Mining Silver

tl;dr: Nate Silver really is a frequentist.

Yesterday, I had the chance to attend a lecture given by Nate Silver on his book The Signal and the Noise. I've written about Silver plenty of times in the past, but this was my first opportunity to see him speak in person.

Nate gave a very good talk. He was refreshingly modest. When he described his 538 work, he called his analysis 'just averaging the polls' (as I've said all along!).

But he was introduced as a 'leading statistician' (I'm sure he doesn't write his own PR blurb, but still...). He didn't talk much about the divide between Frequentist and Bayesian stats, but did put up Bayes's Rule and called it 'complicated math that you'll see in more and more stats journals, if you read such things.'

During the Q&A, I asked him about his attack on Frequentist statistics in his book1, and where he got his viewpoint on statistics. He answered something like, "Mostly from a few papers I've read. Honestly, I don't talk to a lot of statisticians." Then he went on to say that he's 'philosophically a Bayesian' even though most of the methods he uses are Frequentist (like taking polling results and averaging them!). His words, not mine!

He went on a little longer, defending his view. But the main point I took away was that, to Nate, baby stats2 and Frequentist statistics are synonymous, while the Bayesian philosophy is synonymous with anything that works.

This is a fine view to take, and it clearly works for Nate. It has the disadvantage that it's wrong3, but again, it works for Nate.

To recap: Nate is a very smart guy, and he's made a name for himself doing honest work. I just wish that as a "leading statistician" he would get his facts straight about the (very real) debate going on between Bayesian and frequentist statistics.

  1. A taste: "Recently, however, some well-respected statisticians have begun to argue that frequentist statistics should no longer be taught to undergraduates. And some professions have considered banning Fisher's hypothesis test from their journals. In fact, if you read what's been written in the past ten years, it's hard to find anything that doesn't advocate a Bayesian approach." (p. 260 of The Signal and the Noise)

    As a gut check, I looked at all of the articles published in Annals of Applied Statistics, Biostatistics, and Biometrika (three of the top statistics journals) in 2012, and computed the proportion of Bayesian articles. Only 42 out of 213 report on Bayesian methods, about 20%.

  2. On the level of what is taught in AP Statistics.

  3. Modern frequentist statistics is at least as sophisticated as Bayesian statistics, and many people have made arguments for frequentist methods over Bayesian methods.