2. Welfare with inconsistent and variable preferences

Until quite recently, welfare economics as the study of how to conceptualize, justify, model, and operationalize normative standards and policy goals had almost vanished from the agenda of mainstream economics (Atkinson, Reference Atkinson2009; Sen, Reference Sen1987). As the insights of behavioral and evolutionary economics into the complexity of human behavior – giving rise to ‘non-standard models of choice’ – are now increasingly acknowledged in the profession, this has changed dramatically. For many of these new insights raise important policy problems: Is it legitimate to contain status races by taxing certain goods? Is it acceptable to steer people toward certain (e.g., thrifty or environment-friendly) behaviors? Should policy promote public information campaigns, even if they cannot be strictly neutral with respect to people's preferences? All these are contentious issues. They cannot be answered, however, without a thorough inquiry into the underlying concept of welfare. This is a non-trivial task: when individual preferences tend to be inconsistent and subject to endogenous change, traditional preference-based notions of welfare can no longer be applied (Bernheim, Reference Bernheim2009; Sugden, Reference Sugden2004). An agent subject to biased social learning processes does not necessarily choose options that enhance her own well-being. Economists have to think anew about how to define and conceptualize welfare.

This is maybe most apparent in the ongoing debate about the normative foundations of approaches advocating ‘libertarian’, ‘light’, ‘soft’, ‘benign’, or ‘asymmetric’ (henceforth LLSBA) versions of paternalism (see Loewenstein and Haisley, Reference Loewenstein, Haisley, Caplin and Schotter2008, for a survey). According to Thaler and Sunstein (Reference Thaler and Sunstein2003), a policy counts as paternalistic ‘if it is selected with the goal of influencing the choices of affected parties in a way that will make those parties better off’ (ibid.: 175).Footnote ⁸ But what does ‘better off’ mean, i.e., to what notion of well-being or welfare does it refer to? The same problem arises in the context of applying the concept of ‘asymmetric paternalism’ proposed by Camerer et al. (Reference Camerer, Issacharoff, Loewenstein, O'Donoghue and Rabin2003): they require that the net benefit of any paternalistic intervention (e.g., a default rule stimulating increased saving) accruing to irrational consumers should outweigh the aggregate costs to rational consumers (ibid.: 1212, 1219–20). As Loewenstein and Haisley (Reference Loewenstein, Haisley, Caplin and Schotter2008: 12) rightly note, ‘[t]o evaluate costs and benefits . . . once again requires some concept of welfare’.

3. A model of biased cultural learning of consumption norms

Formal models of cultural evolution analyze how cognitive processes of human agents combine with patterns of social interaction at the population level to generate the distributions and dynamics of cultural variants, e.g., the preferences underlying different consumption behaviors (Boyd and Richerson, Reference Boyd and Richerson1985; Cavalli-Sforza and Feldman, Reference Cavalli-Sforza and Feldman1981; Henrich and Boyd, Reference Henrich and Boyd2002; Richerson and Boyd, Reference Richerson and Boyd2005).Footnote ²⁰ We draw on this approach to account for preference learning dynamics in a population of heterogeneous consumers. As shown, we are then able to distinguish different preference learning trajectories and their corresponding welfare implications.

Cultural transmission or learning is biased; people tend to acquire some cultural variants rather than others. We assume two biases here: a role model bias and a content-related, direct bias. First, the choice of a cultural trait can be based on the observable attributes of the individuals who exhibit the trait (Richerson and Boyd, Reference Richerson and Boyd2005: 69). Evidence from social psychology and anthropology suggests that human agents are prone to adopt cultural traits that are shown by role models in their social environment (e.g., Henrich and Gil-White, Reference Henrich and Gil-White2001; Labov, Reference Labov2001). Such a model-based bias includes a predisposition to imitate successful or prestigious individuals. In general, such an indirect bias results if social learners use the value of a secondary character that characterizes a model (e.g., her observable consumption level) to determine the attractiveness of that individual as a model for the primary character (e.g., the preference concerning the level of consumption one should aspire). Hence, one trait of a role model is taken as an indicator whether it is worthwhile to copy another aspect of this individual's behavior or attitude.

Second, individuals are more likely to adopt some cultural variants rather than others based on their content (Boyd and Richerson, Reference Boyd and Richerson1985: 135; Richerson and Boyd, Reference Richerson and Boyd2005: 69). Such a content-based or direct bias can result from cognitive structures that cause people to preferentially adopt some cultural behavior rather than others. Examples in our context are Veblen's (Reference Veblen1898) instinctive preference for status-signaling goods, Ng's (Reference Ng2003) bias in favor of materialism, or Frank's (Reference Frank2008) concern about relative position. In general, a cultural transmission rule is characterized by direct bias if one behavioral variant, e.g., status-signaling consumption behavior, is more attractive than others. While the general strive to signal status is partly innate and partly learned, the kind of commodity suitable to do so depends to a great degree on the cultural environment. A directly biased transmission creates a force that increases the frequency of the culturally transmitted variant that is favored by the bias.

To formally depict some normatively relevant facets of cultural evolution, we draw on a model based on a set of recursion equations that was first applied to quantitative genetics (see Fisher, Reference Fisher1930; Lande, Reference Lande1981; Kirkpatrick, Reference Kirkpatrick1982) and adapt it to our purposes and the specificities of cultural dynamics (for accessible introductions to models of social evolution of this type, see Boyd and Richerson, Reference Boyd and Richerson1985: chap. 8; Cavalli-Sforza and Feldman, Reference Cavalli-Sforza and Feldman1981; McElreath and Boyd, Reference McElreath and Boyd2007: chap. 8). We begin with the effects of biased cultural learning on an individual's preference and consumption levels. In a next step, we will then show how these learning dynamics translate into consequences on the population level. We consider an individual's preference level to be cardinally measurable. In principle, a shift to an ordinal scale should not alter the model's general mathematical properties and thus the argument developed here.

Let x ₀ be the measure of a consumer's initial consumption level and p ₀ the measure of this individual's preferred level of consumption. Moreover, we assume that each individual is exposed to n role models’ consumption behaviors (x ₁, . . ., x_i, . . ., x_n). To allow the relevance of different models (e.g., individuals in different social positions) to differ, we assign different basic weights, denoted by α_i, to them, whereby ∑_i α_i } = 1. Moreover, cultural transmission of consumption behaviors is subject to a direct bias, i.e., individuals choose role models by comparing their notions of an appropriate level of consumption with the observed consumption behaviors of models in the population. We assume that, in the beginning, an individual's preference concerning consumption, p ₀, exceeds her initial consumption level given by x ₀, i.e., the direct bias favors higher consumption levels. This reflects a preference for, e.g., status-signaling goods and a general human concern about status or relative position in a group. This implies that another part of the influence of the ith role model, i.e., beside her basic weight α_i, depends on her revealed consumption level, x_i, and the preference level of the observing individual, p ₀, captured by the function β(x_i, p ₀). In this context, the preference level affects which consumption levels or behaviors an individual finds attractive. The observed consumption level, x_i, that maximizes β(•) is the most attractive one. Role models that exhibit this value will, on average, have the greatest influence in the transmission of consumption behaviors. The farther x_i is away from p ₀, the lower the weight of this role model. Below, we will specify a Gaussian form for β(•) and clarify this idea.

In total, the weight of the ith model in the transmission of consumption behaviors is given by her basic weight, α_i, and the direct bias function, β(•). The recursion equation that determines the individual's level of consumption after cultural transmission, x′₀, is

(1)\begin{equation} x'_0 = \frac{{\sum\limits_i^n {x_i \alpha _i \beta (x_i ,p_0 )} }}{{\sum\limits_i^n {\alpha _i \beta (x_i ,p_0 )} }}.\end{equation}

To represent the evolution of an individual's preference level, p ₀, which is subject to our second, indirect bias, we again assume that the importance of the ith model in cultural transmission depends on a basic weight, α_i, modified by an indirect bias function, θ(x_i, p ₀). As in the case of the direct bias, the latter part of the weight of the ith model is a function of her consumption behavior, captured by x_i, and the observing individual's preference level, p ₀. p_i denotes the n models’ preference levels concerning appropriate levels of consumption (p ₁, . . ., p_i, . . ., p_n). Then, the individual's preference level after cultural transmission is given by

(2)\begin{equation} p'_0 = \frac{{\sum\limits_i^n {p_i \alpha _i \theta (x_i ,p_0 )} }}{{\sum\limits_i^n {\alpha _i \theta (x_i ,p_0 )} }}.\end{equation}

Here, social learners are indirectly biased: they use the value of a secondary character that characterizes a model – the observed consumption behavior – to determine the attractiveness of that individual as a model for the primary character – the preference concerning a certain consumption level or behavior. Therefore, in determining their own preference level, consumers acquire beliefs from their influential role models about who should be imitated, i.e., about which consumption level one should strive for. In addition, the most influential agents – those characterized by a level of consumption close to the preference level of the individual in question – will prefer higher consumption levels than the population as a whole. We assume an individual's level of consumption to be positively correlated with her preference level. Via cultural transmission, agents adopt from their set of role models a higher preference level than they had before entailing a higher aspired consumption level. Furthermore, two individuals exposed to the same set of cultural models will, on average, adopt different cultural variants of traits affected by the role model bias.

Since we aim at understanding the interdependence between individual preference and consumption levels in a group of interacting agents, we take a population-based perspective on the evolution of these two variables. After having defined the recursions that describe how individuals choose their consumption level and how they attain their preferences for a certain consumption level via cultural transmission, we can capture the evolution of the mean values of x_i, , and p_i, , in a population consisting of individuals behaving in the described way. To do so, we look at the changes in these mean values given by the following expressions:

(3)\begin{equation} \Delta \bar x = \bar x' - \bar x = \frac{{\sum\limits_i^n {x_i \alpha _i \beta ( \bullet ) - \bar x\sum\limits_i^n {\alpha _i \beta ( \bullet )} } }}{{\sum\limits_i^n {\alpha _i \beta ( \bullet )} }}\end{equation}

(4)\begin{equation} \Delta \bar p = \bar p' - \bar p = \frac{{\sum\limits_i^n {p_i \alpha _i \theta ( \bullet ) - \bar p\sum\limits_i^n {\alpha _i \theta ( \bullet )} } }}{{\sum\limits_i^n {\alpha _i \theta ( \bullet )} }}.\end{equation}

As is shown in Appendix A (see also Price, Reference Price1970), equations (3) and (4) can be reformulated to

(5)\begin{equation} \Delta \bar x = \frac{{{\mathop{\rm cov}} (x,\beta (x,p))}}{{\bar \beta }}\end{equation}

(6)\begin{equation} \Delta \bar p = \frac{{{\mathop{\rm cov}} (p,\theta (x,p))}}{{\bar \theta }}.\end{equation}

For values of x and p close to and , the bias functions β(•) and θ(•) can be approximated by using a Taylor series approximation (see Appendix B). We then obtain for the changes in the mean values of x and p

(7)\begin{equation} \Delta \bar x \approx {\mathop{\rm var}} (x)\left. {\frac{{\partial \ln \beta ( \bullet )}}{{\partial x}}} \right|_{\bar x,\bar p} + {\mathop{\rm cov}} (x,p)\left. {\frac{{\partial \ln \beta ( \bullet )}}{{\partial p}}} \right|_{\bar x,\bar p}\end{equation}

(8)\begin{equation} \Delta \bar p \approx {\mathop{\rm cov}} (x,p)\left. {\frac{{\partial \ln \theta ( \bullet )}}{{\partial x}}} \right|_{\bar x,\bar p} + {\mathop{\rm var}} (p)\left. {\frac{{\partial \ln \theta ( \bullet )}}{{\partial p}}} \right|_{\bar x,\bar p} .\end{equation}

In this context, cov(x,p) indicates the covariance between a certain level of consumption, measured by x, and the preference level, p, that determines the consumption level perceived as appropriate by a consumer. This is the assumed positive correlation between an individual's consumption level x and her notions about an ‘adequate’ level of consumption, p. The other term, beginning with var(x) or var(p), gives the change due to the direct effect of variation in that variable; for example, how a change in an individual's consumption level affects her influence as a role model. In order to facilitate an evaluation of this two-dimensional system, we need to define particular functional forms for the direct and indirect bias expressions. Plausible forms of β(•) and θ(•) are Gaussian bias functions

(9)\begin{equation} \beta (x,p) = \exp \left\{ { - a(x - p)^2 } \right\}\end{equation}

(10)\begin{equation} \theta (x,p) = \exp \left\{ { - b(x - p)^2 } \right\}\end{equation}

(11)\begin{equation} \Delta \bar x = G_x \left\{ { - 2a\bar x + 2a\bar p} \right\} - B_{xp} \left\{ { - 2a\bar p + 2a\bar x} \right\}\end{equation}

(12)\begin{equation} \Delta \bar p = B_{xp} \left\{ { - 2b\bar x + 2b\bar p} \right\} - G_p \left\{ { - 2b\bar p + 2b\bar x} \right\}.\end{equation}

Next, we solve for and denoting the equilibrium values of the consumption and the preference levels in a population of interacting consumers. It is easy to see that if = , both recursion equations equal zero. Therefore, = is an equilibrium of this two-dimensional dynamic system. Given a graph with on the horizontal axis and on the vertical axis, = is a line through the origin with slope one and any point on this line is an equilibrium. To further analyze the model's dynamic properties, it is helpful to compute its trajectory in such a diagram:

(13)\begin{equation} \frac{{\Delta \bar p}}{{\Delta \bar x}} = \frac{{B_{xp} ( - 2b\bar x + 2b\bar p) - G_p ( - 2b\bar p + 2b\bar x)}}{{G_x ( - 2a\bar x + 2a\bar p) - B_{xp} ( - 2a\bar p + 2a\bar x)}} \\ = \frac{{b(B_{xp} + G_p )}}{{a(B_{xp} + G_x )}}. \\ \end{equation}

If changes one unit as a result of biased cultural transmission, the preference level, , changes $\frac{{b(B_{xp} + G_p )}}{{a(B_{xp} + G_x )}}$ units as a consequence of the correlated effects.

Figure 1 shows two-phase plots of the model's evolutionary trajectories that visualize its dynamics. In both cases, the dashed 45-degree lines through the origins with slope one are the lines of equilibrium, where = . Both, and increase above this line of equilibrium and decrease below it. The other lines in Figures 1(a) and (b) give the trajectories of populations of social learners that begin at different initial points and have slope $\frac{{b(B_{xp} + G_p )}}{{a(B_{xp} + G_x )}}$. Given the case depicted in Figure 1 (a), the slope of the trajectory is less than the slope of the line of equilibrium and the system is stable approaching an equilibrium on the 45-degree line, as is indicated by the arrows. On the other hand, in case Figure 1 (b), if the trajectory's slope is greater, the population evolves toward higher and higher preference levels and respective consumption levels when starting from a point above the line of equilibrium.

Figure 1.

The phase plots of the model's evolutionary learning trajectories for (a) a > b and (b) a < b (B_xp = 3, G_x = 1, G_p = 1).

Therefore, if a(B_xp+G_x)>b(B_xp+G_p), i.e., if the direct bias, whose strength is measured by the parameter a, is larger than the indirect bias effect, measured by b, then the learning dynamic eventually comes to rest at some point along the line of equilibrium. This situation is depicted in Figure 1(a). On the other hand, if the indirect bias effect of role models is greater, i.e., if b(B_xp+G_p)>a(B_xp+G_x), then the mean of the preference level in the population, , is increasing faster than the mean of the consumption level, measured by , and both traits ‘run away’; the system is unstable (case (b) in Figure 1 for initial values lying above the line of equilibrium). As a consequence, the distance of both mean values from the line of equilibrium is increasing from one learning step to the next ( - > - ). Certainly, this process cannot continue forever: some economic or ecological constraints will eventually restrain this dynamic. Some factors not accounted for in the model will eventually limit the evolution of the cultural traits in the population. To conclude, a situation with > (inducing a rise in ) combined with the role model bias function, β(•), that continuously updates the preference levels upward, makes social learners adopt ever more accentuated preferences relative to an ‘appropriate’ level of consumption. This results in a self-augmenting ‘treadmill’ of consumption choices.

Figure 2 offers another way to visualize the system's dynamic properties. It displays the development of a representative individual's preference level, p ₀, and consumption level, x ₀, by iterating the two-dimensional system consisting of equations (11) and (12) for many social learning steps. As can be seen in Figure 2(a), the role model bias in social learning gives rise to a widening gap between an individual's preference and consumption levels. This might also be interpreted as a growing discontent of an agent with her material condition. Figure 2(b), in contrast, assumes that no role model bias continuously updates a consumer's preference level. In this case, an individual's level of consumption finally reaches her preference level. The closing of this gap might entail a lasting satisfaction of this consumer's preferences.

Figure 2.

(a) The development of an individual's preference and consumption levels when indirect bias is greater than direct bias. (b) The development of an individual's preference and consumption levels when there is no indirect bias.

4. Toward a procedural concept of well-being

This section presents an account of well-being that builds upon the insights on processes of cultural learning and preference formation discussed above. The model of cultural transmission informs us about the conditions under which the dynamic interplay of two learning biases can generate a treadmill of self-augmenting consumption cycles. This allows us to draw a line between those learning paths that are in dynamic equilibrium and those paths that depart from this equilibrium in an accelerating way, leading the agent into an extreme state of permanent dissatisfaction where, ultimately, exogenous (economic or ecological) constraints become binding and the latent learning dynamics cannot unfold anymore.

We suggest to use this model for normative purposes in the following way: processes of preference learning can either ‘work’ or ‘fail to work’ effectively. In the former case, the representative agent's gradual updating of her preference level (recall that this is a cardinal variable) settles down in a dynamic equilibrium that allows her to satisfy her preferences by matching her preference path with a corresponding consumption path (again, note that her consumption level, at any point in time, is represented by some cardinal variable). This is the process depicted in Figure 1(a) and it leads us to the following proposition.

Proposition 1: If the preference level of an individual consumer were unaffected by indirectly biased social learning processes, following from equation (11), would be at equilibrium at = _unbiased. Thus, if < _unbiased the mean value of a population's consumption level would increase; if it is greater than _unbiased, it would decrease.

In the latter case, the agent continuously modifies her preference levels in such a way that she systematically fails to successfully satisfy her preference level (and, hence, fails to increase her well-being). If we interpret preferences as instrumental tools that should ideally enhance the agent's well-being, we may say that the agent, driven by the interplay of our two biases, ends up systematically acquiring dysfunctional tools (as shown in Figure 2(a)). This phenomenon is captured by our second proposition derived from the model:

Proposition 2: If the consumers’ mean consumption level, measured by , reaches their mean preference level, , they temporarily meet their corresponding needs. This state is, however, of a transitory nature only if is continuously modified by processes of biased social learning via cultural role models. The individuals then feel deprived again and strive to adapt their consumption level to this new preference level determining the new level of consumption perceived as appropriate. However, the additional resources spent on reaching the new consumption level still only lead to satisfying one's (modified) consumption preference as it did on lower levels of consumption.

Our model of cultural transmission then allows us to operationalize Elster's intuitive idea that in a world of preference endogeneity we should inquire into the historical genesis of the agents’ preferences (see Section 2, above) in order to be able to distinguish between desirable and undesirable ways of preference learning. In Elster's original paper, he mentions ‘adaptive preference formation’ and the ‘deliberative manipulation of wants by other people’ as cases of undesirable preference change (Reference Elster, Sen and Williams1982: 223, 226–230). In his account, the label ‘undesirable’ refers to the presumed lack of ‘autonomy’ in the acquisition of preferences. In contrast, we focus on an empirically particularly relevant case of social conditioning that involves our specific two biases. The model can be used to track the genesis of an agent's preferences by identifying the way these biases interact in driving the agent's process of preference formation. Based on the empirically observed differential impact of these two biases, it is then possible to suggest a normative distinction between two kinds of preference dynamics.

This distinction can be made normatively relevant by suggesting the following notion of individual well-being: rather than being defined as the satisfaction of ‘given’ and perfectly consistent preferences, in our setting well-being should be understood as the individual ability to engage in the effective learning of new preferences over time, with the ‘effective’ proviso reflecting the absence of systematical frustration on the part of the agent. ‘Frustration’ is to be understood in terms of our model of cultural learning, as the systematic non-satisfaction of preferences, due to the ongoing process of preference updating, fuelled by the influence of social role models, in combination with a consumption level that does not grow at the same pace. Consequently, although the individual's level of consumption rises, her preferences remain unsatisfied, since the agent continues to be frustrated by what she is able to attain. In this perspective, the effective learning of new preferences constitutes well-being in a partly procedural sense. It is partly inherently valuable, i.e., independent of the outcomes in terms of preference satisfaction, and its value is independent of the logical consistency of an agent's vector of preferences at any given point in time. Endogenous preferences turn out to be naturally inconsistent, so in our view it does not make sense to treat inconsistency as an argument for discounting the value of an agent's preferences.Footnote ²²

As mentioned above, an alternative way to clarify the contours of our criterion is to frame it in terms of the metapreference approach pioneered by Frankfurt (Reference Frankfurt1971) and elaborated by George (Reference George2004, Reference George2006), among others. We generalize that approach by considering the process that either leads toward an alignment of first- and second-order preferences or away from it. By explicitly modeling the process (or at least one component of it), we are also able to go beyond the assumption that preference change just ‘happens’. Preferences of course often shift in arbitrary and unpredictable ways. But there are also cases where agents develop their preferences further, acquiring valuable knowledge in the process, which tends to make this process irreversible.Footnote ²³

In light of our notion of well-being, status-oriented consumption qualifies as a policy problem to the extent that it generates learning dynamics that ultimately (and, perhaps, paradoxically) jeopardize the individuals’ ability to ‘try out’ new preferences in an ongoing, sustainable way. At the level of instrumental policy-making, it follows that policy should at least not restrict, but preferably foster the ability of individuals to engage in the effective learning of new preferences. In order to achieve this, it seems that two general conditions have to be met: first, agents should obviously be left free to acquire new preferences as they see fit. They should, then, be endowed with the maximum set of opportunities that is compatible with everyone else's opportunity set. Note that this first condition is identical to Sugden's criterion of well-being as ‘opportunity’ (see above). In our framework, it is however not sufficient: our model of cultural transmission predicts that preference formation may be dynamically unstable under certain circumstances. These circumstances are targeted by our second condition: policy should design the institutional framework of markets in such a way that the tendency of preference dynamics to ‘collapse’ by inducing self-augmenting cycles will at least not be promoted, but preferably reduced. Prima facie, this may be achieved by a variety of tools that either involve ‘hard’ paternalism (involving taxation of status goods, say) or milder LLSBA interventions that aim directly at influencing consumers’ preferences.

How can our notion of well-being be justified? We propose to follow Sugden (Reference Sugden2008) in adopting a contractarian perspective that frames the issue of evaluating alternative policy rules in terms of a mutual ‘constitutional’ agreement between the individuals that will be affected by the rules. In order to operationalize this approach, assume that all those individuals are put behind a Rawlsian ‘veil of ignorance’ that deprives them of all information about the preferences they themselves will adopt in the future ‘market game’, where they will be subject to processes of cultural transmission.Footnote ²⁴ Assume furthermore that they do know about the characteristics of the model of cultural learning introduced in Section 3, above, i.e., the agents know about the conditions that may undermine the ongoing learning of new preferences and ‘trap’ them in self-defeating cycles of status consumption. Given these assumptions, we claim that those individuals may agree upon assessing alternative policy rules in light of our notion of well-being. This claim would be refuted if individuals would, through a process of public deliberation, reject the idea to guide and structure policy with the help of such a notion.Footnote ²⁵

This claim is backed by the following proposition: Our notion of well-being satisfies all the criteria that have been posited in Section 2, above. First, it is not detached from what real-world individuals are able to care about and effectively do care about. Being an essential contributing factor to an individual's identity and sense of self, preference learning (and the ability to pursue it) can be argued to be even one of the most important things in life people care about.Footnote ²⁶ Hence, we argue that our concept of well-being satisfies the ‘internalism’ Criterion 1.

Second, our notion of well-being allows for individual preferences to be inconsistent and departing from what is in the agent's ‘best interests’. In our context, this means that under certain circumstances the agent may willingly acquire preferences that turn out to be self-defeating from an ex post perspective. The possible dissociation between preference, choice, and well-being is built into our approach to reasoning about well-being. That is why Criterion 2 is satisfied.

Third, our notion of well-being is explicitly tailored to a world where preferences are not ‘given’ once and for all, but change endogenously. Specifically, they are subject to ‘natural’ processes of social conditioning, which in our model come in the shape of two kinds of biases and their dynamic interplay. In this context, the main advantage of our concept of well-being lies in its procedural nature that reflects the procedural account of preferences supported by our positive background model. Hence, Criterion 3 is satisfied.

Fourth, by virtue of requiring to give agents maximum opportunities to engage in preference learning (subject only to freedom-preserving LLSBA interventions to prevent ‘traps’ in the dynamics of preference learning, see above), our notion of well-being includes considerations related to the instrumental value of individual freedom and opportunity. This allows us to conclude that Criterion 4 is satisfied as well.

Let us also consider an objection that may easily be raised: Doesn't the freedom to learn any new kind of preferences as one sees fit imply that people are free to acquire ‘perverse’ or anti-social (racist or sadist) preferences? To this, we would reply that, first, what counts as ‘perverse’ in a given historical or cultural context is far too malleable for policy-makers to care about (consider the example of homosexuality). Second, in order to try out and learn anti-social preferences, agents will typically have to engage in behavior that is positively harmful and will, then, be legitimately banned anyway.

Next, we return to the problem of status-related consumption. At the level of applied policy-making, our procedural notion of well-being (with its positive background model of cultural learning) may be used to assess the desirability of the usually proposed sets of policy tools to shape people's consumption of status goods. Consider ‘hard’ paternalism. Contrary to a widely shared view in the literature on conspicuous consumption and ‘positional externalities’ (Frank, Reference Frank2008), our model shows that the taxation of status goods is ineffective in reducing the consumption of status-signaling goods and the accompanying participation in ‘status races’. Recall the key relationship between and (Figure 1). If < , a tax imposed on the consumption of the good in question would merely increase the difference between and without removing the consumers’ general motivation to meet their preference level by increasing their personal consumption levels and, thus, finally . The role model-based acquisition of modified preference levels via social learning would remain unaffected. In the case of taxation, ‘status races’ would still take place by means now affordable within the boundaries of the new economic restrictions. Hence, this kind of consumption dynamic may occur on the basis of rather different classes of commodities depending on the available income. The hard-wired motivational factors driving status-seeking would remain unaffected.

Hence, interventions that target processes of preference and social norm formation appear to be the best (and probably only) tools available to counter self-defeating preference dynamics. We will briefly consider two kinds of policies that shape preference formation (the proper realm of LLSBA paternalism), two kinds of policies that shape social norm formation by targeting the ‘social meaning’ of activities (Lessig, Reference Lessig1995), and finally the design of institutional frameworks that satisfy the individuals’ need for self-esteem by providing procedural utility (Frey, Benz, and Stutzer, Reference Frey, Benz and Stutzer2004; Schubert, Reference Schubert2012).

As to the first set of policies, we suggest two kinds of interventions: First, policy may influence agents’ preferences by providing information about the characteristics and ‘side-effects’ of consumer goods that will not be provided in the marketplace. This information provision may proceed on more or less neutral terms, i.e., involve the ‘coloring’ of information by using framing effects or other tricks from the behavioral economics toolbox. Our criterion of well-being would endorse the non-neutral provision of information as long as agents are not made subject to manipulation ‘behind their back’, i.e., as long as they can in principle (i.e., if they explicitly wish to do so) realize the degree of non-neutrality, understand both ends and means of public nudging, and act against the policy-maker's goals.Footnote ²⁷

Second, given that, (1), an important precondition of status consumption and the ensuing status races is the individuals’ ability to incur debt and that, (2), widespread overborrowing is a key effect of people's participation in status races, we argue that LLSBA instruments should be used to discourage people from behavior that leads to excessive indebtedness. This may include the operation of ‘counternudges’ to make nudges ineffective that are used by banks to stimulate the incurring of debt. There is also ample empirical evidence that the manipulation of default options can effectively influence people's attitudes toward saving and dissaving (Thaler and Sunstein, Reference Thaler and Sunstein2008).

As to the second set of policies, targeting processes of social norm formation, we again suggest two specific interventions. They aim at influencing the ‘social meaning’ of activities, which Lessig (Reference Lessig1995: 951) defines as the locally uncontested ‘semiotic content attached to various actions, or inactions, or statuses, within a particular context’. They are derived from socially shared understandings or expectations concerning the appropriate course of action in specific situations. To illustrate, consider ‘what it means’ to be black in the American South in the 1960s, to have a whisky in the office in the early 1960s, to have the same whisky in the office in today's business world, or to smoke as a woman in the early 20th century. Within limits, policy may use the conservation or manipulation of social meanings to achieve certain aims, such as the aim to discourage status-oriented consumption. First, it may regulate advertisement by banning the direct appeal to people's role-model bias in the media. Second, it may exploit the very same role model bias by promoting public campaigns that associate prestigious (prominent and glamorous) role models with an attitude that is immune against the lure of status good consumption (by appealing to people's ‘autonomy’ or ‘self-determination’, say). Third, policy may try to redirect the agents’ status consumption toward goods and services that have less harmful side-effects, in particular in terms of the agents’ resources. For example, status-seeking may be redirected toward activities that are also typically intrinsically valued, such as charitable giving.Footnote ²⁸ While this strategy does not remove the basic motivation to engage in status-related consumption, it may at least curb or ‘slow down’ the concomitant preference dynamics by encouraging activities that have been shown to have an independent and lasting impact on the agents’ hedonic well-being (Frey and Stutzer, Reference Frey and Stutzer2002).

Finally, a third set of policies could be implemented to confront directly one of the key motivational drivers of status good consumption, namely people's needs for self-esteem, respect, and a positive self-image (e.g., Johansson-Stenman and Martinsson, Reference Johansson-Stenman and Martinsson2006). Happiness research tells us that the very same set of needs can also be satisfied by organizing human relationships within the political arena or at the workplace in such a way that people gain ‘procedural utility’ (Dolan et al., Reference Dolan, Edlin, Tsuchiya and Wailoo2007; Frey, Benz, and Stutzer, Reference Frey, Benz and Stutzer2004). This may be a supplementary step toward reducing individuals’ desire to satisfy these needs through expensive status good consumption, hence to decrease the demand for status good consumption, the engagement in ‘wasteful’ status races, and the risk of decreasing individuals’ well-being in terms of their ability to self-improve by learning new preferences.

5. Concluding remarks

While behavioral economics provides ample evidence for the fact that in many economically relevant situations real-world individuals may systematically fail to choose what is in their best interests (as perceived by themselves), little progress has been made to derive convincing policy implications from this. We have argued that this is due to the fact that we cannot formulate policy advice without a concept of well-being or welfare, and that the policy-minded literature in the field still lacks a concept of well-being that could be consistently applied in a ‘behavioral’ setting, i.e., in a setting where individual preferences change and tend to be inconsistent.

Our contribution to the literature has been twofold. First, we have tried to facilitate the quest for a ‘behaviorally adequate’ non-standard concept of well-being by introducing a catalog of four conditions that any such concept ought to satisfy. We have used this catalog to gauge the usefulness of those (three) non-standard notions of well-being that have so far been proposed in the literature. In our view, this exercise already sheds light on the contours of a practically applicable concept (for instance, it tells us that such a concept will certainly be of a procedural nature). Second, following the intuition that additional positive insights into the process of individual preference formation may shed further light on this issue, we have introduced a formal model of preference learning. The model predicts that the interplay of two specific biases can generate self-augmenting learning dynamics that lead agents away from sustainable and welfare-enhancing consumption paths. This prediction allowed us to make a distinction between ‘desirable’ and ‘undesirable’ paths of preference learning. This distinction is, thus, not axiomatic, but grounded in empirical evidence, and it can serve as the basis of a procedural notion of well-being. According to this view, well-being is seen as residing in the individual's ability to engage in the ongoing, not systematically frustrated ability to acquire new preferences. Hence, status races call for policy intervention to the extent that they risk decreasing well-being, thus defined.

We hasten to add an important caveat: this policy intervention should not be organized in a top-down way, by some omniscient social planner. Rather, we have suggested to frame the discussion in terms of the contractarian paradigm: our concept of well-being should be used to inform the individuals themselves (as citizens) about possible ways to think about the problem of status races and their impact on well-being. On the basis of this information, the individuals may agree upon a social contract that stipulates a set of rules helping them avoid the welfare losses associated with status races, while essentially maintaining their personal freedom of choice. Think of it as the analog of a set of traffic rules: they do not prescribe any particular activity (they do not tell you where to go), but they channel interdependent – and potentially harmful – individual behavior in such a way that harmful accidents are avoided as far as possible.