Validation of Simulation Results

Simulation models are designed and used with a goal of learning about a process. Validation is the act of increasing to an acceptable level the confidence that an inference about a simulated process is correct for the actual process. There is no such thing as "the test". The experimenter selects a set of tests from the many possible--a standard decision problem of balancing the cost of testing against the cost of an incorrect inference. This paper outlines possible validation actions and suggests considerations that enter into their choice. Three major classes of actions are finding models with face validity, testing assumptions and testing input-output transformations. Actions range from such highly technical approaches as spectral analysis to behaviorally oriented ones such as the "Turing" test. Validation often becomes more tractable if simulation is viewed as one of several modes of investigation. Complementary research activities--related experiments, empirical analysis, analytic models, or prototypes--are widely used in the physical sciences to increase validity and appear equally appropriate for the management sciences. Some of the ideas and problems of validation are illustrated briefly in an example that involves all-machine simulation, man-machine simulation, related psychological experiments and a field test.