F. J. Díez. Influence diagrams vs. decision trees for medical decision analysis. Harvard School of Public Health. Boston, June 16, 2005.

23 slides (12 pages). PDF (278 KB),  zipped PDF (217 KB).


Influence diagrams (IDs) were developed as a decision-analytic model, alternative to decision trees. In this talk I will discuss the advantages of IDs for modeling decision problems, especially in medicine, such as the explicit representation of causal dependencies and independencies, the use of direct probabilities, the possibility of using canonical probabilistic models (noisy OR, noisy MAX, etc.) and super-value nodes. Additionally, there are specific methods for evaluating IDs, which are in general much more efficient than expanding and rolling-back the equivalent decision tree. I will also discuss the limitations of standard influence diagrams, such as the difficulties for representing asymmetric decision problems, in particular those with unordered decisions, and recent proposals aimed at overcoming such limitations. I will present Elvira, a software tool for building and evaluating Bayesian networks and influence diagrams that has been developed as a collaborative project of several Spanish universities.

Related publications

Extended version of this seminar: See also: