NCJ Number
196845
Journal
Journal of Quantitative Criminology Volume: 18 Issue: 3 Dated: September 2002 Pages: 267-296
Date Published
September 2002
Length
30 pages
Annotation
Multiple-item measures of self-reported offending usually provide the primary outcome measures for individual-level research on the causes of crime and deviance; this article examines the problems presented by the forming of composite scores for these measures and presents a possible solution to these problems.
Abstract
The authors argue that the scaling problems presented by multiple-item measures of self-reported offending and deviance arise because measures of crime and deviance systematically violate the assumption of equal intervals of measurement, both across items (in relation to offense seriousness) and within items (for higher versus lower frequencies of offending). In order to address these scaling problems, the authors propose scaling by means of the graded response model from item response theory (IRT) as a potential means of overcoming the shortcomings of traditional summative scaling and of obtaining useful information about the strengths and weaknesses of the measures. IRT scaling, in the specific form of the graded response model, provides a rational, empirical basis for relating all items and response levels to a shared, interval-level metric. The item parameter estimates support this contention with their coherent portrayal of the relative seriousness of various types of offenses and frequencies of offending. This article illustrates the proposed IRT method through a scale analysis of a 14-item, self-report measure of delinquency, using 3 years of data from the Monitoring the Future Study, which is an annual national survey of high school seniors. The graded response model was found to be consistent with the data, providing results that addressed important substantive questions about self-report measures. 5 figures, 1 table, and 65 references