Properties of Smoothed Design‐Based Variance Estimators from Complex Sample Surveys

MoonJung Cho, John L. Eltinge, Julie B. Gershunskaya, and Larry L. Huff

Abstract

Due to relatively high levels of sampling variability, direct design-based variance estimators are often smoothed before publication. For example, some federal statistics programs publish the medians of a sequence of monthly direct variance estimates, or functions of these medians. The properties of these smoothed estimators depend on several underlying conditions, including sample size; effective degrees of freedom for the direct estimators; correlation of the direct estimators across months; and temporal patterns in the true variances. We compare and contrast these properties with the corresponding properties of generalized variance function (GVF) estimators.