March 4, 2004
Idealisation is the path to the dark side
Brian over at Crooked Timber is thinking about Milton Friedman's defence from the charge that economic modelling is bunk, because empirical evidence shows that people rarely behave they way economic models assume they do.
I think Brian is on the right track by saying that Friedman is partly right to claim that models which generate accurate predictions are useful even when their foundations are known to be crap. However, I think identifying the objects of scientific theories as idealisations is the wrong way to go, and that he's underemphasising something that offers Friedman the possibility of a much stronger defence.
Back in the 19th century, there was some serious economist - I can't remember who, but I remember that he was not considered a crackpot - who thought that the business cycle followed the sunspot cycle. It appeared to fairly accurately predict the British business cycle for a fairly long period. So, he put forward a model that claimed that the sunspot cycle (which was a fairly new discovery at the time) affected agricultural productivity, and that this had a knock-on effect on the whole economy.
This idea was taken seriously, and viewed with some impartiality it isn't ridiculous. In the 19th century, agriculture employed a far larger portion of the population and represented far more of Britain's productivity than it does today. People also spent much more of their income on food and fuel than at the present. So, since they suspected that the sunspot cycle affected weather and climate, they thought it could affect agriculture. There was a plausible theory for how sunspots could cause recessions.
There was only one problem with the theory, and it was a big enough problem to kill that whole line of research. It's complete bollocks.
It turned out that, when they cracked open the books and did an empirical study, they couldn't find any sunspot related variation in agricultural productivity. If Brian's summary of Friedman's arguments are accurate, Friedman is saying that you can't hold the lack of relationship between sunspots and agriculture against an accurate model predicated on just such a relationship existing. It is my understanding that there was enough divergence between the sunspot cycle and the business cycle that any late 18th and early 19th century correlation between the two was almost certainly a coincidence, and that with an additional century and a half of business cycle data, there is no longer any evidence of a relationship between the two.
Brian does point out that a model has more than predictive value, it has explanatory value. Boyle's law - the example usually given of a useful predictive model known to be based on a false assumption - is useful in part because it has explanatory value. Boyle's model assumes that gas molecules in a container collide with the sides of the container, but never with each other. It's a fairly accurate model, because for a lot of kinds of work the difference between the Boyle model and more sophisticated models is quite small, and the calculus involved in Boyle's model is a lot easier.
However, even though Boyle's model includes a known false premise, it still correctly explains why gas pressure changes when volume changes. The basic explanatory value of the model is unaffected by the false premise. Gas pressure is still the force of gas molecules bouncing off the sides of containers, regardless of whether they bounce off each other. Boyle's model is useful not only because it makes fairly accurate predictions, but because it offers a substantial and correct insight into the phenomenon that it's modelling.
The sunspot model of the business cycle, in contrast, doesn't have any explanatory value if its underlying premises are false. Even though sunspots may accurately predict recessions, if sunspot activity has no effect on agriculture, you still know nothing about what causes recessions.
Friedman could have offered a much stronger defence of economic models based on false premises about human behaviour by taking this tack. He might have said, look, the model assumes that humans are rational optimisers and clearly they aren't. But, the real insight of current economic models is that macroeconomic phenomena are an emergent effect of individuals undertaking local, self-centred economic activities. [Okay, in 1953 Friedman would never have used the word "emergent", but there were synonyms he could have used.] Economic models show how individual activity can produce macroeconomic effects, and how macroeconomic policies affect individual behaviour, producing macroeconomic consequences. All this is still true even if humans are half-assed, incompetent utlity optimisers. The kinds of models we produce are just as insightful as Boyle's law, even though - just like Boyle's law - one or more of the premises is nonsense.
I don't think Friedman made that defence. Furthermore, I'm not making any claim about whether or not the kinds of economic models he advanced really are insightful or make accurate predictions.
Brian is right to think poorly of Friedman's claim that facts are irrelevant in the face of theory. (Friedman is starting to sound a bit post-modern to me.) But, I think both Friedman and Brian (and Michael Strevens, who he cites) could make a much better defence of economic models which contain false premises without getting involved in the idea that scientific theories are about idealisations.
Idealisations lead to instrumentalism, instrumentalism leads to behaviourism, and behaviourism leads to the dark side. I'd much rather deal with poorly predictive false theories that tell me something about my real objects of study than highly predictive false theories that tell me about idealised objects.
Posted 2004/03/04 10:06 (Thu)
Jevons is your sunspots man, I believe.
More recent economists seem to have adopted 'sunspot' as a shorthand for a phenomenon that ought to be irrelevant to the economy but which can nonetheless affect it indirectly because people expect that it will have an effect. At least, that's what I think these economists are saying.
Mrs T, that would make the theory performative. Performative theories are a real pain in the ass for modellers, because the only way to effectively discredit them is to show that not only are the premises false, but that people's false belief in the premises is the real cause. Otherwise, it's hard to tell from an ordinary false but potentially insightful theory.
I may be misunderstanding, but I have a way of reading your argument that allows me to use the language of idealisations without slipping into the dark side, at least not in any obvious way.
Your argument seems to me is pointing out that there are idealisations that are useful--simply because the model that one generates predicated on them--has some particularly valuable insight into the more complex phenomenon one is really interested in understanding. There are other idealisations that may produce models that have no explanatory power or little predictive power, or neither.
Perhaps it would help to understand what you mean when you say that idealisations lead to behaviourism?
Pedro, the molecules in the Boyle model aren't idealisations. They are, perhaps, abstractions. They might be symbolic representations - I could live with that. I'm not even particularly bothered by the idea that they might be fictions with no existence outside of the minds of undergraduate thermodynamics students. But, I'd much rather see them as real molecules with poorly understood or misunderstood properties than to see them as idealisations of molecules.
A theory is not necessarily an act of idealisation. It may well involve some abstraction, but its objects should at least seem real to the researcher. A theory in which idealisations work on other idealisations and produce good predictions is grounds for thinking that the contents of the theory are not idealisations but in some sense real things, or at least as real as all the other things you have to deal with in real life. But a theory that tells you how ideal molecules would interact with each other in an ideal world can make no special claim to be helpful in understanding a non-ideal world filled with distinctly non-ideal molecules.
As for idealisations leading to the dark side (you do get that that was a Star Wars joke?), the path goes through instrumentalism. Behaviourism was the very distilled essence of instrumentalism. It claimed that the contents of minds wasn't a subject fit for scientific study, only stimuli and responses. (Okay, that isn't exactly what Skinner said, but that was how it was routinely taken.) Whatever abstractions it made about the contents of minds were purely instrumental in content - no statement about actual mental activity was intended. Any theory produced thereby might well serve to predict behaviour in some way, but it could never explain it.
BTW, Pedro, I realise that this is somewhat different from my position on the philosophy of science some years ago. Just because I've changed my mind doesn't mean that I'm not always right. :^) And, you don't have to be timid disagreeing with me. I've known you long enough to give you permission to say "Scott, you're full of shit" on my blog.
I read Friedman's Intro to Positive Economics some time ago, but I don't have notes here. So I'll wing it.
The example I worked with (maybe from Friedman) was Newton's laws, specifically relating to the solar system. Newton assumed that the sun and earth ar rigid with all their mass concentrated in the center. That was false but harmless, since the adjustments required to make the model more accurate later are small. Furthermore, these adjustments were later made (accounting for tides, irregularities in the sphere, etc.), and without resistance from orthodox Newtonians (AFAIK).
I don't find that to be true of economists. They don't seem always to distinguish between the assumptions of the system and the results. To a lot of them it's "all science", and anyone who questions the assumptions is just unscientific.
Furthermore, some economists' assumptions are not as minor at the physicists' especially because economics is not autonomous but has consequences in non-economic fields (which consequences, of course, economists by definition don't even know about at all -- "trained incapacity".) Examples include the idea that value=price, "revealed preference" = preference, or that resources are either effectively infinite or infinitely substitutable. (In Lomborg and Simon's "cornocupianism" you see the latter assumption asserted as fact).
When arguing policy, almost all economists shift invisibly back and forth between "I'm just and economist and leave other topics to be studied by people in other disciplines" to "Economics is by far the most scientific social science, and results from other so-called social sciences need to be taken with a large grain of salt."
Political science, sociology, anthropology, economics and history all really are competitive explanations of the same subject matter. They do not study "different things". Economics' brilliant success in studying the things it does study does not prove that the things studied less well by other less scientific sciences are not real. It just shows that some aspects of the whole subject are better understood than others. Many scientists come close to believing that if it can't be described in economics, it isn't real.
Similiar things happened after the success of physics. Models from physics have been applied to other areas for centuries, sometimes successfully but often ridiculously. (The ludicrous attempts tend to be aggressively forgotten, making the progress of science look infinitely neater than it really was). But chemistry, biology, and even some area of physics had to adapt or abandon Newtonian principles in order to progress. (Order out of Chaos, Ilya Prigogine.)
By and large I think that economics (like many discourses) is incontestable within its definitions, but problematic at its boundaries (if you ask what would be true if certain axioms were changed). The problems involved are major rather than minor, and almost no economist seems aware of them.
To clarify: economics and political science study the same thing (human society). But they talk about different aspects of the one thing. In that sense they study "different things".
I would even go further to say that the less successful social sciences are less successful in part because they're more inclusive and comprehensive. Economics is an brilliantly-made ideal intellectual system which is enough like the actual system to be powerfully applicable, but too partial to be trustworthy. History is less powerful but -- for that reason -- also less fictitious and more complete.
Economics has a bias toward production, consumption, and exchange, and a bias against all other kinds of human activities and non-activities, and against all other kinds of organizationand also interpretation of human life. We're seeing all the other sciences reconstituted on economic principles nowadays, sometimes with good results and sometimes with bad ones.
ADD: Family life is something that economics has little insight on so far. Economically, having kids is a bad deal for anyone who is in a position to save and invest, and no one in that situation should ever have kids. Only subsistence farmers who are unable to save and will need to be supported in their own age should have kids. I've read only scraps of Gary Becker's economic models of the family, but it seems wildly ridiculous.