Bias and variance reduction in computer simulation studies

The research of several authors was extended to a complex queueing model with eleven responses. Warming-up the system, antithetic variates, and their joint application, were compared with crude sampling. Emphasis was placed on constructed confidence intervals as opposed to the performance of point estimates. Performance was evaluated by means of the coverage probability and the change in confidence interval widths; the results differ from those of the earliest studies. The application of a warm-up period reduced bias; however, this reduction in bias was accompanied by a large increase in the confidence interval widths. Antithetic variates were employed in an attempt to reduce, or eliminate, any increases in confidence interval width caused by applying a warm-up period. The success of the techniques was found to be dependent on the type of model response that was being analysed.