Bayeing At The Moon

It's not a secret that I'm not fond of Bayes. I grew up at a time when Bayesian analysis was an emerging cult, whereas now it seems (in some areas of discourse) to be trite. The problem with Bayes is simply that priors are subjective, and can affect the results immensely. In the final analysis (so to speak), we have a bunch of data, and we wish to assess whether there's an equation (model) which closely predicts these values. The data are concrete, the model is fungible. The data tell the story, but we must be careful in our quest to find the moral.

So, Andrew Gelman stirred up a hornet's nest. Have a read; it's a gas.

In the end, it's always been my feeling that the reason Bayesian stats have become popular is that Ph.D. dissertations from the classical/frequentist arena became increasingly hard to come by. It was simply easier to gin up some Bayesian proposal. And so it was.