An overconfident judgement

As noted above, it would be unusual to find anyone working in the field of behavioural public policy, whether they are of a paternalistic or a more liberal persuasion, who dismisses the notion that people are often inconsistent or, equivalently, incoherent in their choices. However, if a person does choose, say, Good A over Good B in one context, and Good B over Good A in another, how would a third person – including a policymaker – know which choice is the more accurate representation of that person’s preferences (see also Sugden, Reference Sugden2008)? The dominant paternalists in the field – the so-called libertarian paternalists – maintain that we must appeal to a person’s deliberative self, their inner rational agent (or in other words, the criterion of informed desire), rather than their automatic, reflexive choices, to determine what they really want (Thaler and Sunstein, Reference Thaler and Sunstein2008). Smokers, for instance, might say, after a period of reflection, that they really want to abstain from cigarettes, and thus the libertarian paternalist would assume that abstinence reflects their true preferences. However, this is a troubling conjecture, for several interrelated reasons.

For example, people who are asked for their deliberative preferences might simply respond in the way that they believe the interviewer wants to hear – smokers, say, may state on reflection that they wish to quit smoking because they believe that they are being prompted to do so. Moreover, we cannot be sure that the deliberative self is free of the behavioural affects against which the soft paternalists wish to mitigate. For instance, smokers, when asked to reflect on whether they wish to quit smoking, might ignore, or at least downplay, the benefits they reap from smoking. Finally, but most importantly, there is no philosophical or psychological basis on which to argue that an inner rational agent even exists; that the notion of there being an impartial deliberative self that resides within us is a myth (see, e.g. Sugden, Reference Sugden2008; Chater, Reference Chater2019; Bernheim, Reference Bernheim2021). Thus, when people offer (deliberative) justifications for why they do, or do not do, something that they claim to prefer, they are perhaps merely offering reasons – motivated reasons – that do not really drive their choices.

To illustrate the contention that one simply cannot discern which, if any, preference is the ‘right’ one when an inconsistency is observed, consider the following prominent inconsistencies in the behavioural literature: (i) classic preference reversals; (ii) actual pension savings levels versus ex post ‘desired’ pension savings levels; (iii) willingness to pay (WTP) versus willingness to accept (WTA) discrepancies. Classic preference reversals, most closely associated with Sarah Lichtenstein and Paul Slovic (e.g. Lichtenstein and Slovic, Reference Lichtenstein and Slovic1973), show that people have a tendency to prefer high-probability low-outcome bets over low-probability high-outcome bets in pairwise choice tasks, but equate the latter with relatively high-money amounts when asked to separately value them. From these results, we cannot discern whether the preferences indicated in the pairwise choices or the direct valuations are definitively right or wrong; indeed, Lichtenstein, after explaining to respondents with reversed preferences exactly what they had done, found that most of them did not wish to change their answers.Footnote ¹

The alleged inconsistency regarding pension savings is discussed extensively in behavioural public policy (and elsewhere), and centres on the apparent regret that many of those of pensionable age feel in relation to not saving more for their retirements when they were younger. The inconsistency is thus temporal; that is it is between actual savings when young and expressed desired earlier savings when older, and is often attributed to present bias. However, one cannot definitively conclude that the later regret-fuelled preference is the most accurate reflection of an individual’s lifetime desires. The suggestion that the older self does not fully appreciate or remember the desires of the younger self is as plausible as the view that the younger self fails to appreciate or foresee the desires of the older self.Footnote ² In terms of an entire life, a third party – and possibly even the first party – cannot deduce whether more or less saving when young would best reflect an individual’s lifetime preference structure (see Oliver, Reference Oliver2023).

The substantial difference that is often observed between a person’s WTP – their buying price – for a particular good and their WTA – their selling price – for that same good is not accounted for in standard economic theory, and is generally attributed to loss aversion. That is, once a person owns a particular good, their reluctance to give it up – to ‘lose’ the good – is not reflected in their a priori desire to own the good. This endowment effect was perhaps most famously reported by Kahneman et al. (Reference Kahneman, Knetsch and Thaler1990), who found that the median respondent selling price for coffee mugs was twice that of buying prices, an implicitly inconsistent preference but again one in which it is impossible to discern which is the more accurate reflection of ‘true’ preference. To contend that one particular preference is more likely to be the correct one when an inconsistency in choice is observed is a judgement that is often driven by an overconfidence that one’s own preferences are shared by others. To steer people in the direction of that preference from a policy perspective (perhaps without the target population’s permission or even knowledge), even if one maintains that the intervention is non-coercive, is inevitably paternalistic. Given that preferences are apparently highly malleable, it might be that neither preference is a good reflection of what people really value. Or it may be that both offer a good reflection, given the context and way in which they were uncovered (see also Rizzo and Whitman, Reference Rizzo and Whitman2020: 75).

Justifiable inconsistency

In the previous section, I argued that when one observes a choice inconsistency, one cannot generally determine which, if any, of the choices is the ‘right’ one, which counters the paternalistic urge to steer, or even push, people in particular directions. According to Gigerenzer (Reference Gigerenzer2018), the contention that inconsistencies are caused by behavioural biases that are in need of correction has become a truism in much of the behavioural public policy discourse, and yet, he argues, the fast and frugal heuristics that drive much of our behaviour and choices evolved to help us to navigate the uncertainty that is an inevitable feature of almost all of the decisions that we face.Footnote ³ Gigerenzer thus suggests that to view choice inconsistencies as necessarily caused by biases is in itself a bias; beware, he warns, of the ‘bias bias’ (see also Reference Gigerenzer, Gigerenzer, Mousavi and VialeGigerenzer, Forthcoming).

The ‘error error’ that forms part of the title of this article is something akin to Gigerenzer’s bias bias in that I suggest that it is often erroneous to assume that either of the choices that form an inconsistency are necessarily the result of individual error.Footnote ⁴ Rather, it is plausible that we have evolved to be flexible in different contexts because this helps us to achieve our objectives. A person with rigid preferences, living a context-independent life, may, at the very least, suffer a highly regimented, uninteresting existence; at worst, they may find themselves at an evolutionary disadvantage. The argument is thus taken one step further than that outlined in the previous section in that it is postulated here that choice inconsistencies often occur when considerations of outcomes interact with those of context in meaningful ways, such that different values are legitimately placed on the same outcomes in different contexts. In short, the contention is that context dependence is a strength, not a weakness, of the human psyche.

Let us consider this assertion in relation to the Allais paradox, an empirical finding that was foundational to the development of behavioural economics. The paradox shows that people tend to demonstrate inconsistent implicit risk attitudes when choosing between two risky options compared to choosing between a risky option and an option where risk is entirely absent (Allais, Reference Allais1953). In the latter choice, many people place great weight on certainty, resulting in relatively strong apparent risk aversion, an attitude that can in itself be attributed to loss aversion. Although considered a paradox in that it is a direct violation of the independence axiom of von Neumann–Morgenstern expected utility theory and thus descriptively challenges the notion that people maximise their valuation of final outcomes across different choices, one may reasonably ask, what’s wrong with placing a heavy weight upon certainty – with being relatively risk averse – when certainty is an option? Indeed, similar to respondents demonstrating classic preference reversals, Slovic and Tversky (Reference Slovic and Tversky1974) observed that Allais paradox-committing respondents generally did not wish to alter their choices after the reasoning of the paradox had been fully explained to them. Certainty brings the security that one may value in and of itself, because it frees one from the worry and/or realisation of a potentially catastrophic outcome of a risky option; when risk is unavoidable, on the other hand, one may quite legitimately choose the riskier but potentially more rewarding option, particularly when the differential risk between the options available is almost indiscernible.

For a simple hypothetical example of context apparently affecting a preference ordering, consider the archetypal example to which advocates of nudge interventions often refer: the differential placement of fruit and cake on cafeteria shelves. If fruit is placed at the front and cake at the back of the counter, so the argument goes, there will be a greater likelihood that people will choose fruit over cake, and vice versa if the positionings are reversed. Soft paternalists argue that this choice inconsistency is caused by salience; if one were to care about maximising welfare – which, according to nudge advocates if we remember, can be deduced by asking people for their deliberative preferences – then most people will prefer the healthier option, which justifies the placement of fruit rather than cake in the most salient position. That is, it is assumed that, for many people, the choice of cake, when cake is made most salient, is an error in need of correction.

However, as noted earlier, Sugden (Reference Sugden2008) and others maintain that the notion that there is a deliberative self is a fallacy, and thus there is no mode of reasoning that the customer could have used to determine whether they prefer fruit or cake. In these terms, the inconsistency is unproblematic because neither choice can be rationalised. However, an alternative reason why the inconsistency is unproblematic is because the different contexts – that is fruit or cake at the front – may react with the value that consumers place on either desert in meaningful ways, ways that a third party cannot discern, such that neither choice can be viewed as erroneous. For example, placing an item at the front of a shelf, whether it be fruit or cake, may imbue it with a sense of freshness that the customer values, or may incentivise them to focus on how it complements the rest of their meal, etc. With all these features influencing the customers’ choices, we simply cannot say which choice would maximise their welfare, and therefore if planners were to insist that the fruit be placed at the front we can only conclude that this reflects their preferences and not those whom they are aiming to influence.

As an aside, Sugden (Reference Sugden2008) also contends that any dessert ordering that is imposed by a planner does not take into account the preferences or interests of the cafeteria owners. It is likely that the primary objective of the cafeteria owners is to make a profit in a business where profit margins are notoriously tight, and so long as they are not engaged in obfuscation or deception, Sugden maintains that they should be allowed to do so without being subjected to interferences by third parties. If it is more profitable for the cafeteria owners to guide people in the direction of cake rather than fruit, perhaps because the cakes they are selling have a shorter shelf life, then in Sugden’s schema their placing of cake at the front of the counter is a legitimate business practice. However, one can question whether this practice undermines the notion of a free and fair exchange between the owner and the customer.Footnote ⁵ If we, as a society, judge the practice as an unreasonable manipulation of customer buying patterns, then there may be non-paternalistic grounds to regulate against the deliberate salient placement of cake for profit-making purposes (assuming such a regulation is enforceable). This would be a regulation against actions by the cafeteria owners that impose harms on their customers – an externality concern.

Leaving the argument for regulating against externality-generating manipulations to one side (for now), I have so far argued that there are few grounds for correcting choice inconsistencies by appealing to the possibility that one or the other of the choices best reflects the underlying preferences of those targeted for behaviour change. However, this does not mean that there are no grounds at all. Bernheim (Reference Bernheim2021), for instance, who, as noted earlier and like Sugden, questions the notion of an inner rational agent, has in a series of articles postulated a model of behavioural welfare economics where some choice inconsistencies can be classified as errors (e.g.. see Bernheim and Rangel, Reference Bernheim and Rangel2009). The Bernheim–Rangel framework classifies a choice as a mistake if it is predicated on a clear misunderstanding of the available options and consequences, and thus an unallowable inconsistency arises if such a choice conflicts with at least one choice where no such misunderstanding is evident. For example, if an individual expresses a preference for a box of bananas over a box of apples but was mistaken in believing that the box of bananas contained not bananas but peaches, and yet normally, in the absence of what Bernheim and Rangel call a characterisation error, prefers apples over bananas, a third party may legitimately correct the former choice. If there are no characterisation errors, the Bernheim–Rangel model enforces no consistency requirements and thus the model, in such circumstances, adjudges that each of us is the best arbiter of, and will choose that which is most conducive to, our own well-being.

Sugden remains opposed to the notion that policymakers can discern and correct characterisation errors, maintaining instead that they ought to encourage the provision of the most extensive possible range of opportunities from which individuals can choose, and allow individuals free rein to pursue their wants via mutually agreeable exchanges between providers and consumers (Sugden, Reference Sugden2018; see also Sugden, Reference Sugden2021). Bernheim’s (Reference Bernheim2021) view that Sugden’s opportunities framework should offer more scope for considering whether people understand the choices that they face is, to me, convincing. With respect to Bernheim’s specific contention, I also find the argument for correcting characterisation errors appealing, at least with respect to mistakes that seem very obvious (e.g. when people mistake a box of peaches for a box of bananas), if they can be identified and broadly agreed upon. As another hypothetical example, a UK shoe store orders a collection of shoes that it estimates are suitable for its clientele from an overseas manufacturer, not realising that the manufacturer standardises its shoe sizing in US rather than UK measurements. The shoe store’s ‘true’ preference was for shoes in UK sizing, and thus we can conclude the inconsistency between that preference and the order is a genuine error. Ways in which policy intervention might reduce the possibility of such errors occurring will be discussed later in this article.

Unfortunately, cases in which there is general agreement that inconsistencies are caused by characterisation errors might be the less common occurrence, and those where there is little reason to overturn either choice in an inconsistent set may be rife.Footnote ⁶ Take classic preference reversals again, for example. To recap, people have a tendency to prefer high-probability bets with modest payoffs over low-probability bets with high payoffs in pairwise choice tasks, but place a higher money value on the latter than on the former in independent WTA/WTP tasks. It is possible that people demonstrate this inconsistency because we are ill-adept as humans to process these kinds of decision tasks (Gigerenzer, Reference Gigerenzer2018). If people cannot really understand very well the options that they face, we might expect a degree of inconsistency; if so, designing the presentation of the options in ways that people can mentally better process may reduce these inconsistencies. However, if that were the case, the inconsistencies that have been observed are likely to be random rather than systematic in their direction. That is, apparent preference patterns in the direction of choosing the low-probability high-payoff bet but valuing higher the high-probability modest-payoff bet would be as prevalent as choosing the latter but valuing higher the former, but this is evidently not the case.Footnote ⁷ The systematic direction of the inconsistency implies intentionality, compounded by the fact that most respondents do not wish to alter their decisions after their choices, valuations and inconsistencies have been carefully explained to them.Footnote ⁸

If we are sceptical that the notion of a ‘deliberative preference’ exists, as Bernheim and Sugden both are, then we ought to be cautious of any rationale for why people demonstrate these quite complex inconsistencies, whether it be offered by a third party or by the respondents themselves. The most likely explanation for classic preference reversals may be that people tend to anchor on the payoffs when asked to independently value bets, which would favour the bet with the highest payoff, and yet compare each bet’s various aspects in pairwise choice tasks, which may provoke a greater degree of risk aversion. However, even if this is what is happening there is no obvious error in either type of response, and thus, assuming that the respondents are not being manipulated to serve the interests of another party, there is no reason to intervene. It may be difficult to offer explanations for many types of choice inconsistency; it is often difficult to articulate, or even fully comprehend, why we do the things that we do. We might just choose, act and behave differently across different contexts because that is simply what we want to do, and if there is no obvious error in doing so then there is little justification for direct intervention to counter the observed inconsistencies.

Dealing with inconsistency

The conclusion reached so far is that intrapersonal inconsistencies in choice across time or context should not be judged as unreasonable, other than when they are caused by clear and obvious mistakes in individual choice.Footnote ⁹ The question, then, is what are the lessons for policy?

Sugden (Reference Sugden2008), who, as noted earlier, even rejects the concern for characterisation errors, draws on the work of Buchanan (Reference Buchanan1968), who in turn drew on Wicksell’s (1896/Reference Wicksell, Musgrave and Peacock1896/1958) voluntary exchange theory, to reject the notion that an ideal planner can ‘clean’ the choice inconsistencies that are often evident in free exchange. Sugden (Reference Sugden2008: 242) writes that ‘… it is to each person’s advantage that he is able to get what he wants and is willing to pay for, when he wants it and is willing to pay for it’. That is, we ought to look to forms of exchange that individuals recognise as mutually advantageous, each one on their own terms and given their own individual objectives for their lives. Following this line of argument, there is no need for individual preferences to be fully consistent or coherent, and thus the findings uncovered by behavioural economists and psychologists do not challenge the market mechanism. From a policy perspective, we deal with choice inconsistencies by ignoring them. Therefore, aside from regulating against cases of fraud and obfuscation, Sugden’s principal policy recommendation is to maximise the opportunity set and to let people choose and exchange as they desire (Sugden, Reference Sugden2018).

One way to attempt to deal with inconsistencies caused by the characterisation errors identified as important by Bernheim (Reference Bernheim2021) is to improve agentic capabilities, an aspect that Dold and Rizzo (Reference Dold and Rizzo2021) and Dold and Lewis (Reference Dold and Lewis2023) have also, like Bernheim, convincingly (to my mind) identified as an important, missing, part of Sugden’s thesis.Footnote ¹⁰ Agentic capability refers to the ability for people to reflect on their choices in a meaningful way, so that they are more likely to be ‘their own’. There are very general ways in which agency may be enhanced if it is held that there are some ‘goods’ – say, education, health or a basic level of income – that are considered in some sense ‘primary’; primary, that is, if having access to these is considered foundational to a person’s ability to take effective advantage of their own agency. This line of argument offers a classical liberal justification for governments to invest in the provision of (at least) basic universal levels of education, health care and income, mirroring the arguments made by Mill (1859/Reference Mill1859/1969), at least with respect to education, and to some degree Sen (Reference Sen1999) in his capabilities approach.

A more specific set of interventions aimed at identifying characterisation errors fall under the label of boosting, an approach principally associated with Gerd Gigerenzer, Till Grüne-Yanoff and Ralph Hertwig (e.g., see Hertwig, Reference Hertwig2017; Hertwig and Grüne-Yanoff, Reference Hertwig and Grüne-Yanoff2017; Reference Gigerenzer, Gigerenzer, Mousavi and VialeGigerenzer, Forthcoming). Hertwig (Reference Hertwig2017: 143) writes that the ‘goal of boosts is to make it easier for people to exercise their own agency in making choices. For instance, when people are at risk of making poor health, medical or financial choices, the policy-maker … can take action to foster or boost individuals’ own decision-making competencies’. Boosts can be of many different kinds, but those commonly considered include teaching people simple accounting rules aimed at improving their financial literacy, and instructing people on the implications of false negatives and false positives to give them a better understanding of what proportion of those who test positive for medical conditions actually have the illness under investigation. Visual aids can also be used to perhaps help people better comprehend risky options, as might presenting probabilities as natural frequencies (e.g. ‘1 in 100 people are at risk of …’, rather than ‘there is a 0.01 risk of …’). Moreover, boosts can focus on teaching people about the behavioural phenomena such as so-called present bias, loss aversion and the like, so that they can in principle guard against the impact of these phenomena in their own decision-making if they autonomously wish to do so (e.g. they may feel as though salience is tempting them to eat too many sweets, and might therefore choose to place their store of chocolate at the back rather than the front of their food cupboard, an example of what has been termed self-nudging – see Reijula and Hertwig, Reference Reijula and Hertwig2022).

However, there are several possible issues with boosting. For instance, since they generally – although not always – target individual decision contexts, they might not improve agentic capabilities in the kind of blanket way that is potentially offered by public investment in the basic autonomy-enabling goods and services discussed earlier.Footnote ¹¹ Boosts are also labour intensive, and rely on the assumption that people are receptive to, and capable of, absorbing the forms of educational training that are embedded within them. One could also make the claim that boosts change the ‘context’ of choice such that decisions are altered in ways that are unnatural when one considers most real-world decision-making scenarios, and yet the real-world scenarios reveal preferences that are not necessarily illegitimate given their framing. If so, boosts may remove a choice inconsistency when, at least in some circumstances, the inconsistency may offer a better reflection of a person’s preferences given the choice context faced.Footnote ¹² That said, boosts are not intended specifically to correct choice inconsistencies; the advocates of the approach, including Gigerenzer (Reference Gigerenzer2018) as discussed earlier, recognise that such inconsistencies are often a reflection of legitimate preferences. But it may well be that in some circumstances boosts help to correct characterisation errors, thus addressing, at least in part, the concern raised by Bernheim (Reference Bernheim2021). Thus, the careful use of boosts, as a complement rather than as a substitute for the broader institutional interventions that are required to enhance agentic capabilities (e.g. educational and health care services, and a decent minimum income), might be a useful, if marginal, weapon in the policymaker’s armoury.Footnote ¹³

In addition to basic characterisation errors, illegitimate choice inconsistencies can also occur when one party to an exchange manipulates another party through an act of self-interest, which places a further potential qualifier on Sugden’s (Reference Sugden2018) opportunities framework. Countless real and hypothetical examples could be given to illustrate this point, but let us imagine that a shopkeeper notices that her customers are more likely to buy a particular brand of discounted biscuits if a selection of the same – but undiscounted – biscuits are placed next to the discounted biscuits on her shelves. That is, a change in the decision context to some extent causes customers who would not otherwise have bought that particular brand of biscuits to buy them because the introduction of the undiscounted biscuits affects people’s perceptions of the relative ‘value’ of discounted ones – an example of what is known as the ‘decoy effect’. We may conclude from the previous discussion in this article that there is no obvious error in individual choice when preference inconsistencies are observed via the decoy effect. Context can affect people’s preferences for goods in meaningful ways.Footnote ¹⁴ However, what is different here is that the shopkeeper has deliberately altered her customers’ choices through an act of self-interested behavioural-informed manipulation. We can therefore surmise that the behavioural affects are steering choices not in a way that guides customers towards meaningful desires that they wish to pursue, but are instead being deliberately used to cause choices that conflict with their unmanipulated preferences. Due to the very nature of manipulation, customers find it difficult to avoid these influences.Footnote ¹⁵

One may therefore conclude that such acts of self-serving manipulation cause unwarranted choice inconsistencies, the mitigation of which calls for government intervention. One form of intervention would be to boost in the manner mentioned above – that is to educate people about the behavioural affects in the hope that they may then be better able to protect themselves against unwanted interferences. However, even when they are educated as such, boost advocates are perhaps overoptimistic in terms of the capacity and time that people have to notice manipulations that are, by their very definition, well concealed. Thus, the contention here is that there will be occasions when policymakers are justified in openly regulating against particular behavioural manipulations in the exchange relationship. The ‘openness’ clause here is important, since there must by public scrutiny of the legitimacy of any potential form of regulation so as to abide by the rules of deliberative democracy. Not all manipulations of this type will warrant mitigation and thus there cannot be a ‘blanket’ policy response to them. They need to be considered by policymakers on a case-by-case basis, who are likely to conclude that some are relatively harmful infringements designed to attract notice of a firm’s products in crowded marketplaces, and that profit margins in some sectors are so tight that companies may legitimately entice customers to buy their products with the help of these techniques. Yet, as I have noted elsewhere, there are cases where such manipulations are so egregious that policymakers will do well to conclude that some specific, otherwise egoistically fuelled, freedoms have to be constrained if one wants to protect freedom in general (Lippmann, 1937/Reference Lippmann2017; Oliver, Reference Oliver2022). I have in other writings defined government regulations against interferences in a free and fair exchange as ‘budges’ (e.g., see Oliver, Reference Oliver2013).Footnote ¹⁶