1Rational choice theory assumes that individuals have complete and transitive preferences, and models the individuals as acting in an instrumentally rational way to satisfy their preferences. Many experimental works however highlighted that the preferences revealed by the individuals through their choices significantly deviate from the prediction of rational choice theory (see e.g. Kahneman and Tversky (2000) on framing effects, Frederick, Loewenstein, and O’Donoghue (2002) on time inconsistencies, Camerer (2003) and Crawford, Costa-Gomes, and Iriberri (2013) on strategic thinking). The common interpretation of those results is that real humans—unlike the rational Homo economicus populating neoclassical economic models—often make mistakes, and therefore fail to behave in their own best interests (see e.g. O’Donoghue and Rabin, 2003, 186; Thaler and Sunstein, 2008, 9; Ariely, 2008, 239).
2Those findings gave rise to a vast literature advocating new forms of paternalism, such as “libertarian paternalism” (Sunstein and Thaler, 2003), “asymmetric paternalism” (Camerer, Issacharoff, Loewenstein, O’Donoghue, and Rabin, 2003), “light paternalism” (Loewenstein and Haisley, 2008), or “means paternalism” (Sunstein, 2014). The common feature of those approaches is the postulate that non-rational individuals should be helped so as to satisfy their preferences. Sunstein (2014) for instance justifies means paternalism as follows:
More particularly, I have urged that accumulating evidence suggests, more concretely than ever before, that in identifiable cases, people’s mistaken choices can produce serious harm, even when there is no risk to others. The result is a series of behavioral market failures that provide a plausible basis for some kind of official remedy. Behaviorally informed responses need not, and generally should not, attempt to revisit people’s ends. They are focused on correcting mistakes that people make in choosing the means to satisfy their own ends. (Sunstein, 2014, 163-164)
3Following the terminology of Salvat (2014), I will call this new approach to paternalism “behavioural paternalism” (henceforth BP). BP aims at helping boundedly rational individuals to satisfy their own ends, and therefore encompasses both “soft” and “hard” forms of paternalism, such as Sunstein and Thaler’s nudges, and the “coercive paternalism” advocated by Conly (2013). Since Sunstein and Thaler’s libertarian paternalism (henceforth LP) has been the most successful approach so far, this paper will be primarily concerned with LP, although the argument would be exactly the same for all the other approaches mentioned above.
4The object of this paper is to characterise the set of hypotheses about preferences and welfare supporting BP as a new form of paternalism, and then to question the soundness of those hypotheses. My main argument is that BP implicitly endorses Pareto’s model of the Homo economicus (as the “part” of the individual that deals with economic actions), and therefore presents the exact same methodological shortcomings. First I present the three main hypotheses underlying BP: that (i) people have true preferences, (ii) the satisfaction of true preferences is the proper normative criterion (i.e. the supreme goal of normative economics is the satisfaction of those preferences), and finally (iii) it is possible to objectively elicit the true preferences of the individuals (section 1). The following three sections respectively deal with those three hypotheses: I show that the existence of true preferences can be postulated if and only if we accept Pareto’s reductionist account of decision-making (section 2), that the normativity of the satisfaction of one’s true preferences is neither a consequence of Pareto’s theory nor properly justified (section 3), and that economists do not have at their disposal an impartial mechanism that would allow them to elicit the individuals’ true preferences, if such preferences exist at all (section 4).
5Imagine Luke, who just opened an account in a bank that offers a bonus to its new customers. The bank manager, Carla, can either credit €200 on his account, or offer him an asset worth €600, but which is likely to be devaluated with probability 2/3 (its value would then be null). Carla can for instance directly asks Luke to choose between (A) getting the €200 and (B) getting the risky asset. She can also give him the risky asset, and then ask him to choose between (C) reselling the asset at one third of its value (€200) and (D) keeping the asset.
6Suppose that Luke chooses (A) in the first case and (D) in the second (in line with experimental results on loss aversion, Tversky and Kahneman, 1981): since (A) and (C) (as well as (B) and (D)) are identical in terms of monetary payoff (which constitutes the only relevant element of the choice problem from a consequentialist perspective), this set of choices shows that Luke’s preferences changed when the framing of the problem changed. The common interpretation of this preference reversal is that—unlike the standard neoclassical agent—Luke’s choices reveal a loss aversion. Luke is indeed risk averse when facing lotteries involving gains, but risk-seeking when facing lotteries with losses (Kahneman and Tversky, 1979). In the first situation, he prefers the sure gain to the risky one, while in the second situation—when given an initial endowment of €600—Luke prefers the risky option (he is indeed comparing a sure loss of €400 with the possibility of avoiding a €600 loss with probability 1/3). The common interpretation of this kind of situation is that Luke truly prefers one option over the other (either (A) and (C) over (B) and (D), or (B) and (D) over (A) and (C)), but that his actual choice is influenced by the framing of the decision problem: his preferences are therefore context-dependent, i.e. “his preferences depend (irrationally) on factors that ought to be irrelevant” (Hausman, 2012, 20).
7Consider now the choice problem faced by Carla. Suppose that Carla noticed that the way she frames the offer has a significant influence on the final decision of her clients. She therefore knows that framing the problem as a choice between (A) and (B) will induce Luke to choose the safe asset, while he will probably choose the risky one if he must choose between (C) and (D). Thaler and Sunstein (2008, 3) would describe Carla as a choice architect, i.e. an individual who “has the responsibility for organizing the context in which people make decisions”. If Carla is benevolent, then she will choose the choice architecture of the problem such that Luke will be better off, from his own point of view. Carla’s objective is therefore not to propose to Luke what she thinks is good for him, but to help him get what he truly wants. Suppose for instance that, based on Luke’s past financial decisions, Carla can reasonably assume that Luke is actually a risk-lover, and tends to make risky choices. Although she may believe that it would be better for Luke to take more cautious decisions, her benevolence implies that she will respect Luke’s underlying true preferences. She will therefore choose to frame the decision problem as a choice between (C) and (D), since it is more likely that Luke will take the risky asset.
8If we accept the assumption that Luke truly prefers one asset over the other—independently of the way Carla framed the choice problem—and that his actual choices may be influenced by elements he would consider to be irrelevant if he were aware of them (e.g. the choice architecture), then an intuitive normative criterion would be the satisfaction of Luke’s true preferences. The core argument of BP is that people have true preferences they wish to act on, but that most of their decisions are influenced by cues that should be irrelevant, such as the desire of avoiding losses or the greater salience of an alternative (in Sunstein and Thaler’s example of the cafeteria, the individuals tend to choose more prominently-displayed items, independently of their nature). BP therefore distinguishes between the individual’s true preferences (which determines her welfare, a normatively desirable objective) and her distorted preferences (which are revealed through her choices). Thaler and Sunstein (2003) indeed claim that:
We clearly do not always equate revealed preferences with welfare. That is, we emphasize the possibility that in some cases individuals make inferior choices, choices that they would change if they had complete information, unlimited cognitive abilities, and no lack of willpower (Thaler and Sunstein, 2003, 175).
9According to this distinction, the individuals intend to satisfy their true preferences, but are unable to take the appropriate decisions due to psychological biases. They are therefore, in some sense, maximising the “wrong” utility function. Köszegi and Rabin (2007; 2008) suggest for instance understanding individual behaviour as a combination of preference satisfaction and mistakes, and then argue that “preferences often can be revealed by behavior, even when they are not implemented by behaviour” (2007, 479, emphasis in original). Many dual-selves models have then been developed to represent this duality between an “optimising” self and a “psychological” self, but without necessarily stating that the satisfaction of the preferences of the optimising self constitutes the normative criterion (see Davis, 2011, 63-64 for a discussion on the normative or descriptive value of multiple selves models). Individual behaviour is for instance explained as the result of the interaction of two conflicting selves, which are “cold” and “hot” (Bernheim and Rangel, 2004), “deliberative” and “affective” (Loewenstein and O’Donoghue, 2004), “controlled” and “automatic” (Benhabib and Bisin, 2005), a “long run planner” and a “myopic and short run doer” (Fudenberg and Levine, 2006).
10In a situation of choice, individuals are therefore likely to be influenced by elements of the context that would be ignored by a rational agent. Thaler and Sunstein (2008, 8) for instance define a nudge as “any factor that significantly alters the behavior of Humans, even though it would be ignored by Econs”. Bernheim and Rangel (2007; 2009) develop a theoretical framework for behavioural welfare economics (henceforth BWE) that explicitly relies on this duality between a “welfare-relevant domain” and “ancillary conditions”, defined as “[features] of the choice environment that may affect behaviour, but [that are] not taken as relevant to a social planner’s evaluation” (Bernheim and Rangel, 2009, 55). Similarly to Sunstein and Thaler who argue that the choice architecture should be ignored by rational agents, Bernheim and Rangel (2007) claim that “standard economics proceeds from the assumption that choice is invariant with respect to ancillary conditions” and that “[p]ositive behavioral economics challenges this basic premise” (Bernheim and Rangel, 2007, 464-465). The exact same duality is endorsed by Salant and Rubinstein (2008) and Rubinstein and Salant (2012) who assume that “the observed preference ordering is the outcome of some cognitive process that distorts the underlying preferences of the individual” (Rubinstein and Salant, 2012, 375): this distortion results from a “frame”, defined as “a description of details that influence choice behaviour, though it is clear to an observer that they do not affect the individual’s welfare” (ibid., 376). Note however that Bernheim and Rangel (2007; 2009) and Rubinstein and Salant (2008; 2012) do not endorse BP: they develop a theoretical framework for welfare economics based on the postulate that there exist “welfare-irrelevant” events that influence our choices, and therefore that the preferences revealed through our choices are different from welfare (our true preferences). BP could then easily be justified within this framework, but they do not advocate it. On the other hand, the behavioural paternalists mentioned above must endorse BWE’s framework to justify BP.
11It seems therefore that the “anomalies” highlighted by behavioural findings have been understood by behavioural paternalists as “anomalies of behaviour” rather than “anomalies of the theory”: it is not the theory according to which individuals have coherent preferences that is wrong, but the individuals who make mistakes, i.e. whose behaviour is “wrong” with respect to a norm of behaviour. Rational choice theory is therefore normatively correct while actual behaviours are deviations from rationality. This is for instance the case in Kahneman and Tversky’s work on prospect theory: they are indeed talking about anomalies of preferences (Kahneman and Tversky, 1979, 275, 277), suggesting implicitly that “normal” preferences should respect the axioms of rational choice theory (see Hands (2015) on the normativity of rational choice theory, and Heukelom (2014) for a detailed account of Kahneman and Tversky’s position on this question).
12In a discussion in the Journal of Economic Literature with Harstad and Selten (2013) and Crawford (2013), Rabin (2013) explicitly defends this “optimisation approach” (in contrast to a bounded rationality approach, see Harstad and Selten (2013) for a review). He argues the objective is to make:
(imperfect and incremental) improvements over previous economic theory by incorporating greater realism while attempting to maintain the breadth of application, the precision of predictions, and the insights of the neoclassical theory (Rabin, 2013, 528).
13This argues that behavioural economists should take the neoclassical model of behaviour as a benchmark, and progressively integrate new parameters expressing specific psychological inclinations of the decision maker into individual utility functions to improve the predictive power of the theory. The normative standard therefore remains the neoclassical model. I have argued with Infante and Sugden (2016) that BWE (and therefore BP) “uses a dualistic model of the human being, in which an inner rational agent is trapped inside a psychological shell” and that “the inner agent is pictured as the locus of the identity of the human being and as the source of normative authority about its interests and goals” (Infante et al., 2016, 1). Incoherent preferences are seen as an anomaly that should be corrected so as to reveal the underlying “true”, coherent, preferences.
- 1 The main difference with BWE is that Bernheim and Rangel only state that welfare assessments should (...)
14We can now precisely state the hypotheses about preferences and welfare underlying BP. BWE assumes that individuals have true preferences (like the ones we found in neoclassical economics) and treats deviations from the satisfaction of those preferences as mistakes (see Infante et al. (2016) for a detailed account). Most of behavioural welfare economists (a notable exception is Loewenstein and O’Donoghue (2004), for reasons I will expose in section 3) then endorse the satisfaction of the true preferences as the normative criterion. Behavioural paternalists go a step further by stating that the social planner (or the choice architect for LP) should actively1 seek the satisfaction of individual true preferences. Indeed, since the individuals are likely to fail to satisfy their true preferences, paternalism seems to be “unavoidable” (Sunstein and Thaler, 2003, 1175). Behavioural paternalists thus state three relatively explicit hypotheses (which are actually standard hypotheses in conventional welfare economics):
the individual has true preferences (a unique complete, transitive, and context-independent preference ordering which would determine her choices, were she able to reason correctly);
the satisfaction of those true preferences is the normative criterion;
it is possible to identify this preference relation from the choice architect’s standpoint (although various different approaches have been suggested).
15If hypothesis (i) is not true, then we cannot argue that incoherent preferences reveal a mistake, and therefore that real individuals do not act in their own best interests. If hypothesis (i) is true but not hypothesis (ii), then nudging the individuals such that they satisfy their true preferences cannot be considered as normatively desirable. Lastly, if (i) and (ii) are true but not (iii), then it means that, although it would be desirable to nudge the individuals to satisfy their true preferences, the choice architect cannot do this successfully because she cannot elicit what those true preferences are. Therefore, if at least one of these assumptions is not verified, BP is a dead-end and should be rejected.
16In the following sections, I will successively question each one of those hypotheses. My main argument is that BP implicitly endorses Pareto’s methodology of analysis and synthesis and his definition of the Homo economicus, and therefore shares many methodological difficulties with Pareto regarding the definition of one’s true preferences. I will also highlight that, unlike Pareto, BP confers a normative value to preference satisfaction that cannot be justified within Pareto’s approach.
17I will illustrate my argument with the case of Petula’s choice of train: Petula, a pessimistic individual, must take the train to go to work, and has a correspondence halfway (the second train leaves at 8h40). She knows that the first journey takes 30 minutes in normal conditions, but also that delays may happen (incidents actually occur on 1% of the journeys). She has the choice between taking the first train at 7h or a later train at 8h. If she takes the 7h train, then she is almost certain not to miss her correspondence, even if an incident occurs. If there is no incident, she will however arrive too early at the second station, and is likely to regret not having slept an additional hour. If she takes the 8h train, then she will arrive on time for the second train if there is no incident, but she is likely to miss it in case of an incident. Her pessimism inclines Petula to systematically take the 7h train. In addition, suppose also that Norbert, the neoclassical agent, faces the exact same problem than Petula, but after having carefully thought about the pros and cons of taking the 8h train, Norbert decides to systematically take the 8h train. Although he sometimes misses his correspondence, this is compensated by the additional sleeping hour he gets all the other days. From the social planner’s perspective, it seems that Petula is making a mistake when taking the early train (she may believe that the probability of an incident is much higher for instance), since Norbert—Petula’s neoclassical alter ego—takes the later train. The social planner could therefore nudge Petula to take the 8h train; this is possible by informing more explicitly that incidents are relatively rare events, or by offering in a single package the ticket of the 8h train with the ticket of the second train. This kind of policy could induce Petula to take the 8h train, and therefore to better satisfy her true preferences. Solving the Petula case therefore required to assume that: (i) Petula’s preferences can be purified from her pessimism, revealing her true preferences,(ii) the choice architect should help Petula to satisfy her true preferences, (iii) the choice architect can elicit Petula’s true preferences (and therefore identify what Norbert would do).
18In this section, I highlight the similarities between BP’s notion of true preferences (as the preferences we would satisfy if we were rational) and Pareto’s Homo economicus (as the “part” of the individual that chooses in an instrumentally rational way). I argue that defining the true preferences consists in isolating the Homo economicus within the individual, and that this operation requires a reason-based understanding of the true preferences: nothing however guarantees that such true preferences should be context-independent. This could seriously undermine hypothesis (i), according to which individuals have underlying context-independent preferences.
19By opposition to the works of the early neoclassical authors such as Jevons, Edgeworth and Pantaleoni, Pareto ( 1971;  1936) wished to minimize the psychological assumptions used in economics by basing economic theory on principles of rational choice (Bruni and Sugden, 2007, 154). Pareto considered that “[c]learly psychology is fundamental to political economy and all the social sciences in general” (Pareto  1971, Ch.2, §1), and therefore that—in the future—all social phenomena may be deduced from principles of psychology. He however claimed that “natural sciences have progressed only when they have taken secondary principles as their point of departure, instead of trying to discover the essence of things” (letter to Adrien Naville, quoted in Busino, 1964, xxiv). Psychology is probably more fundamental than economics (since the behaviours studied in economics could be explained by psychology), but a scientific theory of economic behaviour “should be deduced from firmly-established empirical propositions about choice rather than sensation” (Bruni and Sugden, 2007, 155; see also Lenfant, 2012, 118-120, on indifference curves as a theoretical rather than empirical construction).
20The foundation of Pareto’s methodology is the method of analysis and synthesis (Bruni and Guala, 2001, 37; Bruni, 2010, 95-96): the investigation of complex social phenomena requires decomposing them into parts, investigating each part separately, and then re-assembling those parts. While science is mainly analytical, practice is a matter of synthesis. Pareto’s criterion for studying human actions (i.e. for decomposing them into parts) is their logic, i.e. (1) whether the action is undertaken (subjectively) in order to satisfy a given purpose, and (2) whether this action is objectively appropriate towards the subjective purpose. Pareto defined “logical” actions as the set of actions for which both criteria are met, i.e. the means are appropriate to the ends both in the mind of the individual and in objective reality. When the objective purpose (what the individual will actually achieve) differs from the subjective purpose (what the individual intends to achieve) the actions are “non-logical”. These non-logical actions can be classified into different categories according to the nature of the difference between the objective and the subjective purpose (Pareto,  1936, §151; see also McLure, 2002, 50-51). Thus logical actions “logically conjoin means to ends not only from the standpoint of the subject performing them, but from the standpoint of other persons who have a more extensive knowledge” (Pareto,  1936, §150).
21Pareto then defined economic actions as those logical actions for which the individual objective is the satisfaction of preferences through the purchase of goods:
We will study the many logical, repeated actions which men perform in order procure the things which satisfy their tastes. [...] Moreover, we will simplify the problem still more by assuming that the subjective fact conforms perfectly to the objective fact. This can be done because we will consider only repeated actions to be a basis for claiming that there is a logical connection uniting such actions. A man who buys a certain food for the first time may buy more of it than is necessary to satisfy his tastes, price taken into account. But in a second purchase he will correct his error, in part at least, and thus, little by little, will end up by procuring exactly what he needs. We will examine this action at the time when he has reached this state. (Pareto,  1971, Ch.3, §1)
22Given this, he also recognized the possibility of producing economic theories without this specific restriction of motives (Pareto  1971, Ch.3, §11), suggesting therefore that economics is the science of logical actions in general, i.e. of the class of actions for which individuals choose in an instrumentally rational way given their preferences. Preferences are however distinct from mere “sensations”, since “it is not an essential characteristic of [economic theories] that a man choosing between two sensations chooses the most agreeable; he could choose a different one, following a rule which could be fixed arbitrarily” (Pareto  1971, Ch.3, §11; see McLure, 2002, 46 on this point).
23As a science, pure economics is only interested in the “economic part” of action, while applied economics should make a synthesis of all the parts (economic, moral, psychological, sociological, etc.). Pure economics should therefore study the individual as a Homo economicus, while applied economics should make the synthesis of the Homo economicus, the Homo religiosus, the Homo ethicus… An actual man is an aggregate of all these different Homines, and apart from a few specific contexts, his behaviour is generally different from the behaviour of the Homo economicus:
For certain concrete phenomena the economic side matters more than all the others. In such a case, one can, without serious errors, restrict himself to the results of economic science alone. There are other concrete phenomena in which the economic side is insignificant, and there it would be absurd to restrict oneself to the results of economic science alone. Quite the contrary, they should be disregarded. There are intermediate phenomena between those two types; and economic science will reveal a more or less important aspect of them. In all cases, it is a question of degree, more or less. (Pareto,  1971, Ch.1, §27)
24Pure economics does not have the ambition of explaining all social phenomena, but only the part in which individuals perform logical actions. Since pure economics only needs a theory of logical actions and not of human actions in general, psychology has no place in pure economics. However, when it comes to practice, the Homo economicus cannot properly describe the behaviour of a real individual, which is conditioned by psychological, sociological, and other factors.
25We can now interpret BP as an attempt of synthesis of the Homo economicus and a Homo psychologicus. BP indeed assumes that individuals want to improve their well-being, as judged by themselves: we find here the first criterion of a logical action, the intention of satisfying a subjective purpose. It is then suggested that since individuals often make bad decisions (either due to incomplete information, limited cognitive abilities, or to a lack of willpower), the choice architect can help them to achieve their true purpose. Reusing the definition of Pareto, we can say that individuals often take actions which logically conjoin means to ends from their own standpoint, but not from the standpoint of the choice architect who is supposed to have a more extensive knowledge (i.e. complete information, unlimited cognitive abilities, and no lack of willpower). In the case of Luke and Carla, we assumed that Carla knew how the choice architecture could influence Luke’s choice, based on the fact that she had already offered similar contracts to other individuals and therefore noticed the behavioural pattern induced by a specific choice architecture, and also that she knew that Luke was actually a risk-lover. Libertarian paternalism is therefore grounded on the will of creating a choice architecture such that the individuals behave as if they were performing logical actions.
26As highlighted in section 1, BP relies on a dualistic model of the human being, by assuming the existence of true preferences (whose satisfaction determines the behaviour of a rational agent) and of “welfare-irrelevant” features of the choice environment, such as “frames” or “ancillary conditions” (which significantly affect the behaviour of real individuals). Similarly to Pareto who considered that the human is an aggregate of different kind of Homines, BP pictures the individual as the aggregation of a Homo economicus and a Homo psychologicus. The Homo economicus is the “neoclassical part” of the individual, and describes the behaviour of an individual when she satisfies her true preferences (she acts in an instrumentally rational way and performs a logical action), while the Homo psychologicus is the “behavioural part” of the individual, describing a non-optimising behaviour (such as following a heuristic or making errors of calculation). BP’s central claim is that individual choice is more accurately described by the Homo psychologicus, while individuals would be better off if they chose as the Homo economicus would choose (and these choices are the ones that are normatively relevant). Proponents of LP for instance argue that choice architects should design the choice architecture so that individuals choose as Homo psychologicus (which are typically not cognitively demanding), and still ensure the resulting choices coincide with what the Homo economicus would have done (Davis, 2011, 62).
27There is however a fundamental ambiguity within this reductionist approach, namely whether the different Homines are substitute or complementary parts of the individual. Pareto indeed did not precisely explain whether the different Homines of the individual simply complete or compete with the behaviour of the Homo economicus. Demeulenaere (1996, 175-177) gives the following example: consider a Muslim who wants to buy a prayer mat; the religious dimension of the transaction does not modify the economic interest for the individual to purchase her good at the lowest price. The Homo religiosus simply completes the Homo economicus by defining her ends. Now consider the same Muslim: the respect of her religious principles does not allow her to lend her money in order to get an interest. There is in this second situation a clear opposition between the behaviour of the Homo economicus and the one of the Homo religiosus. There are therefore two incompatible readings of Pareto’s framework: we can either consider the different parts of the individual as aggregated into the true preferences of the individual—that will give an objective to the instrumental Homo economicus—or as the addition of constraints which may enter in contradiction with the objective of the Homo economicus.
- 2 This issue is particularly striking in the case of intertemporal choices: although it is commonly a (...)
28Of course the very same problem arises with BP: although behavioural paternalists usually see the Homo psychologicus as inducing choices that do not satisfy the true preferences of the Homo economicus, the question of whether the Homo psychologicus may also complement Homo economicus is not tackled. Consider Petula’s choice of train. Being pessimistic may induce her to hold false beliefs about the likelihood of an incident. Her pessimism may also make her strongly averse to risky situations in general (since she is convinced that bad events are very likely to occur). Unlike Petula, Norbert is not pessimistic and objectively assumes the probability that bad events occur. But is Norbert still risk averse? A difficulty arises because Petula’s risk aversion is linked to her psychological perception of risky choices: removing pessimism from Petula’s preferences may also change her true preferences (i.e. what she would have done if she did not have a false belief about the probability of an incident). Defining Norbert as Petula’s neoclassical alter ego has two different interpretations: we could either i) consider that Norbert is just like Petula but without false beliefs, or ii) that Petula’s extreme risk aversion is due to her pessimism and is therefore not relevant for Norbert’s evaluation of the different alternatives. In this second situation, Norbert is not risk averse any more. Being pessimistic can indeed mean that people tend to perceive bad outcomes as more salient: but nothing is said whether there is a “rational” level of risk aversion or not.2
- 3 We can indeed justify the choice of riskless prospects as the expression of loss aversion when the (...)
29We could for instance argue that rational agents are rationally required to be risk neutral: being risk averse can indeed be the result of the tendency to frame choice problems in terms of gains, and therefore be a direct consequence of loss aversion3. In this situation, considering the behaviour of the Homo economicus independently of the behaviour of the Homo psychologicus is problematic; the preferences on which the Homo economicus would act are partially defined by the psychological biases that prevent this same Homo economicus from satisfying them. Assuming the existence of true preferences therefore requires that the different Homines constitutive of the individual—in a reductionist perspective—are substitute rather than complementary.
30The first methodological issue of BP is the implicit assumption that the individual is nothing else than an instrumental entity, whose nature is comparable to a computer. The individual is indeed programmed to act in a predetermined way (satisfying her true preferences), and is sometimes “defective”, in the sense that the action is not undertaken by the instrumentally rational Homo economicus, but by the psychological Homo psychologicus. As underlined by Georgescu Roegen (1971), this computational perspective of the economic agent was already present in Pareto’s analysis:
As Pareto overtly claimed, once we have determined the means at the disposal of the individual and obtained a “photograph of his tastes ... the individual may disappear.” The individual is thus reduced to a mere subscript of the ophelimity function φi(X). The logic is perfect: man is not an economic agent simply because there is no economic process. There is only a jigsaw puzzle of fitting given means to given ends, which requires a computer not an agent (Georgescu Roegen, 1971, 343).
- 4 See Sugden, 1991, 762-763, on the differences between Hume’s instrumental rationality based on pass (...)
31It can be argued that BP is more descriptively accurate than rational choice theory since the Homo psychologicus offers a more realistic description of actual behaviour than the Homo economicus. However, the individual is still reduced to a computer, programmed to satisfy her true preferences (her “tastes” for Pareto). The difficulty of BP is that nothing is said about the content of such true preferences: the true preferences are indeed defined as the preferences that the individual would reveal if she were free of reasoning imperfections, i.e. that the individual would reveal if she behaved as the Homo economicus rather than the Homo psychologicus. The issue is that providing a psychological foundation for the true preferences in terms of desires (i.e. considering that the Homo psychologicus also defines the ends that the Homo economicus tries to satisfy) cannot guarantee that those true preferences are coherent and context-independent: although the Homo economicus is instrumentally rational, the preferences it reveals are coherent and context-independent only if it is also the case of the ends it wants to satisfy. An instrumental account of rationality has however nothing to say about the rationality of the ends of the individual4: this means that the coherence of the true preferences can be postulated only if we accept a reason-based account of preferences, i.e. if we explain the true preferences as the product of the error-free reasoning of the Homo economicus (Infante et al., 2016, 10). When purifying individual preferences from psychological “biases”, we are therefore also emptying the true preferences from their psychological content.
- 5 Infante et al., 2016, 16-17, develop this point in the context of Sunstein and Thaler’s cafeteria e (...)
32However, Infante et al. (2016) argue that—in addition to its lack of psychological explanation—the latent mode of reasoning of the inner rational agent (the Homo economicus) cannot guarantee the completeness and context-independence of the true preferences, even under Sunstein and Thaler’s conditions of “complete information, unlimited cognitive abilities, and no lack of self-control” (Sunstein and Thaler, 2003, 1162). Even if we could define the counterfactual preferences of an ideally rational agent, there are probably choice problems that lack determinate rational solutions (the true preferences would not be complete), and it does not seem that choosing a salient option in this context (and therefore revealing context-dependent preferences) would contravene any principle of sound reasoning5.
33Returning to hypotheses (i), (ii), and (iii) in section 1, it seems therefore that the existence itself of coherent and context-independent true preferences can be seriously questioned. Explaining the true preferences as desires (i.e. considering the Homo psychologicus as the content of the preferences the Homo economicus wants to satisfy) cannot indeed support their coherence, while a reason-based account of the true preferences (considering the Homo psychologicus as a set of forces preventing the Homo economicus from satisfying its preferences) cannot guarantee the completeness of the true preferences.
34Suppose that hypothesis (i) is true: human psychology is therefore reduced to reasoning imperfections without impact on the ends of Homo economicus, and individuals have underlying coherent and context-independent true preferences. The question that follows is whether the satisfaction of those true preferences constitutes a valid normative criterion. It is worth noticing that Pareto only considered the Homo economicus as a descriptive model of human behaviour and a useful abstraction for the study of markets (Pareto, 1971, chap. 3, §65-66, and §87). There is in particular no value judgment about the normativity of coherent preferences. In his attempt to study human actions, Pareto only suggested that there exists a specific set of actions (that occur for instance in repeated markets) for which the individuals tend to be instrumentally rational. Behavioural economics does not particularly contradict this statement, since it only shows that in many situations within the field of study of economics people are usually not rational, in the sense that the preferences they reveal are incoherent; Pareto would probably have agreed with this observation since a huge part of his work consisted of the study of non-logical actions.
35The assumption that individuals should hold coherent true preferences, although central in BP, is however never properly justified. It is certainly clear that in many settings the satisfaction of one’s true preferences is uncontroversially the right normative criterion, for instance when individuals must choose an option among identical offers with complex pricing systems. In this case the true interest of the individual is to choose the option with the lowest price, although her bounded rationality may induce her to choose a costly option (there is in those situations an objective scale allowing the external observer to measure the true preferences of the individual). Several arguments may however seriously undermine the claim that the individuals always benefit from conforming to rational choice theory: (a) non-logical actions are not necessarily mistakes, (b) Homo economicus’s behaviour can be self-defeating, and (c) in a welfarist perspective, we can argue that the feelings expressed by the Homo psychologicus, although not rationally grounded, do actually matter.
36Imagine for instance Sarah, a supporter of the local soccer team. While she enjoys seeing her team win, she is also likely to feel sad when her team loses. Gambling can here be seen as a form of insurance: by betting against her own team, Sarah will obtain a financial compensation when the team loses, i.e. the emotional damage she will endure will be partially compensated by a strictly positive amount of money. In order to satisfy her true preferences (which are affected by the results of her team but also by her monetary gains), we should therefore incite Sarah to bet against her own team.
37Sports gambling is typically a situation in which an individual may become addicted and lose large amounts of money—justifying then paternalistic interventions so as to protect the individual against her addiction. In the above example, gambling seems however to be in Sarah’s own interest (as a form of insurance). It seems nevertheless quite implausible that sports fans would accept a policy inducing them to bet against their own team (they are indeed more likely to bet for their own team, although it is probably not rational). This is a serious argument against the legitimacy of BP: Sunstein and Thaler indeed emphasise that the acceptance of the policy is essential for a policy to be qualified as a form of means rather than ends paternalism (Hausman and Welch, 2010, 126). But if we assume that the only objective of any individual is to act in an instrumentally rational way based on predetermined preferences, then we cannot conceive that some individuals could simply claim the right to act irrationally—and therefore that deviations from rational choice theory are not necessarily mistakes. Sport fans’ preferences are probably incoherent, but if they are perfectly fine with them, we should not interpret this incoherence as a mistake.
- 6 In game situations, the idea that players can benefit from strategic commitments—i.e. voluntary dev (...)
38Another argument against the satisfaction of one’s true preferences as a proper normative standard is that Homo economicus’s behaviour can be indirectly self-defeating (Parfit, 1984). This is typically the case in coordination games (such as the Hi-Lo game; see Sugden, 1991) and games of commitment (such as the Toxin Puzzle; see Kavka, 1983). The counter-intuitive implication of such games is that if an individual wants to satisfy her true preferences, then it is necessary to adopt an apparently non-rational behaviour: being irrational can therefore be rational in those games6, since the individual could be better off by acting in a non-optimising way rather than by following the recommendations of rational choice theory.
39Berg and Gigerenzer (2010, 149) advance a related idea, by advocating a substantive rather than axiomatic form of rationality: rather than explaining behaviour as the satisfaction of given individual preferences, we should investigate whether our heuristics or decision processes are well adapted to the environments in which they are used—ecological rationality would be achieved if this is the case. This suggests that some of the actions we undertake in a social context could for instance be collectively, but not individually, rational (in the sense of rational choice theory).Consider for instance the common claim that individuals do not save enough for their retirement, and therefore that they should be nudged to increase their savings (Thaler and Benartzi, 2004). Although the true interest of the individual would be to unilaterally increase her savings, it is not certain that simultaneously increasing the savings of all the individuals is desirable since this may lead to a low level of aggregate consumption leading to a macroeconomic crisis (Keynes’s “paradox of thrift”). A reason why Americans do not adequately save for their retirement may be that it is collectively rational to do so since this ensures a high level of consumption and then of economic activity. The satisfaction of one’s true preferences also crucially depends on the behaviour of the other individuals; although it is in my interest to choose an option that better satisfies my preferences ceteris paribus, this does not mean that this is still true if we have to nudge simultaneously all the individuals. Correcting individual mistakes is therefore not necessarily a desirable objective, since those apparent mistakes could be the basis of collectively rational behaviours.
40The last argument against true preferences as a valid normative criterion has been exposed by Loewenstein and O’Donoghue’s (2004) discussion of the normative implications of their dual-self model. Although they suggest that satisfying the deliberative self’s preferences (Homo economicus) is a defensible criterion, they also argue that individual welfare depends on our affective self (i.e. Homo psychologicus’s preferences). They illustrate their argument with the choice between driving and flying. Although we know that flying is safer than driving, we are also likely to experience a higher fear when flying. While this fear may not be rationally grounded, we still actually feel it, and this should be taken into account in welfare analysis. This argument is applicable to Petula’s choice (the two others arguments are indeed less relevant for Petula): suppose that Petula has been successfully nudged, and takes the 8h train. Although it was in her best interest to take this train, she is likely to be worried during the whole journey, which makes her travel uncomfortable and should therefore be considered as an additional cost.
- 7 Qizilbash (2012, 649) presents three potential objections to this criterion, the “human limitations (...)
41The second issue of BP is that hypothesis (ii), according to which the satisfaction of the true preferences constitutes the normative criterion, is not properly justified. BP indeed implicitly defends an informed desire account of welfare (Sugden, 2008, 232; Qizilbash, 2012), but seems to take for granted that this criterion is unproblematic7. This is probably due to the progressive recognition of Pareto’s Homo economicus as a normative model of behaviour: economics was indeed defined as the science of logical actions, for which the individuals are rational and manage to satisfy their true preferences. If an individual is not rational in a situation studied by economics, for which she is supposed to be rational, then her behaviour is an anomaly. The progressive extension of the scope of economic analysis to situations in which the individuals do not perform logical actions led economists to the conclusion that the wedge between theoretical predictions and empirical observations is the result of anomalies of behaviour rather than anomalies of the theory.
42Suppose that hypotheses (i) and (ii) are true. I now question the third hypothesis of BP: that it is possible to elicit the true preferences of the individual when they exist. My main argument against the plausibility of this hypothesis is that true preferences are subjective, and that the only criterion to distinguish between true and distorted preferences from the perspective of an external observer is the internal coherence of the preferences (whether they respect the standard axioms of rational choice theory). This implies that an external observer cannot impartially identify the true preferences of the individual when several coherent preference orderings are compatible with the distorted preferences of the individual.
43We know that the preferences displayed by an individual may not correspond to her true preferences due to the influence of “welfare-irrelevant events” on her behaviour. The first possible approach to elicit the true preferences of the individual would be to directly deduce them from the set of actual choices. This approach is for instance endorsed by Bleichrodt, Pinto-Padres and Wakker (2001), who assume that the subjects of their experiment are loss averse and try then to deduce the true and unbiased preferences of the individuals from their stated preferences. Rubinstein and Salant (2012) adopt a similar approach and try to deduce from “behavioural data sets” the true preferences of the individual. However, we cannot know a priori the list of the different biases that influence individuals’ decisions. In the case of Bleichrodt et al., it can for instance be doubtful to assume that individual preferences are only biased by a loss aversion (there is indeed no reason a priori to assume that loss aversion is the only bias, and also that the individuals actually suffer from loss aversion).
44Knowing such preferences is therefore probably insufficient to obtain the true preferences, since there is an issue of identification within the determinants of behaviour between the true preferences and the possible decision flaws that affect the choices of the individual. Although we can identify coherent preferences after removing some biases, we cannot be sure that those preferences are not the conjunction of the true underlying preferences and another bias. Although BP explains individual choice as a combination of preference satisfaction and mistake, the theoretical status of those “mistakes” is not well-identified. They are indeed understood as the deviation from the satisfaction of one’s true preferences, but those true preferences can only be discovered once we have eliminated mistakes and biases.
45Since it seems difficult to directly elicit the true preferences from an external standpoint, an alternative solution would be to help the individuals to discover their own true preferences. After such discussion it would then be possible to extrapolate the preferences observed in this specific setting to other situations, since hypothesis (i) implies that the true preferences are context independent. If (i) is true, then we only need to design a framework in which the individuals tend to perform logical actions in order to be able to discover their true preferences. This assumption is for instance implicit in much experimental work, when individual risk aversion is controlled by a questionnaire (it is assumed that the risk aversion revealed in the questionnaire remains stable when the individual participates in the experiment).
46Several authors—even going back to Pareto (1909, chap. 3, §1)—have suggested that the discovery of the true preferences is the product of learning thanks to the repetition of the situation of choice. A choice architect could therefore perform repeated experiments in order to elicit the true preferences of the individuals. However, since the true preferences of the individual are defined subjectively, there is still an issue of identification since it may not be possible to distinguish between the true preferences and a systematic flaw in the decision-making process. There is therefore the temptation to define a priori the ends of the individual (for instance as selfish ones), and to view the individual as rational and performing logical actions if and only if she is satisfying the preferences expected by the experimenter. This seems for instance to be the case of Binmore’s analysis of the ultimatum game:
Novices offer a fair amount because this is what their currently operative social norm recommends. Novices who are offered unfairly small amounts are programmed to feel resentful and so want to punish the proposer by refusing. But this behaviour changes over time as people dimly perceive that the norm they are using is not adapted to the problem with which they are faced. In the Ultimatum Game, people learn that it does not make much sense to get angry if offered too little, but the mavericks who initially make small offers learn much faster that it does not make sense to demand too much if one is nearly always refused. (Binmore, 1999, F22)
- 8 In the case of experimental economics, we can typically notice that many participants are students (...)
47Binmore considers that the individuals tend to a payoff maximising behaviour in a repeated ultimatum game (since “people learn that it does not make much sense to get angry if offered too little”) even though it is not directly observable since “the mavericks who initially make small offers learn much faster that it does not make sense to demand too much if one is nearly always refused” (the individuals therefore do not converge to the Nash equilibrium). There is therefore the implicit assumption that the individuals respect a social norm because they want to maximise their payoff, although we cannot know what the true motives of the individuals actually are. We can for instance view an individual as respecting a specific norm by conformism (see for instance Asch’s (1955) famous experiment), or—as suggested by Binmore himself (Binmore, 1999, F19)—that the subjects want to achieve what they perceive as the experimenter’s objective8, since this one can be seen as an authoritative figure (Milgram, 1975).
48Since we cannot make a clear distinction between the ends of the individual and the different factors that can influence her decision, it seems quite difficult to design an experiment for which “the time allowed for trial-and-error adjustment is ‘sufficient’ ” (Binmore, 1999, F17). An apparent stable behaviour can indeed correspond to the pursuit of a specific end plus a systematic flaw in decision-making. For instance, in the previous example we can assume that the true objective of an individual is to offer and accept only equal shares, but that she prefers to follow an unfair rule that was implemented during the experiment by conformism. Although the actual behaviour of the individual is accurately predicted by the theory that the individuals want to maximise their payoff, the underlying reasons of her choice are more complex. Furthermore, she is unable to satisfy her true preferences.
49Recall the situation of Carla and Luke: based on Luke’s past financial decisions, Carla believes that Luke truly prefers risky assets. Under the assumption that Luke’s true preferences remained stable over time (and therefore that Luke’s past behaviours can give to Carla some meaningful indications on Luke’s current true preferences), Carla cannot however be certain that Luke is truly a risk-lover. Suppose for instance that the satisfaction of Luke’s true preferences would imply a risk neutral behaviour, but also that, for an unspecified psychological reason, he systematically slightly underestimates his potential losses. In this situation, although his preferences revealed in actions tend to highlight underlying risk-seeking preferences (although he is still subject to framing effects, and therefore chooses riskless options when the choice problem is framed in terms of gains), Carla is unable to distinguish between this systematic mistake and Luke’s underlying true preferences.
50The risk is then that behavioural paternalists impose their own normative views about the true preferences of the individual. Indeed, it is striking that—although BP’s central claim is that people should be helped so as to satisfy their (subjective) true preferences—most of behavioural paternalists claim that we should not have a “too high” body mass index, we should not smoke, we should save a relatively large amount of our incomes, and more generally that we should give a high weight to the long term consequences of our actions. Although those positions seem reasonable, and it can be argued that the reason why many individuals are actually obese or overburden with debt is their lack of self-control, we cannot be certain that an individual cannot truly privilege the short-term benefits of her actions to the long-term costs. If we accept that what fundamentally matters for our own well-being is the long-term consequences of our actions, then we could also argue that individual’s true preferences should induce a vegetarian diet (since livestock farming is very costly in terms of natural resources, and therefore is against our interest in the long term), that individuals should use a bike instead of a car to go to their workplace etc. and it would make sense to argue that the reason why actual people are still eating meat or use their cars is that they have a bias towards the present, and neglect the future costs of their actions. Although those preferences are perfectly defensible and reasonable, it seems dubious to argue that this is necessarily what we truly prefer. The risk is that behavioural paternalists impose in fine as our true preferences what they think it is rational to wish. Since the object of BP is to incite the individuals to follow the choices that would have been done by their Homo economicus, BP implies a standard form of ends paternalism: the ends of the actual individual are indeed replaced by the ends of a counterfactual individual, the Homo economicus.
51To avoid misunderstanding, I must however make clear that my criticism is not about whether ends paternalism is ethically justifiable or not: I am criticising here BP’s claim that we can justify paternalistic policies on the basis that it is what people would want, were they perfectly rational on the basis of their true preferences. What I am questioning here is the democratic legitimacy of behavioural paternalists to decide what we should consider as the true preferences of the individuals, and therefore what should be the objective of a wide range of public policies (such as the level of savings for our retirement). There are many good reasons to argue in favour of paternalistic policies against obesity, the use of polluting vehicles for short distances, etc., but it is not certain that the reason according to which it is what we would do if we were rational is a good one.
52The third difficulty of BP is that, although we assumed that the individual has a unique underlying coherent preference ordering, it does not seem possible to distinguish between those true preferences and a systematic decision flaw. The crucial issue is that the social planner should be able to clearly define what would be the choice of the individuals if they were able to perform logical actions. Logical actions are however defined by conditions such that “sufficient” repetitions and “adequate” incentives: the qualification of “logical” for a specific action is therefore subject to the personal interpretation of the observer. It seems therefore implausible to implement an impartial procedure that could isolate the true preferences from the actual preferences of the individual.
53I have argued that the three hypotheses underlying BP are quite fragile. The existence of true preferences indeed requires accepting a quite implausible theory of human behaviour that is grounded on principles of rational choice and free from any explicit psychology. It also forces us to accept a reductionist account of human beings as a rational agent trapped in a psychological shell, and to treat psychological characteristics as alien elements to our true self. Furthermore, BP imposes the satisfaction of those true preferences as the normative criterion without proper justification. Lastly, it does not seem possible to elicit such true preferences from the behaviour of the individual, since disentangling between true preferences and mistakes at some point depends on the subjective interpretation of the observer. I suggested that these different issues are related to BP’s implicit endorsement of Pareto’s methodology.
54The challenge of behavioural economics probably consists more in questioning the assumptions underlying BP rather than merely questioning the “efficiency” of the individual in satisfying those true preferences. Considering the neoclassical framework as a benchmark for modelling individual behaviour, and progressively integrating psychological factors into the model, may not be the best approach for normative analysis. This approach indeed presupposes that an individual Petula is merely a preference relation, and that her life simply consists in the more or less successful satisfaction of these preferences. These true preferences are necessarily exogenous and remain stable over time: BP models Petula as if she was simply a computer (with some programming imperfections), programmed to satisfy a certain number of predetermined objectives given by her true preferences. This model does not seem to provide a realistic account of personal identity, since it results from the model of the Paretian Homo economicus, which was defined as a representative agent for investigation purposes and the study of market equilibrium (Boianovsky 2013, Lecouteux 2013). In particular, it was explicitly designed so as to avoid integrating within economic models the complexity of individual behaviours. It can here seem a bit ironic that, as behavioural economists, behavioural paternalists implicitly accept the model of the inner rational agent, for which there is no known psychological foundation (Infante et al., 2016, 22)—defining a psychology-free entity was precisely the objective of Pareto, but he knew that this model could not offer a precise comprehension of individual behaviour (Pareto, 1916, §36).
55An implicit assumption supporting the optimisation-based approach for modelling boundedly rational individuals is that, by progressively introducing psychological factors into the model, we will reach in fine an accurate description of how people actually behave. This means that the only difference that exists between a computer and a Human is their power of calculus: the computer is taken as the model of efficiency the Human should tend to. Although this kind of approach can be quite relevant for descriptive purposes, such models run into problems when used to make normative prescriptions on individual welfare. Perhaps behavioural economics does not tell us that Humans are defective computers, but rather that Humans are not computers.
This paper is the third chapter of my PhD thesis. I am grateful to Robert Sugden, Guillaume Hollard, and participants to the conference Economics and Psychology in Historical Perspective (Paris, December 2014) and the 19th Conference of the European Society for the History of Economic Thought (Rome, May 2015). I also thank the editors for their comments and Wade Hands for his detailed comments and suggestions that substantially improved this paper.