Sometimes they come back... I’m a sucker for Jaynes, and I still think that Ap distributions are a much more solid contribution to the subject than any philosophical discussion of Knightian uncertainty or partition of the data, especially when you interpret Ap to be “the probability of retrieving a piece of evidence that will set P(A) = p”.
An interesting discussion—in the comments—about meta-uncertainty or probabilities of probabilities. I say it’s turtles all the way down :-)
Sometimes they come back...
I’m a sucker for Jaynes, and I still think that Ap distributions are a much more solid contribution to the subject than any philosophical discussion of Knightian uncertainty or partition of the data, especially when you interpret Ap to be “the probability of retrieving a piece of evidence that will set P(A) = p”.
Dunno, seems like Gelman is pretty correct to me. Which is to say, there are a finite number of turtles determined by the complexity of your model.
You can always go one meta level higher and ask questions about your model from an outside view.