message size should not exceed :-(

Robert Waldmann

Does anyone hate the limit on comment length as much as I do ?
Kharris had a very interesting comment on a post of mine below. I wrote a long reply.
I can’t post that reply in the comment thread so Kharris’s comment and my reply are is after the jump.

Warning amateur philosophy of science below. Also I’m typing what I always type.

update: Kharris had another comment (which was not too long to post in the thread) which said much of what I said in my reply. I add that comment and my reply to that comment after the jump.

View details

Just a quibble over terms. It might be useful to distinguish between theory and hypothesis in this discussion. And you may want to argue that what you’ve said stands, that hypotheses tend not to be rejected in economics, but I would argue that the hypothesis which is refuted by data in economics is generally treated the same as in physics. Rejected or accepted at the X% level, for the purposes of the particular hypothesis formalized for the particular statistical test.

Now, there are problems with formalizing hypotheses in economics that grow out of theory, and that does tend to screw things up. Starting from the assumption that multipliers are at some level a reality more or less guarantees that you can find multipliers. It is hard to imagine formalizing an hypothesis that “multipliers do not reflect reality at any level” so that it can be tested against the data.

At the level of theory, economics is a very different creature from the physical sciences. Think in terms of Kuhn. His notion is that theories don’t go out of use because they are wrong. They just keep piling up troubling, contrary results until a new theory comes along that better explains the data. There is, however, no economic graveyard. Say’s law still stalks the earth. Freidman admitted that the development of instability in monetary velocity mean his monetary policy prescriptions won’t work, but there seem to be people who didn’t notice him say it. It’s a quibble, I know, but I think what policy makers are willing to accept as approximately true is theory, rather than hypotheses.

Economics can falsify hypotheses left and right. Economics doesn’t seem to be able to falsify theory, except in very specific character. If Kuhn would tell us to bury it, we attach “neo” and keep marching – neo-classical, neo-Keynesian, neo-fascist.

Robert Waldmann

Good discussion. Not a quibble. I will explain how I use the terms (this is really standard in history of science and in natural sciences).

1) a hypothesis is a statement about the world which might be true.

2) a theoryNS is a hypothesis which has been tested repeatedly and not rejected.

NS stands for natural sciences. Note the word hypothesis is often used to mean “testable hypothesis” (for example in statistics). Here I am using it in it’s ordinary English sense so that it might not be testable except jointly with other statements.

By this terminology, I’d say there isn’t any economic theoryNS.

So what do I call that which you call a theory ? I call it the hypothesis of interest. In practice we can’t test the hypothesis which interests us except along with auxiliary hypotheses that we don’t care about.

So a testable hypothesis is a hypothesis of interest and a bunch of statements which don’t interest us. A lot of what happens in economics is testable hypotheses are tested and rejected and economists decide that the part that was rejected was the uninteresting part and the interesting part is like a theoryNS. This is silly. It is not proven false, but the experience of adding auxiliary hypotheses, deriving testable predictions, testing and rejecting sure doesn’t support the hypothesis of interest.

Somehow a statement which has never implied a true prediction gains huge status in the literature, not as something known to be true, but as a default assumption so that you can write a model in which you relax one of the default assumptions but not the others.

It is true that statements which really interest us can’t be tested except along with auxiliary hypotheses which don’t interest us. But there is something strange about an idea, aguess, gaining the status of a theory without ever winning any of its contests with data.

I know of few hypotheses in economics which have explained something, fit data that they weren’t fiddled to fit. One of those few is the pernament income hypothesis.

Below I discuss testing it as an example.

OK I might have something actually useful to say here. Occam said “entities are not to be multiplied without need” (translated from Latin). This is interpreted as saying thatif there are competing hypothesis which fit the data, we should favor the simplest hypothesis, roughly the one with the fewest fudge factors or free parameters. “Favor” means treat as our working hypothesis which means we keep testing it until it is rejected. Maybe it even means we give advice which would be good advice if it were exactly true.

Economists interpretation of Occam’s razor is quite different. In economics some simplicity is more important than other simplicity. Basically there are default assumptions (rationality, no liquidity constraints, price flexibility, perfect competition, non-increasing returns to scale, economic behavior can be understood with utility functions which give utility as a function only of consumption and leisure etc) and relaxing one of them is severely punished. On the other hand, playing with funny utility functions is OK. The list is strange. It includes statements which economists agree are false.

How did this list of statements get to be the default assumptions ? Given that hypotheses of interest can’t be rejected unless you add auxiliary hypotheses, how could the list of default assumptions ever change ? What if anything does this exercise have to do with social science ?

I explain TheoryNS with an example — the usual example.

Take Newton’s model of the solar system. It was simple, 3 laws of motion plus 1 law of gravity plus the approximation that the sun and planets were rigid spheres and no other forces had measureable effects.

This hypothesis fit a huge massive gigantic amount of data. It also worked for far away stars (implications for location and red shift). It was just the most successful hypothesis ever. So it was called a theory. It has been rejected by the data, but it only was called a theory after it had been tested and not rejected with masses and masses and masses of data.

Economists try to test a hypothesis of interest and get nowhere.

Say we are interested in the permanent income hypothesis and we want to test it with aggregate data, so we have to add the auxiliary hypothesis that aggregate demand can be represented as demand by a representative consumer. Oh we also have to assume that utility functions are time separable and separable in consumption and everything else. Now we test the testable hypothesis and obviously it’s false.

The result is that the permanent income hypothesis is a default assumption. You can write a model with liquidity constraints, but, somehow, the economists version of Occam’s razor punishes you.

I bring up the difference between hypothesis and theory in part because in lay and political discourse, the notion of “theory” is often misunderstood. Darwin’s theory gets a bad rap because it’s “just a theory” form people who misunderstand what a theory is, take a theory to be an educated guess. Science attaches “hypothesis” to an educated guess. Theory is what we attach to broad, elaborate explanations of general phenomena. Generally, theory also implies “as well-founded as any explanation we have for now” – but not in economics.

Robert Waldmann

Oooops. Kharris, I responded at great length to your comment above. Js-kit wouldn’t let me post my comment here so now it is a blog post.

The point I was making at gruesome length is almost exactly what you wrote right here

“Generally, theory also implies “as well-founded as any explanation we have for now” – but not in economics.”

I will put your second comment in the post. Also I will quibble.

I think that standard economic theories are as well-founded as any explanation we have for now. That is an easy standard to meet if we have no well-founded explanations at all.

I think that in natural science a theory must not just be the best available explanation (which means fits the data with the fewest fudge factors) but also a clearly good or well founded explanation (which means fits a huge amount of data with few fudge factors).

I think there is a simple operational definition of a theory. A statement is a theory when it becomes impossible to remember all the data which it fits. The usual example is that it is impossible to remember all of the data of where planets were except to recall it all fit Newton’s model except for the precession of the orbit of Mercury. It is impossible to remember all of the anatomical and DNA sequence data that fit’s the theory of natural selection except by remembering that unimportant anatomical details and DNA sequences are what one would expect if the theory of evolution by natural selection were true.

This means that theorists who are making a new hypothesis and want it to fit known data must learn and use the existing theory. So Einstein had to make sure that his theory of gravity had the same implications as Newton’s for moderate densities, velocities, pressures and sheers. So special relativity gives Newtonian mechanics as an almost perfect approximation for speeds far below the speed of light. So the neo Darwinian synthesis (due to August Weisman not just Darwin) had to give the same implications for minor anatomical details as Darwins semi Lamarkian theory.

Theorists bow to old theories and not to raw data, because it is simply impossible to remember all of the raw data correctly summarized by the old theory.

Clearly there is no such thing in economics. There is a bit of it. Old Keynsian equations become stylized facts which must be explained even after theorists turn up their noses on them as theory.