Influence diagrams for Bayesian decision analysis

Influence diagrams for representing Bayesian decision problems are redefined in a formal way using conditional independence. This makes the graphs somewhat more helpful for exploring the consequences of a clients stated beliefs. Some important results about the manipulation of influence diagrams are extended and reviewed as is an algorithm for computing an optimal policy. Two new results about the manipulation of influence diagrams are derived. A novel influence diagram representing a practical decision problem is used to illustrate the methodologies presented in this paper.