Abstract
This paper is concerned with the analyses of ranked set sampling procedures for dealing with the existence of missing observations. We consider sub-sampling the non-respondents strata. The first visit may serve for ranking accurately the observed non-respondents. Two problems are considered: the estimation of a mean and the difference of means The sampling error of the proposed estimators are smaller than their simple sampling with replacement counter parts. Monte Carlo experiments are developed for evaluating the behavior of the estimators. The use of rss appears as the best alternative.