hrubin@odds.stat.purdue.edu (Herman Rubin) writes:> Not necessarily. If one looks at action, Bayesian behavior > with the prior as a weighting measure only is what comes out. > > In the simplest situation, suppose that there are two states > of nature, and several actions. What consistent behavior > implies is that the action to be chosen is that one (modulo > ties) which maximizes a linear combination of the expected > utilities under the two states. The coefficients of this > linear combination, normed, is the "prior". > > This has nothing to do with belief, unless one wants to make > it that. I do not care what to believe, but how to act.
Yes---I actually quite like this approach, though I'm not a serious expert on the foundational issues. Unfortunately I have to go on holiday now :) Otherwise I would love to keep this thread going if only to explain to the guys who don't "get it" why there _is_ a serious potential worry that some kinds of fuzzy logic are at best irrelevant, at worst just wrong. Cheers, w