Mathematics of Risk Classification: Changing Data into Valid Instruments for Juvenile Courts

• NCJ Number: 209158
• Author(s): Don M. Gottfredson; Howard N. Synyder
• Date Published: July 2005
• Length: 44 pages
• Agencies: OJJDP-Sponsored
• Publication Type: Measurement/Evaluation Device

Annotation

This report presents the strengths and weaknesses of various prediction methods used for the risk-screening of juvenile offenders by juvenile courts.

Abstract

Juvenile courts rely on risk assessment instruments to make sentencing decisions about juvenile offenders. The prediction methods used to develop these risk assessment instruments have different utilities and purposes. In this report, the various statistical prediction methods used in juvenile risk-assessment instruments are compared by assessing the validity of the simple classification instruments. Specifically, the report compares the validity of three different types of statistical methods that are most commonly used for selecting and combining variables to assess risk. Data include the records of 9,476 youths who were referred to the Juvenile Court in Arizona during 1990. The steps followed in the development and use of risk assessment measures are identified and the two requirements for any predictor, which are discrimination and reliability, are discussed. Methods of combining predictors, including linear additive models, clustering methods, and bootstrap methods are considered. All of the methods assessed showed a useful degree of validity when they were applied to the sample of youth; validity measures were approximately equal for all of the methods tested. Two methods are recommended for their face validity, simplicity: the Burgess-type scale and the prediction attribute analysis method. Courts are advised to choose the method that best fits their courtroom operations and the population they serve. Tables, figures, footnotes