The methodological reflections of Nobel laureates are not always models of rigour. This is something which should be readily forgiven; winning the supreme mark of achievement in science surely is in part a licence to put a bit more emphasis on personal opinions relative to intersubjective consensus. However, I am pleased to report that Vernon Smith's new volume of opinions amounts to the most important book on economic methodology of the past decade. I say this despite the fact that about a third of it frames its arguments inside a philosophical framework that I believe perpetuates and encourages confusion. But the damage caused by merely philosophical muddles is minor, and massively offset in this case by the clearing away of economic confusion that will occur if a good number of economists – especially younger ones – read Smith carefully and take his advice to heart.

That advice can be summarized succinctly, though the summary hides a wealth of crucial details. Economics as a distinctive science is the study of markets, not the study of the computation of decisions by individual people, which is the province of psychology. This point has bite precisely because neither the static nor the dynamic properties of markets reduce to aggregations of properties of individual computations of choices over allocations of time and capital. There are two main reasons for this. i) Markets encode a great deal of structure to which people accommodate their behaviour. People could not individually compute the rationales for these structures in real time (if at all) but fortunately do not need to. ii) The processing substrate of much exchange behaviour that is individually computed consists of sub-cognitive parallel perception and unconscious modulation of response dispositions, rather than consciously mediated, logically governed decision and choice. These claims do not jointly entail the irrelevance of decision and choice processes to economics, for such processes are indeed essential parts of the production of individual economic behaviour and of market characteristics. But they are parts only; they interact dynamically with, rather than determine, market properties and changes; and they are derivative, not primary, objects of theoretical concern for an economist who is not implicitly switching disciplines and taking up psychology.

Smith argues that these central claims about the basis of economic behaviour were appreciated by Adam Smith, Hume and Hayek. He argues that they are not properly appreciated, and are in some cases missed or denied altogether, by “neoclassical” economists, by behavioural economists and by many applied game theorists. (I put scare quotes around “neoclassical” because the exact designation of that label is endlessly arguable. Smith means by it roughly the tradition now realized in applied general equilibrium theory and rational expectations theory.) What all of these protagonists have in common, according to Smith, is a disposition to try to explain too much economic behaviour by reference to “constructive” rationality and too little by appeal to “ecological” rationality. The former is explicit computation. The latter is akin to what Polanyi (Reference Polanyi1962) called “tacit knowledge” and what Wittgenstein (Reference Wittgenstein1953) called “knowing how to go on”.

I fear that, despite the opening chapters Smith devotes to the topic, the distinction between constructive and ecological rationality is likely to remain obscure for many economists after they read the book. They are likely to equate the distinction either with that between conscious and unconscious processes, or with that between deduced beliefs and “instincts”. As Smith says, Hayek's Sensory Order makes clear that neither of these understandings is correct. However, few economists will want to invest in learning the special-purpose conceptual framework and terminology needed to appreciate Hayek's highly prescient but amateur and half-century old speculative cognitive neuroscience. Fortunately they do not need to. Smith's elaboration, and the comprehension of those who want to follow him further, would benefit from perusal of up-to-date work by some philosophers, notably Andy Clark (Reference Clark1997, Reference Clark2008) and Robert Wilson (Reference Wilson2004), who have elaborated clear structure, informed by current behavioural and brain science, for the distinction that Hayek and Smith are after. The crucial property of “constructive” knowledge isn't that it might be consciously accessible. If I perform a simple numerical calculation (say, subtracting 49 from 57) “in my head” (as people say), there must be part of the process – in the subroutine sense of “part” – that is performed at a level of synaptic transformations that I can only know about theoretically and inferentially. What makes such subconscious operations part of constructive rationality is that they are performed, as Clark puts it, “inboard”, using my individual biological wetware. The operations, though subconscious, are still constructive in Hayek's sense. In fact, however, even this simple operation typically also relies on elements of ecological rationality. Suppose that at the conscious level I form a “mental image” of the two numbers, with the 57 “on top”, and then operate on the “image” so as to “cross out” the 5, “replace” it by 4, and “move” 1 “rightward” to make 17, so that I can subtract 9 and yet get a positive integer. (I put scare-quotes around all imagistic language here to remind the reader that none of this actually involves manipulation of real pictures.) In following this recipe I rely on structures furnished by my culture, based on millennia of collective learning and preservation. The structures in question include the decimal system for representing numbers, and the convention for representation of processes of “carrying” from one decimal place to another. All fully numerate people can carry out such operations, but most could not explain the rationale for their effectiveness. This rationale is “stored” in the historical record and in the knowledge of experts. (An intelligent non-expert could deduce the rationale if she wanted to; such deduction would then make the relevant rational capacity constructive for that agent.) Clark refers to all knowledge that a person draws upon from her culture but doesn't explicitly justify for herself as “scaffolding” for that person. A key property of much scaffolding is that most people don't try to preserve any standing inboard representation of it. For example, I can recite my two cell phone numbers from inboard memory; but my knowledge of the land line numbers at my two houses is instead stored in a few handy devices from which I can quickly retrieve it when needed. These devices are themselves scaffolding for my knowing how to go on when you ask me how I can be reached at home. In an odd-sounding but perfectly straightforward sense, these devices are parts of my mind outside my brain (Clark Reference Clark2004).

Why does this matter to economics? Anti-economists often try to embarrass economists by scoffing at the idea that (for example) a shopper in a supermarket consults his utility function and budget to compute the marginal opportunity cost to him of adding a bag of lettuce to his basket. They succeed in achieving such embarrassment if the economist has no better reply than Milton Friedman's (Reference Friedman1953): I can predict the shopper's behaviour by pretending that the shopper has such-and-such (inboard) representations and performs such-and-such (inboard) computations, and it's none of my concern as an economist that these pretences are obviously false. Never mind the metaphysical mystery Friedman's line conjures up; it is in any case unsatisfying because it provides no guidance whatsoever on what other fantasies the economist might usefully assume, so it conflicts with the pursuit of scientific generalization. In fact, as Hayek emphasized, the shopper's knowledge of (approximate) opportunity costs is encoded in his knowledge of and expectations about (approximate) relative prices of items he himself buys, and his own income. The shopper need not personally construct most of this knowledge. The prices posted all over the store are parts of the scaffolding on which he relies, along with newspaper forecasts about inflation rates, conventions about the meaning of his pay-cheque and bank statements, and, indeed, the whole system of money, property ownership and rules of exchange into which he was enculturated as he grew up.

The “hard-nosed” neoclassical economist might say that the distinction between constructed knowledge and scaffolded processes is none of her business. (See, for example, Gul and Pesendorfer Reference Gul, Pesendorfer, Caplin and Schotter2008.) However, such an attitude is both hasty and arbitrary. In my country, South Africa, many people are only now learning about such important parts of economic scaffolding as the formal credit system, from which their parents were diabolically excluded by the previous regime. Being people, they are learning ferociously quickly. But they are not learning instantly, and this is very important to our current savings rate, interest rate, balance of payments, exchange rate, and every other core input to everyday economic policy. As I write, with credit sharply contracting, the problem could knock us off our political equilibrium path as new members of the black middle class have their new houses and cars repossessed. It is peculiar to imagine that an economist should be obligated by disciplinary division of labour to leave concern for this sort of phenomenon to other kinds of social scientist.

Smith's book does not belabour this point. Much more of his argument is aimed at correcting the over-reaction to neoclassical abstemiousness found in that part (the majority, unfortunately) of the behavioural economics literature that is constantly waving flags for paradigm-shifting revolution. If it is possible for an economist to pay too little attention to actual processes of adaptation to incentive-changing structural dynamics, so it is possible for her to pay too much attention to them. This is for the simple reason that economics is ultimately concerned with economic generalizations. There are principles of cat skinning one will miss if one endlessly and exclusively compares properties of the blades of knives. South Africans are no doubt encoding their appropriation of their new scaffolding in culturally distinctive ways that differ from the parallel learning that has been happening over the same period in Mongolia. For the psychologist or anthropologist these differences might constitute both important data and interesting explananda. The economist might be best advised to abstract away from these differences if what she wants to understand are implications for the deepening of credit (and other) markets in the two countries. I have emphasized “these” because there might be other differences that matter to her. For example, if the two cultures have different traditions for legitimating inter-generational wealth transfer, this could lead to divergent market institutions and behaviour even after learning equilibrates. This empirical possibility, Smith would stress, would best be investigated by performing controlled experiments. The point for the moment is that there can be non-economic differences between groups of people that make a noteworthy difference to their market structures, and other non-economic differences that don't. Behavioural economists often reason as if every difference is automatically their professional concern. The result is a new version of economic imperialism with a hermit crab twist: the economist moves into the homes of psychologists and anthropologists while leaving her former disciplinary shelter uninhabited.

I have provided a didactic gloss, at a very high level, of Smith's general perspective. This is because my space is cramped, not because Smith's book is structured that way. He gets his straight polemical focus established in three short chapters, and thereafter interleaves his methodological morals with analysis of real case studies and experiments. Let me therefore list a sample of the tapas the reader is served, which, by the end of the meal, amount to a filling methodological lesson. There is an argument, based on experimental auctions, against the idea that non-cooperative Cournot-Nash equilibrium can only arise if every player individually computes it; market structures often scaffold the constraints that support it. We get an explanation of why agents in experimental simulations of entrant-incumbent scenarios don't need government sanctions against predatory pricing, even though inboard computation by agents of the conditions for equilibria with accommodation of entry were too subtle for even economic theorists to spot from their armchairs. One of my favorite morsels is a stinging attack on the anti-market rhetoric that Akerlof and Stiglitz have derived from their analyses of information asymmetries (and which the popular business and political press has discovered and swallowed whole). Smith apportions intellectual garlands exactly correctly when he notes that their co-laureate, Michael Spence, has been more careful. Informational asymmetries can be understood as circumstances in which different agents have access to different scaffolding. It is strange to refer to the processes by which markets adapt to such asymmetries as “market failures”; in both natural and experimental settings, they are typically illuminating opportunities to learn how market structures evolve and how people adjust strategies to market structures. What Akerlof and Stiglitz are really getting at with their rhetoric is that welfare asymmetries are often consequences of information asymmetries. It is entirely unclear, as Smith says, why this should be any less generally problematic in environments managed by regulators than in environments governed by relatively laissez-faire policies. The point here is not that Chicago School ideologues should take heart; it is that as input to policy and disinterested scientific understanding ideology from either side of the pop political spectrum is a poor substitute for case-by-case experimental study.

These examples might give the impression that Smith's book is an instance-based campaign for the general superiority of experimental over armchair economics. It is therefore important to point out that he repeatedly scores blows against the dominant, behavioural, version of experimental economics. Because behavioural economists are so keen to prise open black boxes in subjects’ heads, they often deliberately shut off subjects’ access to the sort of scaffolding that is crucial for equilibrium learning in markets outside the lab. Again, this point is made by careful critical analysis of specific sequences of experiments. An example is the literature on attitudes to risk. The search for one general account of such attitudes – based on EUT, Prospect theory, or one of the new-fangled complex alternatives – is usually carried out by trying to put subjects in situations cleansed of all confounds. It is easy enough to see the point of this. But what get treated as “confounds” might often be the scaffolding by which people reduce uncertainty in real markets. The evidence is increasingly persuasive for Smith's conclusion that there is no general, true theory of how people respond to risk using inboard computational processes alone because people don't rely exclusively on such processes unless experimenters force them to. (Compare: How do spiders engineer webs in general – that is, when they have no particular structures on which to anchor them? The question is silly.) This does not mean that there are no generalizations to be had about how people respond to risk given different kinds of market structures.

I said earlier that I have one general complaint about Smith's philosophical framework. This arises in Part III of the book, and concerns his interpretation of the relationship between game theory and theories of human behaviour. Smith follows Myerson (Reference Myerson1991) in incorporating into the foundations of game theory a commitment to the prediction that people will always choose strategies that are strictly dominant with respect to money payoffs. This of course guarantees that game theory will seem to be “refuted” by empirical observation and experiment. It also implies that game theory can't be applied to non-human animals. Some game theorists really have thought this, precisely because they assume that players must solve games by constructively computing their solutions (e.g. by literally running backward-induction algorithms “in their heads”). Because he allows them this assumption, Smith convicts applied game theorists of the neoclassicists’ error of ignoring Hayek's insight about ecological rationality. Smith's philosophical frame here is oddly inconsistent with the over-arching one of his book. Why does he not allow that people can (often) find Nash equilibria, or subgame-perfect equilibria, or Bayes-Nash equilibria, just as they (often) find competitive market equilibria, by relying on scaffolding instead of inboard computation?

It seems clear from the text how Smith would answer this question. At one point he says “I think we have a right to expect utility functions . . . to be tolerably stable across the situations that yield the outcomes, and not to be jumping all around in response to different contexts” (p. 242). Significant space in the book is devoted to resisting recently popular arguments from behavioural economists (e.g. Gintis Reference Gintis2006) for the hypothesis that people have an innate preference for fairness. The main argument for this is empirical, but it is supplemented by a philosophical objection (also registered by Binmore Reference Binmore2005 and elsewhere) to unprincipled adoption of exotic utility functions to handle apparent violations of preferences for dominant outcomes. Smith and Binmore agree that a less ad hoc explanation is that people frame games intended by experimenters to be one-shot as repeated, because one-shot games are ecologically uncommon. (Both also stress evidence that people can learn to play one-shot games.) I certainly think that Smith and Binmore have the right side of this dispute. However, we should resist a false dichotomy between allowing that utility functions are indefinitely contextually variable and insisting that they must range only over outcomes defined with respect to private material payoffs. I think there is strong evidence, for example, that all psychologically healthy people, in all cultures, prefer to be thought to conform to norms of reciprocity, all else being equal (Bicchieri Reference Bicchieri2006). I see no convincing reason not to aim to incorporate this factor into empirically estimated utility functions.

As noted at the outset, this objection seems to me a minor one in the larger context of Smith's vision of economics as a discipline that is empirically grounded, a fecund source of techniques for policy pre-testing, and yet clearly distinct in its subject matter from neighbouring social sciences. I am grateful to the Swedish committee for enhancing the authority with which Smith addresses the profession.