A classification of multiple criteria decision methods

The key objective of all multi-criteria methods is to formalize and capture the trade offs among criteria of interest. We will provide a basic understanding of three broad classes of multi-criteria methods. Table below summarizes the features of these three classes of methods.

Multiple criteria decision methods can be distinguished along several dimensions. From a decision maker’s perspective, a classification scheme based on the information required from him or her to implement the method is useful. In such a classification, the first class of methods is one in which the decision maker supplies no information during the analysis phase. Since the essential information about the decision maker’s trade offs among criteria is not conveyed, these methods provide only a trade off curve (also called efficient frontier, non-dominated solutions, or pareto optimal solutions) among criteria. The decision maker can then choose a preferred point, reflecting his or her inherent preferences, from the trade off curve. Alternatively, formal methods can be employed to elicit the decision maker’s preference function, which can in turn be used to evaluate the points on the trade off curve; the optimal decision is identified as the point that maximizes the utility or value specified by the decision maker.


A classification of Multi-criteria Methods

I None

a) User is not known or multiple users
b) User cannot provide information until some solutions are presented.
c) Problem is simple, or it is convenient to involve user only at the end of the process.

Generation of efficient frontier or trade off curves (Philip, 1972; Zeleny, 1974; Evans and Steuer, 1973)

II Trade offs among criteria or choice between two options sought as needed.

a) Preference function too complex to be stated explicitly
b) User can provide only local information

Interactive mathematical programming (Geoffrion. Dyer, and Feinberg, 1972; Zoints and Wallenius 1976)

III Complete specification of trade offs and preference function

a) Explicit preference function can be obtained.
b) Problem is of strategic nature and, therefore, thinking is focused on values (preferences) desirables before the decisions are evaluated.

Multi-attribute preference function theory (Keeney and Raiffa 1976; Dyer and Sarin, 1979).

In the second class of methods, the decision maker is an integral part of the analytical process. Information is solicited from the decision maker about his or her preferences for various levels of attainment on the criteria of interest. Based on this information, the method progresses to achieve better solutions, with more information sought from the decision maker as it is needed. This sequential approach terminates when an acceptable or the most preferred decision is identified. These methods are interactive in nature, as information from the decision maker is essential for continuing the progress of the method toward the preferred decision.

Finally, in the third class of methods, the decision maker is asked to supply complete information about his or her preferences and trade offs among criteria. A preference function (utility or value function) is then constructed based on the elicited information. Once the preference function is completely known, the selection procedure substitutes the function for an objective function.

Hybrid procedures which combine the strategies employed in the three classes of methods just described, have also been developed. Other classification schemes can also be developed based on whether the outcomes of the decisions are known or unknown whether the time horizon over which the outcomes occur is modeled as single or multiple time periods, whether the decision alternatives are finite or infinite, whether there is a single decision maker or a group of decision makers, whether the competitive reaction is implicitly considered or explicitly modeled.