![IconGems-Print](images/IconGems-Print.png)
445 – Contributed Oral Poster Presentations: Survey Research Methods Section
The Accuracy of a National Generalized Variance Function for Subnational Estimation
Philip Lee
RTI International
Bonnie Shook-Sa
RTI International
Marcus E. Berzofsky
RTI International
Lynn Langton
Bureau of Justice Statistics
Michael Planty
Bureau of Justice Statistics
Generalized Variance Functions (GVFs) approximate the variance of an estimate as a
function of readily available information about that estimate. They can be used to
calculate variance estimates for surveys with complex sample designs, and because they
do not require users to have knowledge of the complex design they are often easier to use
than direct variance estimation techniques such as Taylor Series Linearization (TSL) for
basic analyses. However, the validity of GVFs estimates is only known when they are
applied to estimate types that were used to build the GVF equations.
This paper explores the accuracy of a national GVF when applied to subnational
estimates using data from the National Crime Victimization Survey (NCVS). The NCVS,
sponsored by the Bureau of Justice Statistics (BJS) and conducted by the U.S. Census
Bureau, is a multi-mode, rotating panel design survey of households that produces
nationally-representative criminal victimization estimates for major types of crimes in the
United States. For the NCVS, GVFs created by the Census Bureau were designed to
produce variance estimates at the national level, but their accuracy at the subnational
level has not been evaluated. We assess the accuracy of GVF estimates within
subnational areas based on geographic identifiers on the NCVS Public Use Files (i.e.
Census region, population size, and urbanicity) by comparing them with TSL estimates.
Our analysis found that TSL and GVFs do not provide consistent variance estimates
within these subnational areas and thus, the current NCVS GVFs should not be applied
below the national level.