Wednesday, 30 April 2003

Defending the legacy of Warren Miller

Jane Galt has a lengthy post in which she makes the following statement:

The humanities simply doesn’t have this rigor. In some cases, such as literature, you really can’t, although you can certainly be more rigorous than many of the programs devoted to exposing the obvious truth that Shakespeare and company did not have the same racial and gender sensibilities as 21st century Americans, yawn. In other cases, such as sociology and political science, it’s possible that you could, but don’t yet. That’s why discussions in those courses tend to revolve around the speakers’ opinions on human nature, interesting and possibly right but very difficult to either prove or falsify.

James Joyner took Jane to task for lumping political science in with the humanities, and in comments there the validity of political science as a science is being piled on by James and Steven Taylor. My comments at Jane’s place were a bit less temperate, but some of the other commenters got under my skin too; here’s what I said:

  1. Yes, there is a giant pissing match in our discipline between the empiricists and the non-empiricists. Coupled with that is an internal war among the empiricists between the quals and the quants and the game theorists. But then again political science has really only existed as a separate discipline from history, sociology, psychology, and economics for about 50 years (despite the 100-year pedigree of the APSA), and it’s still trying to figure out what to do with those disparate heritages.

  2. Not everybody plays with 2×2 contingency tables and single-variable explanations of political phenomena. We have this neat concept called “multiple regression” that can deal with more than one independent variable these days; we use that sometimes.

  3. Explanation is more important than prediction in the long run. I’d rather have a wrong prediction than a wrong explanation. If we just wanted predictions, we’d do what insurance companies do: stick a bazillion independent variables in the model and stepwise-regress it. It’ll predict great to your original dataset but (a) will blow up if you apply it to anything else and (b) won’t make any sense anyway (“hey, it turns out people with purple cars and limps have more accidents than others”; yeah, so?). By contrast, I can tell you why almost 400 members of Congress voted for articles of impeachment and procedural motions leading up to it, and I did it with 9 independent variables—and some of the explanations would be quite surprising even to those who followed the public debate and talking heads.

  4. There is no number 4.

  5. I can build you a very simple probit model today that will predict how most voters will behave in the next congressional election, based on nothing other than their demographics. Heck, I already have; read this paper. It ain’t Galileo dropping stuff off the Tower of Pisa (apocryphal), but it’s pretty good for a then-2nd year grad student playing with a Heckman selection model (ok, not so simple) and 50-year-old theories of voter behavior.

Now, granted, there are a lot of snake-oil salesmen running around pretending to have all the answers. Some of them (Larry Sabato *cough*) are in my discipline. Some of them (John Lott, Paul Krugman) are in more “respectable” ones. But just because some Ivy grad whose only exposure to The American Voter was that it happened to be on a bookshelf in someone’s office in his department can’t tell you who’ll win a local school board election doesn’t mean that nobody can.

For the record, no particular Ivy grad was being singled out above. I love all Ivy and non-Ivy grads.