The bootstrap and finite population sampling.

Synthetic populations are used to study methods for adapting Efron’s bootstrap estimation technique to finite population sampling. Of particular interest is the extention of these methods to two-stage cluster sampling. Using simulations based on five artificial populations, two variations of bootstrap estimators and two Taylor series variance estimators for a ratio estimator are compared by mean square errors, stability of the variance estimators, and coverage of the confidence intervals. Generally, there appear to be small differences among the variance estimators, except the bootstrap estimators are somewhat less stable than the Taylor series estimators.