Physics is like Brazil, Statistics is like Chile

If you want to do research in physics, you need to take a bunch of math classes, a bunch of physics classes, then study a bunch more in grad school . . . the research frontier is far away, and it takes a long time to get there. It’s like Brazil: you land on the beach, go through the city and the farms, and it takes a long time before you reach uncharted territory. In contrast, in statistics, the open problems are much more accessible, and it’s possible to do research right away. Statistics is like Chile: you on the beach and it’s just a few steps to the mountains.

In case there was any doubt, this is a metaphor: I’ve never been to South America, and I’m just giving the world-map perspective on these countries, not any realistic appraisal of what it’s like there! The point is that, yeah, in statistics you’ll bump up against live research problems (for example here) all the time, just in what might seem to be routine applied work. It’s just important to be open to the possibility. Remember when I talked about the importance of “the capacity to be upset, to recognize anomalies for what they are”? Historically, this has been the case in physics as well as in statistics, but the hard problems in physics are so hard, and so deep in the field, that it takes much more effort to hope to make progress in them. Statistics is a less mature field; also, it’s engineering as much as it is science, and we’re never far from the research frontier.

P.S. The earliest version of this idea that I could find came from this 2007 paper, “Bayes: radical, liberal, or conservative?” with Alex Jakulin, where where we wrote:

Statistics, unlike (say) physics, is a new field, and its depths are close to the surface. Hard work on just about any problem in applied statistics takes us to foundational challenges, and this is particularly so of Bayesian statistics. Bayesians have sometimes been mocked for their fondness of philosophy, but as Bayes (or was it Laplace?) once said, “with great power comes great responsibility,” and, indeed, the power of Bayesian inference—probabilistic predictions about everything—gives us a special duty to check the fit of our model to data and to our substantive knowledge.

P.P.S. Sometimes I’ve written things that I’ve forgotten. So I googled *Physics is like Brazil, Statistics is like Chile*, with and without quotes, but nothing relevant came up. Then I entered *Brazil Chile* into the search box for this blog, but, again, nothing relevant. Then I googled *Andrew Gelman Brazil Chile* and nothing relevant—there was this post from 2014 on the World Cup but nothing with the above quote. I tried *Gelman Physics is like Brazil, Statistics is like Chile* but I just got some stuff about my cousin (not really) Murray, along with a link to one of my Five Books interviews, which didn’t mention Brazil or Chile. Whassup with that, Google?

Then I tried googling *andrew gelman Physics is like Brazil, Statistics is like Chile*, and that didn’t work either—but it did link to this post of mine, “The Way of the Statistician and the Way of the Physicist,” that I’d completely forgotten about at the blog of the International Statistical Institute. It’s pretty good! I think I’ll repost it on my Substack blog. (I still don’t get why people keep saying I “should have a Substack,” given that (a) it’s just another blog host, and (b) I already do! But, since people keep saying that, sometimes I’ll post new stuff there.

Then I googled *Andrew Gelman Physics Brazil Statistics Chile* and the first link was this video of a talk of mine from 2018, and . . . here’s the transcript (slightly edited):

I like that data science is not an exclusive club. I was a physics major in college and I like to say that that physics is like Brazil and statistics is like Chile in that in physics if you want to get to the frontier you have to study for about 8 years . . . in physics if you you want to make a contribution you need to sort of pack a heavy weight—you get off the beach, you walk for a long way, you take years, then you’re finally at the frontier, you hac your way through the jungle—it takes a long way to get anywhere. In statistics it’s like Chile, you get off the beach and then you’re in the mountains already, so really anyone can make contributions to statistics and data science right away using using their subject matter knowledge so it it really is a very open field and I think that’s a great thing.

The whole talk is a lot of fun! I get into the uncertainty principle—real physicists will get annoyed at that point, but please just think of this as an analogy that helps us understand statistics better—and then there’s a discussion of Bayesian inference. In talk about how by default my prior information is skeptical:

My prior is that I doubt most things are gonna work and the funny thing is that allows you the freedom to explore. If you come into the world as a naif and you say, “I’m gonna believe anything I see that’s statistically significant,” then you’re gonna have to really restrict the range of problems you work on because otherwise you’re gonna start believing everything and then you might as well just start publishing a scientific journal um that’s kind of a joke that last thing because scientific journals publish all sorts of things they shouldn’t . . .

Conversely though if I’m a skeptic I can freely to explore knowing that it will take more information to convince me, so skepticism is really like in some sense a pivot it’s a way for us to use more information . . .

I like this paradoxical idea that an empty openness makes it difficult to operate, whereas a probabilistic skepticism allows us to effectively learn from new information.