Seth points me to these papers:
John P. A. Ioannidis, Effect of Formal Statistical Significance on the Credibility of Observational Associations, Am. J. Epidemiol. 2008 168: 374-383.
Hormuzd A. Katki, Invited Commentary: Evidence-based Evaluation of p Values and Bayes Factors. Am. J. Epidemiol. 2008 168: 384-388.
John P. A. Ioannidis, The Author Responds to “Evaluating p Values and Bayes Factors”, Am. J. Epidemiol. 2008 168: 389-390.
I do not, do not, do not have the energy now to comment on these. Let me just say that what is labeled in the above articles as “Bayesian” is not the only way to do Bayesian statistics. I refer you to Bayesian Data Analysis for exposition of what I consider the more reasonable Bayesian approach, which is based on modeling rather than hypothesis testing and never involves computing the posterior probability that the null hypothesis is true.
I can’t stop people from doing these other things and I wouldn’t even try. But I would like them to be aware of this other, more direct approach. This paper may also help.
Why "so-called Bayesian?" Bayes Factors are Bayesian…I don't like hypothesis testing anymore than the next guy, but if you're going to do it (and many people are) Bayes Factor isn't a horrible way to go.
Richard:
Even if Bayes factors are arguably _a_ Bayesian approach, they're certainly not _the_ Bayesian approach, as is often implied.
Beyond this, I don't think the models underlying Bayes factors make sense (at least, not in any of the applications I've ever been involved in). This weakens them as Bayesian models in my opinion.