Dan Luu writes,

The short version of my question is: A lot of the papers I’ve been reading that sound really interesting don’t seem to involve economics per se (e.g., http://home.uchicago.edu/~eoster/hivbehavior.pdf), but they usually seem to come out of econ (as opposed to statistics) departments. Why is that? Is it a matter of culture? Or just because there are more economists? Or something else?

If you don’t mind getting random unsolicited email just because you write a blog, then there’s the longer version of my question.

I’ve been reading your blog for a couple of years and this post of yours from a month ago got me thinking about studying statistics. My background is mainly in engineering (BS CompE/Math, MS EE). Is it possible to get accepted to a good stats program with my background? I know people who have gone into econ with an engineering, but not statistics. I’ve also been reading some epidemiology papers that are really cool, so statistics seems ideal, since it’s heavily used in both econ and epidemiology, but I wonder if there’s some domain specific knowledge I’d be missing.

I’ve noticed that a lot of programs “strongly recommend” taking the GRE math subject test; is that pretty much required for someone with an unorthodox background? I’d probably have to read a topology and number theory text, and maybe a couple others to get an acceptable GRE math score, but those don’t seem too relevant to statistics (?). I’ve done that sort of thing before – I read and did all the exercises in a couple of engineering texts when I switched fields within engineering, and I could do it again, but, if given the choice, there are a other things I’d rather spend my time on.

Also, I recently ran into my old combinatorics professor, and he mentioned that he knew some people in various math departments who used combinatorics in statistics for things like experimental design. Is that sort of work purely the realm of the math departments, or does that happen in stats departments too? I loved doing combinatorics, and it would be great if I could do something in that area too.

My reply:

1. A lot of interesting papers are being written by economists now, but I’m also a big fan of papers by psychologists. Psychologists will cram the results of ten different experiments into a single article. Instead of just exploring a question with a single dataset, they’ll really try to figure things out, running different experiments to consider different possibilities.

As to the question of why don’t statisticians write more interesting applied papers: I don’t know. I suppose people go into statistics because they like math, not because they care about any particular application. I’ve seen some pretty crappy applied work by well known and respected statisticians–people whose theoretical work I respect a lot. The best scientific work by statisticians is probably not social science but rather in biology, the most famous example being R. A. Fisher’s model unifying the genetics of discrete and continuous traits (I hope I’m not garbling that too much). Also, yeah, there are more economists, and they’re in a more competitive field, so maybe they put more effort into promoting their own work. (Yes, I know, I’m one to talk, seeing as I promote my own work all the time, but maybe I’m not typical of statisticians.)

2. I think it’s easier to get into a top statistics Ph.D. program than to get into a top econ program. I’m a big fan of stat Ph.D. programs, especially if you can work with a good advisor and do some interesting stuff. I don’t know the deal with the math GRE. We used to require it at Columbia but we’d get a lot of applicants who hadn’t taken it, and we’d consider their applications too, so it seemed only fair to diminish it to a recommendation rather than a requirement.

Finally, yes, there’s combinatorics in classical experimental design but I don’t know that much is done in this area anymore. But there is some work, both in probability theory and in applied statistics, on models for networks and trees, for which the mathematics is highly combinatorial in an interesting way. Here are a couple of papers as an example (one of which is in an econ journal!): The mathematics and statistics of voting power and Forming voting blocs and coalitions as a prisoner’s dilemma: a possible theoretical explanation for political instability.