Roth, A.E. "Comments on Tversky's 'Rational Theory and Constructive Choice'," The Rational Foundations of Economic Behavior, K. Arrow, E. Colombatto, M Perlman, and C. Schmidt, editors, Macmillan, 1996, 198-202.
Professor Tversky presents a quick overview of an ever growing body of experimental evidence, to which he has been one of the most influential contributors. This evidence demonstrates that human behavior deviates in systematic ways from the idealized behavior attributed to expected utility maximizers in particular, and to "rational economic man" in general. One of the most striking things about this substantial body of evidence is that, starting at least as early as the work of Allais [1953] and May [1954], it has been collected over the same period of years in which expected utility theory has come to be the dominant model of individual behavior in the economics literature. This adds force to the question Tversky raises in his concluding remarks: what accounts for economists' "reluctance to depart from the rational model, despite considerable contradictory evidence"?
I'll attempt to outline a two-track answer to this question.
First, I'll argue that there are quite defensible reasons for a reluctance to abandon theories of rationality in favor of psychological theories. In particular, I think most economists view the rational model as a useful approximation, rather than as a precise description of human behavior. Experimental demonstrations that people deviate from the model do not strike at the heart of the belief that the approximation is a useful one, since all approximations are false at some level of detail. In view of this, some kinds of evidence, and alternative models, are likely to be more successful than others in attacking the central role of rationality assumptions in the economic literature.
Second, I'll note that, in fact, there is a growing attempt by economists to move away from an overdependence on idealized models of hyper-rationality. My own paper in this volume deals with aspects of this, as do those presented here by Binmore and Weibull.
1. Rational models as useful approximations:
Tversky identifies the essence of the rationality model to be the assumption that an individual can be modeled as having preferences over outcomes. To appreciate the scope of what is being criticized, it will be helpful to consider several increasingly general examples of this kind of model.
The first of these might be called risk neutral economic man. This individual chooses among risky outcomes strictly according to their expected value. He is different from real people in many ways: he never buys insurance, but would be willing to pay any finite amount to participate in Bernoulli's [1738] St. Petersburg paradox (if only he could be convinced that the resources exist to pay off any possible winnings). This model remains a useful, and much used, approximation, although economists mostly agree that it is not a precisely true description of all (or even most) individuals in all, or even most situations.
The more general, conventional model of rational economic man, is that of expected utility maximizing man. Unlike his risk neutral kin, he buys insurance. But, unlike real people, he ignores sunk costs, he always evaluates probabilities in an actuarial way, and is in general the kind of solid citizen admired by economic theorists. This model has by and large replaced that of risk neutral economic man as economists' "canonical" model of individual choice behavior. The reason is that economists became convinced that the phenomena that couldn't be explained by the approximation that individuals maximize expected value were of central importance, and justified a model with unobservable personal parameters of risk aversion. (The phenomena that can't be captured by the risk neutral model include whole industries, like insurance, and large markets, like the futures markets for commodities and currencies.) There is every reason to doubt that expected utility theory would have made such inroads into economics only on the strength of anomalies (from the expected value point of view) like the St. Petersburg paradox (for which Bernoulli in fact proposed a kind of expected utility theory as a resolution).
We should also include in this brief catalog of rationality models a representative example of almost rational economic man, who deviates systematically from expected utility maximization in some ways, but who still has preferences. For example, Kahneman and Tversky's [1979] "prospect theory" models an individual who may not always ignore sunk costs, e.g. because he is sensitive to framing which influences his reference point, and who doesn't handle probabilities like an actuary, but tends to overestimate small probabilities (but sometimes underestimates them, due to "editing"). Nevertheless, once his reference point has been fixed, and his probability curve pinned down, he has preferences. So although he might be comfortable making non- utility-maximizing choices such as those seen in the Allais paradox, he would eschew preference reversals.
In contrast to these models of rational choice, Tversky proposes that we direct our attention to models of what we might call psychological man. Psychological man doesn't have preferences, at least not in the sense that economists customarily think of them. Rather he has a collection of mental processes. And different descriptions of options, different frames and contexts, and different choice procedures elicit different processes. (So he may sometimes exhibit preference reversals because choosing and pricing elicit different mental procedures.) Furthermore, it may be profitable to understand how his different mental processes are elicited, as in Tversky's anecdote about the bread-baking appliances in the Williams-Sonoma catalog.
To understand why economists have not abandoned rationality models for psychological models of this sort, it may be helpful to consider for a moment the class of models of individual choice that seem to be suggested by recent research in brain science and clinical pharmacology. Neurobiological man doesn't (even) have a fixed collection of mental processes, in the sense of psychological man. Instead, he has biological and chemical processes which influence his behavior. Different blood chemistry leads to different mental processes; e.g. depending on the level of lithium (or Valium or Prozac) in his blood, he makes different decisions (on both routine matters and matters of great consequence--even life and death). An understanding of how chemistry interacts with mental processes has proved to be very useful, for instance in treating depression.
One can then pose the neurobiologist's question: What accounts for the [psychologist's] "reluctance to abandon the [psychological] model, despite considerable contrary evidence"? The psychologist's answer (as imagined by this economist) might go something like this: "No one really supposes that an individual's mental processes are fixed and never change. But this is a useful approximation. It breaks down for people who have lithium deficiency, and who (therefore) exhibit abrupt cycles of manic and depressive behavior. But it helps us explain a lot of the phenomena which concern us, without requiring blood tests of our subjects. And, while we are fully persuaded that real people have blood chemistry and brain processes, the compelling evidence that the neurobiologists have assembled on this point does not address the question of how often decisions are affected by normal variations in blood chemistry, and therefore does not address the usefulness of our approximation that people call on fixed sets of mental processes in ways that can be predicted without reference to blood chemistry. (We note that even analysis at the level of blood chemistry is only an approximation to the underlying quantum mechanical processes of the brain.) Finally, the blood chemistry model doesn't seem to bring a lot of explanatory or predictive power to bear on the questions we try to study, like why people exhibit preference reversals."
My point, of course, is that with the natural substitution of terms, an economist's answer to this question could look a lot like the psychologist's. That being said, it should be obvious that evidence about where approximations break down is enormously useful, even when it is not the sort of evidence that causes the approximations to be abandoned. To know that utility maximization may be a weak guide to choices among alternatives with "similar" expected values, or to choices involving probabilities near zero or one, can only enhance the actual (as opposed to the apparent) usefulness of the approximation.
Of course, as more evidence of different kinds accumulates, and as alternative models are developed, the situation may change. It therefore seems appropriate to end this discussion with some remarks on the growing role of non-rational models in economics.
2. The potential for non-rational models in economics:
There has been increased exploration of a variety of "almost rational" models of choice (of which prospect theory is an early example), which generalize expected utility theory but are "plug compatible" with it, in the sense that they are meant to replace it but do the same job, and fit into strategic and market theories in the same way. Experimental comparisons of such theories with each other and with utility theory are reviewed by Camerer [1995]. It remains an open question whether any of these theories organizes all of the data in a way that is clearly superior to utility theory, when the costs of adding additional unobserved personal parameters are counted in. But these theories are not non-rational in the way that Tversky proposes in his paper here.
The same criticism cannot be leveled at the adaptive models of behavior which economists increasingly explore. Whether motivated by simple models of learning, or by biological processes involving natural selection (as in the work of Maynard Smith or Holland), these models treat individuals as entirely non-rational automata who make no conscious choices at all, but whose behavior becomes adapted to their experience. It seems safe to say that most economists also apply these models to human behavior only as approximations. What makes them potential competitors of utility theory is that in some environments it appears that they may lead to the same kind of equilibria as are predicted by standard (rational) game theory. So, to the extent that economists' reluctance to dispense with rational models of individual choice has to do with the fact that their primary interest is in strategic and market phenomena, and that the rational model helps produce useful predictions about such phenomena, these adaptive models indicate that at least some of these predictions may not depend in a critical way upon rationality assumptions. But these models too are quite different from the models of mental processes which Tversky suggests.
What then are the prospects that models of the kind Tversky proposes, which incorporate approximations of particular mental processes, will gain a substantial foothold among economists? It seems to me that proponents of such models have at least two worthwhile avenues to pursue.
The first of these would be to conduct experiments which show not merely that individuals are sometimes systematically non- rational, but which would start to give us some feeling for how important this phenomenon might be. For example, I don't think that any of the preference reversal experiments have yet given information that would allow us to graph the frequency of preference reversals as a function of the difference in expected value (in, say, percentage terms) between the two gambles with which subjects are presented. While this information might not dramatically effect how investigators with very different theoretical dispositions evaluated the data, it would help address the issue of whether we are seeing a phenomenon likely to play an important role in natural environments. Undoubtedly there are more informative experiments to be done than this one; my point is only that investigators who are aiming to assess the size of the disparity between utility theory and observed behavior, rather than merely to show that there is one, might better address the usefulness of the utility maximization approximation.
Second, there is no substitute for careful studies of natural environments. Just as expected utility maximization would not have replaced expected value maximization as economists' typical approximation if it were not for the importance of explaining insurance and related phenomena, economists are likely to find most persuasive those attacks on rational models which produce examples of how psychological phenomena are reflected and exploited in, and shape aspects of, important economic activities. It seems likely to me that advertising, and marketing in general, are areas of economic activity in which this approach will prove fruitful. It is hard to imagine that the volume of advertisements with which we (Americans, at least) are bombarded serve only information purposes. Rather, advertisements and other marketing tools may shape preferences in ways that cannot be accounted for by rational models of individual choice.
Consequently, I feel optimistic that economics has much to gain from the diversity of approaches that are beginning to flourish.
Allais, Maurice [1953] "Le Comportement de L'homme Rationnel Devant le Risque: Critique des Postulats et Axiomes de L'ecole Americane," Econometrica, 21, pp503-546.
Bernoulli, Daniel [1738], "Specimen Theoriae Novae de Mensura Sortis," Commentarii Academiae Scientiarum Imperialis Petropolitanae, 5, pp175-192. English translation in Econometrica, 22, 1954, pp23-36.
Camerer, Colin [1995], "Individual Decision Making," in J. Kagel and A.E. Roth (editors), Handbook of Experimental Economics, Princeton University Press, 587-703.
Kahneman, Daniel and Amos Tversky [1979], "Prospect theory: An analysis of decision under risk," Econometrica, 47, 263-291.
May, Kenneth O. [1954] "Intransitivity, Utility, and the Aggregation of Preference Patterns," Econometrica, Vol. 22, pp1-13.