Files
Abstract
The National Agricultural Statistics Service is developing strategies that limit the amount of sample overlap across
unrelated surveys by using multi-phase sampling principles. In its simplest form, Sample A is selected first, and then
Sample B is chosen from among those members of the population not selected for Sample A. Effectively, Sample B is
selected in two-phases. This two-phase approach extends easily to the coordination of more than two samples, although
meeting accuracy and/or sample-size targets while maintaining strict sample exclusivity is not always possible. Variation
of the basic approach address this problem, but lead to some theoretical difficulties. Sampling weights may be based on
products of conditional selection probabilities rather than on unconditional selection probabilities. Randomization-based
variance estimation likewise may depend on the product of conditional joint selection probabilities. In practice, variance
estimates will be reasonable but may not always be randomization consistent.