26 January 2008

What Use is Game Theory?

In my day job I am a political theorist in a department lopsidedly committed to high-tech game theoretic modelling. This places me in an awkward position because, while I think models are really interesting and useful, I think many of my colleagues are bewitched by a faulty view of why that is the case. In that regard my colleagues hardly are idiosyncratic; they subscribe to much the same faulty view of what makes models useful as the majority of political scientists (regardless of whether they are advocates or critics of the model-making enterprise).

What follows is a passage from a "Freakonomics" column that Steve Levitt wrote in The New York Times a couple of years ago after Tom Schelling finally won the Nobel Prize.
"To my mind, [Thomas] Schelling represents the very best of game theory. He was a pioneer in the field, a man of ideas. Unfortunately for game theory, the simple ideas that are so alluring were quickly mined. What followed was less interesting. Modern game theory has become extremely mathematical, notation heavy, and removed from everyday life. Many of my colleagues would not agree with me, but I think game theory has failed to deliver on its enormous initial promise. I’m not the only one who feels this way. I was recently speaking with a prominent game theorist. He told me that if he knew what he knew and he were just getting started in the profession today, no way would he be a game theorist."
I've italicized what I think is the interesting, unexamined part of the passage. What is game theory (or any other technique of formal modelling) good for in social science? I have views on the topic that are directly indebted to Tom Schelling's writings. They run counter to the canonical (and, I think, if not downright silly, surely unduly narrow) view of the matter among political scientists - namely that we use models to generate predictions/hypotheses that can be tested empirically. I don't want to go so far as to say that that view is not possible; but I do think social scientists very rarely actually approach in their practices. That by no means implies that models are useless. But it does raise the question underlying Levitt's complaint: "What, precisely, is the "promise" of game theory?"

According to Schelling: "A model is a tool." On this instrumental view a model is useful to the extent that it "gives us a head start in recognizing phenomena and the mechanisms that generate them and to know what to look for in the explanation of interesting phenomena." What is interesting, of course, depends on our purposes. And the causal mechanisms underlying recurrent social, economic, and political events (phenomena) are typically not directly observable. Hence we make models to help us depict and examine and talk about causal mechanisms. In that sense, especially when they have a catchy name, models provide "help in communicating."

Notice that Schelling straddles the divide between "realist" and "instrumentalist" views of social inquiry. In so doing he adopts a broadly pragmatist stance. A pragmatist treats models (like concepts and theories and principles and institutions and so forth) as instruments that we use for making our way around in the world. But she also rightly insists that there is a "real" world to be navigated and that that world consists, in part, of causal forces. So, like pragmatists, Schelling thinks of models as tools we use to navigate the real (natural and social) world. It is no surprise that he is not only a superb theorist, but intimately concerned with a range of crucially important policy issues.

Labels: , ,


Blogger Jeffrey said...

I'm not entirely sure what you mean by pragmatism in this context. I was in one of your ICPSR sessions back in 1997. It remains one of my favorite classes at any school, because it helped me connect for the first time two disparate strands of knowledge: rational choice models and classical political theory. For example, you connected Arrow's Theorem to Rousseau's concept of the general will. I still mention this when I teach Rousseau today. Following this class, I began thinking of game theoretic models as being "what if?" simplifications of the world. Thus, the repeated PD becomes a Hobbesian world, and when we compare Locke to Hobbes we have to ask whether the preferences of people in the state of nature have changed. Other philosophers state the collective action problem, which we can then formalize. This is a predictive process, but not really a testable one. It's really about making sure that our assumptions about how the world looks (or ought to look) do indeed imply the political theories which we employ to understand the world.

27 January, 2008 05:04  

Post a Comment

<< Home