Saturday, January 5, 2013

Happy birthday to Bayes' rule

On the 250th anniversary of the posthumous publication of Bayes' rule, this is appropriate. Good introduction to the frequentist/Bayesian debate.

I've been pretty confused by that debate - what exactly is the controversy, as parodied by e.g. xkcd - so this was nice to read since it put methods I'm familiar with in the context of the debate and doesn't focus on settings in which one side or the other is a straw man. Having read this, I still can't say I see what the big deal is. Obviously if you have a reasonable prior, you should use it, and obviously if you don't, some additional assumptions will be required to draw any kind of useful conclusion, and whatever kinds of assumptions you allow yourself will make a difference... both camps make assumptions about forms of distributions, models, etc, so I think the author is overly sympathetic to the frequentists when he says they claim for themselves the 'high ground of scientific objectivity'.

No comments: