Reading back over my last post a few weeks ago, I realized that I sounded like a data-mining zealot or a pure empiricist. The post reads like it was written by someone that has no use at all for Aristotle or Huntington, or any of the other great political theorists… someone who believes that he can somehow pull pure truth out of a collection of social measurements.

No. That’s dumb. In fact, I think that data is worse than useless without theory. The fanciest statistical package in the world cannot make sense of a dataset by itself. Obviously, somebody needs to be gathering data, selecting an appropriate model, and choosing which independent variables to focus on, but that is still an overly facile representation of the problem.

Most packages at this point have automated procedures for selecting the “best” model according to some informational criterion. These come in various flavors and levels of complexity, but almost always boil down to a measurement of model fit–or how closely the mathematical function you’ve cooked up approximates the observed phenomena.

A perfect fit is laughably easy to achieve. All you need to do is throw every explanatory variable you’ve got (at the theoretical limit, this means every measurable quantity in the universe) into the bin marked “factors that somehow cause phenomena xyz” and you’re done. Your graph will connect every single data point you’ve got in a stunning display of super-squiggliness, you’ll have an R-squared of 1, and you’ll be published in every top journal in the land.

Not hardly (and that’s not just because more sophisticated goodness-of-fit measurements penalize you for such “chance capitalization”). You need to be able to explain the process by which you arrived at your magic recipe. In other words, why should we believe that factors a, b and c (and perhaps a*c) cause xyz, and what does that sequence of events look like?

This presents a chicken-and-egg quandary if we are really trying to discover “the truth” about xyz’s causes: if we don’t know what we’re looking for in the first place, we’ll have no idea what data to gather. And when we’re dealing with social outcomes, simply “measuring everything” doesn’t work. These are complex phenomena that demonstrate exceptionally high-dimensional causality. By discipline, and in causal order, political science is built out of mathematics, physics, chemistry, biology, psychology, and economics. We are still working on finding the elementary building blocks of matter; how are we supposed to know what to watch for five or six orders of magnitude later?

Note that I am not making an argument for extreme reductionism here. We’re not supposed to know, we’re supposed to make educated guesses. That’s what a theory is: a carefully considered, logically consistent guess, which in turn ultimately boils down to…. intuition. Yes: when trying to explain what individual people, or crowds of people, or millions people living in a state have done (or are going to do), and why they’ve done it (or are going to do it), I think our best bet is to have a flash of insight.

Where does that kind of insight come from? I have no idea, but reading lots and lots of work by the smartest, wisest, and most insightful people in history is a fantastic place to start.



Political science grad students tend to spend their summers sharpening methods skills at workshops (like SWAMOS, which I attended last summer, or ICPSR), writing or tuning up papers for publication, or simply taking a short break from the vicissitudes of graduate school. It’s unusual to see one stuck in a teenager-packed lecture hall four days a week desperately scribbling down notes on intro calculus and computer programming material. Stranger still for that poor bastard to be there voluntarily. Yes, friends, I am said bastard!

Why would I be subjecting myself to this misery? I have my reasons, above and beyond a well-documented masochistic streak. I’ll start with the general observation that political science and economics, which have always been related, are well into the process of merging. The cutting edge of each discipline is slicing deep into the other.

In 2012, a lot of the top job candidates in political science have an MA in statistics and are  more comfortable building formal models than they are discussing Aristotle–or Huntington, for that matter. This is a natural consequence of the triumph of rational choice theory: departments that routinely hire economists alongside (and often before) political scientists. The writing is on the wall, all over the floor, and spelled out in the sky via smoke-emitting biplane: we need to be data and math people now, or else.

Meanwhile, an increasing number of economists are interested in topics that would have been considered pure political science a few decades ago: political institutions, civil wars, ethnic group salience, legislative politics, and on and on. Although I’m not an economist, my sense is that the old core of the discipline, finance and labor studies and the like, is no longer considered very interesting or sexy from a jobs standpoint.

Economists in general have a better command of advanced quant methods than any other kind of social scientist, and as such, they are well equipped to address and conquer any topic they choose. Political scientists and anthropologists and sociologists of the old guard like to say that economists have no knowledge of theory, or of conditions on the ground in rural Bolivia, or what have you. Maybe not.

But which is easier to acquire–advanced quant or subject/area knowledge? How about advanced quant or political theory? All three are difficult, to be fair, but advanced quant is in relatively short supply. Furthermore, the state of the art in quant is accelerating away from standard-issue quant. So, if you want to be a top-shelf quant guy now, you don’t just need to be able to use statistical packages with competence; you need to be able to program new ones yourself.

Finally, if one looks outside the academia to see what kinds of skills are valued–and many of us graduate students are, given the putrid state of the academic job market–it’s pretty clear what employers are looking for. Big Data is the big dawg. My guess is that Big Modeling will recover from the financial crash sooner or later to form a two-dawg axis.

That there is a high-level explanation of why I’m currently doing what I’m doing to myself. In slightly more detail, I’ve decided that I need another 24 months of math and programming (at a minimum) to pull off my new dissertation idea. I honestly don’t know if I can combine video games and political science in a way that will a) answer interesting political science questions b) in a way that other political scientists will buy while c) being fun to play, but I’m sure as heck going to try!

And if it doesn’t work, that’s okay too. I’ll go start something up in silicon valley.

The Sims: Political Science Edition

Recently, I’ve been considering whether it might not be possible to combine the two areas I am intensely interested in, video games and political science, in a way that won’t get me kicked out of graduate school and might even result in an academic position at some point (or at least won’t totally foreclose the possibility)!

To be sure, political science has become a lot more receptive to advanced computational modeling in recent years, following developments in the natural sciences and more recently in economics. Yes indeed! Some of the more freewheeling practitioners of the dismal science are now writing papers about currency farming and auction house behavior in World of Warcraft.

And, on the flip side of the virtual coin, there are a goodly number of academic refugees now employed in Silicon Valley as big data miners, virtual behaviorists, and the like. The demand has become particularly fierce on the new frontier of video gaming, which lives on the Internet and is fueled largely by “social graphing” and “in-app purchasing.” Zynga, an online games company that makes millions of dollars operating virtual fiefdoms, is hiring data analysts like crazy.

Video games are now computational models that are designed to produce fun and mineable data.

But I digress. A professor of international relations I very much respect, Art Stein, likes to say that the cutting edge in political science methodology runs about a decade behind economics, which in turn runs a decade behind physics and biology. That means that if I finish my PhD around 2016, I might be in very good shape!

The potential applications of game-based simulation methods in political science are endless. This is particularly true in international relations, where direct experimentation is effectively impossible; for instance, we’re not likely to randomly distribute nuclear weapons to countries throughout the world anytime soon. But we can certainly build a model of nuclear crisis and run it tens of thousands of times on the internet, twiddling the knobs to see what comes out.

Is external validity a problem for this kind of experiment? Most definitely. But there are many well-documented issues with scientific inference from observational data (selection bias, anyone? how about endogeneity?), and formal modeling, while appealingly parsimonious, is even more abstract and much less able to deal with the complexity that characterizes the real world. The analytical solution space to N-person cooperation games melts down pretty fast above a handful of players.