Adjustment of importance weights in multiattribute value models by minimum discrimination information

Often decision makers are faced with numerous competing alternatives from which to choose. The selection process usually starts with evaluating the alternatives based on some attributes. A decision maker attaches an initial weight to each attribute, computes an overall score for each alternative, and ranks the available choices. The initial weights approximately reflect the importance of the attribute ranges and may be tentative. An agent wishing to promote an alternative might be interested in knowing how different the weights need be to make this alternative acceptable to the decision maker. The task of the agent is thus to derive some weights that very closely approximate the decision maker’s initial weights and rank the agent’s alternative better than its competitors. It can also guide the agent to determine what kind of changes are necessary in order to make this alternative acceptable. In this paper the authors motivate the problem by discussing some real life situations in which the relative importance of attributes in ranking a set of alternatives is subject of public debate and criticism. A spectrum of decision makers-public administrators, business managers, individuals-are interested to find out to what degree a well publicized ranking depends on the relative importance weights attached to various attributes. The authors use an information-theoretic approach to provide a framework for developing weights that accomplish this task. Specifically, the Minimum Discrimination Information allocation procedure proposed by E.S. Soofi in 1990 is used to adjust the importance weights in the ranking of U.S. metro areas. Application of this technique to product purchasing is also discussed.