Replicate weights can be constructed using different procedures, including replicating some or all stages of sampling and replicating some or all weight adjustments. For the 2011-2012 National Postsecondary Student Aid Study (NPSAS:12), weights were replicated for only the poststratification weight adjustment for the second stage of sampling (student).
We used data from NPSAS:12 to compare design effect estimates computed using two different approaches to replication. The two different replication approaches were 1) replicating only the student-level poststratification weight adjustment (as was done for NPSAS:12) and 2) replicating all stages of the student-level weight adjustments, which encompasses replicating weight adjustments for student-level multiplicity, unknown eligibility, nonresponse and poststratification.
We compared design effect estimates to explore the differences between the variance estimates using the two approaches.
|