2009-04-20

bayesian vs frequentist

Many assume that I am a religious or dogmatic Bayesian, because I use Bayesian statistics. I am not; I use them because (a) they are not ad-hoc, (b) they are justified in the context of probability theory, and (c) for many problems I do they perform better than the alternatives (see, for example, our April Fools' paper). Bovy and I spent some time today doing battle with Kyle Cranmer (NYU) on this point, which was productive inasmuch as we understood the differences between a Bayesian and frequentist approach, but unproductive in that Cranmer thinks he could do much better with a frequentist method (to our Fools' problem). We asked him to put up or shut up, as Gruzinov says. While we were arguing, who should walk in but Yuri Levin (Leiden), the author of the frequentist orbital roulette paper.

1 comment:

  1. >> for many problems I do they perform better than the alternatives

    of course sometimes they do not ...

    ReplyDelete