U.S. flag

An official website of the United States government, Department of Justice.

Statistical Approaches to Assessing Risk

NCJ Number
193444
Date Published
January 2002
Length
2 pages
Annotation
This report summarizes the contents and objectives of a recent OJJDP (Office of Juvenile Justice and Delinquency Prevention) report that provides technicians with a statistical guide for preparing risk scales for practical application in juvenile courts.
Abstract
Risk scales are statistical tools that help decision makers classify juveniles on the basis of expected behavior. After a set of predictive factors is identified, either through examining research or by tapping the expertise of experienced professionals, technical staff must decide how to combine data into a score or risk category that predicts future behavior. The new OJJDP report, "Risk Classification: A Comparison of Methods for Practical Application in Juvenile Courts," details and compares the processes for implementing a number of the most commonly used statistical approaches in risk scaling. This report presents step-by-step procedures for developing a risk scale, using the following statistical methods: a simple method without different weightings of the predictive items, multiple linear regression, discriminant analysis, logistic regression, and predictive attribute analysis. The authors constructed a data set from the automated records housed in the National Juvenile Court Data Archive that held risk and criterion (outcome) information commonly available in most jurisdictions. The authors explain how they applied each statistical method to a random portion of this data set to produce an instrument that divided the referred juveniles into five recidivism risk categories. The predictive capabilities of the risk scales were found to be about the same. The relative benefits of the scales are thus compared according to simplicity, the ease of explaining the statistics to a nontechnical audience, and the ability to control for unwanted influences on the resulting prediction.
Date Published: January 1, 2002