Rationality has something of a bad name these days. Unlearning Economics has an interesting comment on the critique of rationality in economics, and Marc Bellemare has noticed as well. If rationality is problematic for economics, then it is doubly so for the rest of the social sciences. I want to argue in this post that two common critiques of rationality in the social sciences are not as powerful as they might seem, and to highlight a third that has been neglected. I do this not to bury rationalism, but to help rescue it from unhelpful critiques while highlighting where there is constructive work to be done.
To illustrate these points, let us consider the case of the Juggalo.
A juggalo is hard to define, but the Wikipedia link above does a pretty good job. Juggaloes are particularly useful for a critique of rationality in the social sciences. They seem to embody many properties that seem inconsistent with what it means to be rational.
So on to the critiques! Marc describes the popular critique of rationality as employed in the social sciences like this. “Bah! Humbug!…we know people aren’t rational because they do dumb things, like, all the time…”. Let’s call this the 1st Critique of Rationality: people often do things that seem hard to square with the concept of rationality, like wear HatchetGear. Marc reminds us that “dumb” is not inconsistent with “irrational.”
In economics, individual rationality rests on two assumptions:
- Completeness: People can compare two alternatives x and y, and determine whether they prefer x to y, y to x, or whether they are indifferent between y and x, and
- Transitivity: If someone prefers x to y and y to z, then that person prefers x to z.
That’s it; that’s all. When an economist assumes individual rationality, all she is saying is that people’s preferences are complete and transitive.
The point? Rationality doesn’t tell you that a Juggalo can’t be a Juggalo because some observer doesn’t think that makes sense. If our Juggalo friend here has three choices
- Wear HatchetGear all the time (A)
- Wear HatchetGear only at the Gathering of the Juggaloes [NSFW link] (B), and
- Never wear HatchetGear (C)
and prefers A to B and B to C and A to C, then we call that rational.
There is a second common critique of rationality, which is a bit more powerful. This is a behavioral critique of rationality, identified most commonly with both behavioral economics and the pop social science of Malcolm Gladwell and others. Essential here is the idea that neoclassical utility functions do not hold up to careful empirical scrutiny. People are not particularly selfish and quite clearly have other-regarding preferences (see dictator games, norms). Nor do they calculate expected utility particularly well even when they are trying to be selfish (prospect theory, favorite-longshot bias, heuristics and framing). There are many, many other examples. For economists, in particular, the implication is that economics ought to be more like psychology; a good critical review is available from Douglas Gale. Call this the 2nd Critique of Rationality: humans are not selfish calculating machines.
I claim that this 2nd Critique is actually nearly identical to the 1st Critique. The objection is not to rationality, but to a particular set of utility functions. There are many utility functions that one can write down allow other-regarding behaviors, psychological biases, and others. Let’s return now to our Juggalo friend. Maybe he wants to wear HatchetGear all of the time (he prefers A to B), but it turns out, he chooses to only wear HatchGear at the Gathering of the Juggaloes (he appears to choose B over A). Rationality fails?
Or not. Let’s consider that our Juggalo friend has some sort of preference for conformity. So, his utility function looks something like this.
- Wear HatchetGear all the time (A) >> Wear HatchetGear some of the time (B) >> Never wear HatchetGear (C)
- Wear clothing that is the same as everyone around me (X) >> Wear clothing that is different from everyone around me (Y)
There is a parameter λ (0 < λ < 1) that describes the weight that the Juggalo places on (1) versus (2). An choices A through Y have concrete values, so that we can say things like the Juggalo values A twice as much as B. For certain combinations of λ, A, B, C, X, and Y we will therefore indeed observe that our Juggalo friend rationally chooses to go to work in street clothes, and to go to the Gathering of the Juggaloes in HatchetGear. (We might naturally, then, interpret the Gathering of the Juggaloes as a coordination device.)
There’s nothing irrational about this, by the definitions above. More generally, I bet that any particular “behavioral” utility function is just as unlikely to be “the” utility function than is the hyper-selfish and hyper-calculating utility function implied by some critics of rationality. Ariel Rubinstein has a nice comment that matches well with this perspective. It, along with other useful exchanges on behavioral economics, are available here. They apply more broadly to analogous critiques of rationality in other social sciences. The critique of hyper-selfish and hyper-calculating individuals is not really a critique of rationality itself.
But I do think that there is a critique to be made of rationality in the social sciences. I’m not so concerned with rationality itself as I am with the construction of rationalist arguments to explain observed behavior. Because rationality is actually a pretty thin concept—complete and transitive preferences, and that’s it—and because preferences are unobserveable, it is too easy to construct rationalist arguments. Simply define the observed behavior, choice, or phenomenon as DO-X. Define an unobserved behavior as NOT-DO-X. Claim that the preference relationship is DO-X >> NOT-DO-X. Presto: a rationalist argument that accounts for any observed behavior.
The reader will notice here that to rationalize the Juggalo, as above, I did exactly this. I just defined “be a Juggalo” as “wear HatchetGear,” then worked up some preference relations that would generate the observed behavior that I wanted to explain.
I am a fan of rationalism, but this is bad social science. For a fuller and more satisfying explanation of the Juggalo, we need two additional steps. First, we need to understand preferences and preference relations in a non-tautologous way. This is hard, no doubt, and revealed preference theory (which allows us to infer preference relations from observed choices) is one of the crowning achievements of 20th century social science, but the further we get from choices among bundles of consumer goods the less powerful it is. Second, we need true alternatives to the rationalist approach. What if it’s just wrong to think of some observed behaviors as the results of choices? By this I mean more than people do not consciously make choices among alternatives, and more like choice and preference do not describe some social behaviors. What if that were true? How would we know? What would follow? These are questions that rationalists ought to entertain seriously.
Doing these two things does not mean abandoning rationality. But it does mean knowing the limits of what rationality buys you.
Matthew Kocher June 29, 2012
I disagree with some of what you wrote.
1) As a matter of intellectual history, it is not true that rationality just is completeness and transitivity. One of the big problems with rational choice theory is that the set of assumptions is a fast moving target, tailored to the problem at hand. To give one of my favorite examples, in Fearon’s Rationalist Explanations for War, one of the most influential paper in IR in the last 2 decades, rationality is assumed to include the requirement that any 2 individuals will form the same beliefs under uncertainty. The paper’s interesting results mostly flow from this patently absurd and empirically false assumption.
You focus mainly on the distinction between what have been called (by whom exactly I can’t remember) ‘thick’ and ‘thin’ variants of rationality. The latter makes substantive assumptions about preferences, while the former does not.
2) Critiques 2 and 3 should really be thought of as a single critique. In effect, rational choice is on the horns of a dilemma. Go for a thick version and you risk both endorsing false assumptions and failing to explain a great deal. Go instead for the thin variant and you risk — or in my view actually fall prey to — triviality and unfalsifiability (for the reasons you specify in your 3rd critique).
3) Actual people routinely violate both completeness and transitivity.
4) I think there is an even deeper critique, stemming from the happiness research. Actual people turn out to be quite bad at predicting what will make them happy. In other words, there is a deep problem about the very idea of a preference or desire.
lorenzo June 30, 2012
“We need true alternatives to the rationalist approach. What if it’s just wrong to think of some observed behaviors as the results of choices? By this I mean more than people do not consciously make choices among alternatives, and more like choice and preference do not describe some social behaviors. What if that were true? How would we know?”
I think you cannot say that thinking of a given observed behavior as the result of a certain rational choice is true or false. What you can actually say is that it’s more or less useful as for theory-building. We should obviously care that our approach (be it rational or not) is based on a realistic understanding of human behavior, but only to the extent that this piece of information helps us produce more plausible explanations.
Tom June 30, 2012
Thanks, Matt, for commenting. I agree with some of your comments, but not all of them.
As a general point, you are right that there exists a wide variety of scholarship that makes many more additional assumptions about what “rationality” means than the basic building blocks of complete and transitive preferences. But what they call “rational” should not undermine the building blocks, any more than the idea that rationality = hyper-self-interest should. I can’t remember who’s responsible for the thick-versus-thin distinction either, but I agree that this captures this distinction well. Thick, I think, makes lots of substantive additional assumptions about what kinds of preferences count or what utility functions look like. I obviously don’t think that’s productive, or cumulative.
I really think that critique 2 and critique 3 are different. CR2 is a critique of what utility function can look like. CR3 is a critique of how we construct arguments. But nonetheless the description of the dilemma is really a good one: that’s a real insight.
On the violation of completeness and transitivity: I know that we find such violations in the lab. But I didn’t raise this critique because I don’t think that it ends up mattering a huge deal for explaining real world phenomena. It’s like vote cycling, or Giffen goods. I just don’t see any golden opportunity to apply that critique to a real issue.
I think that my last point—what are the real alternatives to preferences or desire—is consistent with your final critique. But I can’t bring myself to cite the happiness researchers.
Tom June 30, 2012
Thanks for reading and commenting, Lorenzo. That’s a common response, to say that we should look at assumptions about rationality as useful or not. I agree with this as a general proposition, but I think that my final critique is that theory-building is tough if you can always work up a rationalist explanation for anything.
There’s also an alternative perspective that says that if rationality does not match the internal mental states of decisionmakers, then it can never be “more plausible explanations.” Perhaps you mean “more accurate predictions?” That’s the “Methodology in Positive Economics” perspective, which does not care about the plausibility of the explanation, but rather predictive power alone.
lorenzo June 30, 2012
Thanks for your reply, now it’s clearer. I guess that the point here is what you consider as a ‘plausible’ explanation and it seems to me you want rationalist explanations not to focus only on predictive power (aka the ‘covering-law account’), but to take into better account also adequate causation (aka the “mechanism-based account’).