Advanced tutorial: Input uncertainty quantification

'Input uncertainty' refers to the (often unmeasured) effect of not knowing the true, correct distributions of the basic stochastic processes that drive the simulation. These include, for instance, interarrival-time and service-time distributions in queueing models; bed-occupancy distributions in health care models; distributions for the values of underlying assets in financial models; and time-to-failure and time-to-repair distributions in reliability models. When the input distributions are obtained by fitting to observed real-world data, then it is possible to quantify the impact of input uncertainty on the output results. In this tutorial we carefully define input uncertainty, describe various proposals for measuring it, contrast input uncertainty with input sensitivity, and provide and illustrate a practical approach for quantifying overall input uncertainty and the relative contribution of each input model to overall input uncertainty.