A restricted multinomial hybrid selection procedure

Abstract

Analysts using simulation models often must assess a large number of alternatives in order to determine which are most effective. If effectiveness corresponds to the likelihood of yielding the best outcome, this becomes a multinomial selection problem. Unfortunately, existing procedures were developed primarily for evaluating small sets of alternatives, so parameters required to implement them may not be readily available or the sampling costs may be prohibitive when a large number of alternatives are present. We propose a truncated, sequential multinomial subset selection procedure that restricts the maximum subset size. Numerical comparisons show that our procedure can be much more efficient than the leading unrestricted procedure. Our procedure requires only one calculated parameter rather than four. We provide extensive tables for cases involving large numbers of alternatives.

When Is Information Sufficient for Action? Search with Unreliable yet Informative Intelligence

“…These conditions usually relate to the relationship between the true probabilities of the two best alternatives (Chen 1988). A good survey of the techniques used in multinomial selection problems appears in Vieira et al (2014). Most selection problems assume a deterministic number of observations.…”