ARD: An automated replication-deletion method for simulation analysis

ARD is an automated replication-deletion procedure for computing point and confidence interval (CI) estimators for the steady-state mean of a simulation-generated output process. The CI can have user-specified values for its absolute or relative precision and its coverage probability. To compensate for skewness in the truncated sample mean for each replication, the CI incorporates a skewness adjustment. With increasingly stringent precision requirements, ARD's sampling plan increases the run length and number of runs so as to minimize a weighted average of the mean squared errors of the following: (i) the grand mean of the truncated sample means for all runs; and (ii) the conventional replication-deletion estimator of the standard error of (i). We explain the operation of ARD, and we summarize an experimental performance evaluation of ARD. Although ARD's CIs closely conformed to given coverage and precision requirements, ARD generally required a larger computing budget than single-run procedures.