Raw Thought

by Aaron Swartz

Tectonic Plates and Microfoundations

In 1915, Alfred Wegener argued that all the continents of Earth once used to fit together as one giant supercontinent, which he later named Pangea. As Wikipedia summarizes:

In his work, Wegener presented a large amount of circumstantial evidence in support of continental drift, but he was unable to come up with a convincing mechanism. Thus, while his ideas attracted a few early supporters … the hypothesis was generally met with skepticism. The one American edition of Wegener’s work … was received so poorly that the American Association of Petroleum Geologists organized a symposium specifically in opposition…. … By the 1930s, Wegener’s geological work was almost universally dismissed by the scientific community and remained obscure for some thirty years.

Today, of course, every schoolchild knows about Pangea. But for a long time the theory was dismissed, not because it lacked evidence or predictive power — it explained why the shapes of the continents fit together, why mountain ranges and coal fields lined up, why similar fossil were found in places separated by oceans, and so on — but because Wegener had no plausible mechanism.

A similar problem happens in the social sciences. Paul Krugman recently noted that while Larry Bartels (in his new book Unequal Democracy) provides solid, convincing evidence that Republican presidents systematically preside over slower growth and increasing inequality, most social scientists don’t believe him because we haven’t yet identified the mechanisms. Krugman:

Now, I’m a big Bartels fan; I’ve known about this result for quite a while. But I’ve never written it up. Why? Because I can’t figure out a plausible mechanism. Even though I believe that politics has a big effect on income distribution, this is just too strong — and too immediate — for me to see how it can be done. Sure, Republicans want an oligarchic society — but how can they do that?

Bartels, for his part, argues that providing the mechanisms isn’t his job — his goal is to highlight the phenomena and encourage many others to research the mechanisms:

How do presidents produce these substantial effects?

One of my aims in writing Unequal Democracy was to prod economists and policy analysts to devote more attention to precisely that question. Douglas Hibbs did important work along these lines … He found that Democrats favored expansionary policies … while Republicans endured and sometimes prolonged recessions in order to keep inflation in check. (Not coincidentally, unemployment mostly affects income growth among relatively poor people, while inflation mostly affects income growth among relatively affluent people.) In recent decades taxes and transfers have probably been more important. Social spending. Business regulation or lack thereof. And don’t forget the minimum wage. Over the past 60 years, the real value of the minimum wage has increased by 16 cents per year under Democratic presidents and declined by 6 cents per year under Republican presidents; that’s a 3% difference in average income growth for minimum wage workers, with ramifications for many more workers higher up the wage scale. So, while I don’t pretend to understand all the ways in which presidents’ policy choices shape the income distribution, I see little reason to doubt that the effects are real and substantial.

When it comes to addressing such arguments more generally, the most famous commentator is Jon Elster. In his classic article “Marxism, Functionalism, and Game Theory”, he insists:

Without a firm knowledge about the mechanisms that operate at the individual level, the grand Marxist claims about macrostructures and long-term change are condemned to remain at the level of speculation.

(To be fair, Elster doesn’t make this as a general argument, but his vehemence has led some of his followers to suggest that it is.)

To be clear, I think discovering mechanisms is important work. All I’m arguing is that it shouldn’t be a necessity for believing in a theory. Instead, I believe it’s an irrational side-effect of an emotional distaste for gaps in knowledge.

As evidence, let me note that such demands for mechanisms never go more than one level deep. Nobody has ever said, “Well, your theory that people are motivated by greed is all very nice, but I just can’t believe it until you can explain how greed is manifested in the brain.” Neuroscience is obviously the microfoundation of psychology, but psychological theories are regularly accepted without neuroscientific microfoundations.

In general, it seems like such commentators support a double-standard. Theories with mechanisms should be judged by their fit with the evidence and predictive power. Theories without mechanisms should be judged by the evidence and predictive power and whether you can think of any plausible mechanisms. I don’t see how this can be justified. There’s no reason mechanism should be privileged in the assessment of knowledge; things are true or false, even if we don’t know why they are true or false.

Indeed, it we typically only investigate the causes of phenomena once we’re convinced that they exist. (Elster admits as much in Explaining Social Behavior, noting that establishing a phenomena’s existence is the first step towards explaining it.) So let’s stop making the mistake of not believing things are true because we don’t know how they happen.

You should follow me on twitter here.

May 14, 2008


Without a mechanism, how can we distinguish between cause and effect? Perhaps people turn to Republicans because of other actually causal phenomena. And frankly, in any social situation, “causality” just won’t be there — too many feedback loops and too many independent actors.

If all someone is arguing for is correlation, then that’s a rather boring thing to argue about. Causation is the thing that is exciting.

Even causation can be a little boring. Causation that can be effected and changed, now that is interesting.

The scientific approach then would be to empirically determine what can cause change. What are electrons, or protons, or quarks? Eh, who knows, but we know how we can manipulate things with them, how we can see the side effects. Social theory often seems more interested in what these things are, as though there is some value to the thing itself, and not so interested in what can be done.

posted by Ian Bicking on May 14, 2008 #

You seem to believe that not having a mechanism limits you to making claims about correlation, not causation. That’s not true at all. Imagine someone gave you a red button and pressing it caused a man to shout “yap!” You can clearly establish a causal connection — do double-blind tests if you have to — even if you have no idea how it makes the man say that.

posted by Aaron Swartz on May 14, 2008 #

I would say it’s sometimes not an easy task to figure out what’s true but unexplained, what seems to be true but is a thinking error, what’s lies that are just widely repeated, what’s correlation but not causation, what’s finding spurious coincidences, etc.

Yes, “things are true or false” - but you don’t know which is which beforehand.

Politics can be especially hard, since there are people who have as their jobs making the false seem true, and the true seem false.

posted by Seth Finkelstein on May 14, 2008 #

Interesting post, which I generally agree with. But this “In general, it seems like such commentators support a double-standard” is incorrect.

There’s one standard: “Theories should be judged by the evidence and predictive power and whether you can think of any plausible mechanisms.” It just happens that for theories with mechanisms you can, in fact, think of plausible mechanisms.

Otherwise it’s like arguing “… a double-standard. Theories without evidence are to be judged by their predictive power and whether you can think of any evidence. Theories with evidence are to be judged by their predictive power alone.”

posted by David Montgomery on May 14, 2008 #

I agree that this is an interesting correlation, and in general that we should look for mechanisms…

But at the same time we should always worry that the effect might not exist. For this one I know of three objections:

  • It needs a one-year lag between president and income. That’s not implasible but it’s one more parameter.

  • It uses “pre-tax family income”, from income tax I think? This is somewhat succeptible to adjustments to minimise tax paid. Higher taxes can scare the rich into sheltering income somehow. (I think this is the same data that leads to an underestimate of inequality in the 70s, for this reason.)

  • Weren’t the republicans supposed to be the party of fiscal prudence, earlier in this data set? In other words the categories shift over time.

posted by improbable on May 15, 2008 #

Shame on you, Aaron. You should know better than to look for awkward correlations between enormous generalizations.

I don’t know enough to argue, but even if there is a correlation between Dem’s and progress, we should be careful what you do with such correlations.

Consider one of those old hand-powered water pumps. Observing someone operating the pump would lead you to conclude that there is a clear correlation between pushing down on the handle and water coming out. Surely, we would be tempted to conclude, there is no point to lifting the handle at all. To maximize the amount of water we get, we should just sit on the handle.

Our economic and political systems are much more complicated than we have so far been able to comprehend, and looking for correlations between changing political labels and imprecise economic measurements is silly.

posted by Andrey Fedorov on May 17, 2008 #

Just FYI, the time-stamp on comments seems to be ~2 days off.

posted by Andrey Fedorov on May 17, 2008 #

People weigh a lot of factors when deciding whether a theory is plausible. If you have a demonstrated mechanism, that makes a theory more plausible than if you’ve just imagined up a possible mechanism.

It’s not “irrational” to include it in the factors weighed, even if it would be irrational to make it a be-all-and-end-all criterion for plausibility.

posted by on May 18, 2008 #

Maybe a concrete example would help. If my computer crashes while I’m depressed, I don’t even seriously consider the possibility that it did so because I’m depressed. If it does so 10 times, and none when I’m cheerful, then I start to wonder, but again, suspect that I’m missing something else that’s actually causing it. If it happens 1000 times, and none when I’m cheerful, and I can’t find any more plausible causes, then I start to accept that it’s caused by my depression, even though it seems unlikely, i.e. despite the lack of plausible mechanism.

posted by on May 18, 2008 #

I write about medical research. Here, people are more likely to believe a drug works if a plausible mechanism is known. But this is almost entirely because a plausible mechanism gives a boost to difficult decision-making under uncertainty. If there is clear empirical evidence, then people will use an important drug whether any mechanism is available or not.

Thanks for letting me know about Unequal Democracy, and the strong evidence of a Democratic/Republican difference. Clearly presidents can do a lot to mess up the country. If they can do it in part by changing widespread public attitudes and assumptions (as seems plausible), then the time frame to make actual changes could be a day or less. Of course the economic statistics will need some time to catch up.

posted by John S James on May 20, 2008 #

You can also send comments by email.

Email (only used for direct replies)
Comments may be edited for length and content.

Powered by theinfo.org.