• Why economics is case-based

    Economic Models as Analogies, by Itzhak Gilboay, Andrew Postlewaitez, Larry Samuelson, and David Schmeidler, is, perhaps,* the best paper about the discipline of economics I’ve ever read. My copy is densely highlighted.

    This gets to something important:

    We argue that viewing economists as generating knowledge that is partly case-based explains the puzzles raised in Section 2. First, one need not wonder why economists feel that they gain insights and understand economics better using models whose assumptions are wrong. In the case-based approach, models cannot be wrong. As long as the mathematical analysis is correct, a theoretical case is valid, the same way that an empirical or experimental case is valid as long as it is reported honestly and accurately. Cases do not make any claim to generality, and therefore they cannot be wrong. […]

    Finally, using the case-based view, one can also understand why economists and psychologists view their models differently. True to the standard, rule based model of science, psychologists try to avoid refutations by being very explicit about the domain of applicability of their models. Economists, on the other hand, often offer models that are merely theoretical cases. These models cannot be refuted, and hence there is nothing to be lost by trying to draw analogies between them and new, remotely connected problems. On the contrary, every problem that may end up being similar to the model increases the model’s popularity. As a result, economists have an incentive to view more real life cases as examples of their models, without risking their theory’s reputation in so doing. [Bold added.]

    I recall an aha moment when I was reading The Theory of Industrial Organization, by Jean Tirole, years ago. You see, I had no formal economics training. My background is physics, a rule-based, not case-based, discipline. I was accustomed to “theory” meaning something that is thought to be true within a specified domain. Then it is tested. If enough credible evidence refutes it, revisions are necessary and, eventually, some are embraced. Without explicitly trying to, Tirole disabused me of the notion that economics worked this way.

    Gilboay and colleagues say economics is clearly not physics, despite the math. It’s not even psychology, usually. Its connection to the real world is much more tenuous. Why? Because in economics, theory is case-based. And case-based models “cannot be refuted!” That one’s model can’t be wrong is likely of some comfort to some practitioners, even if only implicitly.

    Many models are constructed under certain assumptions generally known to be false. (Their sentence, “That the assumptions of economics are false is one of the most poorly kept secrets in science,” is perfect.) As such, in general, they are developed in a way that makes it far less likely for them to be true (or reasonably true-ish) of the world or some clearly articulated sub-domain of it.

    That’s very confusing, because economists do take their models — and the ideas motivated by them — to the world. Then all manner of claims are made about human behavior and ideal policy. Things happen, some of them not good, others just fine.

    When I read Tirole, I learned that you can’t point to anything like “the theory of industrial organization (IO),” despite the title. There’s no Maxwell’s equations of IO. That would be rule-based. Maybe it’s too hard. But do economists even try? Are they, on the whole, sufficiently honest in the discipline’s limitations? Or are they just trying to increase the popularity of this model, or that one? (Not all economists are the same. There is heterogeneity, of course.)

    In IO there are lots of different models of firm behavior (the different cases). Likewise, there are lots of different models relating to the same phenomena in macro and other areas of economics, health economics among them. When and where does each apply? To what extent is each “true”? These things are rarely answered. Indeed, they’re sometimes unanswerable because model assumptions are often so grossly violated and that’s not even the point. Everybody knows it, but engaging that is, to large degree, not what economics is about.

    Needless to say, this was a huge eye opener. The paper by Gilboay and colleagues deserves a full read.

    * “Perhaps” only because I read and forget so much.


    Comments closed
    • Fascinating. Makes a lot of sense now that I think about it. I stopped thinking about economics as a “science” a while ago, but I think this explains it better that just my misgivings. Will have to pass this on to son who is a physics major, but likes econ.


    • I think there are two sides of the issue. The first is that it would be too hard for theoretical economists to be clearer. Even slight modifications of the basic neoclassical theory complicate the math drastically, so as a result, we only consider models with one modification at a time, though in reality all of these models exist together as parts of one incredibly complicated model.

      The second part is that economists just aren’t as clear about the domains of applicability. Physicists frequently use incorrect assumptions. The assumptions of newton’s law of gravity, for example, are incorrect. If you want to send a rocket ship to Andromeda, the wrongness of Newtonian assumptions is probably pretty important. If you want to send a rocket to afghanistan, Newton’s laws are probably fine.

    • So why even bother? If you can’t even guess where and how your theories and models apply, what good are they? I see so much self-serving nonsense being promulgated, to support whatever the rentier who supplies the patronage wants.

    • Theoretical economics is not “case-based.” Economics is an accounting system. It is an elaborate and sophisticated accounting system. It allows us to talk about deadweight loss (sorry Manski). It prevents double-counting. It makes very clear what is and is not being accounted for.

      That is is a good thing. We need an accounting system. But calling it case-based misses the point.

      • You need more than just accounting to get utility functions. At the very least, you need some high-level assumptions like continuity, transitivity, completeness, though any usable model needs much more restrictive assumptions on utility as well. And without utility functions, the concept of “deadweight loss” does not exist.

        • Utility is just the unit of accounting, no? We have developed a theory that connects this to preferences, but we use utility in an accounting sense.

          I am not demeaning or criticising economics. I am just pointing out that economics is best understood as an elaborate, sophisticated, even deep accounting system. Accounting, like a case-based approach, cannot be right or wrong.

          This is why it is so hard to nudge economic theory and why it is hard to incorporate “true” behavior beyond any single instance. You would need to move the entire accounting system, which is not easy.

      • Actually modern economics is often inconsistent with the basic principles of accounting. Parties are often expected to maximize profits and minimize profits at the same time. Sums rarely add up without bogus and fictitious (at least by accounting and “normal human” standards) items. Anyone who looks at a macro model and tries to impose an accounting framework on any of the entities involved with usually wind up with inconsistent results.

    • This reminds me of the question I’m often asked about law as a career choice. My answer: if it weren’t for having to deal with clients and other lawyers, the law would be a great career. If economists didn’t have to deal with (unpredictable) human behavior, economics would be rule-based, economic policy would be done solely by computer, and we’d all be friends.

    • Another thing to consider: in economics, the measurement errors are far greater than in physical sciences. For example, for the National Health Expenditures Accounts, there is actually an economic census, where forms go out to all firms above a certain size, asking for their total revenues. There is going to be measurement error there: some firms won’t respond, some will misstate their revenues by mistake or on purpose, some may state their revenues one way and then there are some accounting issues and they revise their revenues later on.

      Furthermore, the economic census is done every 5 years, as it would be impractical and intrusive to do it yearly. The team working on the NHE accounts interpolates between economic census years using various data sources, all of which come with their own measurement errors. Interpolate is a fancy term for calculating a linear trend between the measurement years.


      Given the larger magnitude of measurement errors compared to other sciences, perhaps economists need to be cautious in deducing theoretical rules from their observations.

    • Arguments presented are actually pretty good; case-basing does an excellent job of reducing -but not completely, the ambiguity in more subject material more encompassing than ‘cased-based’.
      It is almost never appreciated how most of our discussion is opinion -and only very little fact as in scientists actually working in a laboratory or mathematicians at the blackboard. We are, in general, always talking generalities of ‘sorta’ agreement and ‘sorta’ disagreement. Worse still, the whole of language and its various applications is solidly institutionalized or we wouldn’t be able to call ourselves anything from family to nations.

      The whole of this greater material -government, economics, saber-rattling et cetera, is discussed in the new publication at-

    • Well, it is really very difficult to say who is wrong and who is right in this matter. Everybody has explained their views quite clearly. There seems to be no fixed theory regarding a particular thing and things might change from one perspective to another thus making them case specific.