I received the following email:
I’m a PhD student in Political Science at [an excellent U.S. university]. My fields of concentration are comparative politics and political methodology. As someone interested in modeling data from your perspective – and lacking adequate training – I’m writing you asking for advice.
I’m about to start my third year of study. So far I’ve taken the two years methods’ sequence inside the program plus a class on hierarchical models based on your book. This seems a lot but I’m afraid it is not. The sequence is good but somewhat slow: For instance, just at the end of the first year we studied regression analysis seriously (with few matrix algebra) and the class of probability we barely calculate any integral. In the second year we had a excellent applied class on maximum likelihood but, again, with no math (though lot’s of programming). I believe we have some exposure to R and how to program our own models. Yet, we lack a better understanding of what is happening inside the computer. We are also probably unable to really develop new models, if needed. Besides, I’m pretty sure that the training I’m getting doesn’t qualify me to teach statistics in a top political science program.
Thus my basic questions are: How can I improve my methods training? What are the most fundamental stuff that you recommend political science students to learn? Which type of extra class? More math and probability? More programming experience? Switch programs and go to places like Rochester or St Louis, which are really focus on methods? Graduate fast and a try pos-docs like yours before getting the first job?
My reply: Take some probability and theoretical statistics courses from the statistics dept. (Rochester and St. Louis are fine, but the stat dept at your university is excellent.) With your motivation, you’ll be well-prepared to learn some stuff.
I agree with Andrew, take some courses in the stats department. I started the PhD program in political science at UCLA, and after taking the first year stats sequence within the department (which was actually very good), I asked Prof Steve Ansolabehere a similar question and he said "you have an undergrad math degree, go take the stats courses in the math dept" (the stats division was in the math dept at that time). That was some of the best advice I ever received and I ended up switching programs a couple years later.
Maybe some readers of this blog can comment on his or her experience in getting methods training…
Sounds like a good idea to take some courses in the stats department. My brother is also interested in a PhD program in political science. His only concern was with the economy is performing so poorly in term of job opportunities what are likely jobs to be in demand with his PhD?
gail,
program evaluation. the poorer the economy, the more spending on social programs that are required to be evaluated.
I have some thoughts for this student: the first two years' sequence in the stats dept is great, but the econometrics sequence is great too. The perspectives tend to differ (at least at Columbia), and they complement each other. Have a broad perspective: try to follow applications in epidemiology, psychometrics, education stats. The deeper I've gone into methods, the more I've appreciated cross-fertilization across disciplines. Once you start to feel an intuitive grasp of things, try to become involved in statistical consulting. There's been no better way for me to find out how much I don't know, and thus how much I still have to learn, than helping other people solve their problems.
I'd suggest some different things.
First, you're wrong in thinking that if you don't or can't come up with new empirical models, you're unqualified to teach methods at a top program. Being qualified to teach the first methods course at a top program or elsewhere means having a decent command over basic probability and related matters, and primarily just having enough patience not to bite the heads off of grad students who don't immediately get it. Being qualified to teach the applied regression course means only having a good command of OLS, its drawbacks, its corrections, and so on. Being able to come up with new empirical models only affects your ability to teach topical seminars on methods.
Second, if you intend to be a political scientist, you might find your interests better served by honing your skills at research design, drawing inferences, figuring out which ancillary hypotheses are testable, and so on than by acquiring even bigger hammers to hit the same obvious nails with.
Third, a potential problem with statistics programs is that the mind-set seems fundamentally different, much more organized around fitting data than testing hypotheses drawn from explicit theories.