4.1 View A

Recall that we are holding fixed an EU account of rationality. Combining the EU account of rationality with View A of well-being, we have that individual i’s rational self-interested preferences over a given set of choices, under favorable deliberative conditions, can be represented as maximizing the expected value of some utility function $u_{i}(\cdot )$, taking outcomes as arguments. $u_{i}(\cdot )$ is such that: if individual i is EU-compliant and under favorable deliberative conditions self-interestedly prefers outcome x to outcome y, then $u_{i}(x)>u_{i}(y)$. If individual i is EU-compliant and under favorable deliberative conditions self-interestedly prefers action a to action b, then $\sum _{x}{\it\pi}_{a}(x)u_{i}(x)>\sum _{x}{\it\pi}_{b}(x)u_{i}(x)$, with ${\it\pi}(\cdot )$ some probability measure and ${\it\pi}_{a}(x)$ the probability of x given action a.

For short, let us refer to EU-compliant preferences formed under favorable deliberative conditions as “idealized” preferences. Thus, $u_{i}(\cdot )$ is individual i’s idealized self-interested preference-utility function.

For convenience (as is quite common in neoclassical economics), we will assume that $u_{i}(\cdot )$ is temporally separable, so that $u_{i}(x)=\sum _{t=0}^{T\max }D_{t}v_{i}(x_{t})$, with $x_{t}$ the facts about outcome x at time t. $v_{i}(\cdot )$ is individual i’s momentary idealized self-interested preference-utility function. $D_{t}$ is the discount factor for time t, either decreasing with time, or constant (no discounting), typically assumed to take the form $D_{t}=\frac{1}{(1+r)^{t}}$, $r\geqslant 0$. The $v_{i}$ values of time slices, summed (after discounting) over time, add up to the overall utility values $u_{i}$ which in turn represent the self-interested preferences over outcomes that individual i would have, were she to deliberate rationally under favorable conditions.

On View A of well-being, an individual can have “self-interested” preferences for features of the world other than her own mental states (Adler, Reference Adler2012, Chapter 3; Adler, Reference Adler2013, Reference Adler2014). As already mentioned, how to define a “self-interested” preference is a normatively contested topic. While the example of self-sacrifice (see above) suggests that we do need a non-trivial “self-interest” screen – that we cannot simply equate someone’s well-being with her all-things-considered idealized preferences – Robert Nozick’s famous “experience machine” hypothetical suggests that we may find it normatively unappealing to define someone’s “self-interested” preferences as preferences regarding what occurs “in her head.” Nozick writes:

The lesson of Nozick’s hypothetical is not that someone well-being is independent of what occurs “in her head.” That would be absurd. If I am in terrible pain then, ceteris paribus, I am worse off. Rather, the experience machine hypothetical crystallizes the normative case against limiting the sources of someone’s well-being to what occurs “in her head.” It suggests that someone’s well-being should be allowed to depend not only upon her mental states (cognitions, affects, perceptions, evaluations, memories, etc.), but also upon features of her life and the world that are not wholly mentalistic (such as her health, her relationships, the degree to which she has attained her goals, her standing in the community, political liberties, knowledge and education).Footnote ¹⁰

In particular, preferentialists about well-being may plausibly arrive at the following position: We should not define a “self-interested” preference so narrowly that my “self-interested” preferences are required to be preferences regarding my own mental states, and nothing else. However precisely the concept of “self-interest” is defined, that definition should permit me to have a “self-interested” preference for health, liberty, knowledge, goal fulfillment, good relationships, political standing, and other features of my life that are not wholly mentalistic.

Consider again function $v_{i}(\cdot )$ – individual i’s momentary idealized self-interested preference-utility function. Let $M_{i,t}(x)$ be a vector of individual i’s mental states at time t (her feelings, cognitions, memories, perceptions, etc.) and $N_{t}(x)$ non-mental features of outcome x at time t. On the view of well-being under consideration, $v_{i}(\cdot )$ takes the form $v_{i}(x_{t})=v_{i}(M_{i,t}(x),N_{t}(x))$. If self-interested preferences were required to be preferences for mental states, we could use a simpler functional form for $v_{i}(\cdot )$, namely $v_{i}(x_{t})=v_{i}(M_{i,t}(x))$. But View A of well-being declines to adopt such a restrictive understanding of “self-interest.”Footnote ¹¹

Given this view of well-being, is SBPA justified? Let $LS_{i}(x_{t})$ be individual i’s answer to a life-satisfaction survey at time t in outcome x.Footnote ¹² Assume that $LS_{i}(x_{t})$ is indeed good evidence of $v_{i}(x_{t})=v_{i}(M_{i,t}(x),N_{t}(x))$, individual i’s momentary idealized preference utility in outcome x. In expressing his current life satisfaction in a given outcome (x), the individual is expressing the preference utility of his current attributes in x. Adding those preference utilities over time, we get his overall (lifetime) preference utility in that outcome. If this were true – if it were true that $LS_{i}(x_{t})\approx v_{i}(x_{t})$ – then SWB (in particular, life-satisfaction) surveys would have a key role to play in policy analysis, given View A, as indicators of the basic measure of well-being ($u_{i}(\cdot )$ and $v_{i}(\cdot )$). Indeed, one suggestion in the SWB literature is that SWB surveys tell us about preference utility (Clark, Frijters & Shields, Reference Clark, Frijters and Shields2008).

However, for various reasons I have discussed at length elsewhere (Adler, Reference Adler2013), $LS_{i}(x_{t})$ is not reliable evidence of $v_{i}(x_{t})$. One reason has to do with the interpretation of the life-satisfaction question: individuals may “read” the question as asking about feelings of happiness, not levels of preference utility. Second, individual i’s preference utility is a function of a range of mental and non-mental attributes: all the features of the world at time t that individual i under idealized conditions would self-interestedly care about. But individual i, in answering a life-satisfaction question, may be focused on the attributes that are currently salient (for whatever reason attributes become salient). Unless individual i at time t in a given outcome x is carefully paying attention to the full range of inputs into his preference-utility function – to the full range of attributes in $(M_{i,t}(x),N_{t}(x))$ – $LS_{i}(x_{t})$ will be overly responsive to the currently salient subset of attributes in ($M_{i,t}(x)$, $N_{t}(x)$). Third, if $v_{i}(\cdot )$ is estimated by administering life-satisfaction surveys to a group of respondents consisting of different individuals, or the same individual at different times, on the assumption that all respondents have the same preferences – this would be done to increase the number of survey observations that are evidence of $v_{i}(\cdot )$ – problems of scaling arise. Two individuals, or the same individual at different times, may use different numerical scales to express the very same preferences.Footnote ¹³

What role does behavioral economics play in this analysis of the connection between life satisfaction and preference utility? I suggest that behavioral economics weakens the nexus between $LS_{i}(x_{t})$ and $v_{i}(x_{t})$. If behavioral economics is correct, then $LS_{i}(x_{t})$ is even poorer evidence of $v_{i}(x_{t})$ than it would be if individuals, in their actual behavior, behaved as predicted by neoclassical models. Why? $v_{i}(\cdot )$ represents individual i’s rational self-interested preferences under favorable conditions: rational in the EU sense. We ask: how would individual i rank outcomes and choices, were she to have good information, be self-interested, and conform to the axioms of EU theory? But behavioral economics demonstrates that individuals pervasively depart from EU theory in their actual choices. EU choice is not a feature of the automatic, unconscious, System 1 processes that drive ordinary life. In order to come into compliance with EU axioms, individuals need to engage in effortful, conscious, System 2 choice and (very likely) training in probability theory and decision analysis (Kahneman, Reference Kahneman2011).

One could, in theory, imagine a life-satisfaction survey administered only after “debiasing” steps designed to help respondents grasp and satisfy EU norms of rationality. But (as far as I am aware) most life-satisfaction surveys are not preceded by such coaching.

A life-satisfaction survey or some other SWB survey, without debiasing, can only provide evidence about the respondent’s current (non-debiased) attitudes. But behavioral economics tells us that a preference-utility function satisfying EU theory is a constructed object. Individuals do not naturally have this object “in their heads”; they must engage in deliberate, effortful steps, so that their preferences focus on final outcomes (not changes), are complete and transitive, and otherwise measure up to the coherence conditions expressed by the EU axioms. Further, on View A of well-being, the measure of an individual’s welfare is a preference-utility function that satisfies the EU axioms (as well as being self-interested and meeting the requirement of “favorable deliberative conditions”). Thus, if behavioral economics is true (as a scientific matter) and if View A of well-being is adopted (as a normative matter), someone’s response to an SWB survey (without debiasing) can hardly be taken as the measure of her well-being.

The reader might object that $LS_{i}(x_{t})$ needs only to be “good enough” evidence of $v_{i}(x_{t})$ – rather than “good” or “reliable” evidence in some more robust sense – in order to be used by policymakers as the measure of $v_{i}(x_{t})$. However, behavioral economics shows there to be a critical gap between the two. The idealized preference-utility function $v_{i}(\cdot )$ is a numerical representation of preferences that are rational in the sense of being EU-compliant, while – behavioral economics shows us – respondents to SWB surveys without debiasing should not be expected to be rational in that sense. Why think that someone’s verbal evidence expressed in a condition falling short of rationality is “good enough” evidence of what she would want, if rational? Complaints about the “infeasibility” of EU maximization are of no avail, here. First, EU maximization (like algebra) is feasible, if not natural and easy, for ordinary humans.Footnote ¹⁴ Second, if this is untrue (if EU maximization is more like quantum physics than algebra) the upshot should be to drop EU-compliant preferences as part of our conception of well-being – to cease defining well-being in terms of the idealized preference-utility functions $u_{i}(\cdot )$ and $v_{i}(\cdot )$ – and not to take $LS_{i}(x_{t})$ as “good enough” evidence of $v_{i}(x_{t})$.

4.2 View B of well-being

Assume that we are not persuaded by Nozick’s “experience machine,” and find it normatively attractive, on balance, to define “self-interested” preferences as preferences for mental states. (On this possibility, see Adler, Reference Adler2013; Sumner, Reference Sumner1996, Chapter 4.) On this view of “self-interest,” the only inputs to a utility function representing i’s self-interested preferences are i’s cognitions, feelings, perceptions, etc. $v_{i}(\cdot )$ is the momentary utility function representing the preferences that individual i would have, were i to be EU-rational, deliberating under favorable conditions, and self-interested. Thus, on the mentalistic view of self-interest, $v_{i}(x_{t})$ takes the form $v_{i}(M_{i,t}(x))$, not $v_{i}(M_{i,t}(x),N_{t}(x))$. We might (if we are careful) refer to $v_{i}(\cdot )$ as an “experience utility” function – but only with care, since this is “experience” utility only in the sense of being an idealized preference-utility function with an extra structural constraint, namely that it must take the preference-holder’s mental states as its sole arguments.

Assuming $v_{i}(\cdot )$ takes the form $v_{i}(M_{i,t}(x))$ rather than $v_{i}(M_{i,t}(x),N_{t}(x))$, can we now take an individual i’s answer to an SWB survey at time t in outcome x as good evidence of $v_{i}(x_{t})$? In brief, no. To begin, $M_{i,t}$ is a vector of mental characteristics. An individual’s complete “state of mind,” at any point in time, is a multidimensional bundle, comprised of many types of psychological attributes: sensations of pain and pleasure, emotions of different sorts, valuations, cognitions, perceptions, memories, etc. (Bernheim, Reference Bernheim2016, in his contribution to this symposium, makes a similar observation.) Individuals can have heterogeneous rankings of these multidimensional psychological bundles – giving more weight to one or another dimension. Sam might prefer a mental life with lots of pleasure, even if his thinking is fuzzy, his memories disconnected, and his perceptual experiences fairly monotonous. Sheila might care a lot about having crisp cognitions and a rich perceptual life, even at the cost of some pleasure. Griffin provides the following example to illustrate that there is not one dimension of psychological life that necessarily drives human preferences for psychological bundles: “At the very end of his life, Freud, ill and in pain, refused drugs except aspirin. ‘I prefer,’ he said, ‘to think in torment than not to be able to think clearly.”’ (Griffin, Reference Griffin1986, p. 8).

Understanding that $M_{i,t}$ is multidimensional, is an individual’s answer to an SWB survey (in particular, a life-satisfaction survey) good evidence of $v_{i}(M_{i,t}(x))$ – of idealized momentary self-interested preference utility according to View B? There is admittedly one respect in which $LS_{i}(x_{t})$ is better evidence of $v_{i}(M_{i,t}(x))$ as contrasted with $v_{i}(M_{i,t}(x),N_{t}(x))$. With the general form $v_{i}(M_{i,t}(x),N_{t}(x))$, we might worry that an individual will not be paying attention to the $N_{t}$ inputs – to the non-mental attributes that (according to View A) are allowed to be a determinant of self-interested preference utility. On View B of well-being, a failure of attention with respect to non-mental attributes does not undermine the role of life-satisfaction surveys in evidencing $v_{i}(\cdot )$, since only mental attributes are inputs into $v_{i}(\cdot )$.

However, all the other problems discussed above in using a life-satisfaction survey as evidence of $v_{i}(\cdot )$ remain in place. In particular, the key point remains that $v_{i}(\cdot )$ is supposed to satisfy EU theory. But behavioral economics tells us, again, that individuals do not walk around with EU-compliant preference-utility functions already “in their heads.” Rather, arriving at these functions typically involves effort, conscious thought, training, etc. $v_{i}(\cdot )$ would be generated, specifically, by coaching individuals in the norms of EU theory; otherwise providing favorable deliberative conditions; and instructing them to rank outcomes by focusing on the psychological bundles they have in outcomes (how View B defines “self-interest”). Behavioral economics should make us dubious that a life-satisfaction survey administered to non-debiased respondents (even a “tweaked” such survey focusing on psychological life) will tell us much about $v_{i}(\cdot )$.

4.3 View C of well-being

As mentioned above, a long philosophical tradition adopts an objective-good view of well-being (reviewed in Adler, Reference Adler2012, Chapter 3). There is, to be sure, plenty of disagreement within this tradition. “Objective-good” theorists disagree about which goods are on the list. For example, Finnis’ list is life, knowledge, play, aesthetic experience, sociability, practical reasonableness, and religion (Finnis, Reference Finnis1988, Chapter 4). Nussbaum’s is life; bodily health; bodily integrity; the senses, imagination, and thought; emotions; practical reason; affiliation; other species; play; and control over one’s environment (Nussbaum, Reference Nussbaum2000, pp. 78–80). Sher’s is moral goodness; rational activity; the development of one’s abilities; having children and being a good parent; knowledge; and the awareness of true beauty (Sher, Reference Sher1997, p. 201). Griffin’s is accomplishment; autonomy; physical integrity; understanding; enjoyment; and deep personal relations (Griffin, Reference Griffin1997, pp. 29–30).

Objective-good theorists also offer divergent rationales for placing goods on the list. One rationale is in terms of the human essence: goods are the realization of those capacities that are essential to human beings. A second appeals to commonly shared judgments or intuitions of well-being: objective goods are those things which people, under the right conditions, normally perceive as good for human welfare. A third possibility is for the theorist to rely upon her own judgments or intuitions regarding human welfare – without claiming that these are necessarily widely shared.

Leaving aside these (significant) internal disputes, objective-good accounts are alike in severing the conceptual link between a particular person’s welfare and her preferences.Footnote ¹⁵ The list of goods (whatever its precise content) is such that Jim can be on balance better off in y than x even though Jim prefers x – for that matter, even though Jim under favorable deliberative conditions rationally self-interestedly prefers x.Footnote ¹⁶ Note that all three of the rationales for objectivism allow for a divergence between someone’s objective goods and her actual or idealized preferences. This is clear for the first (human essence) and third (theorist’s own conception) rationale. As for the second rationale (shared human judgments), consider the case of idiosyncratic preferences. Jim might actually and/or ideally self-interestedly prefer x to y, even though most humans believe that Jim’s life in y is the better one.

How does behavioral economics enter the analysis? It does so indirectly. By demonstrating the wide gap between someone’s actual preferences and her rational preferences, behavioral economics may undercut the normative appeal of preference-based views of well-being. Preference views are traditionally defended by appealing to notions of individual sovereignty. Shouldn’t Sarah be able to decide for herself what makes her life go best? At least if Sarah is a normal adult human (with the capacity for autonomous choice), shouldn’t Sarah’s well-being depend upon Sarah’s perspective – what Sarah likes, wants, judges, thinks, endorses, etc.? Ceteris paribus, a normatively appealing feature of an account of well-being is that the account is non-paternalistic. The normative evaluation of Sarah’s life defers to Sarah’s own evaluation.

On the other hand, for reasons I adverted to earlier, a preference-based account of welfare should very plausibly be oriented to an individual’s rational preferences. But any definition of well-being in terms of rational preferences will itself be paternalistic, to a certain extent – if rationality is understood in the EU sense. This is what behavioral economics shows us. Consider, in particular, the definition of Sarah’s well-being in terms of Sarah’s EU-rational preferences under favorable deliberative conditions and conditions of self-interest. However we define the latter two components, there will be a gap between Sarah’s well-being and the non-EU-rational preferences that motivate her day-to-day behavior. Sarah’s EU-rational perspective – what she wants when she regiments her preferences so as to comply with EU theory – is not the same as her day-to-day, System 1 perspective. Defining Sarah’s well-being in terms of her EU-rational preferences is in one sense non-paternalistic (since it defers to the EU-rational perspective of Sarah) but, in another sense, is not (since it overrides her day-to-day perspective).

What behavioral economics show us is that an account cannot be non-paternalistic in both senses. And – here’s the rub – by weakening the non-paternalistic credentials of a rational-preference view of welfare, behavioral economics might indirectly support an objective-good view of welfare. That is, the ethical deliberator might find herself at the following juncture in thinking about which view of well-being she wishes to endorse: “I find some appeal in an objective-good view of well-being. On the other hand, I am also moved by the thought that a view of well-being should be non-paternalistic. Behavioral economics now shows me that a preference-based well-being view cannot be fully non-paternalistic (at least if the view appeals to the EU view of rationality, which I also accept, and to rational preferences). It shows me that the considerations in favor of the preference view were weaker than I thought. Balancing the pros and cons of objective-good versus preference views of well-being, I now support an objective-good view.”

While behavioral economics could in this way provide indirect, dialectical support for an objective-good view of well-being, we now need to ask about SBPA. On an objective-good view of well-being (View C), to what extent is an SWB survey good evidence of an individual’s well-being? Well-being, here, means, the overall quality of the individual’s life in light of her attainment with respect to each of the objective goods specified by View C and in light of the appropriate balancing of such attainments (as this balancing is specified by View C).

One possibility is that View C includes some goods that are not mentalistic: goods that are not reducible to an individual’s mental states. Nozick’s experience machine argues powerfully in favor of a View C of this sort.

It is very hard to see how an SWB survey provides strong evidence regarding an individual’s attainment of a good that is not mentalistic.Footnote ¹⁷ Consider the good of knowledge (a standard objective good – see the lists above). Someone’s knowledge depends not merely on what she believes, but on whether those beliefs are true. It depends on her beliefs and the degree of correspondence between those beliefs and the outside world. Asking someone how happy she is, or how satisfied she is overall with her life, or specifically how satisfied she is with her knowledge of the world, is not going to track very well the extent to which she has correct beliefs.

Nor will an SWB survey tell us much about how an individual fares with respect to the overall balance of mentalistic and other goods. That balance will be determined (as per View C) by some criterion external to the person’s own preferences and valuations (such as the human essence, commonly shared judgments of well-being, or some other objective criterion offered by the view).

A second possibility is that View C consists of a list of goods, all of which are mentalistic. For example, one good might be happiness; the second, the quality of someone’s memories (Kahneman, Reference Kahneman2011, part V). SWB surveys or other psychological surveys can be designed to tell us about how someone fares with respect to each good – but the problem of balancing the goods remains.

Psychological surveys would seem to offer strong evidence of well-being only according to a distinctive version of View C: one such that well-being consists in a single, mentalistic good. If the mentalistic good is the kind of affective or valuational state that SWB surveys are designed to measure – feeling happy, feeling satisfaction with life, having a sense of purpose – then someone’s answer to an SWB survey can indeed be good evidence of his well-being as per this version of View C.

The supposition of an objective-good view of this sort is not ludicrous. Consider Benthamite hedonism, understood as follows. A person, at any moment in time, experiences some level of positive affect (pleasure) or negative affect (pain), measurable on a ratio scale with zero for the neutral level, and higher numbers indicating a more favorable affect (more pleasurable if above neutral, less painful if below neutral). This momentary affective level is an introspectible psychological magnitude: individuals can reliably detect whether they are feeling sensations of pain or pleasure, and how intense those sensations are. An individual’s well-being in any given outcome is simply the lifetime sum of momentary affective levels (Bronsteen et al., Reference Bronsteen, Buccafusco and Masur2015; see also Layard, Reference Layard2011, defending monistic hedonism; on the measurement of affect, see Kahneman, Wakker & Sarin, Reference Kahneman, Wakker and Sarin1997).

Benthamite hedonism (thus understood) is clearly mentalistic. Moreover, it is objective: maximizing the lifetime sum of momentary affective levels is posited as the well-being-maximizing course of action for any individual, regardless of whether the individual herself prefers to maximize this affective magnitude – regardless of whether she also cares about other dimensions of mental life (memories, perceptions, cognitions, valuations) or about non-mental attributes. Finally, Benthamite hedonism supports SBPA. Traditional happiness surveys offer some (albeit imperfect) evidence of an individual’s average affective quality during some stretch of time prior to the survey; and even better surveys, using experience sampling or “day reconstruction” to estimate moment-to-moment affects, have been implemented by SWB scholars.

But note carefully the special structural features of Benthamite hedonism that enable SWB surveys to be good evidence of well-being. First, SWB surveys are informing us about the level of individual affects, not about the satisfaction of individual preferences. Second, affective quality is posited as the sole component of well-being. The normative deliberator who adopts this conception of welfare has not only rejected rational-preference accounts (View A and B) in favor of an objective-good account (View C). She, further, rejects both pluralism about goods and the existence of goods that are not mentalistic. Although this version of View C is certainly possible, it is far from obvious why we should find it normatively attractive.