Silver at the Joint Statistical Meeting
I learned yesterday that Nate Silver will be giving the President's Invited Address at this year's Joint Statistical Meeting.
What do I mean? Well, as I've said before (and will probably say again), Nate Silver's methodology, whether for predicting the Oscars or predicting political elections, essentially amounts to averaging many nearly unbiased, nearly uncorrelated estimates to get a better estimate overall. That's smart. But it's not groundbreaking2. And it's certainly not news to a room full of statisticians.
In recent years, the meeting has hosted around 5000 statisticians. I find it really hard to believe that those statisticians don't know the usefulness of averaging. And what else has Nate Silver done? He's popularized the use of quantitative methods, sure. But again, that's not something that a room full of statisticians needs to hear. "Wait, you're telling me that I should be measuring things? All this time..." It seems a bit silly.
I heard recently that Silver might speak at the University of Maryland. Assuming that happens, and assuming I could get into the talk3, I know what issue I would raise during the Q&A: his bald-faced lie in The Signal and the Noise about the current state-of-affairs of statistics4.
Then again, this sort of technical clarification may not have a place in a popular forum. But I would hope that some of the statisticians at the JSM might educate Silver.
Then again, I don't know what a room full of 'professional statisticians' entails. It could be a group of academic statisticians. Or it could be a group of statistical technicians (i.e. those who use statistics, but aren't statisticians). I've never been to the JSM, but if it's anything like the Joint Mathematics Meeting, it should be more of the former and less of the latter.↩
My rants might proceed me. But I doubt someone as famous as Nate Silver cares what a lowly graduate student in applied mathematics thinks about him.↩
Of course, this comes down to the question of 'to Bayes or not to Bayes.' Silver claims all non-Bayesian methods are hogwash. This, itself, is hogwash. (And Silver doesn't show any signs of understanding what 'Bayesian' means, to professional statisticians.) Both frequentist and Bayesian methods have their places. Just like machine learning, in many ways a different approach to the problem of inference, deserves a seat at the table.↩