June 24, 2004 | 16 comments

Occasionally there is some odd comment here or there on this site alluding to “rational choice” models. Now almost nobody in economics uses this phrase, because you don’t need a word to describe what everyone is doing. Yet rationality seems to get some non-economists excited. Why?

The most likely candidate is definitional confusion. Economists make lots of assumptions (so, by the way, do all humans) which may or may not be useful or correct. But most of these are not the infamous “rationality”.

The definition of rationality is a very dull thing indeed, surely nothing worthy of the excitement it generates. A rational agent is one that, for any choice, prefers one of the options or is indifferent. And these preferences must be transitive, so that if A is preferred to B and B to C then A is preferred to C. Lastly, by “preferred” I certainly don’t mean what will make them happy or anything like that, I just mean that that is the one they actually pick if offered a choice.

A model with these characteristics is “rational choice”. There are lots of other assumptions one may or may not wish to make, but we’re done with rationality and on to other, more controversial things, like that people have perfect information or that what one prefers is what makes one happy. All of these additional things are open for debate, but no matter the debate’s resolution, the model is still a rational choice model.

Now I ask you. Why would you ever trust a model of human behavior that relied on agents with incomplete or non-transitive preferences to get its results? Truly that would be a thing worthy of derision.


16 Responses to Rationality

  1. Ryan Bell on June 24, 2004 at 7:53 pm


  2. Jim F. on June 24, 2004 at 8:03 pm

    Worthy of derision, perhaps. But why should I believe that we don’t sometimes have incomplete or non-transitive preferences? (And if that is the assumed model of rational choice, what are discussions of Arrow’s paradox doing in econ?)

  3. Nate Oman on June 24, 2004 at 8:12 pm

    Jim: Arrow’s Impossibility Theorem doesn’t tell us anything about the plausibility of an individual rational actor model because it applies to problems of social choice. The point is that if one assumes individual rationality it is impossible — given some rather weak assumptions — to construct a social decision making function that is transitive.

    A better attack on the rational actor model is something like the ultimatum games in behavioral pyschology that suggest that people’s preferences are not transitive.

  4. Jim F. on June 24, 2004 at 8:18 pm

    This shows what happens when you depend on your memory of what you read a long time ago. Thanks, Nate.

    Your suggested attack mirrors my first question, the one not in parentheses: why should be make the economists’ assumption?

  5. Frank McIntyre on June 24, 2004 at 11:43 pm


    What is the ultimatum example? I am not sure I am familiar with it.


    You can believe we have incomplete and nontransitive preferences to your heart’s content. You can believe that there is no thing as gravity and we are all held to the earth by invisible gremlins. This is your prerogative as a philosopher and human.

    But if you want answers that predict and forecast human behavior, you will do much better if you work with tractable mathematical assumptions that are not wildly off. Rationality can neither be proven nor disproven, but it has been found to be an extraordinarily productive way of thinking about behavior. So two points:

    -you can’t, strictly speaking, disprove rationality from observing behavior.

    -even if you could, if the theory works better than any other, stick with it until you’ve got a better one. This is science, it isn’t truth. People who complain about science also don’t have the truth.

    -a model without transitivity will almost certainly get you nowhere, and so perhaps it belongs with the gremlin theory of gravity…

  6. Kaimi on June 25, 2004 at 12:08 am


    Are you talking about the Kahneman and Tversky stuff from the eighties, with people buying coffee mugs and the like? I’m naturally skeptical of the perfectly rational neoclassical actor — the one Sunstein calls “homo economicus” — but I must confess, I also find the wacky dynmanics of coffee mug markets in economics students to be a little underwhelming as an assault on the rational-actor model. (Then again, it did get them the Nobel, so my feelings of being underwhelmed may not be representative of what most people think).

  7. Jim F. on June 25, 2004 at 12:27 am

    I don’t have a belief about either the completeness of preferences or their transitivity and if I did they would be irrelevant. The question was why economists do, and you answered that: “it allows us to do what we are trying to do.” That’s a reasonable answer.

    But it seems to me that you’re being coy with the final question of your post. You ask why would would trust a model that doesn’t assume completeness and transitivity and you say that any such theory would be risible. However, your explanation of why it would be is much weaker than the claim implied by your question. I presume you have a stronger answer.

  8. Frank McIntyre on June 25, 2004 at 1:50 am


    My last comment is just an extension of the earlier ones. Lack of transitivity quickly turns into having an agent that prefers A to B and B to A. Not that the agent is indifferent, but that they literally prefer both to each other. Any model or logical construct that assumes 1 and not 1 to both be true is going to be useless.

    Incidentally, the model is also amusing in its uselessness. But that is a seperate issue as you note.

    So actually, nontransitivity probably scores lower than the gremlin theory, because the gremlin theory may at least be logically coherent, even if not useful.


    You use the phrase “neoclassical rational actor”. This is exactly the issue. You can argue with the neoclassical assumptions all you want, most interesting economics work violates the simple neoclassical framework in some way since the framework is just a useful starting point. But recognize that it is not rationality you are skeptical of; it is the additional assumptions of the neoclassical model. Those assumptions can be relaxed when necessary. Rationality never really needs to be relaxed.

  9. Thom on June 25, 2004 at 8:33 am

    Having been a few years since I got my BA in Philosophy, can one of you provide me with an illustration of a what you mean by “transitive” or “non-transitive” preferences? I think I follow the rest of the argument, but I’m not clear on what you mean by transitivity. Thanks

  10. Thom on June 25, 2004 at 8:36 am

    Would one of you gentlemen be so kind as to offer an illustration of what is meant by “transitivity” of preferences? My philosophical terminology is a little rusty. Thanks.

  11. Thom on June 25, 2004 at 8:40 am

    Oops, not only did I accidentally double post, but I went to the trouble of rewriting my comment. How embarrassing.

  12. Nate Oman on June 25, 2004 at 11:10 am

    Transitivity: If you prefer A to B and B to C then you prefer A to C.

    Non-transitivity: You prefer A to B and B to C but you prefer C to A.

  13. Nate Oman on June 25, 2004 at 11:29 am

    Here is the description of multi-stage ultimatum games and the possible implications for rationality, taken from the Stanford Internet Encylopedia of Philosophy entry on economics (interesting reading btw):

      A further difficulty with the barebones theory of rationality concerns the individuation of the objects of preference or choice. Consider, for example, data from multistage ultimatum games. Suppose A can propose any division of $10 between A and B. B can accept or reject A’s proposal. If B rejects the proposal, then the amount of money drops to $5, and B gets to offer a division of the $5 which A can accept or reject. If A rejects B’s offer, then both players get nothing. Suppose that A proposes to divide the money with $7 for A and $3 for B. B declines and offers to split the $5 evenly, with $2.50 for each. Behavior such as this is, in fact, common (Ochs and Roth 1989, p. 362). Assuming that B prefers more money to less, these choices appear to be a violation of transitivity. B prefers $3 to $2.50, yet declines $3 for certain for $2.50 (with some slight chance of A declining and B getting nothing). But the objects of choice are not just quantities of money. B is turning down $3 as part of “a raw deal” in favor of $2.50 as part of a fair arrangement. If the objects of choice are defined in this way, there is no failure of transitivity.

      This plausible observation gives rise to a serious problem. Unless there are constraints on how the objects of choice are individuated, conditions of rationality such as transitivity are empty. A’s choice of X over Y, Y over Z and Z over X does not violate transitivity if “X when the alternative is Y” is not the same object of choice as “X when the alternative is Z”. John Broome (1991) argues that further substantive principles of rationality are required to limit how alternatives are individuated or to require that agents be indifferent between alternatives such as “X when the alternative is Y” and “X when the alternative is Z.”

    Obviously, this is not a direct challenge to rationality, because one can define the options so as to maintain transitivity. It does, however, make the theory of rationality — apparently — unfalsifiable, which suggests on at least some views of science that contrary to Frank’s claim, economics is not scientific.

  14. Frank McIntyre on June 25, 2004 at 12:00 pm

    As I noted above and Nate notes here, transitivity is not falsifiable. But all science rests on unfalsifiable assumptions. The idea is to then test secondary assumptions. So I’m not sure why this is objectionable. As you know, without some assumptions at the start, you can get nowhere in any field, scientific or otherwise.

    That said, the proferred violation is not very strong. Clearly the goods offered in the ultimatum game were different because people care about fairness. If you try to limit decision making in this way, you are not testing transitive preferences, you are checking whether people have preferences for fairness. In this case we see that fairness is worth at least 50 cents. Thanks behavioral psychologists!

    A better refutation would hold constant other outcomes and show violations. Any important violations can then be incorporated into a broader definition of preference sets if it is important to the goals of the model.

  15. Kaimi on June 25, 2004 at 1:02 pm


    I think a problem is that when people here “rationality” they don’t just think of transitive preferences, they think of the entire neoclassical model, with all of its baggage. Hence the dislike for “rationality.”

  16. Frank McIntyre on June 25, 2004 at 1:12 pm

    You have made the point I was trying to make. Calling neoclassical assumptions rationality is incorrect. I’m sure you can think of buckets of examples where lay-people say things about the law that are simply meaningless, because they don’t know the meaning of the words they use. If I went to the cucumber thread and said I loved cucumbers because of their yummy chocolate flavor, it would be clear that somebody was confused. We’d want to clarify that confusion. Hopefully, we all now know what rationality is and isn’t.