Considering sample means in Rinott’s procedure with a Bayesian approach

Ranking and selection procedures deal with a number of system designs from which the best must be selected. The indifference-zone type of ranking and selection procedures attempt to guarantee the probability of correct selection, P(CS), while minimizing the simulation computing effort. Many indifference-zone procedures assume the slippage configuration in order to show the statistical validity, which makes the procedures inefficient. There have been many efforts to design more efficient indifference-zone procedures by eliminating the slippage configuration, yet none of them is accompanied by a rigorous mathematical analysis that assures the P(CS) guarantee without using an asymptotic analysis. In this paper, we present a new indifference-zone procedure for ranking and selection. This procedure uses sample mean information instead of assuming the slippage configuration and is therefore less conservative than existing procedures, as it achieves a minimum stated probability of correct selection with lower computing effort. The procedure is statistically valid and mathematically proved using the Bayesian inference model. Experimental work conducted shows that the proposed procedure is robust and that it should be useful to the simulation community.