Regular readers of this weblog are surely aware that my position on economic methodology can be summed up as follows: if what you are doing looks like Popper or Friedman (1953), you are doing something gravely wrong. Friedman’s positive economics essay is so influential, though, that we need continue battening down the hatches as its destructive tempest swirls. The present paper, an oldie but goodie from Bruce Caldwell, offers great assistance in that endeavor. (You may wonder, then, if not Popper and Friedman, then what? I should save this for another day, so as not to scare off the optimists, but basically, social science will not find “truth,” whatever that is. It’s not clear if social science “progresses” where progress is defined as “proving more and more laws about human behavior”. We can, with hard work, provide people intuition about how to think about situations they encounter, but we are at best guides and are certainly not diviners. All this gobbledygook about “predictions” and “true (or false!) assumptions” and pseudoscientific numbers is worthless. And what’s worse: we’ve known this since Mill wrote in the mid-1800s about the multitude of causal factors in social science, and yet we don’t learn. In any case, let’s get back to Friedman.)
If you don’t remember Friedman ’53, the basic argument is that the point of theories in economics is to predict those facts we are interested in predicting, that the “truth” of assumptions is irrelevant to this endeavor, and that if multiple theories work, we should go with the simpler one, somehow defined. Caldwell, among others, notes that this looks a lot like a philosophical position called instrumentalism, which claims that scientific theories are not true or false, but rather are simple instruments like a hammer for understanding and predicting the world. After Friedman’s article was published, a huge literature developed noting all manner of philosophical problems. Lawrence Boland claimed that critics were wrongheaded because they didn’t understand that Friedman was methodologically an instrumentalist, should only be criticized under that lens, and further than instrumentalism had never been refuted as a philosophical position. Caldwell takes aim at that last clause.
So what’s the problem? First, basically no philosopher of science thinks that prediction alone is the goal of science. Rather, explanation is more important. Consider this from an economics perspective. A predictive model might say that if the rate of growth of money is X, then the inflation rate will be Y. But surely no one is satisfied with a paper that gives the result of a regression of the US money supply on inflation, and reports that table. We want to know why such a relationship might hold, what might change the relationship, etc. That is, to restate Hume’s problem, at the time a prediction is made, any prediction about the future is equally accurate. To the extent that we can induct from the past to the future, it must be because we believe some underlying law explaining the relationship in the past and in the future. And how can we judge the explanation if not by appeal to something beyond the prediction itself?
Caldwell goes beyond this point, though, and discusses the common instrumentalist objection to it: if there are an infinite number of theories which equally well “explain” an event, how am I to decide between them? Isn’t the whole concept of “truth” bunk? And if we agree that there is no “true” explanation, we have nothing left but prediction (more on this wrongheaded implication in a second). The problem here is that there is, in fact, some theory T that is true, whether we have found it or not. The noninstrumentalist will believe his theory true by convention, or true for the time being. Without this step, there can be no delineation between science and nonscience (since both make predictions), and there can be no belief that we are reaching a “more true” understanding of the world.
Further, Friedman himself clearly didn’t practice his own methodological system. Caldwell discusses Friedman’s Nobel speech about out understanding of inflation and unemployment, where Friedman speaks about three progressive systems to understand the link, ending with something like the rational expectations framework. But even if the RE theories explain “more”, they certainly do not “explain more with little”; Friedman himself recommends RE on grounds that look a whole lot like support for their richer and more accurate assumptions! There are many similar examples. For one, it doesn’t look to me like Friedman thinks in terms of cardinal utility rather than ordinal utility, despite the fact that ordinal utility is a much more restrictive take on human behavior with little to no added predictive power.
This is all a long way of saying the basic problem of Friedman’s reasoning. He thought that when theories say assumptions A implies consequence B, and A is wrong, we need to save the inference by claiming that we only care about B. But a better philosophy of economics goes a third route: we don’t need to care about A or B, but rather care about the implication itself, the proof, the reasoning that gets us from abstraction A to abstraction B! Take the example of an object falling in a vacuum. The assumption A that whatever object we care about is in a vacuum is untrue. The conclusion, say, that feathers and stones fall at the same speed is also untrue. But the reasoning, the guided intuition, that it is not properties of objects but rather properties of the medium they move through – friction – which causes rates of falls to differ, is nonetheless useful even when assumptions A are not true and predictions B do not hold. That is what good social science looks like. And this matters! As you read this, some woeful economists is having his article rejected because the data do not match the predictions of his model, or because the assumptions of the model are unrealistic (in the sense of being implausible, if not untrue). But these critiques only matter if they affect the guided intuition of the paper in a meaningful way.
http://econ.duke.edu/~bjc18/docs/Critique%20of%20Friedman’s%20Instrumentalism.pdf (Final version from the 1980 Southern Economic Journal. I also came across the following interesting new article on the influence of Weber and Parsons on Friedman’s philosophy of science. Weber’s 1904 methodology text actually contains basically all of the good parts of Friedman’s article, though it leaves out the Popperian notions.)
Very perceptive post about what makes a good theory. I wonder if understanding this methodological difference differentiates good theorists from the also-rans?
What is your take on the validity of empirical studies in the top journals?
Good empirical work is trickier in many ways. I am about to post something along these lines, via a new paper by David Card and friends.