The Quality and User Acceptance of Multiat-Tribute Utility Analysis Performed by Computer and Analyst

Abstract Two potential problems challenge the computerization of decision analysis. First, to what extent can the often ill-defined art of structuring be transformed into software. Second, to what extent is past consumers' satisfaction with decision analysis a function of the formal methods and procedures of the theory and rationale of decision theory, and to what degree do other factors such as personal interaction and the establishment of a rapport account for client approval? We compared multiattribute utility analyses of personal decision problems of undergraduates performed by a human analyst vs. those performed by a standalone 'software package, Multi-Attribute Utility Decomposition (MAUD3). Although subjects favored the analyst session over the MAUD3 session, agreement with and acceptance of the analyst and MAUD3 results implied ordering and most preferred alternative did not differ. We did find that subjects feel better taken care of when more attributes are included in the analysis, but that subjects' holistic ratings are better accounted for by analyses with smaller rather than larger number of attributes. Although the analyst attribute sets were more often judged more nearly complete and better in overall quality, the MAUD3 attribute sets were more often judged more nearly independent, both logically and valuewise. Furthermore, the attribute set of each pair with the greater number of dimensions was overwhelmingly chosen as being more complete, less independent, and of higher quality than the other attribute set. Judgments of overall quality were virtually identical to those of completeness.