A conservative method for selecting the best simulated system

Multinomial data are often produced in simulation experiments designed for comparing alternative system configurations. Since some simulation models are very complex and expensive to run, statistical methods are needed that work well for small sample sizes. The purpose of this article is to provide a conservative lower bound for the simultaneous confidence coefficient for confidence intervals for multinomial proportions of the form &pcirc;i ± s/√(n), i = 1, …, k. This lower bound is conservative in the sense that it does not depend upon the asymptotic distribution of &pcirc;i.