There are two key features to any MCDA intervention, namely the problem structuring and preference modelling to ensure that the analysis is directed at solving the right problem (
of the intervention), and the provision of computationally efficient decision support algorithms (
of the intervention). There are, of course problems where either the computational aspects are unchallenging so that only effectiveness requires the analyst’s attention; or where the problem is in principle well-defined but computationally complex, so that efficiency concerns dominate.
Contexts do arise in which structuring and preference modelling (e.g. identifying criteria, assessing performance in terms of these criteria and aggregation of preferences across criteria) require careful attention, especially when the numbers of criteria are large,
the resulting models are computationally complex. In such contexts the two components of decision support need to work together. On the one hand, the problem structuring and selection of preference models should balance the need to represent decision maker preferences faithfully with the need for a model implementation which is sufficiently responsive and computationally effective to ensure that the decision maker derives useful support. On the other hand, computational methods and approaches must recognize the cognitive limitations of the decision maker in such complex settings. For example, unaided or undirected search across the Pareto set is unlikely to be cognitively meaningful for larger numbers of criteria even with inventive use of graphics.
This paper will focus primarily on reference point methods both for problem structuring and representation, and as a guide to computational identification and exploration of the Pareto frontier. Some comments will, however, be made on the role of other methods of MCDA in this context.