social science can actually predict stuff

Last week, Northwestern poli sci prof Jacqueline Stevens wrote a NY Times op-ed where the main argument is that political science is lousy at predicting things. You can read responses from within the field, like the one at the Monkey Cage.

Stevens makes some good points. For example, social scientists will often dress up common sense in fancy models and jargon. But I do have one major bone of contention with Stevens’ analysis, the social sciences are actually not bad at predicting some things. For example, I can predict, in the future, that college graduates will make a lot more money, on average, than people who didn’t go to college. We’ve known this for decades. This trend will continue. In political science, there are some models that actually predict things pretty well. My favorite is the presidential votes share model, where incumbent two-party vote share is very strongly predicted by the economy.

So what gives? Does a Northwestern professor of political science not know her own field? I can’t speak for Professor Stevens, but I’ll offer a few reasons for Stevens’ skepticism. This may have to do with the lingering conflict in political science over quantitative method.  There’s a bigger issue than arguing over method. Stevens picks on areas of political science where prediction is insanely hard. If you have a system with a few moving parts, prediction may be possible (e.g., predicting the correlation between education and income).

Stevens uses examples like the collapse of the Soviet Union and the rise of al-Qaeda. Global politics is a massively complex system. There’s a lot we don’t know and a lot that is hard to measure. It also has billions of moving parts and shows sudden shifts. Physics is the same way. Two billiard balls? Yeah, we got that. Predicting the next earthquake? Umm… still working on that.

Poo-poohing political scientists for not foreseeing al-Qaeda is like saying that medical science is stupid because doctors didn’t predict AIDS. Superficially true, but so very, very misguided.

Adverts: From Black Power/Grad Skool Rulz

Written by fabiorojas

June 26, 2012 at 12:02 am

3 Responses

Subscribe to comments with RSS.

  1. If one is really interested in the philosophy underpinning her (misunderstood) critique, one could do worse than look at a working paper uploaded at her web site ( “A Popperian Reading of Civil War Studies.”

    In this paper, Dr. Stevens uses a mis-congealed approximation of Popper’s falsification doctrine to criticize scholars who attempt to construct models that explain (not necessarily predict) civil war. That is, she criticizes various models based on the fact that they can be falsified by counter examples. For example, some scholars construct a model of civil war that does not find ethnic tension to be significant- Stevens states that ethnic tension certainly was a factor in conflicts such as the Bosnian civil war and Rwanda- thus, this model is falsified.

    There are numerous problems with this approach: 1) Popper himself did not consider social sciences to be sciences (they occupied the realm of pseudo-science); thus, they were not amenable to the falsification criteria. 2) The scholars creating these models do not claim they are scientific in nature as if they were physicists; they are trying to find key factors that explain why certain events happen- not necessarily predict in the future when civil war will break out.



    June 26, 2012 at 12:19 am

  2. Predicting discrete outcomes from an infinite set is very difficult. A lot of good prediction in the social sciences, for example in demography, predicts continuous individual outcomes within a finite range, such is the fertility or mortality rates. We are very good at that. For example, I recently predicted the number of births with 98% accuracy, simply by guessing that the birthrate would not change ( That is a cheap prediction. More astonishingly, with a simple formula I predicted the number of children named Mary last year to within 22 ( But if I tried to predict *which* family would be the one to name their son Adolf next year, my odds are not so good.


    Philip N. Cohen

    June 26, 2012 at 2:16 am

  3. Fabio, you are right, of course, that social science can predict things. I would argue that we aren’t investing enough effort in prediction precisely because it is really hard. We can, and should, be quantifying how good our predictions are in various realms. That’s why I’m involved with


    Michael Bishop

    June 26, 2012 at 3:06 pm

Comments are closed.

%d bloggers like this: