Reducing input parameter uncertainty for simulations

Parameters of statistical distributions that are input to simulations are typically not known with certainty. For existing systems, or variations on existing systems, they are often estimated from field data. Even if the mean of simulation output were estimable exactly as a function of input parameters, there may still be uncertainty about the output mean because inputs are not known precisely. This paper considers the problem of deciding how to allocate resources for additional data collection so that input uncertainty is reduced in a way that effectively reduces uncertainty about the output mean. The optimal solution to the problem in full generality appears to be quite challenging. We simplify the problem with asymptotic approximations in order to provide closed-form sampling plans for additional data collection activities. The ideas are illustrated with a simulation of a critical care facility