Single-experiment input uncertainty

Abstract‘Input uncertainty’ refers to the simulation model risk caused by estimating input distributions from real-world data, and specifically the (usually unmeasured) variance in performance estimates that this introduces. We provide the first single-run method for quantifying input uncertainty, meaning that we derive our measure of input-uncertainty variance—both overall variance and the contribution to it of each input model—from the nominal experiment that the analyst would typically run using the estimated input models; other methods in the literature require additional diagnostic experiments. Application of our method is illustrated with two examples.