Friday, March 18, 2011

Pulled back from comments

For some reason which I can't understand Adam Ozimek courteously comments on my rude post insulting Ryan Avent and economics (not in that order).


As always, I take your point of view seriously. But I disagree with you here. I think a simplified but reasonable approximation of a "real" science paper is that it has 3 parts: lit review, theory + hypothesis, and empirical test. I take the role of theorists to be specializing in one part, a crucial part, of the scientific method, while not necessarily producing work that in and of itself constitutes science. The same could be said of lit reviews. In short: if you break up the scientific method and intensely specialize, then parts begin to look like not science, but they are.

Is that reasonable?

Also, for an economic hypothesis that has been abandoned because they did not fit the data, how about one of the most famous: the Phillips curve? (At Modeled Behavior I also pointed to 80s RBC models with no money, but Krugman claims people are still doing that, so I'll leave that aside for now).

And as time goes on it becomes increasingly difficult to view economics as a coherent field, but I don't think we're there yet. Most economists come from a grad program that covers a reasonably similar core. What % of programs use Varian, Mas-Collel Whinston and Green, or something similar for the micro core?

I agree certain subfields are largely engaged in un-scientific, but often worthwhile endeavors. Normative pursuits obviously fit here. But I think when critics say econ isn't a science, they're talking about when it's trying to be a science, e.g. when it's making positive statements. So before saying "parts of econ are unscientific", I think it's important to emphasize that the parts that try to be generally are, and some parts that appear not to be (e.g. theory) are part of the scientific process.

# posted by Adam Ozimek : 2:34 PM

Ah, I see you have replied to the Phillips curve in your post below!

# posted by Adam Ozimek : 2:52 PM

Dear Adam Ozimek

Thanks for coming by. I also admire your polite tone (especially given the very rude tone of my post).

Relatively briefly look at the argument that economics as a whole is scientific and that it is one thing with a common core. Yes it has a core "What % of programs use Varian, Mas-Collel Whinston and Green, or something similar for the micro core?" Yes some economists are scientists. However, there is no link between the scientists and the core. Below I ask how Varian has revised Microeconomic Analysis since the first edition in the light of new data. I don't know the answer. Those books are certainly part of the core of modern economics. They do not discuss much the confrontation of testable implications of the theory and data. I am quite sure that they don't point to any empirical successes in which the theory was used to make surprising but correct predictions. Yet it is, indeed, a key part of the core of modern economics.

An aside. Economists always talk about physicists. Why indeed there are many physics textbooks now in use which present theories which have been refuted by the data. People are taught Newtonian Mechanics before they are taught about relativity. People are taught classical physics before quantum mechanics and non-relativistic quantum mechanics before semi-modern quantum mechanics etc. But these are steps in a teaching program. Before becoming a physicist graduate students are explosed to unfalsified models *and* the fact that existing theory is fundamentally not there yet is stressed. Economics graduate students have neither experience.

OK now my long reply to the comment.

I agree that theorists have a role in scientific enterprises. Obvously the classic example is theoretical physicists who do math all day, never conduct and experiment and play a vital necessary role in a very successful scientific enterprise (an aside -- I think one of the worst problems with economists is that the only natural science most of us consider is physics).

But the interaction of theoretical physicists and data is completely different from the interaction of economists and data. In physics, elegant models which contradict the data are abandoned. Many (perhaps most) theoretical physicists look down on experimental physicists (who they rank above non physicists who they rank above sheep). They clearly resent the fact that their brilliant model development is not appreciated, because it happens to not correspond to this universe. But the field moves on.

This just doesn't happen in economics. The theory basically stays the same no matter what the data say. Consider your example -- economists are taught from Varian or Mas Colell Whinston. The first edidtion of Varian's microeconomic analysis was published long ago (I studied from it). How have the data caused later editions to differ from the first ? (I ask for information I have only opened two editions of the book).

The key point is that economists identify the set of theory with the set of models and note that models are false by definition. Testable implications are developed, tested and rejected and nothing happens. The model with false implications is still used. It's status isn't diminished, because economic theory was never a set of scientific hypotheses.

Consider Lakatos's idea of a degenerative research program. Oddly I can't find a good link. The general view is that all research programs have cores and auxiliary hypotheses and rejection of a testable implication generally leads to abandoning an auxiliary hypothesis. The distinguishing feature of a degenerative research program is that, once the core belief is reconciled with the data that rejected the joint core and auxiliary hyotheses, the matter is settled and nothing else changes.

Each confrontation with the data is local and once the core belief is reconciled with that bit of data, no implications for other phenomena are developed. The new auxiliary hypothesis has served only to protect the core belief from the data.

I read something like this somewhere. I thought it was a perfect description of econmics after the death of Lakatos. He was thinking of Ptolomaic astronomy and epicycles, but that is *exactly* what we do.

So when testing the life cycle- PIH we note that the hypothesis does not implay a representative consumer nor utility functions which are separable in consumption and everything else nor time separable utility functions. But when we put consumption into a macro model, we almost always put in a utility function wich is log(c_t) plus other things. That part of the model is rejected, but so what. The auxiliary assumptions are only a problem when the PIH is confronted with data on consumption and not when it is used to develope models which are used to advise poicy makers.

What could possibly be more degenerative than that ?


Anonymous said...

Very interesting post, and I would like to read more of your thoughts on the subject, but …

…what it really comes down to is this: “The theory basically stays the same no matter what the data say.”

That’s the first, second and third reason why I view economics as unscientific.

Anonymous said...

PS: And is there even any theoretical results when we move beyond the individual (without outlandish assumptions)? I think this part is silly. I.e. why not simply state separate assumptions for the aggregate, when you clearly can´t reach a decent theory by summing up the part.

However, I wouldn’t call this silliness unscientific if empirics really mattered (but it sure is ideological).

Matt Austern said...

What you say about the use of falsified models in physics education is true, but it's only part of the story.

The other part is: physics students don't stop studying such models after they've learned the more complete version. Every physics undergrad learns quantum mechanics, but, at least at the school where I went, every physics grad student takes a class in classical electrodynamics. Clearly the classical approximation isn't being used just as a teaching technique; by the first year of grad school the students have already seen Maxwell's Equations and the students have also learned about quantum mechanics. So what's the point of a graduate level class that teaches a classical model, if the students have already been taught that the model has been falsified?

And the answer is that "falsified" is a slightly misleading word. The classical model of electromagnetism is known to be inapplicable in some domains, and it is also known to be an excellent approximation in other domains. We can specify quantitatively the conditions under which the classical model is applicable, and we can explain quantitatively how the more complete theory of quantum electrodynamics reduces to classical electrodynamics in those conditions. Learning approximations is a crucial part of learning physics.

I don't know if there's any very close equivalent in economics.

Robert said...

It is certainly true that physicists use falsified models all the time. You are right, they aren't just teaching techniques. They are also useful approximations.

However, the claim that they are close approximations under the stated conditions is based on massive evidence. They are used because they are useful. They fit a huge amount of data.

This is completely unlike the case of economics. No economic model fits data the way classical physics does. The standard economists claim is that the theory is yes a model not a hypothesis (that is known to be false) but it *might* still be useful. That's where we are at the "might" makes right stage.

I believe that economists are absolutely not in a position such that models which we consider really economic theory (which means they include rational maximizing agents) are actually useful. Rather great effort goes into tweaking those models so that they have the same implications as models which aren't economic theory (because no optimization inside).

Here in comments I'm going to make a strong claim. I don't think that any predictive power has ever been gained by introducing optimizing agents. The reason I think that is that it seems to me that the models with optimizing agents are all always modified until they act like older models without optimizing agents. Then they give the same predictions whihc are OK but nothing special. And zero value has been added.

Before the tweaking they give different predictions which are further from the truth and value has been subtracted.

I challenge economists to present a counterexample (I'll try to think of one myself). I used to think the Permanent income hypothesis was one, not true but giving useful insights and explaining previously puzzling patters. Then I realized that a model without optimization explains those patters.

Adam Ozimek said...


If I understand you correctly, you seem to be complaining that the core theory hasn't changed to reflect empirical results, and that theory in Varian isn't being tested and rejected. But Varian and texts like it present us with the building blocks of economics. But the building blocks are used to construct more nuanced and complex theory which is tested. The fact that the basic building blocks of economics have proven extremely useful and durable over time does not strike me as unscientific.

Anyway, I appreciate your comments, and the question "does the core of micro still present us with the best tools for the final applied work of economic science?" is certainly a question worth asking, and I'm sure on the edges there's always room for improvement. At the very least I'd be surprised if many graduate programs still using just Varian for their micro core, so I think it's fair to say the toolkit has expanded greatly since then. There's certainly a ton more in MGW then there is in Varian.

Robert said...

Dear Adam

Uh oh this is long again. My question about revisions of Varian aimed to argue that the core of economics is not scientific even though a science is done around the edges. I try to explain in a new post.

You assert that the framework is durable and useful. It is certainly durable. I think that when economists try to use it, they relax the assumptions until they have no implications. The framework is useful if one wishes to present one's work in a way which referees who are economists will accept. I don't think it is useful in the sense that someone armed with the tools outpredicts someone not so armed.

OK less important stuff I edited down to here.

One can be considered an economist without taking any micro course (for a grade) after the one that used Varian. I am. I agree that Varian or Whinston Mas Colell is part of the economic canon and one must be familiar with that stuff to be considered an economist. However, I assert that that stuff was developed without reference to data and was not originally scientific.

Since then, much empirical research has been conducted. Generally it has not confirmed the implications of the models in Varian. But there are still part of the canon. Therefore I conclude that economics is not a science.

I have tried to explain my view more or less concisely as another post. To restate here, I agree that economics is a field with a core set of knowledge required of all who claim to be economists. I agree that some economic research is scientific. I claim that the intersection of the core knowledge and the science is empty.

The core knowledge is a set of definitions and theorems -- mathematical knowledge about formal systems. It is not scientific hypotheses which haven't been rejected and have made surprising and accurate predictions.