Empirical input distributions: an alternative to standard input distributions in simulation modeling

The authors investigate the effect of input-distribution specification on the validity of output from simple queuing models. In particular, the use of various kinds of empirical distributions for approximating service-time distributions is studied. It is shown that, when the approximating distributions were compared on the basis of variance and bias in their estimates, the empirical distributions generally did as well as the best fitted standard distributions, and sometimes better. For example, when Weibull was the true distribution, the fitted Weibull and gamma were the best fitting distributions among the standard distributions, with the least bias and variance. The empirical distributions were a good match where both the criteria were concerned, and in some cases had lower variance and bias. >