Ranking and selection techniques with overlapping variance estimators

Recent advances in ranking and selection

“…A number of estimators have been proposed and there are still efforts to develop new variance estimators with better statistical properties such as smaller variance and smaller bias. Experimental results in Healey et al (2007) show that variance estimators with smaller variance and smaller bias can result in meaningful (sometimes huge) savings in the number of observations for the KN+ and KN++ procedures.…”