“As a student of the history and philosophy of science, one of my biggest annoyances is this cultural tendency, in both scientific and popular culture, to by default accord contemporary scientific theories rigor and rationality and all past, disproven theories slipshod irrationality.”

I would have thought that Kuhn drove a stake through the heart of this tendency.

We are all p-zombies now.

Something that is related to parsimony is, I think, the sense that tweaking is like cheating.

Specifically, tweaking is a form of goalpost-moving. (Well, ok, that theory is falsified, but how about this one?) But it’s also a major method of scientific advancement—in some sense, fitting more and more data is what theory is about. ISTM that in science, there’s nothing wrong with being wrong, only with being stubborn about it.

But to come full circle for a moment, there are some economists that are quite well-paid to be stubborn about provably wrong theories. And economics as an intellectual field doesn’t seem to have the will, or perhaps the capability, to cast them out or to draw a line with the same rigor that it is drawn between, say, biology and creationism.

I think so too. I would phrase it as mathematics being part of a narrative since it is the narrative that supplies the initial conditions and boundary values for parameters. I think the process of theoretical economics research goes something like this : you specify a model (e.g. for a particular situation), show that you make as few assumptions as possible, and then show that it is helpful in explaining something (say. that situation). This is not that different from the process in the natural sciences, except that there tend to be a few more assumptions in social science theories. Questioning the assumptions could lead to a better understanding of the limitations of the theories, but in the end, there is still the question of whether this is simply the best thing we have that works. Or in other words, we don’t know which of those assumptions we should drop. Rationality in favor of predictable irrationality? Consistency of preferences in favor of possible racial biases in people?

Seems to me you’re disagreeing over terminology, not concept. Those initial conditions, and boundary values of the parameters specified, constitute the “narrative” of which they spoke.

I would argue that, proceeding from the sort of data you have access to in the social sciences to the sort of hypotheses you end up forming, parsimony is implicitly a claim (and I may be mistaken, but this may also be what Neel is arguing). But why proceed from the data to these parsimonious hypotheses?

In the natural sciences, the answer seems clear : falsifiability. Scientific theories are falsifiable. When contradictions to a scientific theory are seen, the theory is dropped (examples are many – luminous ether, phlogiston, et cetera. Yes, phrenology is a better strawman candidate (in some respects) because the rejection of the theory has not proceeded from falsification, as I argue below). Here, I distinguish the natural sciences from the social sciences as well as the formal sciences (mathematics, logic, some areas of computer science).

In the social sciences, the answer is less clear. This is Milton Friedman’s argument in The Methodology of Positive Economics : if you are going to start dropping social science theories because their assumptions have been falsified, pretty soon you are going to have no theories left at all. (By quoting Friedman, I don’t nescessarily agree with everything that Friedman has said in the paper). So Friedman argues that assumptions in the social sciences cannot be questioned. IIRC, he uses the example of a ball falling through a vacuum to illustrate the impossibility of ever obtaining the conditions nescessary to verify a theory (which was a bad example, in my opinion). To me, this is a key difference in methodology between the social sciences and the basic sciences. Assumptions in the social sciences can be unverifiable and even blatantly false.

I am less sure than Neel is that mathematics is a narrative (and so, I suppose, I disagree with John Quiggin as well. Take problems in engineering or operations management, for example.) Mathematical and computational solutions to problems are “correct solutions” in that they are the best possible solution for the initial conditions in the models and/or the values of the parameters specified. Further, mathematics is the only tool available to keep track of simultaneous and multifacted relationships among entities. Argumentation (and formal logic) cannot be used to describe these relationships completely nor can they be used to solve the problem at hand. This is why engineering models are also “correct solutions” in many respects. They are the correct solution to the problem at hand given initial conditions in the model and/or the values of the parameters specified. Just because there are some problems that cannot be solved using mathematics does not mean that there are no problems that can be solved using mathematics.

Neel Krishnaswaim: “So they can design models to take advantage of qualitative insights (e.g., there was no mass outbreak of laziness in 1929), but there’s no way to falsify their general approach, because you can always add another tweak.”
——
Neel Krishnaswami: “That’s because they know what they are talking about. If you listen to sociologists, you will hear the same thing, only more loudly and larded with rude comments about the folly of economists who are too quick to ignore important effects.”

Do you mean that they know that everybody is tweaking their models? Or that they know that, without appeal to general criteria such as parsimony, virtually any model can be tweaked ad infinitum, and therefore not falsified? The general tenor of what I read is along these lines: You can always tweak the models, so no model can be falsified, so what can we do? I do not read this: You can always tweak the models, so no model can be falsified, so let’s get some discipline.

Neel Krishnaswami: “Parsimony is not a magic wand. It is a claim about the structure of the causal relationships within a system, which can be either true or false, and must be justified with evidence and argument.”

Parsimony is not a claim, it’s a criterion. There is more than one kind of parsimony. Occam’s razor minimizes entities, which I suppose will also cover variables. Minimal algorithmic complexity and minimum description length are modern forms of parsimony. Axioms and postulates can also be minimized.

Now, why parsimony is a criterion is not so easy to say. Is it simply esthetic? More parsimonious explanations are more beautiful. Is it explanatory power? More parsimonious explanations explain more with less. Is it probability? More parsimonious explanations are more probable, a priori.

Something that is related to parsimony is, I think, the sense that tweaking is like cheating. It is as though you are debating with someone and you raise objections to their claims, and they modify or take back or redefine something they said to meet each objection that you raise. You can never pin them down. If falsifiability means anything, it means making clear and definite claims that pin you down. There is an austere discipline to science that I miss in these discussions of tweaking.

Like astronomy, macroeconomics is not a laboratory science. One recent lament that I have seen is that, well, once competing models have been tweaked to cover our current situation, we will have to wait X years to test them. (When, of course, they will be tweaked again. ;)) There is an important anti-tweaking lesson from the field of machine learning: Never tweak your model so that it predicts everything, because you need to avoid overfitting the model to the data, and you need to have a good measure of the error of the model. Tweak the competing models on most of the data and test them against the rest.

Neel Krisnaswami: “Basically, theory in social science is under-determined by the data, and there is no prospect of that fact ever changing. This means that social scientists need lots and lots of yellow books to make even modest claims—you need fancy math to wring conclusions out of weak assumptions.”

That is another problem. Math lets you build magnificent edifices, but to what avail if they rest upon sand?

All scientific theory is underdetermined by the data. Does economics suffer from tweakism?

Math lets you tell a story with the confidence that the story doesn’t have any plot holes in it. But: it remains a narrative

A nice line which I might pinch

Most economists are in serious need of some sociology courses. …

You are perhaps too optimistic on this point. The hottest area in sociology these days is social network analysis, and random graph models of social networks make the rational choice framework look realistic and psychologically insightful.

(I don’t want to rag on either economics or sociology, actually. In both of these cases, facts were thrown out to get something mathematically tractable, and that’s actually a good thing. Math lets you tell a story with the confidence that the story doesn’t have any plot holes in it. But: it remains a narrative.)

And I really had to laugh when he quoted Hayek. I was waiting for him to start blabbering about John Galt. What Cochrane doesn’t get is his metrics will end up being garbage unless he gets beyond a FoxNews level of economic assumptions. Most economists are in serious need of some sociology courses. …

Again, this is true of all science. It is why parsimony is invoked. One thing that I have noticed on economics blogs is how economists do not seem to invoke parsimony. Instead, this is the kind of thing I hear.

That’s because they know what they are talking about. If you listen to sociologists, you will hear the same thing, only more loudly and larded with rude comments about the folly of economists who are too quick to ignore important effects.

Parsimony is not a magic wand. It is a claim about the structure of the causal relationships within a system, which can be either true or false, and must be justified with evidence and argument. It happens that many (but by no means all) physical systems have the property that you can rule out most possible causes, and be left with a model with few enough parameters that you can actually identify a model with a reasonable amount of data.

However, the principle of parsimony does not get rid of enough causes in most social systems. Returning to the example of basketball, we can justifiably appeal to parsimony to rule out considering hemlines or football scores as relevant variables in a theory of basketball. However, we cannot appeal to it to rule out considering the effects of a player’s teammates on the quality of his or her play. Even if you limit yourself to considering one-on-one interactions, the number of interactions grows quadratically, and you quickly end up with too many parameters for the available data to identify a model.

Basically, theory in social science is under-determined by the data, and there is no prospect of that fact ever changing. This means that social scientists need lots and lots of yellow books to make even modest claims—you need fancy math to wring conclusions out of weak assumptions.

Neel Krishnaswami: “This critique doesn’t merely cast doubt on freshwater economics; it rules out the very possibility of any form of social science, as a scientific enterprise. The reason is the world is too complicated, and there is not ever enough data to identify a model.”

As Rabbi pointed out, the same is true for all science. That does not stop the scientific enterprise, however.

“Let’s pick an example less controversial than economics—in particular, basketball. So, suppose we want a theory of basketball, which we want to use to predict the outcome of NBA games. . . . Suddenly you have hugely more variables in your model than the number of players, and the amount of data you need to identify a model with good statistical confidence blows up—you will need more data than any player will ever generate in his whole career.

“So we have no choice but to make use of qualitative insight to prune the space of models, in order to make it small enough that statistics can tell you anything useful. You cannot commit to building only on experimental data because there can never possibly be enough.”

Again, this is true of all science. It is why parsimony is invoked. One thing that I have noticed on economics blogs is how economists do not seem to invoke parsimony. Instead, this is the kind of thing I hear.

“So they can design models to take advantage of qualitative insights (e.g., there was no mass outbreak of laziness in 1929), but there’s no way to falsify their general approach, because you can always add another tweak.”

Does economics suffer from tweakism?

Chris@18: I agree with you about Public Choice theory. I was trying to call it window dressing; sorry for any confusion.

This paper has focused on the question of whether an increase in government debt constitutes an increase in perceived household wealth. The effect of finite lives was examined within the context of an overlapping generations model of the economy. It was shown that households would act as though they were infinitely lived, and, hence, that there would be no marginal net-wealth effect of government bonds, so long as there existed an operative chain of intergenerational transfers which connected current to future generations.

Surely it would have been more circumspect either to say:

It was shown that households in an overlapping generations model would act as though …

or:

An observation of real households acting as though they were infinitely long lived would be consistent with the overlapping generations model presented here …

It’s possible that I missed something crucial in my brief (and not very well informed) reading (I’m fairly confident, though, that the ‘utility functions’ which appear on p. 1100 and following are not derived from any observation of the author). It’s also possible that Barro’s intended audience would (a) never confuse a description of a model with an empirical claim, and (b) possess perfect immunity from any normative effect of his conclusion. Personally, I can’t help thinking that this sort of product seems well suited for the shelves of your typical ‘family office’.