Abstract:
|
The occurrence of nonresponse and its compensation via imputation, maybe by using sampling designs with different stages, pose a challenge to variance estimation. Variance estimates are frequently the basis of accuracy measures such as standard errors of point estimators. Thus, the variance estimation has to be reliable to avoid a biased accuracy measurement of the statistic of interest. An appropriate method to estimate the variance of point estimators using a sampling design with different stages is the rescaling bootstrap of Chipperfield and Preston (2007) which is however based on complete observations. In case of nonresponse and its compensation via imputation, this method has to be modified which is done in this paper. Furthermore, the quality of the proposed method is examined using a Monte Carlo simulation. Particularly, this simulation reveals how the rescaling bootstrap has to apply in comparison to other modifications to obtain an unbiased variance estimator.
|