1974 recession and nationwide increases in the violent crime rate. Tables and 13 references are provided.