Comparison of MCDA Paradigms

Abstract

The underlying concepts of MAUT, SMART, AHP, preference cones, ZAPROS, and outranking methods are compared. Learning systems are considered. The learning view is that decision makers initially do not fully understand all of the criteria that are important. Therefore, rather than uncovering an underlying utility function, what must be uncovered are the full ramifications involved in selecting one alternative over another. This paradigm can involve an evolutionary problem, where criteria can be added or discarded during the analysis. Methods are also reviewed with respect to their psychological validity in generating input data. Past experiments conducted by the authors are reviewed, with conclusions drawn relative to subject comfort in using each method. Subjects typically make errors, in that they have inconsistent ratings of scores across systems, and will occasionally have reversal of relative importance of criteria across systems. This emphasizes the need to be careful of input in decision models, and strengthens the argument for more robust input information. Furthermore, systems based on the same model have been found to yield different results for some. In a study exposing both US and Russian students were compared. Each group found it more comfortable to use systems developed within their own culture.

The concept that seems most attractive is that the analysis needs to focus on the decision maker learning about tradeoffs. A major problem with utility based and outranking methods is that decision makers might consider a wide variety of criteria, but both practicality and mathematics show that only a relatively small set of criteria are really going to matter. Learning methods allow decision makers to focus on these critical criteria and their tradeoffs.