“Despite what the Wikipedia entry says, there’s no objective prior or subjective prior, nor is there any reason to think the Jeffreys prior is a good idea in any particular example. A prior distribution, like a data distribution, is a model of the world. It encodes information and must be taken as such. Inferences can be sensitive to the prior distribution, just as they can be sensitive to the data model. That’s just life (and science): we’re always trying to learn what we can from our data.”
Which, enouragingly, is pretty much what I have been saying to Nic over a long period of time, both in person and perhaps once or twice as a reviewer of manuscripts. I am, however, not optimistic that the message will ever get through to him, as he seems completely impervious. But perhaps the rest of us can just carry on with life regardless.
Edit: having just had another glance at ATTP's post, and the still-growing comment thread, I see no reason to change my opinion about the message getting through...
Edit: having just had another glance at ATTP's post, and the still-growing comment thread, I see no reason to change my opinion about the message getting through...
4 comments:
I remember having a discussion with Pekka about this when it came up a while ago. He was saying that a non-informative prior can lead to ludicrous results, and I was pointing out that the very fact that you think that some results are ludicrous, is itself proof that a non-informative prior is not the right tool. A non-informative prior is what should be used if all results are equally good and none are ludicrous. If there are things that we believe, we should not be using a non-informative prior, we should be using a prior based on what we believe. That's what the prior is. (Sorry to belabor the point.)
Surely none of us approaches the climate sensitivity with absolutely no idea of what it should be. Even someone who has no respect for paleo evidence - which perhaps someone could justify - still ought to get some picture of what's reasonable from basic physics. Not saying I know how to do that, or what feedbacks ought to be included or anything.
A uniform prior is a pretty terrible idea, too.
I saw that on the Andrew Gelman link people were commenting and trying to suggest various priors to try. Has Lewis' work been repeated with various choices of prior? How sensitive is the sensitivity to that? I'd like to know how much of the IPCC's fat tail goes away with _any reasonable prior_; that seems like an important result.
Since MikeR mentions Pekka, I thought I would pass on the sad news that Pekka passed away - according to this - in late November.
I am so sorry to hear it. He was a breath of fresh air in these blogs. Aside from all his remarkable diverse skills, he just didn't care about sides, just who was right on each particular issue. You could _trust_ him: if he said something, it was because he had studied it, himself, with no bias.
Yes, Pekka made useful contributions here, on a number of threads.
Post a Comment