Validation,statistical testing,and the decision to model

Abstract

This paper outlines a decision-theory approach to the validation of simulation models. This approach in volves trading off the chances of incurring the cost of using an invalid model against the costs of additional modelling to increase the Z2-keZihood that the model will fulfill its intended purpose. Importantly, the decision-theory approach provides a conceptual base that has hitherto been lacking for the use of statistical testing in validation. Further, by explicitly relating validation to the decision of whether or not to model in the first place, more soundly based estimates of the costs and probabilities can be obtained than would otherwise be possible. Notwithstanding the formal decision approaches, more subjective methods such as Turing type tests are recommended as also being necessary for a balanced approach to the validation of simulation models.