Validation of Simulation Models : Regression Analysis Revisited

This paper proves that it is wrong to require that regressing a model's outputs on the observed real outcomes give a 45 line through the origin (unit slope, zero intercept). Therefore this paper proposes an alter-native requirement: the responses of the model and the real system should have the same means and the same variances. To test whether this requirement is satisfied, a novel statistical procedure is derived. This procedure regresses the differences of simulated and real responses on their sums. The old and the new procedures are investigated in an extensive Monte Carlo experiment that simulates queueing systems. The conclusions of this experiment are that (i) the old test rejects a valid simulation model substantially more often than the novel test does; (ii) the intuitive test shows 'perverse' behavior in a certain domain: the worse the simulation model, the higher its probability of acceptance; and (iii) the novel test does not reject a valid simulation model too often (its type I error probability is correct), provided the queueing response is transformed logarithmically