On Generalized Variance Functions for Sample Means and Medians

Justin J. McIllece

Abstract

Generalized variance functions (GVFs) and their associated parameters are most often applied to binomial characteristics of complex surveys, usually within the Federal Statistical System. Non-binomial estimates have less GVF support in the general body of literature. This paper presents a framework for modeling the variances of sample means and quantiles and calculating generalized parameters. The models are applied to the mean and median duration unemployed characteristics from the Current Population Survey (CPS), the only two estimates in the primary CPS estimation tables - from The Employment Situation, a Principal Federal Economic Indicator published monthly by the Bureau of Labor Statistics - for which no GVF parameters are produced. The performance of these GVFs is evaluated, and plans for implementation are discussed.