Generalizing the half-width minimization heuristic for mitigating initialization bias

This paper generalizes the initial condition bias truncation heuristic given in White and Minnox (1994) by incorporating a control reference point computed from an analytical queueing approximation. Queueing approximations allow the mean squared error criteria to be used to find a subsequence in the simulation sample path which has the best bias and variance characteristics. The methodology is demonstrated experimentally using a batch means analysis of representative GI/G/m queueing simulations. The results indicate that the incorporation of a control reference point can significantly improve the performance of point estimators for the mean waiting time in a queue