Multivariate Modeling and Standard Error Estimation Using the Current Population Survey: An Examination of Various Approaches

• Activity Number: 76
• Type: Topic Contributed
• Date/Time: Monday, August 9, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Arthur F. Jones*+ and Gestur Davidson and Lynn A. Blewett and Michael Davern and James Lepkowski
• Companies: U.S. Census Bureau and University of Minnesota and University of Minnesota and University of Minnesota and University of Michigan
• Keywords: variance ; income ; poverty ; CPS-ASEC

Abstract

Multivariate statistical analysis of the Current Population Survey (CPS) is widely used in the social science and public health literature as a source of evidence. However, the statistical evidence cited from the CPS not be comparable because researchers use a variety of statistical methods to produce standard errors used in significance tests. We examine five common methods of standard error estimation: (1) the simple random sample estimator, (2) the generalized variance estimator, (3) first order Taylor series robust estimator, (4) a survey design based Taylor Series estimator using the public use file, and (5) a survey-design-based Taylor Series estimator using the internal Census data file. We restrict our analysis to the 2003 Annual Demographic Supplement and we estimate a basic multivariate statistical model to predict personal income, health insurance coverage and poverty. Our findings will help researchers who use the public CPS data for multivariate statistical modeling know how well various approaches to standard error estimation works.