I attended a surprising event on October 19, 2011 at the American Film Institute Silver Theatre in Silver Spring, Maryland. The event was not to see a movie star, but on the other hand, the surprisingly small audience was treated to a event of great star power in the statistical sense, as the symposium honoree Brad Efron, aka the bootrap fame, and also with many other notable roles such as in the play for empirical Bayes and the geometry of likelihood, shined brightly once again. Like always, he gave with an inspirational talk on Bayesian Inference and the Parametric Bootstrap, with a beautiful handout on his slides. He concluded by remarking that improper Bayesian approach should always be evaluated via the frequentist approach. (Jim Berger may be smiling somewhere!)
I remember when first starting graduate school at UNC, Chapel Hill, Brad Efron was invited to give the prestigious Hotelling lecture, and one of his handout was: Why Isn't Everyone a Bayesian? http://www.jstor.org/stable/10.2307/2683105. Now almost 25 years later, Bayesian is really making a comeback. Indeed, just two days later, Sharon B. McGrayne, author of the book: The Theory Would not Die, How Bayes' Rules (Changed the World, my ed) gave a talk at my institute to a surprisingly packed audience of mostly physical scientists, engineers.
Indeed, for Bayesian statistics to be used seriously in practice, such as for uncertainty analysis in metrology, http://www.nist.gov/pml/pubs/tn1297/index.cfm, there is still a long way to go: how to validate and how to relate to classical and traditional ways of uncertainty assessment. Can confidence intervals traditionally being derived based on clasical statistics and frequentist interpretation be really replaced by Bayesian inference? Why and how to make it routine in practice?
Reference: Efron, Bayesian Inference and parametric bootstraphttp://stat.stanford.edu/~ckirby/brad/papers/2011BayesianInference.pdf
No comments:
Post a Comment