Digressions&Impressions

On Ellsberg's defense of Knightian Uncertainty, Arrow, and Ambiguity (II)

In most of this literature, the term ‘ambiguity’ has been treated as a synonym for what Knight called ‘uncertainty’ namely the fact that relative likelihoods are not characterized by well-defined numerical probabilities. The standard method of dealing with ambiguity in decision theory is to endow the decisionmaker with multiple priors as in Gilboa and Schmeidler (1989). This approach may be combined with a variety of preference models, notably including the maxmin model of Gilboa and Schmeidler (1989) and the smooth model of Klibanoff et al. (2005).
For a non-specialist this is puzzling; there is no obvious link to the ordinary meaning (or meanings)1 of ambiguity as a characteristic of propositions with more than one interpretation. In its normal usage, ambiguity is a linguistic concept, but in decision theory it is typically treated as a property of preferences.
The now-standard usage is quite different from that in Ellsberg’s (1961) original article. Ellsberg treated ambiguity, not as a property of preferences or relative likelihoods, but as a property of the information on which judgements of relative likelihoods might be based.

“Responses from confessed violators [of the expected utility (EU) axioms] indicate that the difference is not to be found in terms of the two factors commonly used to determine a choice situation, the relative desirability of the possible payoffs and the relative likelihood of the events affecting them, but in a third dimension of the problem of choice: the nature of one’s information concerning the relative likelihood of events. What is at issue might be called the ambiguity of this information, a quality depending on the amount, type, reliability and ‘unanimity’ of information, and giving rise to one’s degree of ‘confidence’ in an estimate of relative likelihoods.”

---Simon Grant, Ani Guerdjikova, John Quiggin "Ambiguity and awareness: a coherent multiple priors model." (17 April, 2020) [quoting p. 657 of Ellsberg's (1961) "Risk, Ambiguity, and the Savage Axiom]

In Yesterday's post I noted that while writing in defense of Knight, Ellsberg (1961) innovatively identified Knightian uncertainty with a class of decisions that can be specified precisely as empirical violations of the (Ramsey-)Savage axioms.* In particular, in such instances (when the formalization is violated) there is "no way to infer meaningful [subjective] probabilities for those events from their choices, and theories which purported to describe their uncertainty in terms of probabilities would be quite inapplicable in that area." (646; [parentheses added].) This identification was an innovation made possible and intelligible by formalization.  

As I remarked, Ellsberg noted that it is possible to treat the very same formalization as a [A] normative standard for a certain kind of ideal/idealized agent or [B] as a device to make predictions about agents. Ellsberg recognizes that Savage was (primarily) tempted by the normative interpretation whereas Ellsberg was primarily focused on the empirical interpretation. In particular, he wanted to identify classes of choice circumstances in which the possible rationality (669) of one's decision could not be articulated in or explained by reference to the Savage axioms. 

The Savage formalization allows the attribution to agents of precise subjective probabilities about distributions in the world. That is, in certain choice circumstances, the formalism allows for an interpretation of behavior that reveals if not inner states than at least structures that are compatible with such inner states. (It may be useful to compare this with the Samuelsonian strategy of revealed preference.)

With that in place, let's turn to ambiguity. As Grant/Guerdjikova/Quiggin note, in much of decision theory 'knightian uncertainty' is treated as synonymous with 'ambiguity' and has come to mean the absence of well-defined quantitative probabilities.+ And as they correctly note this is not quite what Ellsberg had in mind (and about that more in a second). Before I get to Ellsberg and ambiguity, it is worth noting that the absence of well-defined quantitative probabilities is indeed what Knight meant by uncertainty in the contrast between risk and uncertainty.

However, what the literature means by the absence of well-defined quantitative probability is (and now I quote Gilboa and Schmeidler (1989)) circumstances in which one "considers a set of priors as possible." (emphasis in original). It is important that this set is limited. Now, this is not quite what Knight, generally, intended to mean  by the contrast between risk and uncertainty.  In paradigmatic cases of uncertainty for Knight (and Keynes), we really have no right to make claims about the priors let alone treating them as well defined possibilities.

As an aside, this is not to deny that, as Richard Pettigrew noted to me, that "there are situations in which your evidence and/or current stage of scientific theorising is not going to deliver a unique probability assignment over the states you consider" but in which one has reasons to limit one's subjective probabilities to a set of possible distributions. And if forced to choose one may well come to think of your own (again quoting Pettigrew) as "your best subjective probabilities and use those while knowing that it would be equally compatible with your evidence to adopt an alternative set of probabilities."** If you and the other involved are experts in more or less the same situation, we can call this situation 'reasonable expert disagreement.' In fact, Savage himself noted (without having to contemplate violations of his axioms) that that “we must be prepared to find reasoning inadequate to bring about complete agreement.”  (1954 [1972]:7; see also 3, 67ff.) This is not trivial in the historical context, as I have noted elsewhere, in which the very idea of science involves the arrival of expert consensus. 

Now, as Grant/Guerdjikova/Quiggin note, Ellsberg thought ambiguity something else: namely it is a property of the "nature of one’s information."  And throughout the paper, Ellsberg goes on to deny that this feature "cannot be conveniently described in terms of a sample distribution." (659) Rather the point of using ''ambiguity" is to try to describe circumstances "when there are questions of reliability and relevance of information, and particularly where there is conflicting opinion and evidence." (659) So, one measure , but it is only one such measure, that one's information is ambiguous in Ellsberg's sense is the kind of reasonable expert disagreement [that is, situations one knows an expert's subjective probability about a distribution is compatible with another expert adopting an alternative set of probabilities.] But it is only one such measure because ambiguity itself cannot (to repeat) be reduced to and articulated in one distribution or another. As Ellsberg noted (660), Savage (57-60), too, had tried to characterize the nature of such ambiguity; Savage had explicitly acknowledged that formal theories known to him lacked resources to do so in satisfying ways.

Now, so for Ellsberg ambiguity is a "subjective variable" (660) about the nature of one's information. But the circumstances under which it arises can be characterized objectively. The proxy he uses to characterize these circumstances are violations of the Savage axioms can be predicted to occur and, perhaps, thought (somewhat) reasonable.

But that is not all of it (and this connects to Gilboa/Schmeidler).*** In circumstances of ambiguity one has, according to Ellsberg, some subjective grounds to "eliminate" some "possible distributions," but not all such possibilities. (661) For, Gilboa/Schmeidler (and the Bayesian literature) that is treated as the occasion to look for solutions to decision problems in terms of sets of priors as opposed to unique priors. In so doing, they end up echoing what I have called the displacement strategy of Knightian uncertainty. 

It is worth noting that there is an interesting difference of emphasis between trying to characterize an optimal decision strategy (in the manner of Gilboa/Schmeidler), say because one must choose or wish to guide policy-makers, and between wishing to promote understanding of how to think about or study one's information about possible distributions. (I have jokingly called the latter modal economics.) Ellsberg is clearly less interested in the former, and more interested in the latter.  Ambiguity in Ellsberg's sense does influence choice (664); yet Ellsberg is interested in finding out when and how, and to create "testable propositions" about such states [669], not to tell you what strategies to use in choice.

That is to say, Knight thought of uncertainty in terms of not knowing the (range and nature of) possible distributions. Ellsberg thinks we can develop rigorous thought  about what a theorist taught me to call intermediate circumstances in which we have some information about the possibilities. Somewhat surprisingly, according to Ellsberg -- and here he refers also to Chipman's experiments++ -- this thought has to be developed not from normative first principles, but rather experimentally ((655) within what in reviewing Donald Davidson, Suppes, and Siegel, he calls (in 1957) "the new field of experimental economics;" (1009). Of course, Ellsberg soon got preoccupied by different concerns. But that's really for another occasion.

So, Ellsberg has been remembered as generating a "paradox" in subjective expected utility theory. (This seems to be due to Harry V. Roberts.) This way of thinking about him says more about the economists' attitude toward their own past than Ellsberg. We have seen that alone among the responses to Knight that are still acknowledged by contemporary mathematical economists, Ellsberg wrote to vindicate Knight.+++ But he did so by using the canons of rational decision-making to specify when some circumstances of true uncertainty might occur. We could call them Ellsberg circumstances. These Ellsberg circumstances are characterized by constrained decision options in which information about world has the character of what he called ambiguity. In Ellsberg circumstances one grasps, we might say, the possible contours of the underlying distributions. Whether, in Ellsberg circumstances, it might pay to take bets on these contours is a tough challenge (one may well be in context of what he called "treacherous possibilities." (666)) He then opened the door not to better decision rules, but rather to the epistemology of information, which uniquely he thought ought to be explored experimentally.  

 

 

 

Status Quo Bias and Methodological Paternalism in Data-Mining Economics (or Modal Economics meets Policy Science)

Behavioral biases (such as inattention or myopia) often generate differences between welfare from a policy maker's perspective, which depends on an agent's experienced utility (his actual well-being), and the agent's decision utility (the objective the agent maximizes when making choices). Accounting for these differences between decision and experienced utilities improves predictions about the welfare consequences of policies. The difference between the policy maker's and agent's objectives in behavioral models parallels non-welfarist approaches to optimal policy (Sen 1985, Kanbur, Pirttila and Tuomala2006) and the techniques used to identify agents' experienced utilities resemble those used in the long literature on externalities (Pigou 1920).--R. Chetty (2015) Ely Lecture, “Behavioral Economics and Public Policy: A Pragmatic Perspective”

Raj Chetty (born 1979) was a McArthur ('genius') fellow in 2012, and won a John Bates Clark Medal, awarded by the American Economic Association to "that American economist under the age of forty who is adjudged to have made a significant contribution to economic thought and knowledge” in 2013. The medal is a leading indicator for the Nobel in economics. Chetty, now at Stanford, was (according to his website) “one of the youngest tenured professors in Harvard's history.” Chetty has been on the forefront of methodological innovation in professional economics that is changing the discipline: his work combines (a) data-mining of “administrative datasets with millions of observations” with (b) so-called natural experiments. (I have blogged about his views a few times before [recall here,  here and here].) He is known for his work on labor mobility and teacher quality, especially. He does not have an outspoken political orientation, and I tend to identify him with the technocratic center of professional economics.* His work is creative and fascinating.

I return to the passage above, but let me first quote one of his introductory, methodological remarks of his Ely lecture, which describes his own stance with admirable clarity and frankness. The lecture is purportedly structured around the debate over behavioral economics. (As we will see that turns out not to be the heart of the matter.) He approaches such debates from a,

pragmatic, policy oriented perspective. Instead of posing the central research question as “are the assumptions of the neoclassical economic model valid?", the pragmatic approach starts from a policy question …This approach follows the widely applied methodology of positive economics advocated by Milton Friedman (1953), who argued that it is more useful to evaluate economic models on the accuracy of their empirical predictions than on their assumptions. While Friedman used this reasoning to argue in favor of neoclassical models, I argue that modern evidence calls for incorporating behavioral economics into the analysis of important economic questions.--Chetty (emphasis added).

Leaving aside the mention of Friedman and his famous 1953 essay (lovingly known as, F1953), the core commitment here is that policy application drives the development, even theorizing, of economic models and theories. This is non-trivial, because it makes general purporse (or 'pure') theory dispensable. In fact, the approach evaluates models in terms of their ability to offer guidance on particular policies (recall my post on the old Koopmans vs Vining debate). You may think that I am misrepresenting Chetty, when I say that theory becomes dispensable, because the whole point of his essay is to argue that economics should be willing to use both neoclassical and behavioral models in the context of data-mining and natural experiments. But this is the least significant part of the essay; the whole approach is pragmatic (or opportunistic) about underlying models. Chetty himself notes this at one point when he writes (without apology or concern), that "in some cases, it is not even fully clear exactly what the underlying behavioral model is” that is being tested. For, what's really being tested in his methodology is localized policy efficacy (or some proxy thereof--Chetty is very skilled at deploying proxies). I use 'localized' to note that background institutions are not tested. That is, while the methodology is very policy focused – and thus, encourages experimenting with deviations from the local status quo --, it has huge status quo bias built into its whole project--there is no means toward reflection on to what degree background institutions are better than others.+

Oddly enough, it's precisely on this point -- the huge status quo bias toward the underlying institutions (in Marxist terms, the base and the superstructure)  --, that Chetty's approach deviates from the methodology of Chicago economics at mid-century (when Friedman's essay was written and published). As is well known, Friedman developed his views in discussion and debate with his friend (and fellow Chicago Nobel), George Stigler. Stigler's Presidential Address to the AEA gives a sense of Chicago's ambition on this score:

Economists generally share the ruling values of their societies, but their professional competence does not consist in translating popular wishes into an awe-inspiring professional language. Their competence consists in understanding how an economic system works under alternative institutional frameworks. If they have anything of their own to contribute to the popular discussion of economic policy, it is some special understanding of the relationship between policies and the results of policies. The basic role of the scientist, therefore, is that of establishing the costs and benefits of alternative institutional arrangements. (Stigler (1965) Presidential Address at the AEA, The American Economic Review, 1-2 [emphasis added [recall]]).

It is no surprise that Stigler and Friedman wished to understand "how an economic system works under alternative institutional frameworks" because they wished to combat what they took to be a pernicious status quo (New Deal/Keynesian) bias toward planning and government activism/fine-tuning (not to mention Cold War contrast with Soviet bloc). This is not the place to recount how they went about developing theory to understand alternative institutional frameworks. All I wish to point out, is that if such a "competence" existed (and eighteenth and nineteenth century classical economics was all about developing comparative institutional analysis), it’s being lost in economics because a status quo bias toward the present institutional framework is built into the method (as defended by Chetty). This is a kind of predicable, future Kuhn-loss.

The reason why Chetty's approach has this institutional status quo bias built into the project, is because the data-sets don’t allow projections beyond the institutions that constrain the existing policies (and data). To put the philosophical point simply: modalities cannot be inferred from the data in data-mining. Chetty recognizes the problem, in passing (because he notices that not testing underlying models comes at a cost), and offers a heuristic to get around the problem: “if one can find a domain where agents optimize, one can make robust statements about optimal policy [in a similar domain] that are valid irrespective of the underlying behavioral model.” The problem with this heuristic (and Chetty is aware of it) is that (i) it is often very hard to evaluate when agents are really optimizing, and (ii) this is still  dependent on existing institutions. [There is, thus, also a technical concern about the robustness of this kind of policy science; are the results going to survive changes in technology, norms, institutions, etc.]**

Even though underlying models are not being tested, Chetty’s methodological stance has three pay-offs: (1) it “offers new policy tools…to policymakers;” (2) it provides “better predictions about the effects of existing policies,” and (3) it generates “new welfare implications.” Here I close with a reflection on (3). In particular, Chetty's method is being advertised to help policy-makers deal with the mismatch between their objectives and the objectives of the citizens (or the people studied). The passage quoted at the top of the post addresses this very point.

Chetty's approach makes a sharp distinction between [A] the policymaker or expert, who knows and acts on objective utilities, and [B] the individual agent(s) inside the model, who do not act on objective utilities due to cognitive biases. The agent inside the model is treated as a maximizer; what she is maximizing is not objective utility, but (one may say) subjective utility. This allows one to explore various kinds of policies or nudges that generate a lot of objective utility while accommodating various biases of ordinary agents.

As regular readers know, following the work of David Levy and Sandra Peart, I am a methodological analytical egalitarian, and so I am very suspicious of models that treat agents modeled and the modeler differently. For, it is extremely notable that Chetty fails to model and internalize (a) the known (!) cognitive biases of the policymaker and her advising expert [e.g. expert overconfidence, tacit bias, corrosive incentives,etc.], both of whom are treated  as pure truth-seeker/social optimizer, and (b) the incentives that may influence policymaker and expert. (I am always amazed that economists fail to treat their own incentives seriously.) Note that Stigler gets around this point by positing that agents modeled and the experts agree on tastes/values (recall this point about regulatory capture). While one may, perhaps, ignore the expert's biases and incentives in context of ideal or pure theory, it is a remarkable oversight when one is aiming to optimize policy science.

 

 

Arrow Reviewing Rawls, and Discovers an Old Nemesis (with a note on Harsanyi)

Rawls, however, starting from the same premises, derives the statement that society should maximize min ui. The argument seems to have two parts: first that in an original position, where the quality of an entire life is at stake, it is reasonable to have a high degree of aversion to risk, and being concerned with the worst possible outcome is an extreme form of risk aversion; and, second, that the probabilities are in fact ill defined and should not be employed in such a calculation. The first point raises some questions about the meaning of the utilities and does not do justice to the fact that, at least in Vickrey and Harsanyi, the utilities are already so measured as to reflect risk aversion. [...] The second point is a version of a recurrent and unresolved controversy in the theory of behavior under uncertainty; are all uncertainties expressible by probabilities? The view that they are has a long history and has been given an axiomatic justification by Ramsey and by L. J. Savage. The contrary view has been upheld by F. H. Knight and by many writers who have held to an objective view of probability. KJ Arrow (1973) reviewing Rawls's Theory of Justice in Journal of Philosophy, 251. [I thank MA Khan for renewing my interest in the review.--ES]

It says something about the then-impoverished state of political philosophy that it was an economist who reviewed Theory of Justice in Journal of Philosophy. In the famous review of Rawls, Arrow recognizes that Rawls has embraced a key insight of Frank Knight's; the denial that probabilities can be assigned to certain decisions under conditions of genuine uncertainty. What exact bits of Rawlsian text Arrow has in mind is left unclear, but I think it is obvious he must have in mind the following two features of Rawls’s approach (i) there is some role for epistemic uncertainty in the origin position, where one has no knowledge of “the likelihoods of the possible circumstances” (1971: 134); and (ii) that for Rawls it is crucial that one is “taking a chance” (on which person one will be) in the original position (1971: 165). [1]

Observing Rawls's debts to Knight must have annoyed Arrow because twenty years before in a very important review article (it's been cited over 700 times), he had ridiculed Knight's approach; rather than treating Knight's position -- that there is a genuine distinction between unmeasurable uncertainty and improbable risk -- as respectable and distinctive, Arrow (1951) claimed that “Knight’s uncertainties seem to have surprisingly many of the properties of ordinary probabilities, and it is not clear how much is gained by the distinction [between risk and uncertainty]”. (417) 

In his review of Rawls, Arrow implies that Knight's defense of the distinction (between uncertainty and risk) relies on a controversial assumption: that the only legitimate way to conceive of probability is as something objective. Granted this assumption, Arrow can simply respond (as he already done in 1951: 428-432) by suggesting (echoing Savage) that for decision-making purposes we can treat probabilities as subjective. As an aside, unlike Arrow, Savage, worries in The Foundations of Statistics, that with this move “we must be prepared to find reasoning inadequate to bring about complete agreement.” (1954: 7).

Again, Arrow can't really bring himself to really grapple with Knight's position even when confronted with Rawls's reliance on it. For, Knight's position -- and here (recall) Knight is in substantial agreement with Keynes (who has dropped out of the discussion [despite then still being the towering figure of the public image of economics]) --  is to argue that there are plenty of circumstances in which assigning even subjective probabilities is utterly ungrounded. (You may say in the language of contemporary decision-theorists, these are cases where it would be irrational to bet.) By implying that the embrace of genuine uncertainty relies on an objective notion of probability (which can be treated as unsophisticated and old-fashioned), Arrow can pretend that his position is more sophisticated (with a "long history" in its back).

Arrow (correctly) notes that even Rawls cannot rely too much on Knight's position because Rawls, too, embraces a decision procedure (not unlike the one familiar from Arrow's own "particular tradition of intellectual thought: that of welfare economics" (246)) with a representative agent-model in which some consensus is generated. (Recall from Rawls (who engages with earlier Arrow  in context), "By assuming certain general desires, such as the desire for primary social goods, and by taking as a basis the agreements that would be made in a suitably defined initial situation, we can achieve requisite independence from existing circumstances. The original position is so characterized that unanimity is possible; the deliberations of any one person are typical of all." Theory of Justice) But this procedure is unstable according to Arrow because -- irony of history -- Knight was right after all: "empirical knowledge is after all uncertain, and even in the original position individuals may disagree about the facts and laws of the universe." (255) That is to say, "Rawls transfers the problem to the area of factual disagreement." (255)

One might think that the previous paragraphs need not be articulated. For the conventional wisdom is that in his famous (1975) review, Harsanyi shows that Rawls position cannot withstand technical scrutiny. But this is only so because Harsanyi (like many Bayesians after him) assumes away what even Arrow grudgingly had to concede that there is a "recurrent and unresolved controversy." To be sure: Harsanyi allows that in Rawls "the participants would be uncertain about what their personal circumstances would be under any particular institutional framework to be agreed upon." (594) But, rather than acknowledge the existence of (Knightian) uncertainty, Harsanyi goes on to conflate "the two parts" of Rawls's argument that Arrow had correctly diagnosed. In effect, Harsanyi simply ignores the second part;  thereby, Rawls's use of maximin is not the (purported) one that Harsanyi effectively demolishes.

 

The Technocratic Conception of Politics (The Economists and Rawls redux)

By a 'technocratic conception of politics,' (recall) I mean to capture the following three features of an enduring image of politics present in (social) science: first, it is characterized by the ideal that with social knowledge and its progress, substantial political disagreement can be eliminated. Here is a very striking instance of the claim from Milton Friedman's (1976) Nobel lecture: "Many countries around the world are today experiencing socially destructive inflation, abnormally high unemployment, misuse of economic resources, and, in some cases, the suppression of human freedom not because evil men deliberately sought to achieve these results, nor because of differences in values among their citizens, but because of erroneous judgments about the consequences of government measures: errors that at least in principle are capable of being corrected by the progress of positive economic science." The context of the passage is the furor surrounding his association with the so-called 'Chicago Boys' in Chile and the assassination of his Chilean critic Orlando Letelier (recall).

In particular, second, this ideal of conflict-free politics presupposes (as is clear from the quoted passage in Friedman's Nobel lecture) considerable value-unanimity in society. So, for example, in a famous article, the Chicago-school economists, Stigler and Becker write, "one may usefully treat tastes as stable over time and similar among people;" establishing this point "is the central task of this essay." When value-unanimity is granted then ordinary welfare economics is possible as a kind of (social) engineering science (recall).

Third, the conception requires an image of science in which one of the central aims of policy scientists is to achieve consensus (or lack of disagreement). In modern economics this idea goes back at least to Sidgwick's (1887) Principles of Political Economy (recall; for some prehistory on Hume). It was a central feature of Lionel Robbins's influential methodological restatement (1932) of what economics is: "Surely, for the sake of securing what agreement we can in a world in which avoidable differences of opinion are all too common, it is worth while carefully delimiting those fields of enquiry where this kind of settlement is possible from those where it is not to be hoped for —it is worth while delimiting the neutral area of science from the more disputable area of moral and political philosophy." This image of science was made widely popular by Thomas Kuhn in Structure (1960), but via Talcott Parsons's Weberianism and others it was already well known in mid-twentieth century social science.

It is worth emphasizing that the expectation of consensus was by no means universal. Friedman's sometime co-author, Savage, insists in The Foundations of Statistics, that "we must be prepared to find reasoning inadequate to bring about complete agreement." (1954 [1972]:7; see also 3, 67ff.) But this view became a minority position (cf. Aumann).

On Carnap, Luck and Necessity in Samuelson's *Foundations* of Economics

By 1983 we were all, so to speak, mathematical economists...This is as it should be. Soft and hard sciences are cumulative disciplines. We each bring our contributions of "value added" to the pot of progress. In Max Planck's much-quoted words: Science progresses funeral by funeral. Inside tomorrow's physics treatise will be the lasting truths of Isaac Newton and also of the professor who works down the hall from you. Something of the same goes for economics, where often the dance must proceed Two Steps Forward and One Step Back.--Paul A. Samuelson 1998 (1378-9)

Had I remained on the Chicago Midway, I might well have missed out on the three revolutions that remade mainstream economics: the Keynesian Revolution, the Imperfect-Competition Revolution, and the Mathematical-Economics Revolution. Instead I had a front-row seat.--Paul A. Samuelson (1379)

Had I been a strict constitutionalist, I might well have stopped there. The result would have been a shorter 200-page book with one fully integrated theme.--Paul A. Samuelson (1384)

Reading widely, I was a child of my time, but it was the internal logic of the economic puzzles that guided Foundations' growth.--Paul A. Samuelson (1375 [emphasis in original])

The historian of economics, Roger Backhouse, has long been prodding me to pay closer attention to Paul Samuelson, who was the economist's economist for much of the second half of the twentieth century. In particular, in commenting on my work on the pre-history of Kuhnian ideas within Chicago economics, he insisted that I should not underestimate the significance of Samuelson (the very anthithesis of a Chicago economist, despite having done a BA there). And, indeed, in reading Samuelson's reflections on his own achievement on his seminal Foundations of Economic Analysis a half century later, I was struck by the Kuhnian imagery, including his self-presentation as a would-be-revolutionary (rather than "strict constitutionalist") and the focus on how puzzle-solving  (rather than external influences) led to the breakthroughs. 

But I wondered to what degree the evidence from 1998 wasn't itself a consequence of Kuhn's influence. Nudged by Backhouse I was directed to the following passage in Samuelson's Foundations:

[A]t a time when pure economic theory has undergone a revolution of thought—from statical to dynamical modes. While many earlier foreshadowings can be found in the literature, we may date this upheaval from the publication of Ragnar Frisch’s Cassel Volume essay [1933--ES] of only a decade ago. The resulting change in outlook can be compared to that of the transition from classical to quantum mechanics. And just as in the field of physics it was well that the relationship between the old and the new theories could be in part clarified, so in our field a similar investigation seems in order. (Samuelson 1947, p. 284)

So, Samuelson was very explicit about that fact that he was participating in a scientific revolution that was initiated before him. In an "age of specialization," Samuelson thought of himself as the 'last generalist' because after his framework setting work, the end of "upheaval," there would only be a need for esoteric specialist work. In Samuelson's hands the revolution has an air of inevitability.

On Keynes, Chicago Economics, and Social Engineering

"It is quite another thing to conclude that therefore ends of good policy are beyond the realm of scientific discussion. For surely the primary requisite of a working social system is a consensus on ends. The individual members of society must agree upon the major ends which that society is to seek. If any large share of the population actively disagrees with the society's ends, and in particular if it believes that the system is unfair by the group's criteria, the social system will surely disintegrate, probably with violence.

This is almost axiomatic in modern social theory…

At the level of economic policy, then, it is totally misleading to talk of ends as individual and random [as new welfare theorists do—ES]; they are fundamentally collective and organized.

If this conclusion be accepted, and accept it we must, the economist may properly exceed the narrow confines of economic analysis. He may cultivate a second discipline, the determination of the ends of his society particularly relevant to economic policy. This discipline might be called, following J.N. Keynes, applied ethics.7

7 See his Scope and Method of Political Economy (4^th edition: London 1930) especially chap. II. I did not realize how neglected this excellent book has become until, when I recently referred to it, several friends expressed surprise that I did not know Keynes's middle name was Maynard! –GJ Stigler (1943) "The New Welfare Economics" American Economics Review, 33(2): 357-8.

At the very start of his famous (1953) methodology paper, Milton Friedman calls attention to a three-fold distinction he inherits from the logician-economist, J.N. Keynes (yes, father of that more famous philosopher-economist, J.M. Keynes): "In his admirable book on The Scope and Method of Political Economy, John Neville Keynes distinguishes among "a positive science ... a body of systematized knowledge concerning what is; a normative or regulative science...[,] a body of systematized knowledge discussing criteria of what ought to be...[,]an art... [,] a system of rules for the attainment of a given end"; comments that "confusion between them is common and has been the source of many mischievous errors"; and urges the importance of "recognizing a distinct positive science of political economy."* Friedman admits that he is primarily interested in articulating the nature and methodology of positive economics, and along the way he wishes to argue that "positive economics is in principle independent of any particular ethical position or normative judgments." (By contrast, according to Friedman normative economics presupposes positive economics.) In what follows, I accept these distinctions by stipulation.

Neither Stigler nor Friedman, follow Keynes slavishly (recall this post, too). In particular, in the passage quoted at the top of this post, Stigler offers a world-historical innovation over Keynes that will have far-reaching consequences in economics and the way it understands itself as a policy science. In Keynes, "applied ethics" (that is, normative or regulative economics) is the business of philosophers not economists. And one of the main reasons it is the business of philosophers (and not economists) is that Keynes does not expect consensus over the ends of society to emerge. As he writes, "There is a further reason why a positive science of political economy should receive distinct and independent recognition. With the advance of knowledge, it may be possible to come to a general agreement in regard to what is or what may be in the economic world, sooner than any similar agreement is attainable in regard to the rules by which economics of individuals and communities should be guided. The former requires only that there as to the facts; the latter may be prevented by conflicting ideals." (50-51) So, Keynes's proposed strategy is to remove (applied) ethics as much as possible from economics so that economics can quickly hasten to become a mature (in a proto-Kuhnian sense) science.

Piketty, Ray, and Economic Explanation

Inequality has been rising, and to see it well, one should study “top incomes,”: those of the top 1% or even, in a variant which we might call SuperOccupy, the top 0.1% or 0.01%. This is an extremely important observation. There are lots of people in the top 1% (more in India, for instance, than in a good-sized European country), and they cast a long and enviable shadow...Even a large sample survey will often fail to pick these people up, so Piketty’s meticulous examination of tax records (along with co-authors) in different countries is to be applauded. This is data work at its best, with a well-defined reason for doing it, and when I read the papers with Emmanuel Saez and Tony Atkinson that put these findings together, I felt I had learnt something.--D. Ray, on Piketty's Capital. [HT Hülya Eraslan]

One of the most important economists today that you probably have never heard of, Debraj Ray,* wrote a careful, amusing, and -- as one ought to expect -- very insightful response to Piketty's Capital. Ray matters in economics: he is the co-editor of the leading journal. Ray is one of the intellectual leaders of the new generation of 'theorists' that combine mathematical sophistication and data with a focus on policy relevance.* Ray is also an important figure in development economics, which he has helped move to the center of the discipline. You might think such sociology is irrelevant, but as Ray notes, Piketty's Capital has a self-presentation and narrative which is a form of "positioning" that appeals to his once-inside now purported outside-status in the profession. Part of Piketty's rhetoric is to rail against "simplistic mathematical models" (16; 574, and, especially, the autobiography on p. 31-2 culminating in an indictment against "petty mathematical problems.") This is not just a matter of mere positioning within economics; for, -- and Ray does not comment on this -- at crucial junctures, Piketty relies on a contrast between the esoteric mathematical models of the experts and democratic decision-making under conditions of transparency (e.g., 480, 513; recall also this post; Vallier also picks up on this [3E/3P] and will blog about in the future). But it is also a matter of positioning oneself as being the privileged expert with the public. Even if Ray's irritation were wholly self-interested, he has put his finger on on a non-trivial feature of Piketty's rhetorical strategy. (Don't get me wrong lots of economists become such privileged experts with worse strategies: i.e., they are bought by some interested party, or they get chosen by a politician who likes what s/he hears, etc.)

That Piketty's is, in part, a rhetorical strategy is clear from two facts: first, when it suits him he, too, relies on fairly simple mathematical models (often he is disarmingly upfront about this, e.g. 364). Second, he relies in non-trivial ways on existing economics. I had pointed to this in my second post on Piketty, but I am really an amateur and Ray's post adds a lot more detail, insight, and depth to what I had said. In particular, what becomes clear is that in some non-trivial respects Piketty is working with commitments that are found in the old economics work-horses (that is, the descendants of the "models of Harrod and Domar") and some so-called "dynamic efficiency" assumptions (see Ray's slightly more technical appendix here.)  Piketty is a world-class economist, so this should not surprise. But as I pointed out before, it is unclear what constraints are driving his partial appropriation of existing economic theory.

Now, the core of Ray's piece is actually applied philosophy of science:

Philo of Economics: Scientific Integrity in Econometrics.

In the past 20 years, researchers have begun to embrace the idea that identification is not an all-or-nothing matter and that a set of plausible assumptions that does not deliver point identification can still contain useful information about parameters of interest.

This partial identification view has been motivated by the fact that point identification is not the objective by itself and in essence takes us back to Koopmans & Reiersol’s (1950) dictum whereby the specification of a model ought to be based on [A] the underlying economics, [B] prior knowledge (such as the linearity of variable cost that people have established for this industry), or [C] other assumptions with universal or [C*] almost universal acceptance, but [D] should not be geared primarily toward point identifying the parameters. “Scientific honesty demands that the specification of a model be based [B] on prior knowledge of the phenomenon studied and [E] possibly on criteria of simplicity, but not [D] on the desire for identifiability of characteristics that the researcher happens to be interested in” (Koopmans & Reiersol 1950, pp. 169–70). Once the structure is specified, the model can either have no information about the parameter of interest, restrict the parameter of interest to a nontrivial set, or point identify the parameter of interest. This is exactly the domain of identification analysis, with partial identification taking the view that identification is not only about verifying whether the third case holds, but also determining the extent of information contained in the second and linking this to the type of assumptions that the researcher proposes in the model.--Elie Tamer (2010). [Emphasis and Letters added to facilitate discussion. --ES]*

In quoting Koopmans, who won a Nobel in 1975, and Reiersol (hereafter K&R), Tamer calls attention -- en passant -- to the significance of the role of virtue in ordinary scientific practice. It goes beyond general rules of integrity, by appealing to an explicit rule ('do x or y, but not z') in model specification that prevents abuse of a scientific method. The need to articulate the rule suggests that already back in 1950 K&R  were aware of the danger of (what we might label) fishing expeditions. While reflecting on the fact that K&R, focus exclusively on individual virtue (and ignore incentives, institutions, and detecting mechanism), I noticed something curious about Tamer's passage.

 

Criteria [A] through [C*] are neither obviously identical to each other nor identical with [E]. Moreover, criteria [C] and [C*] are together with criteria [B] essentially conservative: appealing to them prevents exploring unusual, unpopular, or wacky ideas. Now, these criteria prevent researcher's personal idiosyncrasies from generating (ad hoc) results, and (useful when computer power was expensive and scarce) focuses the search space. But it also makes it hard for adventurous and speculative research. Now, I was not surprised by this because I had already claimed that Koopmans should be credited for promoting mathematical econometric techniques and tools – and more generally inferential technologies that produce univocal and stable figures in calculating the implications of policy alternatives – within economics and, of course, to policymakers; this made economists attractive as policy experts (as opposed to say, sociologists, lawyers, anthropologists, and historians).