Replicate Variance Estimation in a Two-Phase Sample Design Setting: Simulation Study with National Survey of College Graduates Data

• Activity Number: 122
• Type: Topic Contributed
• Date/Time: Monday, July 30, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Survey Research Methods
• Author(s): Michael White*+ and Jean Opsomer
• Companies: U.S. Census Bureau and Colorado State University
• Keywords: NSCG ; ACS ; two-phase sample design ; variance estimation ; replication ; simulation

Abstract

The 2010 National Survey of College Graduates (NSCG) selected its sample from 2009 American Community Survey respondents, creating a two-phase sample design. This creates variance estimation complexities when using replication methods, as there are two sample designs to account for. One solution is to create a set of replicates for each sample design phase but this can be unwieldy as it leads to a large number of replicates. Another solution, which we pursued, is to use replicates from the first-phase sample and adjust them to account for the second-phase sample design. In particular, we used a Reweighted Expansion Estimator (REE) that post-stratifies the second-phase sampling weights back to the first-phase estimated totals within each of the second-phase sampling stratum. We conducted a simulation study to evaluate the performance of the REE estimator with different replication methods, including successive difference, grouped-jackknife, and balanced repeated replication. Initial results showed poor performance with some replicate variance estimates due to the inability of the replicates to capture the systematic sample selection used in the second-phase sample design. Accounting for the systematic second-phase sample with post-stratification resulted in good performance for the REE estimator with all the replication methods and the successive difference replication method was ultimately chosen as the 2010 NSCG production method.