A simple rule for mitigating initialization bias in simulation output: comparative results

Results are presented for experiments with four heuristics for determining the number of observations to truncate from the beginning of an output sequence generated by a steady-state, discrete-event simulation, with arbitrary initial conditions. The performance of the marginal confidence rule (MCR) developed by White and Minnox (1994) is compared with that of enhanced implementations of practical heuristics proposed by Conway, Gafarian et al. (1978), and Fishman (1972). All rules are applied to output sequences generated by ten runs each of four representative queuing simulations. Results confirm the significance of the start-up problem. Results also demonstrate that simple truncation heuristics can solve this problem, if carefully designed and applied. While all four heuristics are shown to provide improved accuracy without undue loss of precision, in all of the experiments the performance of the MCR dominated that of the other rules.