NCJ Number
249479
Date Published
December 2015
Length
43 pages
Annotation
This guide provides documentation for users of the National Crime Victimization Survey (NCVS) who calculate variances of estimates from complex sample designs by using generalized variance functions (GVFs),which are models of the relationships between estimates, their characteristics ( i.e., crime type, year, and subpopulation size), and their variances.
Abstract
Using these formulas, users can calculate variance estimates without knowledge of the sample design and without specialized software. This is the method traditionally used for NCVS variance estimation. This guide provides documentation for users in estimating variances with GVFs and incorporates guidance on the three key NCVS estimate types found in most Bureau of Justice (BJS) statistical reports. These estimate types are victimization totals, proportions, and rates. The guide includes descriptions of the calculation of the estimates, as well as how to conduct significance testing between pairs of estimates. Section 1 of the guide summarizes the data structure of the NCVS public use files (PUFs). Section 2 summarizes how the GVFs were developed, compares GVF with direct variance estimation, and provides guidance on when each method is appropriate. Section 3 outlines how to apply the GVFs to calculate variance estimates for the NCVS. Section 3 discusses how to apply the GVFs to calculate variance estimates for the NCVS; and Section 4 provides detailed examples for various types of estimates. Appendixes contain the GVF parameters for 1993 -2012 and the 3-year correlations by crime type for 1993-2012. These are needed for pooled-year estimation. A separate user guide is available for those who use direct variance estimation to estimate the frequency and characteristics of criminal-victimization data for the United States as indicated by the NCVS. 2 tables