Re: Bayes vs. Popper

From: mal
Affiliation:
Address:
Date: 14 Dec 2006
Time: 07:40:41

Comments

Eliezer - I am not an expert on Bayesian analysis, but I have never been aware of a formalized preference for simple models over complex ones, at least not in the Ockham's razor sense. As in your coin flipping example, Bayesian analysts may tend to impose simplicity through their prior probabilities, but there is nothing in the Bayesian approach that forces them to do so. They might just as well impose complexity by adopting different prior probabilities. With regard to your "relativistic theory of gravity" example, it isn't clear to me whether you are referring to Einstein's confirmed theory, in which case we may reasonably assume that the predictions about the comet are valid without direct observation, or whether it is some new relativistic theory that has not yet been confirmed. In the later case, we cannot assume that the comet will behave as the model predicts because the model hasn't been confirmed. Confirmation (validation) requires that we be in a position to make observations that have the potential to invalidate the model.

Disclaimer