Restricted Subset Selection Procedures for Simulation

This paper develops two procedures for screening a set of normal populations with unknown moments in order that the final subset of selected populations satisfies the following requirements involving the user specified parameters P*, δ, and m: With probability at least P*, the selected subset will contain a population whose mean lies less than the distance δ from the largest mean. Although the size of the selected subset is random, at most m populations will finally be chosen, where m is usually taken small enough to reserve adequate resources for a more intensive follow-up study. The exact procedure V E is a two-stage random sampling scheme that is designed to compare transient or steady-state simulation models based on independent replications of each model. The heuristic procedure V S is a modification of V E that is designed to compare steady-state simulation models based on a single prolonged run of each model. A complete justification is given for V E together with tables of the constants required by both procedures and an algorithm for computing these constants. Experimental results are summarized to gauge the robustness of V E against departures from normality and to evaluate the performance of V S on several types of stationary stochastic processes.