ABSTRACT
Experimental design of simulation experiments is virtually identical to the design of experiments for real systems, whether laboratory experiments or ˆeld studies. The main difference is that with simulation experiments, the experimenter has complete control over the experiment. He or she can control all the variables in the experiment, whereas in real-world experiments there are always factors that cannot be controlled by the experimenter. In simulation experiments, variables represented by input parameters are called factors whereas the resulting dependent variable outcomes are called responses. Experimental designs usually are developed after parameter estimation has been completed so that the appropriate parameters and their levels can be incorporated into the design. Usually, not all parameters in a model would be included as factors. Only those factors we suspect will have the greatest effect on the response, or are of particular interest, need to be included. Likewise, there may be several response variables of interest. The particular design is constructed to maximize the amount of information with a minimum amount of effort (Martin 1968). Therefore, we are interested in designing ef‚cient simulation experiments, i.e., determining the fewest number of experiments that will provide the necessary data for statistical analysis.
