Analysis of Generalized Variance Function Estimators from Complex Sample Surveys

MoonJung Cho


For applied work with generalized variance function (GVF) models for sample survey data, one generally seeks to develop a model that produces variance estimators that are approximately unbiased and relatively stable. Through simulation, we evaluate the bias and variance of model coefficients, and the bias and variance of the GVF estimator. In addition, we compare and contrast confidence interval coverage rates and widths of the GVF estimator to design-based estimators. We study these properties with varying degrees of freedom for the GVF estimators and a refined bias adjustment factor for nonlinear transformations in the lognormal model. Our simulation study is based on the data from the U.S. Current Employment Statistics (CES) survey.