Abstract
Let π1, …,πk be k independent populations with continuous distributions F1, …, Fk respectively. Hollander and Proschan (1972) proposed as a measure of the NBU-ness of Fiand used it to develop a test of exponentiality versus non-exponential NBU alternatives. In this paper the k populations are compared in terms of ▵i and we are interested in selecting the population with the largest ▵ value which is defined as the least NBU population. Our procedure is based on the U-statistics Jn(Fi. Which is the unbiased estimator of ▵i and its large-sample properties. Some applications of the selection procedure to gamma, Weibull and linear failure rate families are discussed.