Abstract:
A question that commonly arises in longitudinal surveys is the issue of how to combine differing cohorts of the survey. The different cohorts can represent disjoint populations, a single population, or overlapping populations. In this paper we present, under a superpopulation approach, a course of action for combining different cohorts in a longitudinal survey with a repeated-panel/rotating-panel design; namely the Survey of Doctorate Recipients, conducted by the U.S. National Science Foundation. In this case the cohorts represent non-overlapping populations. The procedure builds upon the Weighted Generalized Estimation Equation method existing in the literature for handling missing waves in longitudinal studies. Although our method is set up under a joint-randomization framework, which takes into account the superpopulation model, our simulations show that the method also performs well for estimating welldefined finite population quantities, as well as superpopulation parameters. We also propose a design-based, and a joint-randomization, variance estimation method.
Keywords:
Estimating change; Finite population parameters; Replication variance estimation; Rotating panel surveys; Superpopulation parameters; Weighted Generalized Estimating Equations
tr180.pdf