Anonymous asked:
su3su2u1 wrote that Bayesian estimates aren't calibrated, so you can't expect a 95% Bayesian interval to contain the truth 95% of the time, he seems to claim this is well known. A google search then turned up a LW post by cousin_it called Bayesian Flame that seems to agree. You are the only person I've seen arguing with su3su2u1 about statistics, so I thought I'd ask- is this true? What should I make of it?
I hope I never gave the impression that I’m able to argue with su3su2u1 about statistics. I don’t really understand the field in a formal way beyond a couple of tidbits I have picked up to try to understand scientific papers better. I’ll publish your ask in the hopes of enlisting some of the other people here to address this.
But I bet it has something to do with priors. Every time someone gets into a fight about Bayesianism, they end up concluding “Sure, that would work great if you could get good priors, but you probably can’t”.
