Prediction: an Overview of Selected Methodological Issues (From Prediction and Classification: Criminal Justice Decision Making, P 21-51, 1987, Don M. Gottfredson and Michael Tonry, eds. -- See NCJ-116250)

• NCJ Number: 116251
• Author(s): S D Gottfredson
• Date Published: 1987
• Length: 31 pages
• Sale Source: University of Chicago Press
• Publication Type: Issue Overview
• Language: English
• Country: United States of America
• Note: *This document is currently unavailable from NCJRS.

Annotation

This survey of statistical methods available for making criminal justice predictions concludes that the present ability to predict either decisions of criminal justice functionaries or offender behavior is modest.

Abstract

Factors affecting the accuracy of predictions, regardless of method, are discussed: measurement reliability, base rates, selection ratios, sample representativeness, and cross validation. Research suggests that predictions of decisions are more accurate than are predictions of behavior. In addition, statistical predictions outperform intuitive or clinical predictions. The following methods of prediction are described: multiple regression, the Burgess method, clustering models, and multidimensional contingency table analysis. The article concludes that predictive accuracy can be enhanced by improving data reliability and measurement techniques, by employing both statistical and subjective bootstrapping models, and by using theory-driven approaches. Finally, investigators should pay more attention to ethical considerations. Approximately 70 references. (Author abstract modified)