Resampling Inference With Complex Survey Data

From MaRDI portal





The main purpose of this paper is to demonstrate that the bootstrap method can be extended to general sampling designs provided independent and identically distributed samples. The explanation begins with defining a correct bootstrap method for stratified samples. The main idea is to resample with smaller size \(m_ h\) \(n_ h\) in each stratum. Even for nonlinear parameters the estimators have the shape of a function of sample means. The choice of \(m_ h\) is discussed, too. The special case of combined ratio estimator is studied a little bit deeper. Section 3 provides bootstrap confidence intervals and studies their coverage probabilities for known strata variances. A summary of a simulation study is presented in the last paragraph of this section. The next section contains a brief extension of the method to stratified SRS without replacement. Several resampling methods for unequal probability sampling without replacement are presented in Section 5, the main result being a generalized random repeated replication. Another generalization - namely to two-stage cluster sampling without replacement - is dealt with in Section 6. In both these cases the method is justified by showing that in the linear case the variance estimator reduces to the standard one.




Cited in
(82)








This page was built for publication: Resampling Inference With Complex Survey Data

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3800888)