A Methodology For Validating Multivariate Response Simulation Models By Using Simultaneous Confidence Intervals

This paper deals with the substantiation that a multivariate response self- or trace-driven simulation model, within its domain of applicability, possesses a satisfactory range of accuracy consistent with the intended application of the model. A methodology is developed by using simultaneous confidence intervals to do this substantiation with respect to the mean behavior of a simulation model that represents an observable system. A trade off analysis can be performed and judgement decisions can be made as to what data collection budget to allocate, what data collection method to use, how many observations to collect on each of the model and system response variables, and what confidence level to choose for producing the range of accuracy with satisfactory lengths. The methodology is illustrated for self-driven steady-state and trace-driven terminating simulations.