Uncertainty, sensitivity and scenario analysis: how do they fit together?

Abstract

Dealing with uncertainty is becoming increasingly important in model-based decision support. Various methods have been developed in order to do this, including uncertainty, sensitivity and scenario analysis. Although these different methods serve their purpose, the availability of a large number of methods can make it difficult for practitioners to understand the similarities and differences between them and when the use of one is more suitable than another, resulting in confusion. In addition, researchers often identify with belonging to a group dealing with a particular approach, which can lead to a lack of crossfertilisation and understanding. In order to assist with bridging the gap between researchers working on different approaches to dealing with uncertainty and eliminate confusion for practitioners, the objective of this paper is to examine the relationship between uncertainty, sensitivity and scenario analysis in the context of model-based decision support, and to take the first steps towards establishing common ground between these methods and assess the contexts under which they are most suitable. This is achieved by conceptualising the various methods as different approaches to “sampling” the hyperspace of model inputs, although this is done from different perspectives and for different ends (Figure 1). It is therefore also necessary to think about the assumptions each method is making about the space being explored, and there are benefits to be gained in thinking about how best to sample the space for each purpose. The approaches identified in this conference paper provide a first level of coarse characterisations. Further refinements in categorisation is possible (with the differentiation between narrative and stress testing scenarios as a first example), and likely to be useful. There are connections to be made to other disciplines, such as philosophy and decision theory, regarding the assumptions each method makes.