Validation of multivariate response models using Hotelling's two-sample T2 test

Abstract

A procedure is given for using Hotelling's two-sample T2 test to test the validity with respect to the mean behavior of a multivariate response simulation model that represents an observable system. A tradeoff analysis can be made to help the modeler decide on a data collection method and budget, the number of observations to collect, and suitable model builder's and user's risk levels for testing the validity under a given set of experimental conditions. A model of an M/M/1 queuing system with two performance measures serves as an example for the validation procedure.