| In discrete-event simulations, one of the long-recognized challenges is the fact that initial observations may not be representative of the real-world system being studied. When this is the case, bias in output statistics computed from the simulation may lead to incorrect conclusions. The most common means of mitigating this biasing effect is to truncate a sequence of initial observations and exclude these from subsequent analysis. The challenging aspect of this approach is in deleting enough observations to reduce bias without removing observations that are not inducing bias. Removing too many observations unnecessarily reduces the sample size and, thus, the power of the statistical experiment.;The efficacy of the MSER algorithm at automatically managing the tradeoff between truncating too few and too many initial observations has been demonstrated in literature. However, it is still not widely adopted. We believe that this is due to a lack of theoretical foundation for MSER and, more generally, lack of intuitive understanding within the discrete-event simulation community about MSER.;In this dissertation, we present findings from a research program that included deriving the expected behavior of MSER, making a link between MSER and the standard error of fitted from regression literature, demonstrating the superiority of MSER over other algorithms in reducing the mean-squared-error in the estimate of process mean, and comparing MSER with an accepted unit-root stationarity test from time-series literature. An equally important component of our program that was focused on increasing understanding of MSER within the discrete-event simulation research community was the creation and presentation of a teaching story on MSER. We present that teaching story here, as well. |
