Scientific Theory as Predictive Machinery

A reply to Milton Friedman's economic methodology...


Milton Friedman’s 1953 paper is an attempt to rescue economics from scientific realism. He argued that the primary criterion of validity for economic models was not the “realism” of the assumptions, but the accuracy and importance of the predictions generated by its implications. Realism has many uses in the language of science. Its primary use, of course, is in the interpretation of “unobservables” in scientific theories, that the success of science involves the status of unobservable entities talked about by scientific theories. If one is a scientific realist she says the unobservable things talked about by science are little different from ordinary observable things (such as tables and chairs). For the realist with respect to economics, these unobservable entities include the assumptions of economic theories, such as that businessmen supply commodities up to the point where marginal cost equals marginal revenue.

Friedman’s paper takes a different position than that of the physical scientist. His paper is methodologically committed to instrumentalism, while being epistemologically pragmatist. The two are closely related. An epistemological pragmatist is oftentimes a methodological instrumentalist, much like John Dewey. It is also clear that Friedman places little confidence in the reasons the realist might proffer for believing significant portions of what a theory says about unobservables, and in other words is an antirealist with respect to the ontological status of the unobservables in economics.

While it was commented that the philosopher might not find any “new” ideas with Friedman paper, the idea of positive economics, as I will argue in this paper, is given new meaning. Friedman is knowingly influenced by Popper (although he is no Popperian, as I shall argue) when he links positive economics to the Popperian ideal of science as conjecture and refutation. In Friedman’s version the conjectures do not necessarily need to be realistic, but only that they give rise to predictions that can be empirically verified. This formula, when substantially followed up by empirical testing, gives the economist a powerful device for developing scientific machinerey with strong predictive records.

“Viewed as a body of substantive hypothesis, theory is to bejudged by its predictive power for the class of phenonmenon which it is intended to ‘explain.’” (Friedman 1953)

Friedman’s instrumentalism fundamentally denies that theories are truth-evaluable, and that they should be treated like a so-called black box into which you feed raw observed data—“brute sensations”—and through which you produce observable predictions. As a study in philosophy, it is apparent that the Friedman method then requires a distinction between theory and observation, and within each type a distinction between terms and statements. This has been covered extensively elsewhere, and for that reason will not be repeated here. But it is interesting to note that the theory-language of Friedman’s methodology, if explicitly developed, would be very close to the theory language developed by the logical positivist school. If Friedman’s methodology were thoroughly exhausted, we might end up with something quite like the protocol sentences and observation sentences found in the positivist literature.

Friedman is curiously obsessed with factual evidence. He espoused the view that “factual evidence can never ‘prove’ a hypothesis; it can only fail to disprove it.” This sounds admirably Popperian. But Friedman’s “fail to disprove” is not synonymous with falsification. With every failed attempt to disprove a theory, one actually lends weight to the confirmation of that theory. (One could arue that was Popper’s project too.) But the theory itself does not depend on being “disproved” by a single shred of evidence, as in Popper. It’s not obvious whether Friedman even encourages the falsifying of theories, but it’s clear that he’s concerned about the verification of certain theories he approved of. On this account, one could simply ask those who are working on problems in support of your ideological position to “please work harder at verifying them.” In contrast, Popper would ask of the same people to “please attempt to falsify your work if you really think you’re on to something.” But Friedman’s verificationist methodology is wrapped neatly in a somewhat naïve account of empiricism.

“The only relevant test of the validity of a hypothesis,” says Friedman, “is comparison of its predictions with experience.”

This sounds, again, admirably Popperian in its repeated stress on testing. But “the only relevant test of validity” is, essentially, verification with the economist’s brute sense experience. (That is, the verification one finds in economic indexes and journals.) This is a very interesting philosophical position indeed because, together with his rejection of realism as a test of theory, meant that the problem of induction was simply done away with. Friedman apparently does not hear David Hume rolling in his grave. The problem of induction is not the most interesting of problems in economics. It is of immense importance, however, to its own philosophically empiricist foundations. The problem classically involved the epistemic disconnect between what was observed in the past and predicting what will be observed in the future. Justifying induction on grounds that it has worked in the past, Hume says, is essentially begging the question—a fallacy students of philosophy are perhaps all-too-familiar with.

Friedman is fundamentally un-Popperian in this sense due to the fact that Friedman is arguing from a naïve inductivist position. His position begs the question of the validity of induction. Popper sought to resolve this problem by placing it within a scientific context: science does not necessarily argue from induction, but rather, science argues by deductive principles. Popper essentially makes “modus tollens”—proof by contrapositive—the centerpiece of his methodology. And on this account, of course, one should pay more attention to that which potentially falsifies the argument rather than what confirms it.
Moreover, according to Popper, the more the theory “exposes its chest” (i.e. the more risk the theory is and the more it tells us) the more vulnerable it is, the more it is likely to be accepted. But with Friedman’s instrumentalism, there is no room for the theory’s neck to stick out. It cannot test its own assumptions. It’s only testable in a non-comparative way. We cannot tell whether one theory’s neck is sticking out further than any other’s, since we have no adequate account of what a realistic/unrealistic assumption is.

Friedman suggests that the testing of assumptions themselves is not important. The assumptions are not the explanandum, (“that which must be explained”) they are the explanans (“that which does the explaining”). Friedman does not care too much about whether that which does the explaining is “realistic”. Its simplicity is to be taken into consideration, he says, but what counts as simple is not adequately elaborated in the paper. In this sense, Friedman’s unobservables are decidedly un-Popperian. They’re strictly axioms and the scientist should not trouble himself with their bothersome philosophical justification. They are, in a word, instruments of science.

We can see how Friedman’s methodology is unlike Karl Popper’s. How, then, is it like the methods espoused by those of the logical postivist school? The simplest answer is that one can see Friedman’s method suffered from the same problems as the positivist school method suffered. In principle, then, Friedman is positivist since his method adds little novelty to the subject, and hence could not solve the problems of earlier methods. Friedman is consciously aware of Popper, and he covers his verificationist residue in language the Popperians would find pleasing. For example, his repeated stress on testing of hypotheses with sense experience. This is not, as I have argued earlier, the same thing as falsificationism.

Friedman’s view is that concepts and theories are merely useful instruments whose worth is measured not by whether the concepts and theories are true or false (or correctly depict reality), but by how effective they are as predictive machinery. In fact, however, it is a positivist account.

Besides their concordance with brute sense experience, theories must also be judged by another criterion: whether the alternatives to the theory are also acceptable. But just as the logical positivists could not choose the better theory between Freudian and Adlerian psychology, Friedman cannot choose the better theories in economics.

I want to stress that Friedman is a relativist with respect to the unobservables in the theory. Since these unobservables could, in fact, be wildly diverse and Friedman has no adequate method with respect to the reality of the unobservables themselves. The method says that whatever does have the strongest predictive record is in some sense worthy of the title “scientific theory”. But on this account one could have a theory of rational behavior interpreted instead through, say, a Christian account of sinful nature. Friedman says choose the theory which predicts most and puts to use the least unrealistic assumptions. But this does not solve the theory choice problem, since we cannot test whether one is putting to use more or less unrealistic assumptions. Moreover, on this account, the scientists cannot even tell us what a realistic assumption looks like. Friedman simply plays on our intuitions as to what a “realistic” and “unrealistic” assumptions looks like. We somehow know this already, and thus Friedman again begs the question. And in the last analysis, all Friedman arguments against what he railed most, Paternalism, is merely an elaborate “intuition pump” as Daniel Dennet says. The entire argument is built into the assumptions.

Further, what would Friedman be able to say about the demarcation problem between science and non-science? His theory does not adequately guard against pseudoscience, as Popper noted of the logical positivists. If a non-science model could arrive at higher predictive records, then Friedman has nothing to demarcate between the two. In fact, he ought to prefer the non-science model since it has a better predictive record.

As a final note, Friedman and the positivists share essentially the same notion of what the job of the philosopher is. This is particularly interesting to me, as one who studies philosophy. Hands writes in Reflections that the logical positivist movement redefined the job of the philosopher as turning philosophy into a kind of “conceptual cleanup operation.” The philosopher is to point out to the aspiring scientific community what was (and was not) meaningful discourse. Likewise, Friedman redefines the job of the philosopher in a similar fashion. Formal logic, Friedman says, is a tautology, and these philosophical tools are “aids in checking the correctness of reasoning, discovering the implications of hypotheses.” (P186.) This should come as no surprise, as I have argued that Friedman’s methodology is simply a lapse into positivism. The philosopher’s duties once again include janitorial duties like sorting out all this brute empirical data and storing it in some vast “analytical filing system.” The philosopher, then, is once again a discourse-Nazi and a merely the secretary of science.

What Friedman was trying to establish was a framework in which economic knowledge could not be attacked at the foundations. It could not be attacked on the basis of realism (its most-feared opponent) because the task of economic model-building is not to aim the predictive machinery towards cranking out realistic assumptions. The Friedman case for an instrumentalist methodology is built into the argument presuppositionality, and thus should be refuted before it can get off its feet. It comes as no surprise to Popper that this method should be tautologically irrefutable, but this lends far less credibility than is often believed.