Robustness of Additive Value Function Methods in MCDM

A simulation model is constructed of choice between a discrete number of non-dominated alternatives. The long-run goals of the decision maker are assumed to be consistent with a hypothetical preference structure which satisfies assumptions of completeness, transitivity and additive independence for an ideal set of criteria. The use of additive value functions to aid the decision maker in this choice is simulated for a variety of contexts and under a number of non-idealities such as the omission of criteria, confounding of criteria and inconsistent responses. It is found that ideal preference orderings are well identified by additive value functions provided that the non-idealities are moderate and that sufficient effort is put into modelling changing marginal values for different levels of performance. One of the potentially most sensitive areas is that of shifts in the decision maker's reference points as a result of the types of preference information asked.