Sensitivity Analyses in Interval Decision Modelling

have resulted in a decision tool capable of handling imprecise information in complex decision situations. Some of the research has been directed towards developing decision analytical representation models and algorithms. The decision tool takes intervals as well as comparative relations as input, and incorporates a level of sensitivity analyses embedded into the representation instead of applying separate analyses on top of the evaluation procedure. Further, besides the embedded sensitivity analysis, an explicit sensitivity analysis is useful in order to point out the most critical probabilities, values, or weights to the decision under consideration. This paper deals with the theory and implementation of sensitivity analyses in a decision tool supporting interval probabilities, values, criteria weights, as well as comparative relations. I. INTRODUCTION 1 In the market of decision analytic tools today, see e.g. [1] for a survey, most of the software packages are not capable of handling imprecise information. Instead, the input needs to be specified in precise numbers, which is often considered less realistic for decision-makers. The models and software based on them mainly consist of some straightforward set of rules applied to precise numerical estimates of probabilities and values. Then, sensitivity analyses are often not easy to carry out in more than a few dimensions at a time. The decision analytic tool discussed in this paper, DecideIT, coincides functionally with regular decision tree software, such as Precision Tree [2] or TreeAge Pro [3], but also with multi-criteria software such as Expert Choice [4] since it handles probabilities, values, and weights all in the same framework. However, DecideIT also handles imprecise information, not requiring a pointwise precision in the input parameters, but instead being capable of handling intervals as well as comparative statements between parameters, such as " greater than " , " more important than " , " equal to " or " between 20 and 40 more valuable than " [5].