Handling imperfect user statements in real-life decision analysis

Using established techniques from operations research (linear programming) and well-known measures of probability and value, we present a computational representation and evaluation of imperfect, imprecise user statements in decision analysis. The presentation starts with the structure of a decision problem and a model of the situation is discussed. The courses of action are represented by consequence sets in the decision structure. Statements could have an interval form to reflect the translation of the imperfect input data. Statistical decision theory is utilized to discriminate between alternatives. Dealing with imprecise statements means encountering decision situations where different alternatives are to prefer in different parts of the consistent solution space to the constraints. Consequently, selection rules based on traditional admissibility are not enough to indicate preferred choices. New admissibility concepts and a procedure for analyzing such situations are discussed. An algorithm for optimizing bilinear functions that does not even require LP is presented.