Simulation and regression: Dealing with the assumption of a common error variance

Abstract

In the application of regression analysis to discrete event simulation the assumption of a common error variance is usually questionable. Procedures are suggested for testing this assumption, for dealing with departures from it within the regression protocol, and for avoiding the difficulty by controlling the length and/or number of simulation runs at each experimental point. Passive approaches are discussed in which it is assumed that a set of simulation runs are made and are then subjected to statistical analysis. More dynamic approaches are then considered which take advantage of the ready availability of additional data in the simulation context.