A sensitivity analysis procedure for Bayesian decision-making

In an effort to see how analytical model outputs change with respect to variations in model inputs, sensitivity analysis procedures have been widely used in applications such as mathematical programming and classical optimisation. However, until recently, sensitivity analysis has seen only limited application in the area of decision theory and support. This paper investigates the use of sensitivity analysis in the realm of classical Bayesian reasoning, where the probabilities of the states of nature are revised based on additional information. These updated probabilities only become useful, however, if they lead to an optimal decision different from that obtained on the basis of prior probabilities. This paper develops a novel sensitivity analysis procedure for Bayesian decision-making and proposes a set of criteria for the ranges of the model input parameters over which the current solution will remain optimal.