Using intervals for global sensitivity and worst-case analyses in multiattribute value trees

Sensitivity analyses have long been used to assess the impacts of uncertainties on outcomes of decision models. Several approaches have been suggested, but it has been problematic to get a quick overview of the total impact of all the uncertainties. Here we show how interval modeling can be used for global sensitivity analyses in multiattribute value trees, and a nuclear emergency case is used to illustrate the method. The approach is conceptually simple and computationally feasible. With intervals, the decision maker can include all the possible uncertainties and quickly estimate their combined impact. This is especially useful in high-risk decisions where a worst-case type of sensitivity analysis is essential. By varying the intervals one can also examine which uncertainties have the greatest impact and thus need the most consideration. Global sensitivity analysis reveals how the outcome is affected by many simultaneous variations in the model.