Resumen
We propose and investigate an alternative definition of robustness for estimators of the variance of the sample mean arising in steady-state simulation experiments. The alternative definition implies that of all known batching estimators-overlapping-batch-means estimators have both optimal mean squared error and optimal robustness.