Evaluation of Direct Variance Estimation, Estimate Reliability, and Confidence Intervals for the National Crime Victimization Survey

• NCJ Number: 249242
• Date Published: December 2015
• Author(s): Rick Williams; David Heller; Lance Couzens; Bonnie Shook-Sa; Marcus Berzofsky; Hope Smiley-McDonald; Chris Krebs
• Length: 55 pages
• Agencies: BJS-Sponsored
• Publication Type: Research (Applied/Empirical); Report (Study/Research); Report (Grant Sponsored)

Annotation

This study examined the feasibility of using direct variance estimation (DVE) on data obtained by the National Crime Victimization Survey (NCVS), and it offers recommendations regarding the calculation of confidence intervals (CIs) and the identification of unstable estimates.

Abstract

Examines the feasibility of using direct variance estimation for the National Crime Victimization Survey (NCVS). It compares generalized variance function (GVF) estimates to two direct variance estimation methods (Taylor Series Linearization and Balanced Repeated Replication). It also provides recommendations for calculating confidence intervals and identifying unstable estimates. A companion user's guide has been developed to describe how to implement the direct variance techniques detailed in this report using different software packages. Access to GVF User's Guides and other documentation is available online.