SBatch: A spaced batch means procedure for steady-state simulation analysis

We discuss SBatch, a simplified procedure for steady–state simulation analysis that is based on spaced batch means, incorporating many advantages of its predecessors ASAP3 and WASSP while avoiding many of their disadvantages. SBatch is a sequential procedure designed to produce a confidence–interval (CI) estimator for the steady–state mean response that satisfies user–specified precision and coverage–probability requirements. First SBatch determines a batch size and an interbatch spacer size such that beyond the initial spacer, the spaced batch means approximately form a stationary first–order autoregressive process whose lag–one correlation does not significantly exceed 0.8. Next SBatch delivers a correlation–adjusted CI based on the sample variance and lag–one correlation of the spaced batch means as well as the grand mean of all the individual observations beyond the initial spacer. In an experimental evaluation on a broad range of test problems, SBatch compared favourably with ASAP3 and WASSP.