Random non-response on study variable versus on study as well as auxiliary variables

Abstract

In most practical situations in sample surveys, random non-response, for some units, is present on both the study and auxiliary variables. In other situations, it may be present on study variable only since information on the auxiliary variable may be easily obtainable. In this paper, we compare the two situations by considering the usual ratio estimator for the population mean. The former is found to provide a more biased, more efficient and more non-response robust estimator in comparison to the latter whenever it is advisable to use the ratio method of estimation over simple random sampling in the absence of random non-response.