Working smarter when developing linear simulation metamodels

We propose that proper metamodel specification is far more important than the level of simulation effort used in developing a metamodel to estimate the expected response of a given system or simulation. In particular, M/M/k queues with various configurations of arrival rate, service rate, and number of servers were simulated using different levels of simulation effort. The average queue length for each configuration was computed, and then two different metamodels were fit to the simulation data. Analysis of the residuals from the fitted metamodels indicates that metamodel specification has a significant effect on the statistical quality of the estimated expected queue length while the level of simulation effort used in fitting the metamodel has virtually no such effect