A Procedure for Building Confidence Interval on the Mean of Simulation Output Data

Abstract

One of the existing methods to build a confidence interval (c.i.) for the mean response in a single steady state simulation system is the batch means method. This method, compared to the other existing methods (autoregressive representation, regenerative cycles, spectrum analysis, standardized time series), is quite easy to understand and to implement and performs relatively well. However, the most serious source of error is the underestimation of the variance of the sample mean due to the correlation of batch means. In the new method, which is essentially based on the batch means approach, weights are assigned to the batch means such that they become uncorrelated. Then based on the assumption that the output data are obtained from steady-state simulation, as expression is developed for obtaining the confidence interval on the mean.