Decision analysis provides a theoretically appealing approach to making decisions in the face of uncertainty and multiple conflicting objectives. Unfortunately, most decision makers (DMs) have been unwilling to adopt this rationale approach to important decision problems, in part, due to the stringent information demands it places on them in applying its concepts and methods. This research project involves the development of a novel interactive approach and prototype decision support system (DSS) for performing decision analysis which relaxes its stringent measurement requirements by precluding the need for the DM to assess his/her beliefs and tastes directly, precisely, and/or completely. The philosophical basis of the approach is based on robutness and is aimed at making decision analysis a potentially more accessible and viable tool for decision making in practice. The investigators' initial effort focused on making decisions under uncertainty based on imprecise, incomplete beliefs or tastes. They now extend our approach (1) to include making decisions under uncertainty when both beliefs and tastes are imprecisely and/or incompletely specified, (2) to include multiple criteria decision problems when either or both tradeoffs and individual attribute value functions are imprecisely and/or incompletely specified, (3) to incorporate knowledge and experience-based intelligence into our RID methodology and DSS for prediction and learning to reduce the cognitive demands on the DM and accelerate the solution process, and (4) to develop the system principles and prototype design for an integrated, intelligent DSS that can simultaneously support decision making under uncertainty and multiple criteria decision making. The contribution of the research will be both theoretical and pragmatic. For example, on a theoretical level, research in decision analysis under conditions of both imprecise probability and utility functions should reveal new insights regarding the dynamics of the relationship between imprecise beliefs, tastes, and choice preferences when making decisions under uncertainty. Further, the treatment of attribute weights and conditional value/utility functions as random variables in multiattribute utility analysis should allow us to specify the probability function of the imprecisely induced aggregated value/utility intervals associated with alternatives. Such information permits the use of stochastic dominance concepts to filter inferior alternatives. From a pragmatic perspective, the results of this research should lead to the development of a methodology and integrated DSS that should allow more pervasive and successful applications of decision analysis in practice.
