LAUSR

ABSTRACT The paper compares the accuracy of using cluster samples and using stratified samples to estimate a population total. Several clustering algorithms are used to partition a finite population into strata or clusters. Several variants of stratified sampling designs and one-stage cluster sampling designs, including those dependent on various inclusion probabilities, are taken into account. The accuracies of the estimators are compared using simulation experiments. The results of this paper let us conclude that partitioning a population into clusters could significantly improve the accuracy of estimating the total using sampling dependent on inclusion probabilities proportional to the aggregated auxiliary variable. Moreover, the considered estimators based on cluster sampling designs could be easily used in practice.