Experimental Designs for Computer Simulation Experiments

Abstract

Experimental designs have found widespread application in the biological and physical sciences. The purpose in using these designs is not only that they are economical in terms of the number of experimental trials required, but more importantly, they provide a structure for the investigator's learning process. Experimental designs are available for (1) investigating the relationship of the response to the factors in order to determine the underlying mechanisms governing the process under study and (2) finding the combination of factor levels at which the response variable is optimized. Since a computer simulation experiment is indeed an experiment, it follows that one should give careful attention to experimental design questions prior to conducting a simulation experiment. Within the context of two simulation experiments with an inventory model, this paper discusses factorial, fractional factorial, rotatable, and response surface designs. We conclude by outlining a number of possible pitfalls and contingencies.