Modeling Variances to Determine Sample Allocation for the Current Population Survey Current Population Survey (CPS)

Reid Rottach and Gregory Erkens

Abstract

The Current Population Survey employs a two-stage rotating panel design, with each month’s sample being made up of eight replicate second-stage samples. Panel correlations for state and national unemployment estimates are modeled, allowing us to make predictions for sample allocation which account for proposed changes in the sample design and changes in the underlying population. We use the panel correlations to estimate variance components. Calibration and composite estimators are considered.