Numerical vs Cardinal Measurements in Multiattribute Decision Making: How Exact Is Enough?

• Published in: Organizational behavior and human decision processes 1995-10, Vol.64 (1), p.9-21
• Main Authors: Larichev, O.I., Olson, D.L., Moshkovich, H.M., Mechitov, A.J.
• Format: Article
• Language: English
• Subjects: Behavioral decision theory; Decision making; Decision making models; Decision support systems; Statistical analysis
• ISSN: 0749-5978; 1095-9920
• DOI: 10.1006/obhd.1995.1085

Multiattribute decision making can involve consideration of both quantitative and qualitative measures of criteria attaintment. Some decision support systems (decision aids) to help multiattribute decision making quantify value functions. One of the most popular of these systems, multiattribute utility theory (MAUT), requires two types of input. Decision makers need to express the relative value of different attainment levels on each criterion, as well as express the relative importance of these criteria. Some systems (such as DECAID) require simple direct graphical input of value and criterion importance. Other systems (such as LOGICAL DECISION) use more complex means of expressing relative value. Either way, MAUT converts expressions of criterion importance into quantitative form. This study compares the relative stability of numerical results obtained through two decision support systems, DECAID and LOGICAL DECISION (LD), used in the task of evaluation of multiattribute alternatives. Additionally the relative stability of results was measured by comparison with results obtained using an ordinal method, ZAPROS. ZAPROS is a decision support system for construction of a partial order over the set of alternatives. It does not require conversion of qualitative measures into quantitative form. The relations among alternatives are close to those based on ordinal dominance. The results of experiments show that ordinal relationships between task parameters are much more stable than those obtained from quantitative measures. Results from DECAID and LD are much less coincident with each other than with results obtained through ZAPROS. Many inconsistencies were found in subject responses. It is concluded that more attention should be given to the means of testing judgment consistency, and that in some cases, attempts to solve decision tasks through more "exact" judgments of value function parameters may lead to erroneous results.