## did I mean endemic? [pardon my French!]

**D**eborah Mayo wrote a Saturday night special column on our Big Bayes stories issue in *Statistical Science*. She (predictably?) focussed on the critical discussions, esp. David Hand’s most forceful arguments where he essentially considers that, due to our (special issue editors’) selection of successful stories, we biased the debate by providing a “one-sided” story. And that we or the editor of *Statistical Science* should also have included frequentist stories. To which Deborah points out that demonstrating that “only” a frequentist solution is available may be beyond the possible. And still, I could think of partial information and partial inference problems like the “paradox” raised by Jamie Robbins and Larry Wasserman in the past years. (Not the normalising constant paradox but the one about censoring.) Anyway, the goal of this special issue was to provide a range of realistic illustrations where Bayesian analysis was a most reasonable approach, not to raise the Bayesian flag against other perspectives: in an ideal world it would have been more interesting to get discussants produce alternative analyses bypassing the Bayesian modelling but obviously discussants only have a limited amount of time to dedicate to their discussion(s) and the problems were complex enough to deter any attempt in this direction.

**A**s an aside and in explanation of the cryptic title of this post, Deborah wonders at my use of *endemic* in the preface and at the possible mis-translation from the French. I did mean *endemic* (and *endémique*) in a half-joking reference to a disease one cannot completely get rid of. At least in French, the term extends beyond diseases, but presumably *pervasive* would have been less confusing… Or *ubiquitous* (as in Ubiquitous Chip for those with Glaswegian ties!). She also expresses “surprise at the choice of name for the special issue. Incidentally, the “big” refers to the bigness of the problem, not big data. Not sure about “stories”.” Maybe another occurrence of lost in translation… I had indeed no intent of connection with the “big” of “Big Data”, but wanted to convey the notion of a big as in major problem. And of a story explaining why the problem was considered and how the authors reached a satisfactory analysis. The story of the Air France Rio-Paris crash resolution is representative of that intent. (Hence the explanation for the above picture.)

June 26, 2014 at 9:09 am

A collection of random thoughts on this ans the linked post:

– the idea that bayesian methods can be used to force a bad model to “fit” is odd to me (the reference to “when bayes shatters” is complicated as it depended on the infinite dimensional nature, the topology and the ability to measure data exactly. It’s less a lesson and more a specific measure under which bayesian inference for a very specific class of problems has bad properties. Attempting to generalise from this is fairly brave. For this class of problems bayes “works” or “doesn’t” based on how you define works, which suggests that everything is problematic and the insight probably won’t stretch to simpler problem classes)

– big data is the least interesting big bayes problem.

– there seems to be an implicit idea that non-bayes solutions need to be advocated. I don’t agree with that, in the sense that they don’t lack proponents (see, for example, machine learning). But the idea that only bayes will solve the problem is a bit whiffy… “bayesian modelling” + non-bayesian computing (eg map estimates) would probably work just as well.

– there is also an idea in Mayo’s post (or specifically the quoted passages) that focussing on one strategy (or even advocating it) negates all the others. I think that’s rubbish. Anyone who ever does applied work starts with easy things and works their way out. The inference method is driven by the questions you need to answer (or fail to answer). It’s not uncommon to make bayes methods that have frequentist properties, or make frequentest methods that are bayes inspired or any other hybrid/bastard/pragmatic inference method. And that’s a good thing. And that’s why the idea of dismissing entire approaches seems dumb to me: it’s fantastically short sighted.

Anyway. Those are my feelings, such as they are.