Wednesday, 7 August 2013

No octagon!

One silly thing I really like about Canada is their street sign for "no stopping" $-$ it always makes me think that Canadians really dislike octagons and are not ashamed to let their national sentiment known to all visitors...

But, enough stupid geometry; while I'm in the airport departures area waiting for BA to graciously open their desks, I'll write down some random thoughts about the JSM. In fact, I'll divide my impressions in two broad categories: "cool" and "not-cool" things.

Cool
• The session on RDD was nice, I thought. I got some interesting comments by Guido (Imbens, who was discussing the session); I'll upload my talk here. He was a bit surprised to see that our instrument was weak (in other words: that the GPs by and large do not obey the prescription guidelines for statins). As I said in the talk, we base this on a limited subset of the data, so it is possible that this is just an artefact. On the other hand, we think that something like that really may be going on. We'll see in the next few months.
• I saw a very, very interesting session on Hamiltonian Monte Carlo $-$ I particularly liked Matt Hoffman's talk (he's part of the Stan development team). During the session, Mark's work has been cited quite a lot (obviously). What was really funny is that each of the three speaker pronounced his surname in a different way (roughly: Jeròlamee, Jerolàmee and Geeròlamee). In Italian, the first version should be the right one $-$ I'll have to ask him what he does call himself...
• Another very good session was the one dedicated to the 20-th anniversary since the "introduction" of MCMC to the statistical community. I really liked Alan Gelfand's talk.
• Jon Wakefield and Håvard were also really good and I enjoyed their session too. I left before the third talk (whose title seemed quite interesting too), because it clashed with another session I wanted to see. In retrospect, I could have stayed put.
• To see the book on display at the CRC booth was also really cool. I couldn't really bring myself to have a picture taken in front of it $-$ although I was really, really tempted. I was kind of hoping someone would offer to do it, but nobody did...
• On the other hand, recording the promotional video interview was kind of cool. Joanna, the interviewer, said that I was a "natural", but I think she was just being nice, really... We'll see when the final version is edited. Hopefully it won't be too embarrassing.
• Meeting quite a good bunch of nice people and having a couple of nice dinners. At the CRC reception, somebody was staring at my badge and then they said: "do you have a blog? I think I've read it!"

Not-cool
• A couple of horrible Bayesian talks $-$ given by people who clearly didn't know why on earth they were being forced to use a Bayesian model (my guess: their supervisor, who was not doing a very good job, though). In one case, I tried really hard not to be mean or impolite, but I had to ask a question (among other things, they had logistic regression coefficients in the magnitude of 10 and with very large standard deviations $-$ convergence problems hadn't even crossed their mind).
• If I'm not missed anything (which I don't think I have), only a couple of talks on health economics or statistical models for cost-effectiveness, in general. This is not news (last year was pretty much the same). But more importantly, you get the feeling that the level of these talks is a good 15 years behind what we do in Europe. The models are not terrible (the one I saw today was a microsimulation comparing several screening strategies). But only ICERs were reported and there was no sensitivity analysis whatsoever. I asked a question, and the presenter told me that in fact they had done some SA (which, as I probed further, turned out to be deterministic and was not reported anyway). I asked about CEACs and EVPIs, but she basically tried to change the subject. She also suggested (in reply to my question) that they had used some "distributions" to account for uncertainty in the parameters $-$ but there was really no evidence of that in what she presented... I am seriously toying with the idea of submitting a proposal for a short course on statistical methods for health economics to next year JSM; it will be in Boston (which is a plus). But I'm not sure I'm biting way more than I can really chew off for next year...
• Nate Silver's talk. It wasn't bad per se. But the rumours are that they had to pay a ridiculously high fee (I mean, seriously: think of a ridiculous fee for a scientific conference. And then double it. And then probably add something else. And that's probably slightly less than the figure that was whispered in the dark corridors of Montreal's Palais des Congres). This really pisses me off, if you pardon my French (which is just appropriate, given we were in Quebec), because the registration fee is really steep and I don't think that we've got our money's worth, if that fee is really the truth (which I think it may well be!). Perhaps the organisers could think of different ways to entertain the audience (who, let's not forget, are nerdy statisticians, so probably enjoyed way more David Spiegelhalter's talk, anyway!) and keep the costs down.