When reading (many) economists’ take on methodology, I feel the urge to mutter “Popper!” in the tone that Jerry Seinfeld used when Newman walked in the room. His long and wrongheaded shadow casts itself darkly across contemporary economics, and we are the worse because of it. And Popper’s influence is mediated through the two most famous essays on methodology written by economists: Friedman’s 1953 Positive Economics paper, and Paul Samuelson’s response. I linked to the paper in this post, from a 1973 AER, because it gives a nice, simple summary of both (all economists should at least know the outlines of this debate), and argues against Samuelson – this is rare indeed, since it is quite a struggle to find anyone with a philosophic background who supports Friedman’s take. Easier is to find a philosopher who thinks both Samuelson and Friedman are mistaken: more on this shortly. I should note that this post will, in all likelihood, be the only time anything negative will be said about Samuelson (aka, the GOAT).
Friedman is basically an instrumentalist. This means that he sees theories in economics as a generators of predictions. Good theories predict what we care about well. Bad theories predict that poorly. Wrong assumptions do not matter, only wrong predictions about what we care about do. (A quick note: Kevin Hoover argues that we should think of Friedman as a “causal realist” rather than a pure instrumentalist. To Hoover, Friedman’s methodological stance is that economists should look for the 1) true 2) causal mechanisms underlying 3) observational phenomena. He is firmly against a priori axioms, and also thinks that the social world is so complex that theories are by necessity limited; these two facts mean that the relevant “assumptions” of a theory are generated in a cyclic process by which data gives rise to assumptions which are held until we get better ones. Testing assumptions is done by testing the data. There’s something to this point, but I think Hoover is missing just how much Friedman emphasizes prediction as goal in his ’53 essay: prediction is everywhere in that essay!)
Samuelson responded with his famous “F-Twist,” the F being Friedman. Imagine that assumptions A in a theory B lead to conclusions C, where C-, subset of C, is what we care about. Imagine that C+ is true. If that is the case, then a better theory B is one which uses assumptions A-, subset of A, in theory B-, to make only predictions C-. In such a case, B- is a representation theorem, and A- and C- are logically equivalent: A- implies C- and C- implies A-. But then how can one say the reality of assumptions do not matter but the reality of conclusions do? False assumptions necessarily imply false conclusions. Samuelson, like many modern decision theorists, sees economics as a tool for description only. We show that two sets of statements – say, SARP-satisfying revealed preference and ordinal utility theory – are logically equivalent. The best theories are the simple statements that are logically equivalent to some complex social or economic phenomena. Theory is a method of filing. Lawrence Boland, in a few papers, has argued that Samuelson and Friedman are not that far apart. They are both “conventionalists” who do not think the best theory is the “true” one, but rather judge theorems based on some other convention, prediction in Friedman’s case and simplicity in Samuelson’s.
Wong’s point in this essay is that Samuelson’s ignores difficulties in parsing theories. Going from A to A- can be a tricky thing indeed! And, in Friedman and Popper’s worlds, it is not even that important, since the relevant predictions won’t change. And once you think theory should, in practice, be more than just a logical equivalence, you can see that theories have much more content that their assumptions alone: the truth of A can imply the truth of C, but the truth of C does not necessarily imply the truth of A. To Wong (and presumably Friedman), statements like “any deductive proposition is tautological; we can only say that some are less obvious or less interesting than others” (Solow) miss the point that logical equivalence is not the only theoretical relation.
Now pure instrumentalism is pretty ridiculous. As Harry Johnson noted, “the demand for clarification of the mechanism by which results can be explained is contrary to the methodology of positive economics with its “as if” approach.” Restricting the use of theory to making good prediction goes against everything writers on the methodology of social science have argued since the 19th century! In particular, there are three huge problems. First, many models can predict the past well, so will be evaluated at the present equally by an instrumentalist. How shall I choose between them? Second, in a stochastic world, all theories will make “wrong” predictions – see Lakatos. Third, “lightning never strikes twice” in social science. The social world is ever changing. To the extent that we think there are deep tendencies guiding human behavior, or the concatenation of human behavior in a market, theory can elucidate the implications of such tendencies even if nothing is explicitly predicted. Equilibrium concepts in games can help guide your thinking about all sorts of social situations without presuming to “predict” what will happen in complex social environments. I’ve got one more methodology post coming shortly, from the Cartwright/Giere style of philosophy of science, which discusses why even pure scientists should (and do!) reject “the one true model goal”, or the search for a true (in a realist sense) description of the world. The one paragraph argument against the one true modelers is Borges’ great parable “Del rigor en la ciencia”.