Evaluating Various Methods of Standard Error Estimation for Use with the Current Population Survey's Public Use Data

• Activity Number: 363
• Type: Contributed
• Date/Time: Wednesday, August 6, 2003 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Michael E. Davern*+ and James M. Lepkowski and Gestur Davidson and Lynn A. Blewett
• Companies: University of Minnesota and University of Michigan and University of Minnesota and University of Minnesota
• Keywords: Current Population Survey ; variance estimation ; public use data ; generalized variance ; Taylor series linearization ; health insurace coverage

Abstract

Standard error estimation is a problem for analysts of the Current Population Survey (CPS) because most of the variables needed to construct standard errors are not available on the public use file. We use four methods of standard error estimation in this paper: (1) the basic "simple random sample" (SRS) approach, (2) the "generalized variance" (GV) approach contained in the Source and Accuracy Statement of the CPS, (3) the "robust variance" (RV) estimation approach, and (4) a Taylor Series Linearization (TSL) approach with both a stratum and clustering variable. We restrict our analysis to: Income, Poverty and health insurance coverage. For income the GV, RV, and TSL on produce similar standard errors and they are significantly larger than those produced through the SRS method. For poverty, the GV approach produces the largest standard errors and the SRS and RV estimation produce the smallest standard errors. For health insurance coverage the GV estimates are the smallest with the SRS and RV estimates a close second and third. The TSL estimates for health insurance coverage are the largest.