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.