Variance estimation by jackknife method under two-phase complex survey design

摘要

The paper uses a weighted jackknife approach. In the literature, a weighted jackknife procedure is used to address the lack of balance caused either by varying probabilities of selection or by varying sizes of the units being selected in the sample. The paper focuses on deriving a (new) better jackknife variance estimator (NJVE) of the two-phase version of the generalized regression estimator (GREG). The intent is to smooth out the imbalance in the sample caused by varying probabilities of selection and to make the variance estimator dependent on pair-wise inclusion probabilities. The paper compares, using simulation exercises, the performance of the NJVE with that of the traditional one and also with that of the variance estimator based on the linearization technique. The paper demonstrates that the performance NJVE is better than that of either of its rival estimators with respect to certain chosen performance criteria.