Treatment of unit non-response in two-stage sampling with partial replacement of units

Abstract

Three separate estimators, for the estimation of the population total, based on two-stage sampling design on two successive occasions with partial replacement of secondary stage units only have been presented. For these estimators it is assumed that there is unit non-response. Hence, Hansen and Hurwitz (1946) technique has been used to adjust for the non-response. Empirically, it has been found that the estimator that uses individual weights within first stage units perform better than the other estimators that use common weights in estimating the population total. Theoretically, it has also been shown that the larger the ratio of the within first stage unit variability of the non-respondents to the between first stage unit variability the higher the gain in efficiency of the proposed estimators over the estimator obtained when there is no partial matching of units.