Selecting a good system: procedures and inference

We present two-stage experiment designs for use in simulation experiments that compare systems in terms of their expected (long-run average) performance. These procedures simultaneously achieve the following with a prespecified probability of being correct: (i) find the best system or a near-best system; (ii) identify a subset of systems that are more than a practically insignificant difference from the best; and (iii) provide a lower confidence bound on the probability that the best or near-best system will be selected. All of the procedures assume normally distributed data, but versions allow unequal variances and common random numbers.