Thursday, December 30, 2010

Brian Skyrms on the interpretations of probability theory

BS: Most scientists (and some philosophers) are frequentists and don’t trust the Bayesians.

Many philosophers (and some scientists) are Bayesians or subjectivists “all the way down”, and many don’t believe in objective probabilities.

VR: That would explain a lot.


David Parker said...

This might be old hat for you, but I have just examined Nagel's argument for permitting some intelligent design discussion in the classroom.

It looks to me like he is taking a Bayesian approach, which will be rejected by most scientists if Brian is correct.

Anonymous said...

Happy New Years Eve!

As to Probabilities, it is highly probable that John Loftus will be drunk on his ass by midnight CST!


Just a little joke!

Victor Reppert said...

Don't make this thread about Loftus, too.

Blue Devil Knight said...

You are right about scientists. We tend to think that the probability of this coin coming up heads is 1/2regardless of your subjective evaluation of the probability. I personally tend to go with the propensity interpretation, but frankly it doesn't matter all that much. The same theorems hold no matter what your interpretation of probability theory. In practice it's like a physicist worrying about whether numbers are real. It is likely to make her a worse physicist, not better.

Blue Devil Knight said...

I never met a philosopher who was a serious Bayesian, other than perhaps using probabilistic models of belief formation, in which the probabilities in these models of others' belief states were taken in their subjectivist sense. But that's a very different monster than someone who actually advocates for Bayesianism.

How will my statistics, as a scientist, be changed if I were to become a Bayesian? That is, in practice what difference will it make?

Someone once tried to convince me that I wouldn't be able to treat parameters of distributions as random variables with a probability distribution unless I'm a Bayesian (and doing this helps in some statistical inference problems). But that was wrong (he was right that early frequentists didn't like to treat parameters like that, but it isn't a consequence of it isn't as if the choice is between frequentism and subjectivism).