OrgTheory.net has a great post today assessing the merits of game theory. The author makes some excellent points. In particular, he points out that some of the more interesting outcomes we see in the social sciences are off-equilibrium–inherently, then, these are the ones that we can’t explain with game theory. Additionally, and more generally as a criticism of rational choice, individuals don’t necessarily make decisions that conform to what we typically think of as the rules that describe rational actors.
Check out the post at OrgTheory.net, and the preceding entry by the same blogger (Michael McBride) about game theory’s potentially misleading name, and meet me back here after the jump.
First, the point about off-equilibrium outcomes is great. When we have defined pretty effective rules by which social interactions tend to follow, it’s always interesting to try and explain why it doesn’t happen like all indicators say that it should. I don’t think this is necessarily a drawback or hurdle of game theory, though. When we enter the realm of social sciences, we forfeit the ideal of perfectly predictable outcomes. Not that we shouldn’t use methods we think can help us explain as much variance in social phenomena as possible, but game theory is only one tool at our disposal.
Second, the notion that real people aren’t always rational actors is a common criticism of rat choice (and of game theory, by extension). In the specific context of rational choice, I would counter that a strict application of rational actor assumptions is actually not very useful. McCubbins and Thies’ 1996 paper in the Japanese journal Leviathan (also available from SSRN) advocates a slightly looser definition of rationality. The authors reduce the rational actor assumptions to their most basic parts and set forth three basic tenets of rational choice models:
1. Rationality — individuals make reasoned decisions.
2. Component analysis — only small parts of the system are important in predicting any type of human behavior.
3. Strategic behavior — individuals ‘play games’: and take into account what other
individuals are likely to do before making decisions.
This last point draws us into a third reaction to the McBride’s post: game theory is inherently interactive. We can’t expect to see perfectly rational behavior by individual players in a game when we extract them from the context of that interaction. The purpose of game theory is to help us elucidate strategic interactions among individuals, and we don’t really care where their preferences come from–only that they act upon their separate preferences within the game. In the more general discussion of the rational choice research program, the rational actor assumptions are also eloquently summarized by McCubbins and Thies in pages 3 through 11 of their paper.
McBride’s other post is about the potential disservice to game theory done by its own name. While it perhaps doesn’t give it the weight of a sophisticated method used in most of the social sciences, I think the name of game theory is just fine. Rather, "rational choice theory" is more problematic. First, it’s not a theory; it’s a set of assumptions. Second, it naturally lends itself to the above criticism that individuals aren’t always "rational" in common parlance, and implies that only those who believe that individuals are always rational, as such, have any business taking this perspective on strategic interaction. McCubbins and Thies suggest "positive political theory." I still have problems with "theory" as opposed to, perhaps, "research program" or "perspective," though, but this renaming helps the authors make their point that all social research inherently adheres to the first two tenets of PPT listed above. Add the third tenet and you have instant game theory at your disposal.
I really appreciate these sorts of posts, primarily because they get my mind working along a dimension that is removed from my day to day research. I am a consumer (and I’d like to think of myself as a producer) of multi-method research. Just the same, I get wedded to my methods of choice and am too quick to dismiss other approaches.
But I often find myself an apologist for game theory. Indeed, I find myself arguing against the positions I myself expressed in opposition to the approach some years ago.
Cynthia, to riff off of your point about off-equilibrium results. I think (and maybe game theorists, generally, would agree) that these off-equilibrium results are wonderful results (not as damnatory as the critics would argue). They are wonderful not because they fit the rigid and clearly identifiable strictures of rational choice assumptions, but because they deviate in regular ways. The fact that cooperation exists (even in PD situations), that there is evidence supporting prospect theory, that there is a well-developed and still growing field of behavioral law and economics, that we see a systematic over-optimism that plagues human psychology; these are wonderful research agendas that have game theory and rat cho to thank for their insights because they were founded in recognition of rat choice’s empirical limitation. The fact that game theory and rat cho are incomplete is not as damning as some might think. It can be a wonderful springboard for a fruitful research agenda. I don’t think this is a nail in the rat cho coffin.
Upon reading your post, I immediately thought of theories of light. Is it a wave? Is it a particle? It is neither (?) but behaves like both. Is it too far of a stretch to think that human behavior is neither perfectly rational, nor devoid of rational self-interest? Might the answer depend on what we are asking? And if the question we ask is not fully answered by our initial expectations, does this not provide us with tremendously valuable information?
I think the existence of off-equilibrium behavior is a product of “model misspecification” rather than a problem with game theory. Just like any statistical model, if you mispecify a game theoretic model, the results will be inaccurate. Ways to misspecify a game theoretic model: (1) Not including all relevant actors; (2) Not correctly identifying all the choices “real world” actors make; (3) The sequence of actions are wrong; (4) The payoffs are wrong; (5) Our assumptions about information are inaccurate. If we make any of these mistakes, the model can produce unrealistic equilibria.
I the one shot PD is a perfect example of model misspecification when applied to actors who continually interact. Since “Defect, Defect” is the only equilibrium, it cannot explain why two actors in such a scenario would cooperate. Once we take into consideration that actors have repeated interactions in the PD, we can conclude that cooperation is possible.
Quote I heard today that reminded me that we have to make sure we identify all the relevant choices people make when we create game theoretic models:
“Everyone comes to a fork in the road. Be spontaneous and walk through the middle.”