13 February 2014

Bayes At the Moon

There's been an increased (hopefully, seasonal like some Sam Adams beer and the Groundhog) run of p-value bashing mashed up with Bayes Uber Alles meme-ing.

Enough already. If Bayesian analysis were all that much better at avoiding the "yesterday was like the day before, today is like yesterday, so tomorrow will be like today" pit which run-of-the-mill quants fill into when they gave us The Great Recession, then the Bayesians would have been trumpeting their superior analyses. I didn't see any. And, from a structural point of view, Bayesians are more likely to fall in the pit: the emphasis on "knowing" a prior, aka "yesterday's data". He's an old guy who died centuries ago, let him be.

No comments: