Self-discrepancy theory

The first layer of the theory comes from self-discrepancy theory (Higgins, Reference Higgins1987). Simplistically, self-actualisation is driven by attempts to align our ‘actual’ self with our ‘ideal’ and ‘ought’ selves. Our actual self is who we are right now, including certain innate motivations, talents and parameters like height. We may have an organic disposition towards science and adventure, for example. Our ideal self is who we would like to be, like an astronaut. The actual self encourages and constrains the ideal self. Someone predisposed to exercise will have an easier time becoming an athlete, for example. Equally, someone short will struggle to become a competitive basketballer. Finally, our ought self is who we feel we have a responsibility to be. For example, a religious individual might ideally like to complete several pilgrimages but feels that they ought to stay home and provide for their family. In reality, the self is fragmented and compartmentalised into many more ‘multiple selves’ than these three concepts (Showers & Zeigler-Hill, Reference Showers, Zeigler-Hill, Leary and Tangney2012), but the concepts nonetheless provide a useful and parsimonious framework through which to understand a central mechanism of preference formation. The ideal and ought selves have some similarities with what Callard (Reference Callard2018) calls as ‘aspirational self’. They are valued identities that we only understand vaguely and reason towards proleptically, calibrating them as we learn about what is valuable and what would suit us through choice and associated feedback.

The ideal and ought selves are high-level goals of some sort, like becoming an astronaut. Various lower-level goals are then derived from this higher-level goal in a nested fashion. For example, to become an astronaut, one would need to be knowledgeable in physics and fit enough to withstand a space shuttle launch. To become fit, one would need to attend the gym regularly and eat well. A good diet would consist of vegetables and proteins while being low in fat and sugar. The individual would therefore prefer a chickpea salad over a burger and chips combo for lunch. In this way, the high-level goal informs preferences down to a minute level, including even seemingly trivial consumption choices, and the ideal self comes to constitute an entire identity. Note that neither the ideal nor ought self is necessarily ambitious. We use the example of an astronaut because it is straightforward to work with in our analysis, but many people have ideal and ought selves that are more mundane. Someone may ideally like to be a homebody with a well-kept garden, or a reliable father, or a good boss in small local firm.

The process of harmonising the actual, ideal, and ought selves through behavioural choices is guided by affective signals (Showers & Zeigler-Hill, Reference Showers, Zeigler-Hill, Leary and Tangney2012). Notably, discrepancies between the actual and ideal selves trigger depression, while discrepancies between the actual and ought selves trigger anxiety (Silvia & Eddington, Reference Silvia, Eddington, Leary and Tangney2012). These psychopathologies are especially acute when the actual self aligns with the ‘feared self’, a conceptualisation of who the individual does not want to be (Woodman & Hemmings, Reference Woodman and Hemmings2008). Conversely, coinciding with your ideal self brings positive affect. We feel joy when we achieve our goals (Emmons, Reference Emmons1986). We feel exhilarated when we recognise that we are progressing dynamically towards those goals. And we feel vitality or a sense of easy motivation when we are working towards valued ends (Sheldon & Elliot, Reference Sheldon and Elliot1999). Note that affective signals come when choices are realised, which is often a process rather than an event. A burger might taste delicious when ingested, but some of the affective signal associated with consuming the burger comes later when you're looking at your gut and feeling sad about your lack of progress towards your aesthetic goal. Learning and preference formation thus encompasses much more than decision utility.

Self-actualisation through harmonisation of the self-constructs proceeds by way of choice, feedback and adjustment (illustrated in Figure 4). The individual begins by developing some conception of the ideal self that is amenable to the ought self and not ruled out by parameters of the actual self. They then try to become this ideal self by choosing in accordance with the values, behaviours and symbols inherent to that ideal self. In other words, they try to affirm an identity. For example, the wannabe astronaut might take steps in that direction by joining a gymnastics club and taking a course in astrochemistry. Their choice behaviour over time will reveal their actual self and make apparent whether they are progressing dynamically towards their ideal self. Imagine that the wannabe astronaut always finds excuses to skip out on gym training but loves their astrochemistry course. They will be revealed in their actions as not being their ideal self, namely on-track to be an astronaut. The individual must introspect on the affective signals that will accompany this information. If they become depressed, as predicted by self-discrepancy theory, it suggests that this ideal self-concept (astronaut) might be appropriate for them, but they must work harder or smarter to realise it. Perhaps they need to swap from gymnastics to swimming. If they are not upset, then perhaps they have little motivation for being an astronaut and should consider an alternate ideal self that is more inspiring. Given the positive feedback they get from astrochemistry, maybe a planet-bound research role will suit them better. They should pursue these feelings further by deepening the associated identity, perhaps by seeking employment as a research assistant, or taking similar courses.

Figure 4. The iterative process of self-actualisation.

Self-determination theory

The second layer of our theory of preference formation comes from self-determination theory (Ryan & Deci, Reference Ryan and Deci2017; SDT), which helps to describe the nature of the actual, ideal and ought self-constructs in terms of motivational differences. It also provides some explanation as to why some values and behaviours ‘fit’ an individual better than others.

SDT posits a spectrum of motivation running from intrinsic on one end to extrinsic on the other (see Figure 5). Intrinsically motivated behaviours are engaged in for their own sake, often spontaneously. They are thus self-determined. In contrast, extrinsically motivated behaviours require some degree of self-regulation because resistant parts of the psyche must be suppressed to undertake such behaviours. The most extreme form of extrinsic motivation is duress, where an outside influence coerces behaviour – these are controlled behaviours and are not engaged in autonomously. One step closer to intrinsic motivation is introjected behaviours. These are extrinsically motivated, but unlike duress, here the individual administers the control upon themselves. An illustrative example is behaviours done solely to earn parental approval. Identified behaviours are recognised as valuable by the individual, but they are not intrinsically motivated. For example, many people exercise to improve their health, but have only limited motivation for the exercise itself. Extrinsic, introjected and identified behaviour all involve degrees of self-regulation because they are not undertaken spontaneously as a result of intrinsic motivation. Integrated behaviours have been connected to other intrinsically motivated behaviours. They thereby become easier to motivate and move into a more central position within an individual's identity. For example, someone who joins a soccer club for exercise may find that ‘doing soccer’ becomes connected over time to socialising, watching soccer and other behaviours and values that are intrinsically motivated for them. This makes ‘doing soccer’ easier to motivate.

Figure 5. The motivation spectrum in self-determination theory.

The actual, ideal and ought selves map neatly but imperfectly onto SDT's motivation spectrum (see Figure 6). The ought self involves regulating one's behaviour to conform to ethical objectives, including other-regarding preferences. It thus corresponds to introjected and identified motivation. The ideal self is something you want to become. It is made up of identified behaviours that become integrated over time. For example, an aspiring astronaut might identify with fitness and physics literacy. Pursuing these behaviours is often arduous and requires self-regulation (i.e., willpower), especially in the early stages (Besser-Jones, Reference Besser-Jones2014). The individual must master difficult mathematics and train up to an athletic standard well beyond their current capacity. However, as the individual's competence improves and these behaviours become assimilated into the individual's life and routines, they become integrated. The individual might incorporate cycling into their commute to the lab, for example, and see the world in a more enchanting way thanks to their newfound mathematical literacy. This makes their motivation easier. Finally, once the individual becomes their ideal self, the associated behaviours may become fully intrinsic. Once they are a skilful physicist and apex athlete, the associated tasks will no longer seem arduous. Indeed, the individual will likely get a rush of positive affect when executing these tasks because of feelings of competence. We can see here that varieties of motivation are themselves a kind of affective signal.

Figure 6. Self-concepts on the spectrum of motivation.

SDT argues that the objective of motivation is the nourishment of basic psychological needs for autonomy, competence and relatedness. Autonomy is the sense of being volitional in one's life. Competence is the sense that one is skilful at behaviours that promote one's flourishing. And relatedness is the sense of being loved and cared for, especially by valued others, and of belonging. Experimental studies across multiple national and cultural contexts have evidenced the claim that nourishing basic psychological needs increases subjective well-being (SWB: a combination of positive affect and life satisfaction), vitality and self-esteem and ensures ease of motivation, as well as reducing depression, anxiety, compartmentalisation, defensiveness and personal rigidity (Sheldon et al., Reference Sheldon, Elliot, Ryan, Chirkov, Kim and Wu2004, Reference Sheldon, Abad, Ferguson, Gunz, Houser-Marko, Nichols and Lyubomirsky2009; Church et al., Reference Church, Katigbak, Locke, Zhang, Shen and de Jesus Vargas-Flores2013; Chen et al., Reference Chen, Vansteenkiste, Beyers, Boone, Deci, Van der Kaap-Deeder, Duriez, Lens, Matos, Mouratidis, Ryan, Sheldon, Shoenens, Van Petergem and Verstuyf2015). These indicators of need satisfaction, many of which are feelings, are analogous to experienced utility.

Introjected and identified behaviours can yield situations where the expected decision utility associated with a particular behaviour was larger than the actual experienced utility (Kahneman et al., Reference Kahneman, Wakker and Sarin1997). We might expect that dressing fashionably will have high utility because it will make us popular (an introjected motivation), but this turns out to be incorrect because our new friends are as shallow as the identity we express in our clothes. Similarly, we might think that classical music is enriching (an identified motivation) but we find it boring in practice.

Such discrepancies between expected decision utility and actual experienced utility in the SDT framework and their relationship to motivation, behaviour and well-being are a stark theme of studies of intrinsic and extrinsic pursuits. Empirically, goals associated with extrinsic and contingent rewards like money, fame and image provide relatively weaker payoffs than goals associated with intrinsic rewards like enjoyable experiences, personal growth and intimacy (Kasser & Ryan, Reference Kasser and Ryan1993, Reference Kasser and Ryan1996). This is because extrinsic pursuits do not nurture basic psychological needs (or, when they do so, then only to a smaller degree).Footnote ¹ Sheldon has expanded these insights into a more general theory of goal self-concordance (Sheldon & Elliot, Reference Sheldon and Elliot1999). He finds that people derive greater psychological well-being from goals that fit their personalities, or more daringly, their innate selves (Sheldon & Vansteenkiste, Reference Sheldon, Vansteenkiste, Strathman and Joireman2005). The ‘innate self’ implied here is very minimalist. It corresponds only to some elementary predispositions, aptitudes and biological parameters like height and intelligence. It is a close conceptual analogue of the ‘actual self’ at the onset of adulthood before self-actualisation starts to change the actual self as it harmonises with the ideal and ought selves. The innate self of SDT certainly has little relationship to the notion of ‘true’ preferences in behavioural welfare economics (Dold, Reference Dold2018).

Social feedback

Alongside affective feedback, agents can also receive social feedback from choice and behaviour. Indeed, the two varieties of feedback can be bound up together. Guilt, shame and self-esteem are often described as ‘social emotions’ that communicate to us when we are doing things that will be negatively perceived by our peers (Leary & Baumeister, Reference Leary and Baumeister2000; Haidt, Reference Haidt2012; Leary, Reference Leary, Van Lange, Kruglanski and Higgins2012). Strictly social feedback can come in a variety of forms including dis/approval from our peers, encouragement and advice. Social appraisals have been found to exert more influence on self-appraisal when the perceiver is considered by the perceived to be relevant to their self-concept, an in-group member, desirable, valued or otherwise important (Alicke et al., Reference Alicke, Guenther, Zell, Leary and Tangney2012; Wallace & Tice, Reference Wallace, Tice, Leary and Tangney2012). Neuroscience studies align with this result. They show that when an appraiser is from a group you care about, their appraisals of you will activate the self-assessment part of your brain. This is not the case when they are from a group you don't care about or a random stranger (Devos et al., Reference Devos, Hyunh, Banaji, Leary and Tangney2012, p. 158).

One of body of literature in the ‘social self’ space within psychological science that is especially relevant to self-actualisation is ‘self-verification’ theory (Swann, Reference Swann, Van Lange, Kruglanski and Tory-Higgins2011). It conjectures that people seek confirmation of their self-views from others. One way we come to know whether we have aligned with our ideal and ought selves is if our peers avow that we have. Empirical studies in self-verification theory find that self-views guide social interaction and, provided they are stable, make an individual's behaviour more predictable to others. This predictability stabilises the way others respond to the individual, which makes it easier to verify one's self-view through social interaction. Stable self-views thus encourage the emergence of a stable, coherent, social environment and vice versa, leading to a virtuous cycle wherein both self-concept and social environment become clearer and better fitted to each other. An important, empirically validated hypothesis that emerges from this theory is that people prefer social appraisals that align with their self-view even when these appraisals are negative (Swann & Buhrmester, Reference Swann, Buhrmester, Leary and Tangney2012). People move away from both incorrect and correct-but-negative appraisals over time towards groups that are both affirming and accurate in their social appraisals.

Social feedback is especially relevant to the ought self. Guilt and shame are morally coded emotions, and social approval of conduct is a central theme in many accounts of ethical behaviour, such as Adam Smith's Theory of Moral Sentiments. The introjected ‘responsibilities’ that make up much of the ought self are typically tied to social expectations. We feel a duty to conform to certain ethical precepts, and deviation from these precepts can result in anxiety owing to an actual–ought self-discrepancy. This is one way in which other-regarding preferences enter into self-actualisation, but such preferences need not be introjected (Besser-Jones, Reference Besser-Jones2014). Individuals can identify with certain ethical values and other-regarding preferences and pursue these authentically and autonomously as part of self-actualisation. This appears to be the case with converts to many ethical systems, both religious and secular, such as the growing membership of the effective altruism movement. Such individuals desire to be ‘good people’. Over time, such identified values can even become intrinsically motivated through the process of internalisation. Indeed, some people may even possess intrinsically motivated other-regarding preferences as part of their innate selves. Experiments in game theory, for example, have identified ‘supercooperators’ who seem to easily engage in generous behaviour (Novak & Highfield, Reference Novak and Highfield2011).

Agentic preferences

The idea that affective signals guide behaviour and shape preferences over time is central to both self-discrepancy theory and SDT. In the former, discrepancies between self-concepts give rise to negative affect, while coincidence between the actual and ideal selves is punctuated by positive affect. In SDT, intrinsically motivated activities are associated with positive emotions and the absence of psychopathology, in contrast to extrinsically motivated activities. These affective signals encourage the individual to detach from extrinsic pursuits and unsuitable conceptualisations of the ideal self over time and comport towards identified and especially intrinsically motivated activities that characterise a self-congruent ideal self. This is how agentic preferences form and come to be understood and acted on by an individual.

We describe the preferences that emerge from the self-actualisation process outlined above as ‘agentic’ to emphasise their relationship to valuation, self-expression and autonomy. We take our notion of agency from Sen (Reference Sen1985, pp. 203–204), who defines it as ‘the freedom to achieve whatever the person, as a responsible agent, decides he or she should achieve’; he continues in saying that ‘[a] person's agency aspect cannot be understood without taking note of his or her aims, objectives, allegiances, obligations, and – in a broad sense – the person's conception of the good’. Our analysis speaks to Sen's themes. Autonomy and the pursuit of intrinsic motivations and identified values speak to ‘freedom to achieve whatever the person … decides he or she should achieve’. The introspection and calibration that is fundamental to our notion of self-actualisation implies reflective endorsement of preferences, the proffering of reasons for those preferences and taking ‘responsibility’ for choices on the basis of those preferences. ‘Aims and objectives’ are preferences, ‘allegiances’ come in through the need for relatedness, and ‘obligations’ are present in the ought self. Harmony between the actual, ideal and ought selves requires alignment between the person's behaviour and their ‘conception of the good’. Note that we are not making any claims about what is required for ‘agency’ here. We are merely justifying our use of the term in a descriptive capacity.

Strategies for working with agentic preferences

First, policymakers could observe experienced utility in the form of affective feedback from choices made with and without nudges wherein cognitive biases are suspected. If a cognitive bias causes people to choose out of line with their agentic preferences, then they will feel bad after choosing. This is distinct from them feeling good or ambivalent after being nudged, as suggested by Paul and Sunstein (Reference Paul and Sunstein2019) in their ‘as judged by themselves’ reformulation of the welfare principle. Note, however, that negative feedback is only a necessary condition for introducing a nudge because it implies that the ‘wrong’ preferences were satisfied by a choice, but it is not sufficient. It suggests that a bias might be interfering with the satisfaction of an agentic preference, but it does not establish the existence of said bias.

Proving the existence of an interfering bias and the effectiveness of a nudge would require experimental conditions of choice without bias, choice with bias and choice with bias and nudge. In the absence of bias or nudge, individuals should choose their preference. The introduction of the cognitive bias, say by placing confectionary near the check-out, should then interfere with this preference satisfaction. Participants will report negative feedback from their choices. Implementing the nudge should then return choice to where it was without the bias or the nudge. Respondents will report no meaningful feedback because their expected utility from choosing in line with their preference aligns with their experienced utility. This approach is similar to how behavioural experiments in the nudge tradition have proceeded, except that agentic preferences are used here rather a neoclassical notion of ‘rational’ preferences. This is messier but more realistic – it takes psychology seriously.

A second strategy is pre-hoc assent. If you can demonstrate to people that they will likely fail to satisfy their agentic preference at the moment of choice due to a cognitive bias, many will assent to a nudge being implemented. For example, imagine an organisation like a gym whose members have agentic preferences for exercising. Now imagine that they are then shown compelling evidence from the cognitive biases literature that they are more likely to take the stairs if those stairs are painted like piano keys and make sounds when stepped on. We might reasonably expect the members to recognise this as a good outcome from the point of view of their agentic preferences for exercising and assent to the stairs being redesigned. This is a similar proposal to those involving strong ‘transparency’ within the libertarian paternalist paradigm. This principle requires policymakers to inform people about the nature and mechanism of a nudge before implementing it. In an experimental study, Paunov et al. (Reference Paunov, Wänke and Vogel2019) found that more stringent transparency can actually increase the uptake of nudges.

An important issue for policy applications of agentic preferences is that they may be robust to some forms of cognitive bias, but still susceptible to others. This can inform which nudges they are relevant to. For example, an individual committed to their agentic preferences may have the integrity to hold to them despite peer pressure and is thus less likely resistant to social desirability bias. However, this same conviction may make the individual more susceptible to confirmation bias and motivated reasoning when a desire to protect one's agentic preferences or self-image can distort the recruitment and evaluation of evidence (Nickerson, Reference Nickerson1998; Epley & Gilovich, Reference Epley and Gilovich2016). Additionally, the organismic drive to self-actualise (Ryan & Deci, Reference Ryan and Deci2017) has also evolved in the same context that produced the various biases of statistical reasoning catalogued by behavioural economists, such as conjunction fallacy, base rate fallacy, authority bias and framing effects (Kahneman, Reference Kahneman2011). Finally, self-actualisation is an iterative process that takes place over potentially medium to long-time horizons, which makes agentic preferences acutely sensitive to biases relating to poor prospecting, such as default bias and hyperbolic discounting. An important inference from these observations is that agentic preferences do not exhaust the grounds on which a nudge could be justified. Nudges to overcome default bias in pension plans, for example, are more straightforwardly justified by appeal to a traditional ‘rational’ account of preferences (Hausman, Reference Hausman2012).

Considerations for policy practice

Our conditional defence of nudging raises at least two practical issues for nudges in policy practice related to cost. Recall that one source of legitimacy for nudges is their cost-effectiveness. They must be demonstrated to ‘work’ through experimental empirical designs, and they must be cheap to implement. Our proposed logic undermines this cost-effectiveness in two ways. First, it requires policymakers to exert far more effort than they presently do to establish the existence of agentic preferences and some cognitive bias that is interfering with the satisfaction of those preferences. Second, it requires policymakers to conduct more surveys than they currently do, either to establish negative feedback as evidence of some interference from cognitive bias, or to establish pre-hoc assent to nudges going forward.

We do not see the first issue as especially problematic; quite the opposite. It is unethical to introduce nudges, at least under the auspices of libertarian paternalism, without first checking for agentic preferences and the existence of some cognitive bias. Doing so runs the risk of simply nudging behaviour to align with the preferences of policymakers, perhaps even by creating a bias rather than removing one. For example, as part of a partnership with local councils in the UK, environmental organisation Hubbub installed ‘ballot bins’ that allow smokers to vote in spot polls using their cigarette butts. This nudges them to bin their butts rather than littering them. A good outcome for the council. But what preference do the smokers have? To litter owing to convenience, presumably. And what cognitive bias is being addressed? It seems that Hubbub introduces a cognitive bias around competitiveness to steer the behaviour of smokers in the direction Hubbub prefers. We have chosen this example deliberately because the outcome seems so ‘good’. Smokers amuse themselves, councils have less cleaning to do, and streets are cleaner. Yet this policy seems to deliberately alter preferences and thus contravenes the welfare principle of libertarian paternalism. If a psychological intervention leverages cognitive bias rather than removing them then it is not a nudge, just paternalism, or even pure manipulation. It may be possible to justify such policies, but not under the auspices of libertarian paternalism; it would rather be a political justification of nudging (Guala & Mittone, Reference Guala and Mittone2015). Notably, litter is a classic case of a negative externality where rational behaviour on the part of the litterers leads to net social costs. Hicks-Kaldor efficiency, among other normative principles, can be used to justify interventions against litter, including manipulative ones like ballot bins.

The additional costs associated with surveying can be minimised through effective sampling. If a representative sample (at least enough to provide statistical power) of the target population assents to the introduction of a nudge, then it can be rolled out to the population at large. For example, imagine a large university conducts a trial involving a representative sample of student volunteers that establishes agentic preferences for moderating caloric intake among a supermajority of those students. The study further establishes that cognitive biases emerging out of large plates in the cafeteria are interfering with the satisfaction of these agentic preferences. The university could now publicise the results of the trial to the entire student population and ask them, as part of a poll, whether they would like cafeteria designs nudged to make healthy eating choices easier. While far from free, these are low-cost methods for establishing legitimacy. In some cases, it may be straightforward to survey the entire population, as in the case of small and medium enterprises considering the implementation of nudges for their staff. Such spending seems appropriate when choice architects are trying to justify paternalistic manipulation. Indeed, Schmidt (Reference Schmidt2017) has previously argued that such quasi-democratic accountability upon nudgers is appropriate and could actually legitimise the more widespread use of nudges. We make no comment here on what sample size and representativeness in a survey/vote would be required to establish assent, though a simple majority strikes us as too low a bar to implement libertarian paternalistic interventions.