My brother Andy pointed me to this climate science blog. I don’t know much about climate science, but I do know about probability and data modeling, and I like the way this guy writes about them. For instance, he has a really nice piece illustrating how you can model the climate at a variety of levels of complexity from very simple up to massively complex simulations. The point of this post is to debunk the notion, which seems to be widespread, that the only reason scientists believe in climate change is because of complicated black-box simulation codes. He illustrates how you can see the big picture very easily from much simpler models.
His most recent post is about his born-again Bayesianism. It’s generally very sensible and worth reading, although I want to point out one important distinction that I think he blurs a bit. The word Bayesian can describe two different (but overlapping) kinds of people:it can refer to
- People who use a specific set of statistical techniques, or
- People who have a certain philosophical stance about the meaning of probability.
Personally, I think that you absolutely have to be a Bayesian in the second sense of the term: the frequentist notion of probability strikes me as utterly incoherent. But I think you should be completely agnostic as far as the first point is concerned. Bayesian and frequentist statistical techniques are just tools. They’re both perfectly sensible, and you should use whichever tool is more convenient for the problem you’re trying to solve at the moment.
I think that some people think that being a Bayesian in sense 2 means that you have to be a strict Bayesian in sense 1 — that is, that you can never calculate a confidence interval again. Fortunately, it just isn’t so. For instance, I cowrote a paper quite a while ago in which we analyzed the same data set from both Bayesian and frequentist points of view to illustrate the relation between the two.