25 – Recent Research on Current Population Survey
Understanding Variance Estimator Bias in Stratified Two-Stage Sampling
Khoa Dong
U.S. Census Bureau
Tim Trudell
U.S. Census Bureau
Yang Cheng
U.S. Census Bureau
Eric Slud
U.S. Census Bureau, Center for Statistical Research & Methodology
The Current Population Survey (CPS) is one of the oldest surveys in the United States and the source of numerous high-profile economic statistics, including the national unemployment rate. Balanced repeated replication (BRR) is a commonly used variance estimation method at the U.S. Census Bureau when there is no design-based variance estimator available. Due to its sampling design, CPS requires collapsing of non-self-representing (NSR) strata to make pseudo-strata in order to implement BRR. These pseudo-strata should ideally contain exactly two perfectly matched primary sampling units (PSUs). This paper examines properties of BRR estimator when estimating variance of response rate of eligible housing units (HUs) in CPS NSR strata. In addition, we will present a bias study of the BRR variance estimator using simulations based on CPS data.