Improvement of Performance of Variable Probability Sampling Strategies.

摘要

Precision of estimators used with variable probability sampling may be improved either at the design stage or at the analysis stage following selection of the sample. Improvement at the analysis stage is obtained by employing a Horvitz-Thompson difference estimator, which may improve precision even when the response variable is negatively correlated with the inclusion probabilities. In general, optimization of the Horvitz-Thompson difference estimator is achieved by employing bootstrapping to facsimile populations constructed from the sample data; in some cases optimization can be achieved analytically. This strategy focuses sampling considerations on population description in order to improve the facsimile population representation of the real population under study.