Validation and statistical-analysis procedures under the common random number correlation-induction strategy for multipopulation simulation experiments

This paper provides a validation procedure and statistical analysis methods for multipopulation simulation experiments conducted under the common random number (CRN) correlation-induction strategy in order to fit a metamodel of interest between the mean response and a selected set of input factors. Before conducting the statistical analysis, we need to validate certain key assumptions made for the CRN strategy, which, if violated, render the proposed statistical analysis invalid. Validation is composed of a three-step statistical procedure. The first step tests for multivariate normality, the second step tests the structure of the covariance matrix between responses, and the third step tests for the adequacy of the proposed metamodel. The proposed statistical analysis provides optimal estimates for the unknown metamodel parameters and gives optimal simultaneous confidence intervals on linear combinations of these parameters. Both the validation procedure and statistical analysis methods are illustrated with an example of a hospital simulation study.