Short communication: selecting input distributions for use in Monte Carlo simulations

Monte Carlo analysis is increasingly employed as a means of quantifying uncertainties in risk assessment. However, the output of any Monte Carlo analysis is extremely sensitive to the input distributions used. In this paper we present a simple set of guidelines for the selection of input distributions. The guidelines involve allocating variables with unknown distributions to appropriate distributional "classes," based on the range of possible values that could be obtained for the data and known characteristics about the behavior of the variable. If the class of distribution is not identifiable, then variables can be related to a specific distribution using goodness-of-fit techniques. If these identification schemes fail, bootstrapping techniques may be used to generate a sampled distribution from the available data. Finally, "default" distributions may be employed using analogous data if site-specific data are unavailable.