| Abstract: | The problem of uncertainty in the outcome of decisions has tended to be glossed over by many MCDM methods, with the exception of multiattribute utility theory (MAUT). MAUT does, however, require quite complicated preference elicitations and knowledge of the full multivariate distribution of outcomes. Results from a series of simulation studies indicate that the preference orderings of MAUT are only minimally changed when using a simple additive aggregation of marginal utilities, especially in relation to the natural imprecisions inherent in preference elicitation. It is shown in the simulations that by far the most critical aspect of multicriteria decision analysis under uncertainty is not the form of aggregation, but the correct elicitation of marginal utilities which properly represent decision maker preferences over gambles. We relate the results obtained here to other results on the approximation of distributions by three- or five-point discrete distributions, and suggest that the use of deterministic MCDM methods of any form (not necessarily value function techniques), applied to an extended formulation in which each criterion measure is repeated for three or five ‘scenarios’, can be justified. |
