Malcolm Forster

University of Wisconsin, Madison

"The Likelihood Principle and the Reversibility of Causal Arrows"

Abstract:

The Likelihood Principle (LP) is the founding principle of Bayesian and Likelihoodist statistics; it says, roughly, that likelihood is all we need to describe the relationship between theory and evidence. Although it is appropriate for some problems, the paper argues that LP is not always adequate for the purpose of model comparison. The issue has arisen explicitly within the context of causal modeling in recent years, so Bayes nets are chosen as examples. But the same point raises broader questions about how statistics might be expanded into a rigorous theory of scientific reasoning.

Back to Talks Page