Replicate Variance Estimation for the National Survey of Parents and Youth Using Eigenmatching

• Activity Number: 196
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Louis Rizzo*+ and David R. Judkins
• Companies: Westat and Westat
• Keywords: jackknife ; BRR ; finite population corrections ; multistage sample

Abstract

This paper describes a case study in replicate variance estimation from the National Survey of Parents and Youth, a nationally representative sample of families with youths 9-17 years old. The replicate weights are derived using a method first proposed by Fay for providing consistent replicate variance estimators for complex sample designs. This method defines modified jackknife replicate weights to provide replicate variance estimators which exactly match the sample-design unbiased variance estimators for totals, by "eigenmatching": algebraically reproducing the eigenvalue-eigenvector structure of the sample-design unbiased variance estimator (including the exact finite population corrections at each level of sampling). These replicate variance estimators will then be consistent for sufficiently smooth functions of unbiased estimators of totals such as ratios, regression coefficients, etc. (as is achieved for with-replacement sampling using unmodified jackknife or BRR methodologies). The NSPY application is a multistage stratified sample design with primary sampling units as the first stage of selection.