Improved Techniques for Assessing the Accuracy of Recidivism Prediction Scales: A User's Guide to the Machine- Readable Files and Documentation and Codebook

• NCJ Number: 146228
• Author(s): J Cohen; S Zimmerman; S King
• Date Published: 1992
• Length: 236 pages
• Grant Number(s): 86-IJ-CX-0039
• Agencies: NIJ-Sponsored
• Publication Type: Measurement/Evaluation Device

Annotation

These five volumes present background and coding information for a data set from a study designed to measure the validity or accuracy of four instruments that predict criminal behavior by using a method, the Relative Improvement Over Chance (RIOC), which overcomes some limitations posed by other traditionally used validity measures.

Abstract

The four predictive instruments include the INSLAW, RAND, SFS81, and CGR scales. By using RIOC as a tool for measuring scale accuracy, the investigators also sought to examine the RIOC measure as an indicator of the accuracy of criminal behavior predictive instruments. In 1990, the researchers used longitudinal data from five existing, independent studies to assess the validity of the four predictive measures. Each data file is composed of its own set of samples and represents a different geographical area and distinct stage in the criminal justice processing system. Each file contains criminal records and demographic information regarding individual offenders. For all five files, original data were collected in the 1980's. The samples for the data files include 771, 1,022, 177, 1,602, and 908 offenders respectively. The data from these files were recoded when necessary and applied to each of the four predictive scales to produce individual scores for all the samples. Variables for each of the five files include a selection of the original variables, recoded variables used to construct the predictive scales, final scores for these scales, and follow-up data on subsequent offending. Tables and coding information