Variance reduction in the simulation of call centers

We show via concrete illustrations how the variance can be reduced in the simulation of a telephone call center to estimate the fraction of calls answered within a given time limit. We examine the combination of a control variate and stratification with respect to a continuous input variable, and find that combining them requires care, because the optimal control variate coefficient is a function of the variable on which we stratify. In a setting where we compare two similar configurations of the center, we examine the combination of stratification with common random numbers. We show that proper use of common random numbers reduces the convergence rate of the variance of the difference of performance measures across the two systems.