Rotation Designs and Composite Estimation in Sample Surveys
Part 1. Motivating Their Use
Patrick J. Cantwell
KEY WORDS: Repeated sampling, panel surveys, change over time, panel conditioning, internal consistency.
ABSTRACT
One can think of a rotation design as a compromise between a complete sample overlap and taking independent samples. Each extreme has advantages and disadvantages. By using a rotation design, one hopes to realize some of the variance reduction of the complete sample overlap, while reducing its excess burden. In this paper, we start by motivating the use of a rotation design and composite estimation to improve the estimator of current level of a parameter, θt , then look at compositing to improve the estimator of change, θt − θt-1. Some consideration is then given to doing both: estimating level and change simultaneously. Finally, we briefly discuss other practical issues that influence the choice of designs and estimators, including generalizing the estimators, panel conditioning, cost, the mode of data collection, and respondent burden.
CITATION:
