Three-Toed Sloth

"Frequentist Accuracy of Bayesian Estimates" (This Week at the Statistics Seminar)

A speaker who needs no introduction (but will get one), on a topic whose closeness to my heart needs no elaboration (but will get it):

Bradley Efron, "Frequentist Accuracy of Bayesian Estimates"

Abstract: In the absence of prior information, popular Bayesian estimation techniques usually begin with some form of "uninformative" prior, intended to have minimal inferential influence. Bayes's rule will still produce nice-looking estimates and credible intervals, but these lack the logical force attached to genuine priors, and require further justification. This talk concerns computational formulas that produce frequentist accuracy assessments for Bayesian estimates. Both encouraging and cautionary examples will be presented.

Time and place: 4:30--5:30 pm on Monday, 18 March 2013 in the McConomy Auditorium, University Center

Memo to self: ask Efron what he thinks of Fraser's "Is Bayes Posterior Just Quick and Dirty Confidence?" (arxiv:1112.5582).

Enigmas of Chance; Bayes, anti-Bayes