Some modest proposals for simulation software: design and analysis of experiments

Simulation software has made great advances in recent years along the dimensions of modeling capabilities, animated graphics, and ease of use. There have also been real improvements in terms of both generality to model a wide variety of systems, and in specialization for quick and accurate modeling in specific application domains. And, of course, the dramatic improvement in computing/cost ratios has rendered truly commonplace what were just a few years ago impossibly time- and memory-consuming simulations. However, the statistical design-and-analysis capabilities of simulation software have not kept pace with the modeling, graphics, and ease-of-use advances; nor have they kept pace with ongoing research on the underlying methods. This paper discusses a wide variety of design-and-analysis capabilities that the author feels should be as available and as easy-to-use as are current modeling and graphics functions. The list includes only methods that are in existence today, so is not just a "wish list" of perhaps impossible dreams. The paper concludes with speculation on why this imbalance exists and suggests how it might be effectively addressed.