Exploring the Effects of Changes in Design on the Analytical Uses of the NCVS Data

• NCJ Number: 211568
• Journal: Journal of Quantitative Criminology Volume: 21 Issue: 3 Dated: September 2005 Pages: 293-319
• Author(s): David Cantor; James P. Lynch
• Date Published: September 2005
• Length: 27 pages
• Publication Format: Article
• Publication Type: Report (Study/Research)
• Language: English
• Country: United States of America

Annotation

This paper examines the effects of the 1992 redesign of the National Crime Victimization Survey (NCVS) across population subgroups that have been found to be important in previous analyses of the NCVS.

Abstract

The authors discuss three changes to the NCVS that were expected to improve the completeness of the survey's crime estimates. The first change was a revision of the screening interview used to obtain information on potential crime events. The new design used a "short-cue screener" logic that provided a much larger number of cues to the respondent. The second major change was the use of computer assisted telephone interviewing (CATI). Prior to 1992, the majority of interviews were conducted by telephone, but not in centralized facilities, and they were not computer-assisted. The third change was the modification of the procedures for recording series incidents. This made it more difficult to treat multiple victimizations as a series incident. The analysis of the impact of these changes focused on whether they introduced artifacts into the data that will result in distortions in substantive models of victimization risk. The authors estimated and examined the effects of the design changes on substantive, multivariate models of victimization risk. The analysis suggests that although there were increases in reporting between the old and new designs, they did not substantially change the effects observed across subgroups. Multivariate models of victimization risk were not significantly influenced by the design change. In most cases, coefficients retained their direction and magnitude. Where there were significant interactions between respondent attributes and design, the effect reduced the slope estimated for particular independent variables. Future research should examine the implications of the direction of the interactions that remained unexplained. 3 tables and 29 references