Practical and Theoretical Considerations of Panel Effects for the Current Population Survey’s Composite Estimator

Gregory Erkens

Abstract

Labor Force surveys are often designed with rotating panels of households, where households remain in the survey for a set number of months and then rotate out of the sample. Rotation patterns often induce correlations that can increase an estimate's efficiency. To further improve precision, the US Current Population Survey (CPS) uses an AK Composite estimator that consists of two primary components: a calibrated estimate, and an estimate of change using overlapping panels between adjacent months. Panels may exhibit effects in their Labor Force estimates based on the number of months a panel remains in the survey--often called month-in-sample effects. Panel effects impact the Composite estimator in a particular fashion. These effect can change over time, and it's important to consider carefully how panel effects influence a Composite estimator. This paper looks at some practical aspects of panel effects on CPS estimates, and how practical considerations interact with theoretical ones. In particular, we look at the following topics: considerations for MSE, choosing a Composite estimator, estimation of inputs such as panel effects, and optimizing for multiple estimates.