The Reconciliation of Decision Analyses

A decision problem can be structured in many different ways. This paper addresses a few of the problems that arise as a result of this diversity; for instance, is a particular structuring method worth doing: is the effort likely to be rewarding? Further considerations arise when two or more methods are used and must be combined. Any rule of combination must attach weights, that depend on errors in the analyses, to the results. Consequently, the questions of errors both at random and decision nodes require discussion. We show that errors at decision nodes produce biases. A basic idea underlying our argument is the concept of true utilities and probabilities: we include a discussion of this concept.