1

Do People Seek to Maximize Happiness?

Evidence from New Surveys*

Daniel J. Benjamin

Cornell University, Institute for Social Research, and NBER

Ori Heffetz

Cornell University

Miles S. Kimball

University of Michigan, Institute for Social Research, and NBER

Alex Rees-Jones

Cornell University

*** PRELIMINARY AND INCOMPLETE DRAFT: PLEASE DO NOT CIRCULATE ***

Abstract

We present survey respondents with hypothetical, pairwise decision problems. We ask them which option they

would choose, and we also ask which option they believe would give them the highest subjective well-being

(SWB). We find that individuals‘ expected choices are strongly predicted by which option they expect will lead

to greater SWB, but we also find that their choice ranking of the options deviates systematically from their

SWB ranking. In some of our studies, we also ask respondents which option they believe would lead to higher

levels of other aspects of their lives besides SWB, such as health, control over life, and sense of purpose. We

find that people‘s rankings of the options based on these other aspects explain their choices, even controlling for

the SWB ranking. However, we also find that SWB ranking explains variation in choice better than all of these

non-SWB factors combined. On the one hand, our results support the view that individuals do not seek to

maximize happiness alone (at least the way happiness is typically measured); they are willing to make trade-offs

between happiness and other aspects of their lives. On the other hand, since people‘s prediction of what will

give them the greatest SWB is the best single predictor of choice, our results also suggest that the burgeoning

use of happiness data as an empirically-measurable proxy for welfare may nonetheless be justified in many

cases.

JEL Classification: D03, D60

Keywords: subjective well-being, SWB, happiness, life satisfaction, utility, choice

* We are extremely grateful to Dr. Robert Rees-Jones and his office staff for generously allowing us to survey their patients and to

Cornell‘s Survey Research Institute for allowing us to put questions in the 2009 Cornell National Social Survey. We thank Gregory

Besharov, John Ham, Benjamin Ho, Erzo Luttmer, Michael McBride, Ted O‘Donoghue, Matthew Rabin, Antonio Rangel, and Robert

J. Willis for especially valuable early comments and suggestions. We are grateful to participants at the CSIP Workshop on Happiness

and the Economy and at Cornell‘s Behavioral/Experimental Lab Meetings, Junior Faculty Lunch, and Behavioral Economics

Workshop for helpful comments. We thank Kristen Cooper, Isabel Fay, John Farragut, Geoffrey Fisher, Sean Garborg, Jesse Gould,

June Kim, Nathan McMahon, Greg Muenzen, John Schemitsch, Elizabeth Traux, Charles Whittaker, and Brehnen Wong for their

research assistance. We thank the National Institute on Aging (grant P01-AG026571/01) for financial support.

E-mail: db468@cornell.edu, oh33@cornell.edu, mkimball@umich.edu, arr34@cornell.edu.

2

Economic researchers are increasingly using measures of subjective well-being (SWB), such as self-

reported happiness or life satisfaction, as an empirical proxy for utility. The hope is that these measures can

enable economists to conduct welfare evaluations that are difficult to conduct with standard revealed preference

methods, such as measuring the negative externality from neighbors‘ higher earnings (Luttmer, 2005), the

average American‘s tradeoff between inflation and unemployment (DiTella, MacCulloch, and Oswald, 2003),

and the effect of health status on the marginal utility of consumption (Finkelstein, Luttmer, and Notowidigdo,

2008). In order for this strategy to assess welfare as conceived in the standard economic approach, a necessary

assumption is that an agent‘s preferences over alternatives and her measured SWB under each alternative lead

to identical rankings of the alternatives. This paper provides evidence for evaluating that assumption. In other

words, the goal of this paper is to evaluate whether individuals seek to maximize SWB.

In pursuing this goal, this paper empirically addresses a recent theoretical literature on the economics of

happiness. There are two alternative views regarding the relationship of SWB to utility that give different

guidance regarding the use of SWB data in empirical work.1 The first view, reflected at least implicitly in a

large number of papers (e.g., Gruber and Mullainathan, 2002; Oreopoulos, 2007), is that SWB data represent

idealized revealed-preference utility in the sense of what individuals would choose if they were well-informed

about the consequences of their choices for SWB. The second view, explicitly laid out in Kimball and Willis

(2006) and Becker and Rayo (2008), is that even well-informed agents will be willing to trade off SWB with

other things they care about, making SWB and utility distinct.2 The present paper attempts to empirically assess

these two views by measuring the concordance between happiness and preferences.

In the economics literature, some of the interest in using measures of happiness has been to identify

possible mistakes that people make due to mispredicting happiness. In that literature, it is assumed that people

are trying to maximize SWB, and divergence between choice and ex post SWB are used to infer that people

mispredict how their choices will affect their SWB. Unlike that literature, we wish to test the assumption that

people are trying to maximize SWB. Doing so requires comparing an individual’s choice behavior with her

beliefs about how each possible choice would affect measured SWB.3 Our strategy is to confront individuals

with empirically-relevant choice situations to see whether they choose the alternative that they predict will

1 Some researchers adopt yet a third view, the philosophical position that happiness per se is the appropriate maximand for a

policymaker (e.g., Bentham, 1789; Layard, 2005; Bronsteen, Buccafusco, and Masur, forthcoming). See Bernheim (2009a, 2009b) for

a categorization and discussion of alternative philosophical approaches to welfare economics. Throughout this paper, we adopt the

standard economic perspective (common to the two views mentioned in the text) that the appropriate maximand for welfare analysis is

an individual’s preferences. However, the value of our results does not hinge on this perspective. Even if one believes that observable

preferences are not the appropriate maximand (such as in cases where someone feels social or moral pressure not to reveal their true

preferences; see Koszegi and Rabin, 2008), understanding the link between choice and subjective well-being is still valuable to a

policy maker. 2 For elaborations of this view, see also Kimball, Levy, Ohtake and Tsutsui (2006), Kimball, Ohtake and Tsutsui (forthcoming), and

Kimball, Nunn and Silverman (2010) 3 In the terminology of Kahneman, Wakker, and Sarin (1997), this would amount to comparing ―decision utility‖ with ―predicted

utility.‖

3

maximize their own SWB.4 However, because of the difficulty of empirically measuring ex ante beliefs

regarding the SWB consequences of important real-world decisions, our strategy is to confront survey

participants with hypothetical decisions, and to ask the participants for which alternative they expect greater

SWB, as well as which alternative they would choose. At the empirical level, the question we are posing is:

How good is predicted SWB as a proxy for predicted preferences?

In this paper, we analyze choice and predicted happiness data for a variety of hypothetical decision

scenarios from 538 undergraduates, a convenience sample of 1,495 adults, and a representative sample of 1,000

Americans. Each scenario has two alternatives. For example, one scenario describes a choice between a job

that pays more but allows less sleep versus a job with lower pay and more sleep. We assess participants‘

predicted happiness (their beliefs about how happy they would be) using questions based on each of the three

measures of subjective well-being that are most commonly used in the empirical literature: life satisfaction,

happiness with life as a whole, and felt happiness. We have two main results.

First, we find that participants‘ predictions about happiness are a powerful predictor of their choices.

The strength of the relationship between predicted happiness and predicted choices depends on the measure of

happiness. For example, in one study, we find that life satisfaction predicts about 65% of the variation in

preferences, while happiness with life as a whole and felt happiness predict about 59% and 55% of the variation

in preferences, respectively. Of course, when the same participants are asked to assess predicted SWB and to

make choices, the relationship between the two may be inflated by participants‘ desires to appear consistent.

However, when we ask one group of participants to make choices and another group to assess predicted SWB,

we find the opposite; the difference in mean ratings is smaller and less statistically distinguishable in all

scenarios. This suggests the possibility that when we ask participants to rate the options only in terms of

predicted happiness, they report their overall assessment of the options, even if this assessment actually

incorporates aspects of the options other than happiness. To address this possibility, in another study we

measured happiness as part of a list of things one might care about: own happiness, happiness of one‘s family,

health, romance, social life, sense of control over life, spirituality, fun, social status, boredom, physical comfort

and sense of purpose. In that study, the difference in predicted happiness between the two alternatives explains

43% of the variation in intensity of preference for one alternative over the other when happiness is measured in

isolation, and explains 31%-39% of the variation when happiness is measured as part of the list, depending on

the phrasing of the happiness question. In comparison, the characteristic we measured that was second best at

predicting choice—predicted control over one‘s life—explains 8%-20% of the variation in choices in these

4 A large experimental psychology literature finds that people often make choices that ex post do not end up maximizing measured

happiness (Gilbert, 2006). Therefore, preferences do not always line up with ex post measured happiness. Relatedly, there is evidence

that predictions of happiness do not always accord with ex post measured happiness (Gilbert, 2006). However, as far as we are aware,

the question we focus on—whether predicted happiness lines up with preferences—has received relatively little attention. Becker and

Rayo (2008) propose empirical tests of whether things other than happiness matter for preferences, but their tests rely on ex post

measured happiness and hence assume that people correctly predict the happiness consequences of their choices.

4

surveys (when happiness is measured as part of the list). The third best characteristic was predicted sense of

purpose, explaining 11%-12% of the variation. Regardless of the survey manipulation, happiness out-performs

the other characteristics as a proxy for decision utility. Our main takeaway message from this variety of ways of

measuring happiness and other attributes is that measured happiness is a strong predictor of preferences and

may be the single best empirical proxy.

Our second main result is that predicted happiness is not the only characteristic of the choice alternatives

that enters preferences. The fact that the other characteristics we measured have predictive power for choice

does not by itself imply that they enter preferences; it is possible that they matter only because they are

correlated with predicted happiness. However, in regressions of choice on predicted happiness and a single

other characteristic, the coefficient on the non-happiness characteristic typically remains statistically significant

despite controlling for predicted happiness. That finding may suggest that non-happiness characteristics enter

preferences independently of measured happiness. Consistent with the idea that individuals make trade-offs

between happiness and other objectives they care about, we also find that predicted happiness correlates more

strongly with choices for participants who indicate a higher ―willingness to pay‖ for happiness in a different

survey question.

Our results suggest that common SWB measures in general, and measured life satisfaction in particular,

may be good empirical proxies for preferences. However, since predicted SWB is not the only characteristic of

a choice alternative that enters preferences, caution is warranted in drawing welfare conclusions exclusively

from measured SWB.

To take a step toward providing practical guidance to empirical researchers about in which kinds of

situations measured SWB may be a better or worse proxy for preferences, we analyze whether our results differ

across scenarios. We find suggestive evidence hinting that SWB may be a bigger factor in everyday, minor

decisions (such as eating an apple versus orange, or doing menial work versus paying a fee) than in more major

decisions (such as taking a job that is interesting versus preparatory for career, or going to a college that is

selective versus one that promises a good social life). Our results are qualitatively similar across the subject

populations we study, and we do not find observable demographic characteristics of respondents (e.g. age, race,

income, gender) that predict how much they weight SWB in their choices.

We subjected our results to a variety of robustness checks, two of which we mention here. First, we

were worried about possible self-control problems: participants may forecast a different choice than the one that

they expect would maximize SWB because they anticipate a failure of self-control. To explore this possibility,

in addition to asking participants which alternative they would choose, we also asked them which alternative

they would want themselves to make. We find our results are largely similar when we restrict the sample to

observations where respondents said they were choosing what they would want themselves to choose. Second,

we cross-checked our main results by eliciting participants‘ willingness to pay for various services, each of

5

which could improve one aspect of life, such as happiness, health, and so on. We find evidence that respondents

with higher willingness to pay for happiness have fewer instances where they choose an option which they do

not predict would maximize happiness.

Our work is related to a literature in philosophy that poses thought experiments in extreme hypothetical

scenarios in order to demonstrate that people‘s preferences encompass more than their own happiness.5 Unlike

that literature, by focusing on realistic choice situations, we seek to address to what extent measured happiness

is a good proxy for preferences in empirically-relevant choice situations. As far as we are aware, the most

closely related evidence to ours is one of the experiments described in Tversky and Griffin (1991). In a

between-subjects experiment with 66 undergraduates, participants were presented with a hypothetical choice

between a job with a higher salary where co-workers earn even more versus a job with a lower salary where co-

workers earn even less. Like we find in a similar choice problem, Tversky and Griffin found that participants

were more likely to choose the higher absolute salary but said they would be happier with the higher relative

salary. Tversky and Griffin interpret the result as supporting their theory that payoff levels are weighted more

heavily in choice, while contrasts between payoffs and a reference point are weighted more heavily in

subjective well-being judgments.

We wish to emphasize that our focus in this paper is the relationship between preferences and measures

of happiness typically used in empirical research. We do not address the question of whether a sufficiently

broad conception of happiness that is not captured by typical measures might be indistinguishable from

preferences. That being said, our methodology could be applied to test whether some measure of a broader

notion of happiness predicts choice better than the subjective well-being measures typically used in current

empirical research.

The paper is organized as follows. Section I presents a theoretical framework to clarify what can be

inferred about preferences from our surveys. Section II discusses the survey design and subject populations. In

Section III, we analyze the main results, and Section IV presents additional results and robustness analyses.

Section V concludes. Details of the experimental materials are described in the Appendix, and details of all

pilots are in the Web Appendix.

I. Theoretical Framework

Preferences may depend on happiness as well as on other factors. We assume that an individual‘s

preferences can be represented by an increasing, concave, and differentiable utility function,

(1) U(H(X, G), X, Q),

5 For example, in one famous thought experiment, a reader is asked to introspect whether he or she would choose to be hooked up to

an ―experience machine‖ that would guarantee complete happiness for the rest of life, without awareness that the experience is not real

(Nozick, 1974, pp.42-45). If a reader would refuse the machine, he or she must not care exclusively about happiness.

6

where H is happiness, X is a vector of observed factors that both affect happiness and enter preferences directly,

G is a vector of unobserved factors that affect preferences only through happiness, and Q is a vector of

unobserved factors that enter preferences but do not affect happiness.

Do people make tradeoffs between happiness and other factors? To examine whether people make such

tradeoffs, we test whether there exist components of X (e.g. health, or sense of purpose) that enter preferences

directly and not just through their effects on happiness. Therefore, one central hypothesis we seek to test is

whether UX > 0. To do so, we present survey participants with a sequence of scenarios. Each scenario involves

making a choice between two options. Option 1 can be described by the levels of inputs it provides, (X1, G1,

Q1), and Option 2 can be described by (X2, G2, Q2). For a given scenario, define ∆X ≡ X1 – X2, ∆G ≡ G1 – G2,

∆Q ≡ Q1 – Q2, and ∆H ≡ H(X1, G1) – H(X2, G2). Assuming ∆H, ∆X, and ∆Q are small, the individual will

choose Option 1 if and only if

(2) ∆U ≈ UH ∆H + UX ∆X + UQ ∆Q ≥ 0

(as a tie-breaker, we assume an individual chooses Option 1 if she is indifferent). Assuming the unobserved

term UQ ∆Q is distributed normally with mean µQ and variance ζQ2, we can rewrite (2) as

(3) Pr{choose Option 1} = Φ(α + βH ∆H + βX ∆X),

where α = µQ / ζQ, βH = UH / ζQ and βX = UX / ζQ are normalized marginal utilities, and Φ is the cdf for a

standard normal random variable. For each scenario, we ask participants to predict how happy they would be

under Option 1 relative to Option 2, as well as how Option 1 ranks relative to Option 2 on improving a number

of other factors, such as health, physical comfort and sense of purpose. These survey questions give us measures

of ∆H and ∆X. To test whether UX > 0, we estimate probit regression (3) and test the null hypothesis that βX = 0.

Our key identifying assumption is E(∆Q | ∆H, ∆X) = 0. This assumption would be violated if we neglected to

measure factors important for choice that are correlated with the factors we did measure.

We also ask participants their intensity of preference for Option 1 relative to Option 2, and we use this to

run a linear regression. Assuming our intensity-of-preference measure is monotonically related to the utility

difference between the two options, P = Π(∆U(H(X, G), X, Q)) with Π′ > 0 and (0) = 0, then P ≈ Π(0) + Π′∆U

= Π′UH ∆H + Π′UX ∆X + Π′UQ ∆Q. We can estimate

(4) P = γ + δH ∆H + δX ∆X + ε,

7

where γ = µQ, δH = UH Π′ and δX = UX Π′ are normalized marginal utilities, and ε ~ N(0, ζQ2 (Π′)

2). Under the

same identifying assumptions, we can test the null hypothesis that δX = 0. While the OLS regression (4) allows

us to use more information provided by the participants (namely, the intensity of their preferences) than (3)

does, it also requires the additional approximation that Π(∆U) is locally linear across the entire range of

decision scenarios.

How good is happiness as an empirical proxy for utility? Another central question we seek to answer is

how well measured happiness proxies for utility. To do so, we examine the fraction of scenarios where the

participant chose Option 1 when ∆H > 0, and Option 2 when ∆H < 0. We also run a univariate linear regression

of P on ∆H, which we interpret simply as a forecasting regression. The R2 from this regression captures how

well measured happiness predicts (intensity of) choice.

We return to this model in Section IV, where we extend it to provide theoretical basis for robustness

checks aimed at validating our main findings (the latter are reported in Section III). Before doing so, in the next

section we use the framework developed above to describe our survey design.

II. Survey Design and Subject Populations

We conducted eight different surveys among 3,033 respondents in total. In this section we describe the

design of four of the surveys, which ran among 2,280 participants and comprise the sample for our main

analyses. Table 1 provides complete details of these four surveys, which we call Studies 1-4: participant

population, sample size, scenarios used (see II.A below), type of questions asked (see II.B below),

randomizations of question order, etc. The other four surveys are described in the Web Appendix, which gives a

complete account of what we found and why we changed the questions from one survey to the next. We use

those other four surveys mainly for robustness checks.

II.A. Scenarios

At the core of our surveys is a sequence of hypothetical scenarios. In each scenario, respondents face a

choice between two options. The choice highlights a tradeoff. For example, one of the scenarios is:

Say you have to decide between two new jobs. The jobs are exactly the same in almost every way, but

have different work hours and pay different amounts.

Option 1: A job paying $80,000 per year. The hours for this job are reasonable, and you would be able to

get about 7.5 hours of sleep on the average work night.

Option 2: A job paying $140,000 per year. However, this job requires you to go to work at unusual hours,

and you would only be able to sleep around 6 hours on the average work night.

The full set of scenarios is given in the Appendix. In choosing the scenarios and the tradeoffs, we were guided

8

by three considerations.

First, we wanted to collect evidence that would be as relevant and as close in subject matter as possible

to the happiness literature. For this purpose, we picked scenarios that highlight tradeoffs that the literature

suggests might be important determinants of SWB. Hence, in some of our scenarios respondents face choices

between jobs and housing options that are more attractive financially versus ones that bear a financial cost but

allow for more sleep (Kelly, 2004), a shorter commute (Stutzer and Frey, 2008), being around friends

(Kahneman, Krueger, Schkade, Schwarz, and Stone, 2004), or making more money relative to others (Luttmer,

2005; see Heffetz and Frank, 2008, for a survey).

Second, since some of us were initially unsure we would find any divergences between predicted SWB

and choice, in our earlier surveys we focused on choice situations where individual SWB may not be the only

consideration. Hence, in some scenarios respondents choose between a career path that promises an ―easier‖ life

with less individual sacrifices versus one that promises posthumous impact and fame, or between a more

convenient and ―fun‖ option versus an option that might be considered ―the right thing to do.‖

Third, once we found divergences between predicted SWB and choice, in our later surveys we wanted to

assess the magnitude of the relationship between predicted SWB and choice in the most empirically-relevant

scenarios. In order to pick scenarios that highlight choices and tradeoffs that are relevant and important in the

lives of our respondent populations, we asked a sample of 102 University of Chicago students to list the three

most important decisions they made in the last day, the last month, the last 2 years, and in their whole lives.

Consistently major themes in their responses were studying, sleeping, choosing classes, and choosing which

college to attend. We used their answers to construct scenarios that seem representative of ordinary decisions

for respondent college-age population, and we subsequently confirmed these themes in an independent sample

of 156 Cornell students.6 Hence, in the resulting scenarios, individuals have to choose between socializing and

fun versus sleep and schoolwork; traveling home for Thanksgiving versus saving the airfare money; attending a

more fun and social college versus a highly selective one; and following one‘s passion versus pursuing a more

practical career path. We added to these scenarios additional ones that focus on economic decisions that are

prevalent in almost everyone‘s life, like choosing one food item versus another, or trading time for money.

The three considerations above resulted in a pool of 13 scenarios, and respondents in different surveys

are asked different subsets of this pool. Additionally, some surveys come in different versions where the

scenarios appear in different order. Finally, some scenarios are asked in different variations (e.g. different

wording, different quantities of money/sleep/commute time, etc.) and some scenarios are tailored to different

respondent populations (e.g. while we asked students about school, we instead asked older respondents about

6 The most frequent responses to these questions involved basic decisions to accept a job or internship (4% of responses), decisions of

which college to attend (13% of responses), and decisions on maintaining contact with family (3% of responses). These responses

motivated the addition of scenarios 7-10 on these subjects (see the Appendix for full phrasing). In addition, 3% of responses involved

sleep tradeoffs and 5% of responses involved decisions about maintaining friendships, validating our previously formulated scenarios

on these topics (scenarios 1 and 2). Full details are in the Web Appendix.

9

work).

II.B. Main Questions

Once subjects are presented with a scenario and its two associated choice options, they answer a few

questions. Our surveys include four types of questions, which we refer to in the text and in Table 1 as questions

about (a) choice; (b) meta-choice; (c) SWB; and (d) other aspects of life.

(a) Choice question

In all surveys and about each scenario, respondents are asked: ―If you were limited to these two options,

which do you think you would choose?‖ They answer by checking one of six checkboxes, horizontally

presented as a scale that moves from one option to the other: definitely choose Option 1; probably choose

Option 1; possibly choose Option 1; possibly choose Option 2; probably choose Option 2; and definitely choose

Option 2. We convert this six-point scale to an intensity-of-choice variable, ranging from 1 to 6, which can be

collapsed into a binary choice variable (―choose Option 1‖ versus ―choose Option 2‖).7 We treat answers to this

question as measures of preference.

(b) Meta-choice question

In all of the surveys except the CNSS, the choice question above is followed by the question: ―If you

were limited to these two options, which would you want yourself to choose?‖ Again, there are six possible

answers (definitely want Option 1; probably want Option 1; possibly want Option 1; possibly want Option 2;

probably want Option 2; and definitely want Option 2), which we convert to either an intensity variable (ranging

from 1 to 6) or a binary variable (―want to choose Option 1‖ versus ―want to choose Option 2‖). We defer a

detailed discussion of the purpose of this question to Section IV.B below, where we argue that one could view it

as a measure of meta-preference.

(c) SWB question

We asked three different versions of the SWB question, modeled after what we view as the three

―families‖ of SWB questions that are most common in the literature:8

(i) life satisfaction: ―Between these two options, which do you think would make you more satisfied

with life, all things considered?‖;

7 Responses to the Cornell National Social Survey (CNSS) were given only in this binary format instead of the six-point scale. This

format was chosen because the CNSS was conducted by telephone instead of on paper. The binary response scale was both briefer for

interviewers to convey and easier for respondents to understand. 8 See the Web Appendix for a list of SWB measures and the studies in the literature that use them, classified by the family of questions

they use.

10

(ii) happiness with life as a whole: ―Between these two options, taking all things together, which do you

think would give you a happier life as a whole?‖; and

(iii) felt happiness: ―Between these two options, during a typical week, which do you think would make

you feel happier?‖

In some survey versions, described below, we ask respondents about 12 different aspects of life, of which (one‘s

own) happiness is only one. There, we slightly modified the wording of questions (ii) and (iii) to be comparable

with the other aspects we asked about. These modified versions of (ii) and (iii) are:

(iv) own happiness with life as a whole: ―Between these two options, taking all things together, which

option do you think would make your life as a whole better in terms of … [your own

happiness]‖; and

(v) immediately-felt own happiness: ―Between these two options, in the few minutes immediately after

making the choice, which option do you think would make you feel better in terms of …

[your own happiness].‖

From the perspective of psychological theory, ―felt happiness‖ may be the cleanest measure of SWB

because it is more directly accessible than complex cognitive judgments of overall well-being (e.g., Schwarz

and Clore, 1983). However, from the perspective of economic theory, a ―felt happiness‖ measure may be

deficient because it encourages participants to focus on how they will feel in the near-term aftermath of the

decision; decisions that increase felt happiness in the short term but reduce it in the long term might be

predicted to increase felt happiness even if they decrease overall well-being. The ―life satisfaction‖ and

―happiness with life as a whole‖ measures might be expected to correlate better with choice because they

require participants to integrate SWB over one‘s lifetime.

(d) Question about other aspects of life

As mentioned, in some survey versions, instead of asking only about predicted happiness, we ask

respondents to rate the two options in terms of these 12 aspects of life: your own happiness, your family‘s

happiness, your health, your romantic life, your social life, your control over your life, your life‘s level of

spirituality, your life‘s level of fun, your social status, your life‘s non-boringness, your physical comfort, and

your sense of purpose. While the list is limited by the length of the survey, we chose these aspects of life in an

attempt to capture what else, besides one‘s own happiness, might matter for preferences. In choosing these

items, we were guided by economists‘ and philosophers‘ enumeration of ―capabilities‖ (Sen, 1985; Nussbaum,

2000), non-hedonic components of SWB proposed by psychologists (White and Dolan, 2009), and our own

introspections. In some survey versions, we separate ―own happiness‖ from the other 11 aspects, and ask

respondents first just about own happiness in each scenario, and then, re-presenting each scenario, we ask about

the other aspects. We then call the happiness sub-question an ―isolated‖ measure of SWB. In other versions, the

11

12 aspects appear together, either in the order specified above—in which case we call the happiness sub-

question ―first in a series‖ measure of SWB—or in reverse order, starting with ―your sense of purpose‖—in

which case the happiness sub-question is a ―last in a series‖ measure of SWB.

II.C. Additional Questions

Our main survey questions outlined in II.B (a)-(d) above can be summarized as follows. All subjects are

asked the choice question (a). For some subjects, this question is followed by the meta-choice question (b). All

subjects are asked a SWB question (c), but different variations of this question are asked to different subjects.9

The SWB question is asked either in isolation—where subjects are only asked to predict one SWB measure

under the alternative scenarios—or in series—where subjects are asked about twelve aspects of life, as

explained in (d).10

In order to validate the evidence from these scenario-based main questions, we added to our surveys

three types of additional questions. First, subjects were asked to assess the current levels of SWB (e.g.

happiness) and other aspects of their life. Second, they were asked questions aimed to elicit self-reported

willingness-to-pay (WTP) in time (rather than in money) for improving different aspects of their life. As

explained below, combining these two sets of questions allows us to construct a measure of the importance

individuals assign to the different aspects in their utility function. We can then examine how this measure

correlates with our scenario-based measures.

As a third type of validation questions, our survey participants were explicitly asked for interpretations

of their own behavior. For example, they were asked to explain their choices when these could not be explained

as maximizing predicted SWB.

We postpone a full discussion of these three types of additional questions to Section IV, where we

describe them in detail, explain the rationale behind them, provide the theory they are based on by further

developing the theory from Section I and, most importantly, use the evidence they provide to assess our main

findings. Before doing so, in the next section we first discuss our main findings.

9 The only exception is study 2 which employs a between-subject design, in which subjects are asked either the choice question or a

SWB question. 10

As mentioned above, Table 1 details this and other information by our four main surveys. In addition to reporting each survey’s

population type, size, and cell sizes, the table indicates which of the thirteen scenarios are used, which of the subjective well-being

questions (i)-(v) above is used, and whether it is used in isolation or in series. It also indicates the format of the choice question (6-

point scale versus binary question), whether a meta-choice question is included, and whether all subjects are asked both the choice and

the subjective well-being questions (a within-subject design) or half the subjects are asked the choice question and the other half the

subjective well-being question (a between-subject design). Finally, the table details the ordering of scenarios in different versions, as

well as the ordering or the SWB question versus the choice and meta-choice questions.

12

III. Main Results

Table 2 reports the distribution of responses to our surveys‘ main choice and SWB questions.11

For each

scenario in each study, we count how many subjects favored Option 1 over Option 2 in the choice question, and

how many subjects favored Option 1 over Option 2 in the SWB question. The joint frequencies are reported in

four cells (for the between-subjects scenarios in Study 2, the (separate) distributions are reported in two cells).

For each study and scenario, the table reports in an additional cell the p-value from a two-sided nonparametric

equality-of-proportions test, comparing the proportion of subjects who favored Option 1 in the choice question

with the proportion who favored Option 1 in the SWB question. In studies and scenarios that employ a within-

subject design, the test is the paired Liddell exact test, while in the between-study scenarios in Study 2, the test

is the unpaired Fisher exact test.12

Examining Table 2‘s leftmost column (titled ―Income vs. Sleep‖) in Study 1 illustrates some of our main

findings. First, the top two cells reveal that 87 percent of respondents predicted that they would choose the same

option that they predicted would be better in terms of SWB. This evidence can be interpreted as suggesting

that—at least in the current context—the SWB question is a good predictor of the choice question. Second, the

bottom two cells show that a non-trivial 13 percent of our respondents predicted a choice that does not align

with their predicted SWB under the alternative options.

A quick look through the rest of the columns in Study 1 verifies that Scenario 1 is rather typical in this

respect: in the rest of the scenarios, 10 to 19 percent of respondents‘ predicted choice does not align with their

predicted SWB difference. This evidence is consistent with the idea that some respondents are willing to trade

off SWB for other aspects of their lives. Of course, this evidence is also consistent with alternative explanations

(e.g. measurement error). These are addressed below.

In the rest of this section we analyze in detail the response patterns above, and show that they cannot be

entirely accounted for by explanations that focus on measurement error, special features of the questionnaires,

the survey design, or the sample. Specifically, we rule out classical measurement error as an explanation of our

findings, and provide evidence that respondents reacted systematically differently to the choice vs. the SWB

questions and that they carefully read the details in the scenarios. We further examine the replicability of our

results in a between-subject design and, separately, in a nationally-representative sample. We also examine

different versions of the SWB question, and show that the exact SWB measure used makes a difference and that

some measures might predict choice better than others. Finally, we explore how subjects‘ choices are related to

their rankings of the options in terms of non-SWB aspects of life. We find that (a) the other aspects statistically

11

Non-response was generally low. In the CNSS, less than 5% of respondents answered ―Do not know‖ or refused to answer on any

given question. In the Cornell sample, non-response was rare; virtually all questions had a non-response rate less than 2%. Only one

Cornell response was excluded due to obvious confusion with instructions. Due to the less-structured sampling method used in our

doctor‘s office sample, some questions from those studies had non-response rates as high as 20%. However, the majority of this non-

response is driven by respondents being called in for their appointments, alleviating concerns of selection bias. 12

In study 4, where respondents could indicate SWB indifference, responses indicating indifference were dropped from the tests.

13

significantly help predict choice beyond SWB, and (b) subjects‘ rankings of the options in terms of SWB far

outperforms their ranking in terms of other aspects for predicting choice.

III.A. Evidence on systematically different responses to the choice vs. the SWB questions

Our first test is a nonparametric equality-of-proportions test. It tests the null hypothesis that respondents

answered the SWB question the same way they answered the choice question. That is, we test the null

hypothesis that the ―treatment effect‖ of asking the choice question instead of the SWB question is equal to

zero. Looking at the Liddell test row for Study 1 in Table 2 reveals that the null is strongly rejected in five out

of the six scenarios. Indeed, that the proportion of subjects who preferred Option 1 choice-wise is, in most

scenarios, different from the proportion of subjects who preferred Option 1 SWB-wise is easily seen by

comparing study 1‘s ―Higher SWB: Option 1/ Chosen: Option 2‖ cells with the ―Higher SWB: Option 2/

Chosen: Option 1‖ cells. This finding suggests that the mean preference ranking across the two options differs

from the mean SWB ranking, but it is also consistent with the possibility that the measurement error (i.e., the

noise component of responses) has different variance for the choice question than for the SWB questions.

III.B. Evidence on systematic reactions to scenario content

Having established that our respondents respond differently to the choice vs. the SWB questions, we

now proceed to show that these different responses react in systematic (and expected) ways to the explicit

figures—e.g. dollar amounts—in the scenario questions.

To check whether participants responded sensibly to the questions, in Study 2—where subject

population was the same as that of Study 1—we asked four different versions of Scenario 4 (career legacy

versus income). In all versions, the annual income for the alternative of being an artist who will have a lasting

legacy was held constant at $40,000. The only difference across versions was the annual income for the

alternative of being a commercial artist, which equaled $42,000, $60,000, $80,000, or $100,000.

We found that reply patterns responded to annual income levels in the expected direction. For example,

at $42,000, 20% of respondents predicted they would choose to be a commercial artist and 19% predicted it

would make them more satisfied with life, compared to 71% and 63% at $100,000, respectively (p < 0.001).

Figure 1 shows that participants are monotonically more likely to predict choosing being the commercial artist,

as well as to predict that the alternative of being a commercial artist would lead to greater SWB, when annual

income is higher, consistent with non-random responding to the decision scenarios. Furthermore, importantly,

23 percent of the respondents predicted a choice that would not lead to greater SWB in at least one of the four

annual income levels above (including 3 percent who behaved this way in all four income levels). This finding

would be implied by a model in which different individuals are inconsistent between choice and SWB at

different income levels: as income for the commercial artist option increases, more subjects switch toward the

14

commercial artist option, but it takes a higher income for most subjects to rank that option as giving higher

SWB than it does to induce them to choose it. However, we cannot rule out the possibility that different

individuals exhibit choice-SWB inconsistencies at different income levels simply because measurement error

(in choice ranking, SWB ranking, or both) is drawn independently for each of the four income-level questions.

III.C. Evidence from within- vs. between-subject design

As mentioned above, the rest of the scenarios in Study 2 were conducted in a between-subject design:

while half the respondents only answered choice questions regarding the scenarios, the other half answered only

SWB questions (specifically, life satisfaction questions; we return to this point below).13

This was done to

verify that our findings in Study 1 are not entirely due to the within-subject design. Our main worry was that

subjects who are presented with both a choice and a SWB question might attempt—consciously or not—to

respond in what they perceive as a consistent way.14

Looking at the results for Study 2 in Table 2 reveals that our worry was not justified. In the three

scenarios that were kept identical between Study 1 and Study 2, we cannot reject cross-study equality of

proportions of subjects who preferred Option 2 in either the SWB or the choice question in five out of six cases

(see the third and fourth rows in study 2). And although our confidence intervals are too wide to allow us to

reject small differences, in two out of the three scenarios results in Study 2 look almost identical to those in

Study 1. Scenario 1 is an example: while in Study 1, 30 and 41 percent of respondents ranked Option 2 higher

on SWB and choice respectively, in Study 2 the corresponding percentages are 34 and 44. Equality of

proportions between the choice and SWB question is rejected in both studies (by the Liddell and Fisher tests,

respectively), although the p-value is larger (and the sample size smaller) in Study 2. Another example is

Scenario 3, where we cannot reject equality of proportions (choice vs. SWB) in either Study 1 or Study 2, and

where Option 2 is favored in the SWB and choice questions by 48 and 46 percent of respondents in Study 1 and

by 51 and 48 percent in Study 2.

On the other hand, SWB proportions in Scenario 13 change dramatically, from 36 percent in Study 1 to

54 percent in Study 2 (while choice proportions change from 48 to 53 percent, a difference that is not

statistically significant). This means that while equality of proportions between the choice and SWB questions

in scenario 13 can easily be rejected in Study 1, it cannot be rejected in Study 2. Overall, then, we interpret

Study 2 results as strongly rejecting the hypothesis that the within-subject design of Study 1 creates artificial

consistency in answering the two questions. Indeed our evidence may suggest that if anything, it is possible that

13

Notice that Scenario 4, which was presented in Study 2 four different times in a within-subject design, was presented only at the end

of the questionnaire (see the related footnote in Table 1 regarding the scenario order in that study). Thus, when responding to the other

scenarios’ questions, Study 2 subjects could not have imagined that they were answering only one of a pair of possible questions,

maintaining the integrity of the between-subject design. 14

As discussed in Section IV below, in Study 4 we explicitly asked respondents, at the end of the questionnaire, whether they thought

they had included such consistency considerations when answering the scenario questions. See discussion there.

15

replies in the between-subject design were more consistent. This interpretation may be consistent with Study 2‘s

results for additional two scenarios (2 and 12), which differed from their versions in Study 1 and hence cannot

be compared across studies, but failed to produce statistically significant proportion differences between the

questions.

III.D. Evidence from a national sample

In Study 3, we used Cornell‘s National Social Survey (CNSS) to present Scenario 1 to a nationally

representative sample. Looking at Study 3‘s results in Table 2 suggests that although the proportions in the four

cells differ from those in the two studies above—which is not surprising, given the different populations—our

main results are not driven by peculiar features of the respondent population in Studies 1 and 2. As in Study 1

above, a large share of subjects gave the same answer to the choice and the SWB questions (in fact, the share of

these subjects grew from 87 percent in Study 1 to 92 percent in Study 3). And, also as above, respondents who

answered the questions differently have done so in the same systematic way in which they have in Study 1:

almost all of them predicted that while Option 1 would be better for SWB, they would choose Option 2. This is

shown statistically by a strong result in the Liddell test.

III.E. Evidence regarding specific SWB question framing/language

As discussed above, participants‘ predictions of their SWB are in general highly predictive of their

choice. Averaging across all of our surveys, participants, scenarios, and SWB questions, 86 percent of the time

participants predict making the choice they predict will generate greater SWB, which is statistically

distinguishable from 50 percent (one-sample difference-of-proportions test statistic=59.53, p = 0.000)

Table 3 provides regression evidence on the relationship between predicted SWB and choice within each

survey. The left panel of the table displays probit regressions with scenario fixed effects, where the dependent

variable is a dummy for choice of one of the alternatives, and the independent variable is a dummy for

predicting higher SWB in that alternative. The right panel displays OLS regressions where the dependent

variable is the participant‘s intensity of choice (on a 6-point scale, demeaned at the scenario level) in favor of

one of the alternatives, and the independent variable is the participant‘s rating of the likelihood of being happier

in that alternative (also on a 6-point scale). The regressions are conducted separately for each of the five

versions of the SWB question (see Section II.B. (c) for details) and, in study 4, separate regressions are run on

isolated vs. in series SWB measures. Finally, since Study 3 included only Scenario 1 (―Income vs. Sleep‖), for

ease of comparison the table reports results both for all scenarios together, and for Scenario 1 separately.

Looking at the results for Study 1, Table 3 reveals that the isolated SWB measures (i), (ii), and (iii)

differ from each other in how predictive they are of the choice question. For example, looking at the

(demeaned) OLS regressions, where R2‘s are accounted for by only the SWB measure, the regressions for study

16

1 suggest that the (isolated) life satisfaction measure (i) is statistically significantly a better predictor of choice

(with R2 = 0.65, s.e. = 0.03 for all scenarios, and R

2 = 0.74, s.e. = 0.06 for scenario 1) than the (isolated) felt

happiness measure (iii) (R2 = 0.55, s.e. = 0.03 for all scenarios, and R

2 = 0.60, s.e. = 0.06 for scenario 1).

15 At

the same time, evidence for the happiness with life as a whole measure (ii) varies a little more. Compared with

the felt happiness measure, it is a better predictor in the OLS regressions than it is in the probit regressions (but

remember that the probit regressions for all scenarios include scenario fixed effects). Across studies, however,

when looking at only scenario 1, happiness with life as a whole is a better predictor in study 3 than in study 1,

although the difference is not statistically significant.

III.F. Evidence regarding isolated vs. in-series SWB measures

Having found that the language of the SWB measure could matter, we explore, in study 4, whether the

position and context of the SWB measure could matter as well. Looking at study 4‘s OLS regression results in

Table 3 reveals that in that study, the two SWB measures we use—own happiness with life as a whole (iv), and

immediately felt own happiness (v)—explain the same amount of choice variation (R2 = 0.43) when asked in

isolation, but possibly explain different amounts (which are slightly below the above R2) when asked in series.

To further explore this, we aggregate across the two own happiness measures (iv) and (v) and run OLS

regressions (not reported) separately for an isolated happiness measure (when happiness is the only

characteristic of the alternatives rated by participants), a first-in-series measure (when happiness is the first in a

list of characteristics rated by participants), and a last-in-series measure (when happiness is listed last). For

comparing R2‘s, we calculate bootstrapped standard errors, based on 100 resamplings of the data.

We find that own happiness is most predictive of choice when asked in isolation (R2 = 0. 43, s.e. = 0.02).

When asked first-in-series, R2

is 0.34 (s.e. = 0. 04), compared to 0.37 (s.e. = 0.04) as last-in-series. Although

statistically, these differences in R2 are not significant, our finding that an isolated own happiness measure

seems to predict choice no worse than an in-series measure may suggest that respondents‘ interpretation of

SWB measures might be context-dependent. For example, it is possible that when the SWB question is

presented in isolation, respondents interpret it as aiming at a more inclusive notion of SWB, whereas when it is

presented as part of a series of questions regarding other aspects of life, it is interpreted as a ―stripped down‖

notion of SWB, or as ―SWB controlled for other aspects of life.‖

III.G. Evidence that aspects of life other than SWB help predict choice

Finally, Tables 4-8 analyze our findings from Study 4. In that study we elicited from respondents the

relative rankings of the two options of the scenarios in terms of a list of twelve aspects of life, only one of

15

The standard errors we report above for R2‘s are not reported in the table. They are calculated using bootstrapping.

17

which was a SWB measure (―own happiness‖). The aspects-of-life measures came in two versions—a ―life as a

whole measure‖ and an ―immediately felt‖ version, so that the ―own happiness‖ measure came in two versions,

corresponding with SWB measures (iv) and (v) in II.B(c) above. Since these SWB measures specifically

measure own happiness, in this subsection we interchangeably refer to them as measures of SWB or as

measures of happiness.

Table 4 and Table 5, present evidence on the additional predictive power of aspects other than

happiness, and the comparative ability of each aspect to explain variation in choice. The dependent variable is

intensity of choice (the choice rating on a 6-point scale), and the dependent variables are predicted happiness

and other aspects of life (on a 7-point scale). The top panel estimates regressions of choice on happiness and

each other aspect individually. The bottom panel estimates regressions of choice on each aspect by itself. Data

are demeaned at the scenario level to control for scenario-specific fixed effects.

The two tables are identical in structure, and while Table 4 shows data for the isolated SWB measures

(see II.B(d)), Table 5 shows them for in-series measures of SWB. Since typical SWB survey-based data are

collected using isolated SWB measures, the results in Table 4 answer the question: ―By how much can we

improve the typical isolated SWB measures as predictors of choice?‖ At the same time, the results in Table 5

answer questions such as ―If happiness were measured in series along with measures of other aspects of life,

would it become a better or a worse predictor of choice compared with its isolated version?‖ and ―Will it

outperform in-series measures of other aspects of life?‖

Looking at the leftmost column of Tables 4 and 5 (in either panel, which for this column is identical)

reveals that own happiness is a strong and statistically significant predictor of choice. These univariate

regressions‘ R2 can directly be interpreted as a measure of the predictive power of the regressor. In the leftmost

column, it is 0.43 in Table 4, and 0.36 in Table 5, which suggests that an isolated version of own happiness is a

better predictor of choice than its in-series equivalent. The percentage of the variation in choice explained by

SWB is significantly lower in study 4 compared to studies 1 and 3. This difference might be due to one of

several factors, such as the different response scale used in study 4 (allowing indifference), the different sample

populations, or the different scenarios asked across different studies. These latter two possibilities are addressed

below, in sections IV.D and IV.E.

Each one of the rest of the columns of panel (a) (in either table) adds one in-series measure of an aspect

of life other than own happiness to the regression. The results reveal two interesting patterns. First, non-SWB

aspects of the options seem to matter. Most of the aspects—family happiness, health, control over life, life‘s

level of spirituality, social status, physical comfort, and sense of purpose—are statistically significantly

predictive of choice, controlling for either isolated own happiness (Table 4) or in-series own happiness (Table

5). Second, even when an additional aspect is highly statistically significant, R2s do not improve much. Looking

18

at the ―Incremental R2‖ column in panel (a) in either table shows that R

2‘s improve by no more than 0.02 (for

sense of purpose in Table 4, and for control over life in Table 5).

Looking at the rest of the columns at the bottom panel, all of the aspects are highly predictive of choice.

However, in terms of predictive power, own happiness is unambiguously superior to the rest on the list. In

Table 5, where all twelve aspects are measured in-series and hence are most directly comparable, control over

life is the aspect of life that has the highest R2 after own happiness, at 0.14, followed by sense of purpose (R

2 =

0.12), life‘s level of fun (R2 = 0.11), and life‘s non-boringness (R

2 = 0.11). In other words, we find that as a

single predictor of choice, among in-series measures, own happiness outperforms the rest by a wide margin.

While Tables 4-5 analyze how each of the non-happiness aspects compares with happiness individually,

Tables 6-8 break down the results by scenario, and analyze the relative explanatory power of happiness and the

non-happiness aspects when all are controlled for simultaneously. The three tables are again identical in

structure, and while Table 6 pools all the data, Tables 7 and 8 separate the results for isolated happiness

measure and for in-series happiness measure, respectively.

As is evident from, e.g., Table 8 (which can be compared with Table 5), the coefficient on happiness is

somewhat smaller when all of the aspects are controlled for. Nonetheless, it remains by far the most predictive

of choice, as assessed either by magnitude of coefficient or by the incremental R2 from adding all of the non-

happiness aspects to the happiness-only regression. Since subjects‘ rankings of the options in terms of one

aspect are correlated with their rankings in terms of other aspects, the aspect coefficients are quite different in

the fully-controlled regressions (Tables 7-8) compared with the one-aspect-at-a-time regressions (Tables 4-5).

This however does not change our main findings above. Far example, examining the results pooled across

scenarios in column 1 (Table 8), ―control over your life‖ has the second-largest coefficient after happiness. The

next most highly predictive aspects, not statistically different from control over life, are social status and sense

of purpose. These aspects‘ coefficients are well below the coefficient on own happiness.

IV. Robustness and Additional Results

In this section, we discuss supplementary analyses that are helpful for interpreting and elaborating on

our main findings. We outline the theory underlying these additional tests, describe the survey instruments used

to carry them, and report our findings.

IV.A. Participants’ interpretations of their own behavior

We would like to infer from discrepancies between participants‘ choices and their predicted SWB that

factors other than SWB enter preferences. However, a concern in any survey is that participants may interpret

the questions they are answering differently than we the researchers interpret them, potentially invalidating our

inferences. To obtain evidence about participants‘ own interpretation of their behavior, we asked them what

19

they were thinking when their choice differed from the option they predicted would maximize SWB.

Specifically, after participants finished all the decision scenarios, but before the WTP questions, we asked:

At times in the earlier scenarios, you might have chosen an alternative which you did not think would

make you happier.

If you made such a choice, do you believe it was a mistake? That is, if you could go back and change your

answer now, would you want to? (Please circle one).

YES / NO / I never made this kind of choice.

If you made such a choice, do you think you would regret it? That is, if your chosen alternative actually

occurred, do you think you would later wish you had instead chosen the other option? (Please circle one).

YES / NO / I never made this kind of choice.

If you made such a choice, please explain your reasoning: [blank space]

Most subjects think of their choice-SWB discrepancies as intentional. In response to the question whether such

a discrepancy ―was a mistake,‖ 7% said yes, 73% said no, 19% said they never made such a choice, and 1% did

not respond. In response to the question whether they ―would regret‖ such a discrepancy, 23% said yes, 57%

said no, 19% said they never made such a choice, and 1% did not respond. Moreover, our conclusions from

Section III remain essentially the same if we exclude participants who believed they were making a mistake or

believed they would regret their choices. The first column of Table 9 repeats our preferred specification from

Table 6, while the second and third column compares the results when we exclude ―mistake‖ subject and

―regret‖ subjects, respectively.

IV.B. Self-control problems

A related concern is that these discrepancies may reflect a self-control problem (e.g., as in Laibson,

1997; O‘Donoghue and Rabin, 1999), rather than a preference for non-SWB aspects of the options. In most of

our decision scenarios, it is not obvious how a self-control problem would be implicated, but there are a few

where it is. For example, in one decision scenario, one of the alternatives was staying out later with friends,

while the other was going to bed earlier to feel better and be more productive the next day. A participant who

correctly anticipates having a self-control problem might say she would choose to stay out late, even though her

welfare would be maximized by going to bed earlier.

In order to assess the potential importance of self-control problems, in some versions of the survey, in

addition to asking participants what they would choose, we asked them what they would want themselves to

choose (their meta-preference). We reasoned that if a participant preferred Option 1 but would choose Option 2

due to a self-control problem, the participant would indicate a greater likelihood of choosing Option 1 but a

20

meta-preference for Option 2. Aggregating across all surveys where we asked about meta-preferences, we

found discrepancies between choice and meta-preference in 25% of cases.

However, while self-control problems may explain some of subjects‘ choices, our main results from

Section III appear to be robust when we exclude these observations. The fourth column of Table 9 shows that

when we regress choice on SWB and all the non-SWB aspects, the results for the sample restricted to subjects

whose preference and meta-preference always coincide are qualitatively the same as the results for the full

sample.

IV.C. Artificial consistency between choice and SWB responses

One explanation for why SWB predicts choice so well is that subjects think they ought to give the same

answer to the two questions. In Section III.C, we discussed evidence from a between-subjects study (Study 2),

which suggests that this kind of ―artificial consistency‖ cannot fully account for our results. In Study 4, we also

collected additional data relevant for evaluating the importance of ―artificial consistency‖ in our studies. After

participants finished all the decision scenarios, we asked them:

Throughout this survey, we asked you to choose between two alternatives, and we also asked you to rate

the options in terms of how they would affect various aspects of your life. When you made these ratings,

were you trying to make your ratings consistent with what you chose? (Please circle one.)

(A) I rated the aspects independently from what I chose.

(B) When I rated the aspects, I tried to be unaffected by what I chose, but I was probably affected to some

extent.

(C) When I rated the aspects, I tried to be consistent with what I chose.

(D) Other (Please specify) ________________________

In this question, 30% of our respondents answered (A), 49% answered (B), 20% answered (C), and 1% did not

respond. These responses suggest that an attempt to be consistent may have consciously or (perhaps more

commonly) unconsciously played a role in subjects’ responses. Nonetheless, the fifth column of Table 9 shows

that our results from Table 6 remain qualitatively very similar when we exclude from the analysis all subjects

who did not respond either (A) or (B). This evidence reinforces the impression from Section III.C that artificial

consistency does not drive our main conclusions.

IV.D. Heterogeneity across decision scenarios

In a regression of choice on SWB and non-SWB aspects, the coefficients may vary depending on which

scenarios we presented to participants. Here we examine three dimensions of heterogeneity.

Local vs non-local tradeoffs. As outlined in Section I, the interpretation of the regression coefficients as

capturing marginal utilities assumes a linear approximation to the utility function. This approximation will be

21

more valid to the extent that the options in the decision problem involve small changes from the subject‘s status

quo. In Table 6, we label three of the scenarios as containing ―non-local tradeoffs.‖ These are all cases where

one of the options asked the subject to imagine having a particular level of income, which could well be quite

far from the subject‘s current income. In Table 10, we compare the regression coefficients for the scenarios with

non-local tradeoffs with the regression coefficients for the full sample. The first column of Table 10 shows the

regression of choice on SWB and the non-SWB aspects (repeating our preferred specification from Table 6),

while the second column restricts the sample to scenarios with non-local tradeoffs. Despite the theoretical

possibility of different results, the coefficient estimates are in fact for the most part quite similar.

Representative scenarios. When we started this project, some of us were unsure whether we would find

any discrepancies between predicted happiness and choice. In order to maximize the chance we could detect a

discrepancy, we chose our initial scenarios with a happiness tradeoff in mind. For example, the artist scenario

was designed to generate a tradeoff between happiness and a posthumous legacy that may contribute little to

happiness during life.

Once we found clear discrepancies between predicted happiness and choice, we became interested in

estimating the role of predicted happiness in the most empirically-relevant settings. To address this question,

we would ideally ask participants about a representative sample of decisions they face. We generated some

decision scenarios that we believed were representative, such as a basic tradeoff between time and money

(scenario 6). As discussed in Section II.A, we also generated decision scenarios that are common in our

participants‘ lives by asking students at the University of Chicago and at Cornell University to tell us about

some of the most important decisions they face.

The third column of Table 10 restricts the sample to scenarios chosen to be more representative of

important decisions for college students. Despite our concern that we initially chose scenarios that highlighted a

tradeoff with SWB, the regression coefficient on SWB is actually smaller in the representative scenarios than in

the full sample (the first column).

Everyday, minor decisions. From a theoretical perspective, our findings indicate that while SWB (as it is

typically measured in applied work) is a major component of preferences, it is not a direct measure of utility.

However, from a practical perspective, an important question is whether the marginal utility of happiness varies

across scenarios. If so, then empirical work that uses measured SWB as a proxy for utility is more valid in

scenarios where the marginal utility of happiness is high. On the other hand, in decision situations where the

marginal utility of happiness is low, welfare conclusions based on analyses of measured SWB may be

misleading. A hypothesis we can test with our data is whether SWB plays a more important role in everyday,

minor decisions, where other factors such as sense of purpose are simply not relevant. In Table 6, we categorize

six of our decision scenarios as being ―everyday/minor decisions.‖ The fourth column of Table 10 repeats the

regression from Table 6, but with the sample restricted to observations from decision scenarios featuring

22

everyday/minor decisions. We do indeed find that SWB has a higher coefficient here than in the full sample.

Nonetheless, we continue to find other aspects as statistically significant predictors of choice—including sense

of purpose and control over life, both of which one might have expected to be irrelevant in this kind of decision.

IV.E. Preference heterogeneity across individuals

Our main results can be interpreted as an examination of average or representative preferences; here we

examine the degree of heterogeneity seen across individual preferences. We regressed 6-point choice on 6-point

SWB, subject-level characteristics, and interactions between SWB and subject-level characteristics (results not

shown). These subject-level characteristics were age, gender, log income, race dummies, and—only in Study 4,

where we administered a personality questionnaire—scores on the ―Big 5‖ personality traits. None of the

demographic variables had statistically significant coefficients. Of the personality traits, only neuroticism had a

significant coefficient; a higher rating of neuroticism is associated with lower predictive power of SWB ratings.

V. Conclusion

In a variety of decision scenarios and with a variety of subject populations, we test whether individuals

think they would choose—or would like to choose—the same option they predict will maximize SWB. Our

findings suggest that choice and SWB are conceptually distinct, and that SWB appears to be only one among

several arguments in the utility function. However, SWB appears to be a uniquely important argument in the

utility function that predicts choice far better than other factors that enter preferences.

23

References

Becker, Gary S., and Luis Rayo. 2008. ―Comment on ‗Economic growth and subjective well-being: Reassessing

the Easterlin Paradox‘ by Betsey Stevenson and Justin Wolfers.‖ Brookings Papers on Economic

Activity, Spring, 88-95.

Bentham, Jeremy. 1789. An Introduction to the Principle of Morals and Legislations.Bernheim, Douglas B.

2009a. ―Behavioral Welfare Economics.‖ Journal of the European Economics Association, 7(2-3), 267-

319.

Bernheim, Douglas B. 2009b. ―Behavioral Welfare Economics.‖ Presentation at the First Annual Behavioral

Economics Conference, University of California--Berkeley, May 22.

Blanchflower, David G. and Andrew J. Oswald. 2004. ―Money, Sex and Happiness: An Empirical Study.‖

Scandinavian Journal of Economics, 106(3), 393-415.

Bronsteen, John, Christopher J. Buccafusco, and Jonathan S. Masur. Forthcoming. ―Well-being analysis.‖

Georgetown Law Journal.

DiTella, Rafael, Robert J. MacCulloch, and Andrew Oswald. 2003. ―The Macroeconomics of Happiness.‖

Review of Economics and Statistics, 85(4), 809-827.

Ellison, Christopher G. 1991. Religious involvement and subjective well-being. Journal of Health and Social

Behavior,32, 80–99.

Finkelstein, Amy, Erzo F.P. Luttmer, and Matthew J. Notowigdo. 2008. ―What good is wealth without health?

The effect of health on the marginal utility of consumption.‖ NBER W.P. 14089, June.

Gilbert, Daniel T. (2006). Stumbling on happiness. New York: Knopf.

Gruber, Jonathan, and Sendhil Mullainathan. 2002. ―Do cigarette taxes make smokers happier?‖ NBER W.P.

8872.

Heffetz, Ori, and Robert H. Frank. Forthcoming. ―Preferences for Status: Evidence and Economic

Implications.‖ Handbook of Social Economics, Jess Benhabib, Alberto Bisin, Matthew Jackson, eds.,

Elsevier.

John, Oliver P., and Sanjay Srivastava. 1999. The Big Five trait taxonomy: History, measurement, and

theoretical perspectives. In Lawrence A. Pervin and Oliver P. John (Eds.), Handbook of personality:

Theory and research (2nd ed., pp. 102-138). New York: Guilford.

Laibson, David I. 1997. ―Golden eggs and hyperbolic discounting.‖ Quarterly Journal of Economics, 112(2),

443-477.

Layard, Richard (2005). Happiness. Penguin Press.

Luttmer, Erzo F.P. 2005. ―Neighbors as Negatives: Relative Earnings and Well-Being.‖ Quarterly Journal of

Economics, 120(3), 963-1002.

24

Kahneman, Daniel, Alan B. Krueger, David A. Schkade, Nobert Schwarz, and Arthur A. Stone. 2004. ―A

survey method for characterizing daily life experience: The Day Reconstruction Method.‖ Science, 306,

1776-1780.

Kahneman, Daniel, Peter P. Wakker, and Rakesh Sarin. 1997. ―Back to Bentham? Explorations of experienced

utility.‖ Quarterly Journal of Economics, 112(2), 375-405.

Kelly, William. 2004. ―Sleep-Length and Life Satisfaction in a College Student Sample.‖ College Student

Journal, 38(3).

Kimball, Miles S., Helen Levy, Fumio Ohtake, and Yoshiro Tsutsui, 2006. ―Unhappiness After Hurricane

Katrina.‖ University of Michigan mimeo.

Kimball, Miles S., and Robert J. Willis. 2006. ―Happiness and Utility.‖ University of Michigan mimeo.

Koszegi, Botond, and Matthew Rabin. 2008. ―Choices, Situations, and Happiness.‖ Journal of Public

Economics, 92, 1821-1832.

Nozick, Robert. 1974. Anarchy, State, and Utopia. New York: Basic Books.

Nussbaum, Martha. 2000. Women and Human Development: The Capabilities Approach. Cambridge:

Cambridge University Press.

O‘Donoghue, Ted, and Matthew Rabin. 1999. ―Doing it now or later.‖ American Economic Review, 89(1), 103-

124.

Oreopoulos, Philip. 2007. ―Do dropouts drop out too soon? Wealth, health and happiness from compulsory

schooling.‖ Journal of Public Economics, 91, 2213–2229.

Schwarz, Norbert, and Gerald L. Clore. 1983. ―Mood, misattribution, and judgments of well-being: Informative

and directive functions of affective states.‖ Journal of Personality and Social Psychology, 45(3), 513-

523.

Sen, Amartya. 1985. Commodities and Capabilities. Oxford: Oxford University Press.

Stevenson, Betsey and Justin Wolfers. 2008. ―Economic Growth and Subjective Well-Being: Reassessing the

Easterlin Paradox.‖ NBER Working Paper No. 14282.

Stutzer, Alois and Bruno S. Frey, 2008. ―Stress that Doesn't Pay: The Commuting Paradox,‖ Scandinavian

Journal of Economics, 110(2), 339-366.

Stutzer, Alois. 2007. ―Limited Self-Control, Obesity and the Loss of Happiness,‖ IZA Discussion Papers 2925,

Institute for the Study of Labor (IZA).

Tversky, Amos, and Dale Griffin. 1991. ―Endowments and Contrast in Judgments of Well-Being.‖ .‖ In

Richard J. Zeckhauser, Strategy and Choice. Reprinted in Choices, Values, and Frames.

White, Mathew P., and Paul Dolan. 2009. ―Accounting for the richness of daily activities.‖ Psychological

Science, 20(8), 1000-1008.

25

Figure 1.

26

Table 1: Study-specific Information Study 1 Study 2 Study 3 Study 4

Sample Population Doctor‘s Office Doctor‘s Office CNSS Cornell Students

N 497 569 1000 214

Scenarios Presented

(See Appendix for text of

scenarios)

1, 3, 4, 11, 12, 13 1, 2, 3, 4, 12, 13 1 1-10

Choice and SWB variables†

Sample size of each cell in parenthesis

Life Satisfaction (i)

Isolated (164) (569)

Happiness with Life as a

Whole (ii)

Isolated (162) (1000)

Felt Happiness (iii)

Isolated (171)

Own Happiness with Life

as a Whole (iv)

Isolated (52)

In Series (52)

Immediately Felt Own

Happiness (v)

Isolated (55)

In Series (55)

Choice Question Format 6-point 6-point Binary 6-point

Meta-Choice Question? Yes Yes No Yes

Choice vs SWB: Within- or

Between- Subject Design

Within Between Within Within

Randomizations

Scenario order 4-1-11-12-13-3 1-2-12-13-3-4 1 1, 2, …, 9, 10

3-13-12-11-1-4 3-13-12-2-1-4‡

Choice Question order Choice-Meta-SWB Two opposite

orderings of

aspects of life SWB-Choice-Meta

SWB asked in isolation or

in series

SWB asked in

series

SWB asked in

isolation Notes: The different variations between studies 1-4 are presented above. In the first panel of the table, the number of responses in each cell is indicated in parenthesis.

Refer to section II.B for the text of the choice question and meta-choice question. ―Choice vs SWB: Within- or Between- Subject Design‖ indicates whether each subject was asked to predict both choice and SWB for a given scenario (the within condition), or only one of the two (the between condition). The sequence of numbers

under question order corresponds to the numbering of the scenarios found in the appendix. † The exact phrasing of SWB questions i-v above is explained in section II.B.

‡ Question 4 is asked last in both question orders because it is the only question in study 2 where we ask both choice and predicted SWB. In order to have a clean

between-subject design, we did not want subjects to know we were interested in both choice and SWB until after subjects were done with the rest of the questions.

27

Table 2: Summary of Scenario Responses Scenario Number (For Phrasing, see Appendix)

1 2 3 4 5 6 7 8 9 10 11 12 13 Tradeoff implicated in

decision problem: Income

vs

Sleep

Concert vs

Dinner

Abs. Inc. vs

Rel. Inc.

Income vs

Legacy

Apple vs

Orange

Income vs

Labor

Socialize vs

Sleep

Income vs

Dinner

Education vs

Social life

Interest vs

Career

Concert vs

Duty

Rent vs

Commute

Friends vs

Income

STUDY 1 Version 1 Version 1

Higher SWB: Option 1 Chosen: Option 1

58% 48% 24% 16% 52% 50%

Higher SWB: Option 2 Chosen: Option 2

29% 42% 60% 65% 32% 34%

Higher SWB: Option 1 Chosen: Option 2

12% 4% 14% 7% 5% 14%

Higher SWB: Option 2 Chosen: Option 1

1% 6% 2% 12% 11% 2%

p-value of Liddell Exact Test

0.000 n = 425

0.35 n = 420

0.000 n = 422

0.024 n = 422

0.002 n = 425

0.000 n = 422

STUDY 2 Version 2 Version 2

Higher SWB: Option 2 34% 86% 51% 54% 54%

Chosen: Option 2 44% 84% 48% 55% 53%

p-value of difference in proportion choosing option

1 between studies 1 and 2

0.430 0.640 --- --- 0.212

p-value of difference in

proportion with higher

SWB under option 1

between studies 1 and 2

0.398 0.342 --- --- 0.000

Higher SWB: Option 1 Chosen: Option 1

41%

Higher SWB: Option 2 Chosen: Option 2

44%

Higher SWB: Option 1 Chosen: Option 2

11%

Higher SWB: Option 2 Chosen: Option 1

4%

p-value of Fisher Test 0.020 n = 525

0.716 n = 524

0.433 n = 525

0.930 n = 526

0.793 n = 525

p-value of Liddell Exact

Test

0.000

n = 2852

28

Scenario Number (For Phrasing, see Appendix)

1 2 3 4 5 6 7 8 9 10 11 12 13

Tradeoff implicated in

decision problem:

Income

vs

Sleep

Concert

vs

Dinner

Abs. Inc.

vs

Rel. Inc.

Income

vs

Legacy

Apple

vs

Orange

Income

vs

Labor

Socialize

vs

Sleep

Income

vs

Dinner

Education

vs

Social life

Interest

vs

Career

Concert

vs

Duty

Rent

vs

Commute

Friends

vs

Income

STUDY 3

Higher SWB: Option 1 Chosen: Option 1

74%

Higher SWB: Option 2 Chosen: Option 2

18%

Higher SWB: Option 1 Chosen: Option 2

7%

Higher SWB: Option 2 Chosen: Option 1

1%

p-value of Fisher Test 0.001 n = 1968

p-value of McNemar Test 0.000 n = 980

STUDY 4 Version 2

Higher SWB: Option 1 Chosen: Option 1

28% 23% 35% 40% 30% 39% 51% 62% 51% 28%

Higher SWB: Option 2 Chosen: Option 2

44% 44% 37% 28% 32% 26% 14% 15% 22% 31%

Higher SWB: Option 1

Chosen: Option 2

17% 10% 2% 12% 1% 2% 6% 14% 3% 29%

Higher SWB: Option 2 Chosen: Option 1

1% 3% 12% 8% 1% 10% 16% 4% 19% 3%

Percent Indicating Indifference for SWB

10% 20% 14% 12% 36% 23% 13% 5% 5% 9%

p-value of Fisher Test 0.000 n =402

0.167 n =380

0.033 n =393

0.362 n =397

0.912 n =345

0.025 n =368

0.013 n =395

0.003 n =409

0.001 n =411

0.000 n =401

p-value of McNemar Test 0.000 n =192

0.008 n =170

0.000 n =183

0.280 n =187

1.000 n =135

0.001 n =160

0.003 n =185

0.001 n =200

0.000 n =201

0.000 n =191

Notes: Response distribution by study and scenario. For the complete text of each scenario, see the appendix. If a questions phrasing changed meaningfully between surveys, the

version of the question is indicated in the first row of the study block. For between-subject comparisons, we report the Fisher Test p-value testing the null-hypothesis that mean

response to choice question = mean response to SWB question (an unpaired equality-of-proportions test). For within-subject, we report the analogous Liddell Exact Test p-value (a

paired equality-of-proportions test). In cases where respondents could indicate SWB indifference, responses indicating indifference were dropped from these tests. In study 2, we

report the p-value of fisher exact tests of the difference of proportions between SWB (or choice) responses in study 2 and SWB (or choice) responses in study 1 to provide a check

of cross-sample consistency and comparability. The values for the Income vs Legacy scenario in study 2 are pooled across the four payoff amounts faced.

29

Table 3: Regressions of Choice on SWB

Probit Estimates OLS Estimates

Study 1 Scenario 1: Income vs Sleep

Study 1 Study 4 Scenario 1: Income Vs Sleep

Study 1 Study 3 Study 1 Study 4

Life Satisfaction (i)

Isolated 0.78

(0.02)

R2 =0.53

n=424

0.88

(0.04)

R2 =0.73

n=141

0.81

(0.02)

R2 =0.65

n=424

0.88

(0.04)

R2 =0.74

n=141

Happiness with Life as a Whole (ii)

Isolated 0.70

(0.03)

R2 =0.39

n=411

0.70

(0.03)

R2 = 0.40

n=137

0.86

(0.02)

R2 =0.51

n=980

0.79

(0.02)

R2 =0.59

n=411

0.91

(0.05)

R2 =0.69

n=137

Felt Happiness (iii)

Isolated 0.70

(0.03)

R2 = 0.40

n=433

0.76

(0.05)

R2 = 0.42

n=147

0.73

(0.02)

R2 = 0.55

n=433

0.81

(0.05)

R2 = 0.60

n=147

Own Happiness with Life as a Whole (iv)

Isolated 0.59

(0.03)

R2 =0.43

n=519

0.46

(0.09)

R2 =0.32

n=52

In Series 0.52

(0.03)

R2 = 0.31

n=510

0.48

(0.09)

R2 = 0.36

n=51

Immediately Felt Own Happiness (v)

Isolated 0.58

(0.03)

R2 =0.43

n=538

0.58

(0.08)

R2 =0.50

n=54

In Series 0.59

(0.03)

R2 =0.39

n=549

0.70

(0.07)

R2 =0.68

n=55

Notes: Standard errors in parentheses. The left panel reports probit regression results, predicting the option chosen from the option with higher predicted SWB. The right panel

reports OLS regression results, predicting the intensity of preference on the intensity of predicted SWB. For both types of regressions, we report the estimates from all scenarios in

the first columns, and restricted to the income vs sleep question (question 1 in the Appendix), which was the only scenario asked in all of our surveys. * p < 0.05,

** p < 0.01,

*** p < 0.001

30

Table 4: OLS Regressions of Choice on Happiness and Each Aspect (Study 4, happiness measured in isolation) Dependent Variable: 6-point Choice Own happiness 0.576*** 0.550*** 0.558*** 0.560*** 0.567*** 0.546*** 0.562*** 0.557*** 0.559*** 0.579*** 0.565*** 0.536***

(0.020) (0.021) (0.021) (0.021) (0.021) (0.022) (0.020) (0.021) (0.021) (0.022) (0.021) (0.021)

Family happiness 0.131***

(0.028)

Health 0.103**

(0.033)

Life's level of romance 0.123***

(0.034)

Social life 0.0403

(0.028)

Control over your life 0.116***

(0.029)

Life's level of

spirituality

0.153***

(0.034)

Life's level of fun 0.0957***

(0.028)

Social status 0.103***

(0.027)

Life's non-boringness -0.0122

(0.027)

Physical comfort 0.0663*

(0.027)

Sense of purpose 0.156***

(0.025)

Observations 1057 1057 1057 1056 1053 1055 1053 1056 1056 1052 1056 1057

R2 0.43 0.44 0.43 0.44 0.43 0.44 0.44 0.44 0.44 0.43 0.44 0.45

Incremental R2 0.01 0.00 0.01 0.000 0.01 0.01 0.01 0.01 0.000 0.01 0.02

Notes: Standard errors in parentheses. OLS regression coefficients are reported. The dependent variable is 6-point intensity of choice described in section II.B.a, with a higher value indicating

a higher probability of choosing option 2. The dependent variables are the 7-point responses described in section II.B.b,c, with a higher number indicating a higher predicted value of that aspect under

option 2. All variables are demeaned at a scenario-level, equivalent to including scenario-specific fixed effects. Incremental R2 is calculated as (R

2 of choice on happiness and another

aspect)-( R2 of choice on happiness), capturing the additional predictive power added by the additional aspect.

* p < 0.05,

** p < 0.01,

*** p < 0.001

OLS regression of Choice on Each Aspect (Study 4, happiness measured in isolation) Dependent Variable: 6-point Choice Regressor Own

happiness

Family

happiness

Health Life‘s

level of

romance

Social life Control

over your

life

Life‘s

level of

spirituality

Life‘s

level of

fun

Social

status

Life‘s

non-

boringness

Physical

comfort

Sense of

purpose

Regression coefficient 0.576*** 0.319*** 0.342*** 0.333*** 0.222*** 0.367*** 0.333*** 0.291*** 0.233*** 0.220*** 0.249*** 0.340***

(0.020) (0.042) (0.054) (0.060) (0.044) (0.047) (0.061) (0.046) (0.040) (0.046) (0.036) (0.039)

Observations 1057 1062 1065 1066 1059 1064 1063 1063 1060 1060 1057 1063

R2 0.43 0.05 0.04 0.03 0.02 0.05 0.03 0.04 0.03 0.02 0.04 0.07

Notes: Standard errors in parentheses. Each column reports the OLS regression coefficients of 6-point intensity of choice on the 7-point rating of one aspect. All variables are

demeaned at a scenario-level, equivalent to including scenario-specific fixed effects. * p < 0.05,

** p < 0.01,

*** p < 0.001

31

Table 5: OLS Regressions of Choice on Happiness and Each Aspect (Study 4, happiness measured in series)

Dependent Variable: 6-point Choice Own happiness 0.555*** 0.527*** 0.525*** 0.540*** 0.542*** 0.495*** 0.538*** 0.536*** 0.531*** 0.527*** 0.519*** 0.508***

(0.023) (0.025) (0.025) (0.025) (0.025) (0.025) (0.024) (0.027) (0.024) (0.026) (0.025) (0.025)

Family happiness 0.097**

(0.033)

Health 0.124**

(0.038)

Life's level of romance 0.071

(0.044)

Social life 0.058

(0.037)

Control over your life 0.174***

(0.032)

Life's level of

spirituality

0.089*

(0.044)

Life's level of fun 0.051

(0.036)

Social status 0.132***

(0.032)

Life's non-boringness 0.082*

(0.035)

Physical comfort 0.129***

(0.032)

Sense of purpose 0.144***

(0.031)

Observations 1049 1048 1049 1046 1047 1048 1047 1049 1048 1049 1048 1048

R2 0.36 0.36 0.36 0.36 0.36 0.38 0.36 0.36 0.37 0.36 0.37 0.37

Incremental R2 0.00 0.00 0.00 0.00 0.02 0.00 0.00 0.01 0.00 0.01 0.01

Notes: Standard errors in parentheses. OLS regression coefficients are reported. The dependent variable is 6-point intensity of choice described in section II.B.a, with a higher value indicating

a higher probability of choosing option 2. The dependent variables are the 7-point responses described in section II.B.b,c, with a higher number indicating a higher predicted value of that aspect under

option 2. All variables are demeaned at a scenario-level, equivalent to including scenario-specific fixed effects. Incremental R2 is calculated as (R

2 of choice on happiness and another

aspect)-( R2 of choice on happiness), capturing the additional predictive power added by the additional aspect.

* p < 0.05,

** p < 0.01,

*** p < 0.001

OLS regression of Choice on Each Aspect (Study 4, happiness measured in series) Dependent Variable: 6-point Choice Regressor Own

happiness

Family

happiness

Health Life‘s

level of

romance

Social life Control

over your

life

Life‘s

level of

spirituality

Life‘s

level of

fun

Social

status

Life‘s

non-

boringness

Physical

comfort

Sense of

purpose

Regression coefficient 0.555*** 0.365*** 0.424*** 0.421*** 0.362*** 0.441*** 0.398*** 0.413*** 0.315*** 0.411*** 0.381*** 0.401***

(0.023) (0.037) (0.043) (0.049) (0.041) (0.033) (0.051) (0.037) (0.038) (0.037) (0.036) (0.034)

Observations 1049 1049 1050 1047 1048 1049 1048 1050 1049 1050 1049 1049

R2 0.36 0.09 0.09 0.07 0.07 0.14 0.06 0.11 0.06 0.11 0.10 0.12

Notes: Standard errors in parentheses. Each column reports the OLS regression coefficients of 6-point intensity of choice on the 7-point rating of one aspect. All variables are

demeaned at a scenario-level, equivalent to including scenario-specific fixed effects. * p < 0.05,

** p < 0.01,

*** p < 0.001

32

Table 6: Choice on Happiness and All Aspects, with Incremental R2 (Study 4, all data pooled)

Decision problem: Pooled 1 2 3 4 5 6 7 8 9 10

Tradeoff implicated in decision

problem:

Income

vs

Sleep

Concert

vs

Dinner

Abs. Inc.

vs

Rel. Inc.

Income

vs

Legacy

Apple

vs

Orange

Income

vs

Labor

Socialize

vs

Sleep

Income

vs

Dinner

Education

vs

Social life

Interest

vs

Career

Non-local tradeoff?

Everyday/minor decision?

Dependent Variable: 6-point Choice

Own happiness 0.487*** 0.463*** 0.543*** 0.538*** 0.469*** 0.779*** 0.587*** 0.322*** 0.510*** 0.361*** 0.343***

(0.018) (0.054) (0.054) (0.058) (0.058) (0.060) (0.066) (0.058) (0.057) (0.056) (0.059)

Family happiness 0.076** 0.024 -0.079 0.190* 0.068 0.314 0.139 -0.109 -0.005 0.176** 0.220**

(0.023) (0.053) (0.119) (0.081) (0.067) (0.216) (0.095) (0.091) (0.077) (0.064) (0.070)

Health 0.018 -0.088 -0.031 -0.033 0.148 -0.083 0.071 0.241* 0.091 -0.069 -0.215*

(0.030) (0.080) (0.122) (0.129) (0.100) (0.103) (0.120) (0.093) (0.103) (0.081) (0.102)

Life's level of romance 0.008 0.091 -0.042 0.232 -0.076 -1.532*** -0.159 -0.084 -0.028 -0.022 0.280*

(0.034) (0.094) (0.101) (0.136) (0.111) (0.379) (0.136) (0.102) (0.113) (0.090) (0.110)

Social life -0.045 -0.025 0.018 -0.105 -0.011 0.256 0.071 0.010 -0.017 -0.112 -0.137

(0.027) (0.086) (0.072) (0.093) (0.093) (0.304) (0.112) (0.105) (0.096) (0.088) (0.084)

Control over your life 0.079** -0.000 0.075 -0.093 0.142* 0.107 0.124 0.110 0.019 -0.026 0.171**

(0.024) (0.068) (0.095) (0.095) (0.064) (0.143) (0.090) (0.083) (0.092) (0.068) (0.060)

Life's level of spirituality -0.009 -0.090 0.017 -0.116 0.064 -0.177 -0.071 0.046 -0.060 0.104 -0.119

(0.032) (0.077) (0.085) (0.140) (0.089) (0.313) (0.128) (0.125) (0.114) (0.087) (0.100)

Life's level of fun 0.046 0.028 0.124 0.026 0.036 -0.210 0.046 -0.182 0.082 0.138 0.202*

(0.028) (0.066) (0.080) (0.102) (0.078) (0.198) (0.107) (0.142) (0.103) (0.098) (0.093)

Social status 0.077*** 0.070 -0.026 0.091 0.031 0.611 0.059 0.191 -0.022 0.092 0.181*

(0.023) (0.059) (0.076) (0.067) (0.063) (0.364) (0.096) (0.101) (0.105) (0.055) (0.073)

Life's non-boringness -0.076** -0.020 -0.141 0.269* -0.029 0.062 -0.086 0.178 0.003 -0.074 -0.236*

(0.027) (0.061) (0.092) (0.125) (0.075) (0.177) (0.093) (0.105) (0.100) (0.091) (0.094)

Physical comfort 0.034 0.175** 0.013 -0.071 0.037 0.180 -0.038 0.019 -0.127 0.096 -0.111

(0.022) (0.062) (0.094) (0.085) (0.063) (0.107) (0.077) (0.081) (0.072) (0.070) (0.074)

Sense of purpose 0.101*** 0.118 0.161 0.091 0.039 0.453** -0.083 0.148 0.178* 0.136* 0.125*

(0.021) (0.062) (0.082) (0.072) (0.063) (0.139) (0.084) (0.077) (0.085) (0.065) (0.051)

Constant 0.345* 1.282*** 1.300** -0.931 0.008 0.475 0.335 -1.150* 1.273** -0.131 1.187*

(0.141) (0.344) (0.495) (0.495) (0.443) (0.631) (0.597) (0.575) (0.406) (0.421) (0.517)

Observations 2089 210 209 208 209 210 208 208 207 211 209

R2 0.50 0.52 0.49 0.56 0.46 0.66 0.43 0.34 0.42 0.44 0.38

Incremental R2 0.03 0.06 0.04 0.08 0.05 0.07 0.02 0.11 0.03 0.10 0.17

Notes: Standard errors in parentheses. OLS regression coefficients reported. The dependent variable is 6-point intensity of choice described in section II.B.a, with a higher value

indicating a higher probability of choosing option 2. The dependent variables are the 7-point responses described in section II.B.b,c, with a higher number indicating a higher

predicted value of that aspect under option 2. The first column reports the regression pooling all scenarios and including scenario fixed effects. The remaining columns report the

regression restricted to a specific scenario. For the full wording of the scenario, see the Appendix. * p < 0.05,

** p < 0.01,

*** p < 0.00

33

Table 7: Choice on Happiness and All Aspects, with Incremental R2 (Study 4, happiness measured in isolation)

Decision problem: Pooled 1 2 3 4 5 6 7 8 9 10

Tradeoff implicated in decision

problem:

Income

vs

Sleep

Concert

vs

Dinner

Abs. Inc.

vs

Rel. Inc.

Income

vs

Legacy

Apple

vs

Orange

Income

vs

Labor

Socialize

vs

Sleep

Income

vs

Dinner

Education

vs

Social life

Interest

vs

Career

Non-local tradeoff?

Everyday/minor decision?

Dependent Variable: 6-point Choice

Own happiness 0.534*** 0.459*** 0.636*** 0.558*** 0.482*** 0.802*** 0.702*** 0.305*** 0.596*** 0.400*** 0.430***

(0.022) (0.072) (0.065) (0.070) (0.071) (0.072) (0.075) (0.077) (0.077) (0.075) (0.079)

Family happiness 0.122*** -0.042 0.075 0.293** 0.167 0.307 0.087 0.174 0.051 0.341*** 0.237*

(0.031) (0.082) (0.153) (0.105) (0.086) (0.288) (0.107) (0.149) (0.105) (0.095) (0.108)

Health -0.018 -0.094 -0.266 -0.024 0.094 -0.092 0.135 0.176 -0.026 -0.050 -0.271

(0.039) (0.116) (0.159) (0.170) (0.119) (0.143) (0.135) (0.144) (0.146) (0.117) (0.145)

Life's level of romance 0.032 0.181 -0.124 0.207 -0.106 -0.864 -0.048 -0.092 0.032 -0.072 0.291*

(0.042) (0.129) (0.129) (0.165) (0.138) (0.787) (0.164) (0.147) (0.146) (0.124) (0.134)

Social life -0.053 -0.140 -0.032 -0.225 -0.087 0.146 -0.057 0.066 0.000 -0.120 -0.146

(0.033) (0.121) (0.097) (0.145) (0.113) (0.335) (0.111) (0.133) (0.135) (0.117) (0.108)

Control over your life 0.018 0.001 0.140 -0.061 0.101 0.045 0.074 0.067 -0.121 -0.196* 0.167

(0.033) (0.099) (0.120) (0.123) (0.087) (0.216) (0.103) (0.125) (0.131) (0.097) (0.086)

Life's level of spirituality 0.019 -0.001 0.011 -0.087 0.002 0.579 -0.003 0.126 0.026 0.128 -0.089

(0.042) (0.106) (0.107) (0.187) (0.119) (0.515) (0.155) (0.192) (0.161) (0.128) (0.146)

Life's level of fun 0.128*** -0.010 0.250* 0.143 0.108 -0.737* 0.067 -0.089 0.221 0.381** 0.186

(0.036) (0.101) (0.108) (0.134) (0.096) (0.348) (0.120) (0.187) (0.140) (0.124) (0.131)

Social status 0.048 0.075 0.056 -0.026 0.080 -0.113 0.032 0.120 -0.118 0.180* 0.084

(0.030) (0.088) (0.099) (0.090) (0.086) (0.648) (0.112) (0.164) (0.144) (0.077) (0.098)

Life's non-boringness -0.149*** -0.078 -0.317* 0.217 -0.184 -0.084 -0.149 0.012 -0.082 -0.185 -0.269*

(0.034) (0.089) (0.127) (0.156) (0.095) (0.276) (0.106) (0.142) (0.149) (0.132) (0.126)

Physical comfort -0.002 0.248** -0.133 -0.083 -0.007 0.112 0.020 -0.071 -0.133 0.096 -0.133

(0.029) (0.093) (0.122) (0.105) (0.079) (0.171) (0.079) (0.100) (0.102) (0.105) (0.091)

Sense of purpose 0.124*** 0.089 0.323** 0.160 0.111 0.673*** -0.055 0.179 0.177 0.013 0.076

(0.028) (0.101) (0.107) (0.089) (0.078) (0.155) (0.093) (0.104) (0.130) (0.089) (0.070)

Constant 0.337 1.714*** 1.038 -1.004 0.720 0.312 -0.125 -1.689* 1.649** -0.604 1.582*

(0.182) (0.481) (0.659) (0.642) (0.592) (0.941) (0.624) (0.848) (0.510) (0.627) (0.707)

Observations 1040 105 104 103 104 105 102 103 104 106 104

R2 0.55 0.51 0.63 0.66 0.51 0.72 0.63 0.38 0.52 0.54 0.45

Incremental R2 0.04 0.11 0.11 0.12 0.08 0.13 0.00 0.10 0.04 0.20 0.15

Notes: Standard errors in parentheses. OLS regression coefficients reported. The dependent variable is 6-point intensity of choice described in section II.B.a, with a higher value

indicating a higher probability of choosing option 2. The dependent variables are the 7-point responses described in section II.B.b,c, with a higher number indicating a higher

predicted value of that aspect under option 2. The first column reports the regression pooling all scenarios and including scenario fixed effects. The remaining columns report the

regression restricted to a specific scenario. For the full wording of the scenario, see the Appendix. * p < 0.05,

** p < 0.01,

*** p < 0.00

34

Table 8: Choice on Happiness and All Aspects, with Incremental R2 (Study 4, happiness measured in series)

Decision problem: Pooled 1 2 3 4 5 6 7 8 9 10

Tradeoff implicated in decision

problem:

Income

vs

Sleep

Concert

vs

Dinner

Abs. Inc.

vs

Rel. Inc.

Income

vs

Legacy

Apple

vs

Orange

Income

vs

Labor

Socialize

vs

Sleep

Income

vs

Dinner

Education

vs

Social life

Interest

vs

Career

Non-local tradeoff?

Everyday/minor decision?

Dependent Variable: 6-point Choice

Own happiness 0.455*** 0.478*** 0.397*** 0.534*** 0.450*** 0.720*** 0.488*** 0.402*** 0.388*** 0.419*** 0.208

(0.030) (0.083) (0.093) (0.112) (0.110) (0.106) (0.116) (0.102) (0.098) (0.096) (0.108)

Family happiness 0.033 0.113 -0.090 0.041 -0.046 0.022 0.204 -0.299* -0.087 0.017 0.207*

(0.035) (0.072) (0.194) (0.139) (0.108) (0.397) (0.164) (0.131) (0.135) (0.088) (0.101)

Health 0.056 -0.040 0.117 0.060 0.340 0.124 -0.207 0.339* 0.208 -0.087 -0.153

(0.046) (0.113) (0.208) (0.210) (0.199) (0.168) (0.244) (0.135) (0.160) (0.116) (0.158)

Life's level of romance -0.065 0.021 -0.064 0.297 -0.100 -1.883* -0.516* -0.094 0.037 -0.003 0.234

(0.056) (0.142) (0.175) (0.237) (0.193) (0.857) (0.252) (0.156) (0.212) (0.154) (0.231)

Social life -0.015 0.070 -0.044 -0.074 0.045 0.828 0.546 -0.051 -0.096 -0.035 -0.105

(0.046) (0.122) (0.110) (0.143) (0.160) (0.764) (0.287) (0.197) (0.170) (0.160) (0.167)

Control over your life 0.118*** 0.013 -0.095 -0.165 0.180 0.263 0.194 0.147 0.155 0.121 0.174

(0.036) (0.092) (0.150) (0.154) (0.096) (0.205) (0.153) (0.124) (0.136) (0.101) (0.093)

Life's level of spirituality -0.047 -0.081 0.001 -0.195 0.120 -0.978 -0.062 -0.075 -0.117 -0.034 -0.106

(0.050) (0.118) (0.135) (0.228) (0.138) (0.775) (0.235) (0.201) (0.177) (0.133) (0.160)

Life's level of fun -0.045 -0.147 0.138 -0.089 -0.045 0.229 -0.068 -0.382 0.069 -0.089 0.174

(0.047) (0.105) (0.126) (0.195) (0.144) (0.321) (0.223) (0.246) (0.177) (0.170) (0.154)

Social status 0.108** 0.104 -0.092 0.290** -0.011 0.939 0.086 0.329* 0.161 0.017 0.263*

(0.035) (0.082) (0.124) (0.107) (0.102) (0.513) (0.174) (0.154) (0.165) (0.081) (0.127)

Life's non-boringness 0.041 0.153 0.060 0.219 0.175 -0.234 -0.116 0.328 0.053 0.037 -0.133

(0.044) (0.090) (0.137) (0.222) (0.128) (0.313) (0.182) (0.186) (0.150) (0.136) (0.153)

Physical comfort 0.075* 0.072 0.176 0.059 0.069 0.308* -0.009 0.053 -0.147 0.069 -0.071

(0.036) (0.086) (0.146) (0.151) (0.116) (0.147) (0.167) (0.155) (0.114) (0.098) (0.139)

Sense of purpose 0.084* 0.088 0.110 0.021 -0.047 0.467 -0.016 0.138 0.150 0.198* 0.178*

(0.033) (0.081) (0.131) (0.122) (0.111) (0.714) (0.153) (0.121) (0.119) (0.098) (0.083)

Constant 0.301 0.608 1.796* -1.109 -0.993 0.315 0.724 -0.879 0.256 0.416 0.598

(0.223) (0.487) (0.734) (0.860) (0.685) (0.892) (1.239) (0.867) (0.790) (0.620) (0.885)

Observations 1049 105 105 105 105 105 106 105 103 105 105

R2 0.48 0.60 0.43 0.52 0.49 0.65 0.34 0.39 0.38 0.44 0.37

Incremental R2 0.03 0.07 0.03 0.08 0.11 0.06 0.08 0.17 0.06 0.08 0.19

Notes: Standard errors in parentheses. OLS regression coefficients reported. The dependent variable is 6-point intensity of choice described in section II.B.a, with a higher value

indicating a higher probability of choosing option 2. The dependent variables are the 7-point responses described in section II.B.b,c, with a higher number indicating a higher

predicted value of that aspect under option 2. The first column reports the regression pooling all scenarios and including scenario fixed effects. The remaining columns report the

regression restricted to a specific scenario. For the full wording of the scenario, see the Appendix. * p < 0.05,

** p < 0.01,

*** p < 0.00

35

Table 9: Robustness checks Full Sample Excluding people who

believed they would regret

the cases when they did not

maximize happiness

Excluding people who believed they made a

mistake when they did not

maximize happiness

Excluding people whose answer for choice was not

the answer they wanted

themselves to choose

Only people who responded A or B in question 14

Own happiness 0.487*** 0.497*** 0.503*** 0.577*** 0.462***

(0.018) (0.020) (0.018) (0.019) (0.021)

Family happiness 0.076** 0.080** 0.071** 0.021 0.082**

(0.023) (0.027) (0.024) (0.026) (0.026)

Health 0.018 0.015 0.027 0.026 0.032

(0.030) (0.034) (0.031) (0.034) (0.033)

Life's level of romance 0.008 0.022 0.030 0.042 -0.007

(0.034) (0.037) (0.035) (0.038) (0.038)

Social life -0.045 -0.040 -0.040 -0.114*** -0.045

(0.027) (0.031) (0.028) (0.032) (0.030)

Control over your life 0.079** 0.088** 0.064* 0.077** 0.082**

(0.024) (0.028) (0.025) (0.028) (0.028)

Life's level of spirituality -0.009 0.012 0.001 -0.065 -0.002

(0.032) (0.037) (0.033) (0.037) (0.036)

Life's level of fun 0.046 0.049 0.046 0.080* 0.065*

(0.028) (0.032) (0.029) (0.032) (0.031)

Social status 0.077*** 0.053* 0.071** 0.085** 0.092***

(0.023) (0.026) (0.023) (0.026) (0.025)

Life's non-boringness -0.076** -0.091** -0.095*** -0.096** -0.073*

(0.027) (0.030) (0.028) (0.030) (0.030)

Physical comfort 0.034 0.012 0.032 0.045 0.036

(0.022) (0.026) (0.023) (0.026) (0.025)

Sense of purpose 0.101*** 0.108*** 0.098*** 0.145*** 0.117***

(0.021) (0.025) (0.022) (0.025) (0.024)

Constant 0.345* 0.326* 0.292* 0.178 0.183

(0.141) (0.156) (0.146) (0.141) (0.169)

Observations 2089 1617 1932 1429 1658

R2 0.50 0.51 0.50 0.63 0.48

F-statistic and p-value of test of different slope coefficients

from the full-sample results

1.00

p=0.444

1.99

p=0.022

13.35

p=0.000

1.18

p=0.295

Notes: Standard errors in parentheses. OLS regression coefficients reported. The dependent variable is 6-point intensity of choice described in section II.B.a, with a higher value indicating a higher

probability of choosing option 2. The dependent variables are the 7-point responses described in section II.B.b,c, with a higher number indicating a higher predicted value of that aspect under option

2. The first column reports the regression pooling all scenarios and including scenario fixed effects. The remaining columns report robustness checks, excluding portions of the sample population.. For

full details, see section IV. * p < 0.05, ** p < 0.01, *** p < 0.00

Table 10: Results Grouped by Scenario Type Full Sample Non-local Tradeoffs Representative Scenarios Minor Decisions

Own happiness 0.487*** 0.490*** 0.379*** 0.550***

(0.018) (0.033) (0.028) (0.026)

Family happiness 0.076** 0.060 0.109** -0.020

(0.023) (0.036) (0.035) (0.042)

Health 0.018 0.029 0.019 0.087*

(0.030) (0.055) (0.044) (0.044)

Life's level of romance 0.008 0.092 0.011 -0.089

(0.034) (0.061) (0.049) (0.052)

Social life -0.045 -0.077 -0.048 0.010

(0.027) (0.049) (0.043) (0.042)

Control over your life 0.079** 0.042 0.100** 0.074

(0.024) (0.041) (0.036) (0.041)

Life's level of spirituality -0.009 -0.028 -0.015 -0.028

(0.032) (0.054) (0.050) (0.052)

Life's level of fun 0.046 0.034 0.072 0.040

(0.028) (0.045) (0.048) (0.046)

Social status 0.077*** 0.081* 0.091* 0.036 (0.023) (0.034) (0.036) (0.044)

Life's non-boringness -0.076** -0.004 -0.097* -0.059 (0.027) (0.043) (0.045) (0.042)

Physical comfort 0.034 0.077* -0.017 0.019 (0.022) (0.038) (0.034) (0.035)

Sense of purpose 0.101*** 0.073* 0.125*** 0.127*** (0.021) (0.037) (0.032) (0.038)

Constant 0.345* 0.253 0.495* 0.350

(0.141) (0.224) (0.219) (0.225)

Observations 2089 627 835 1042

R2 0.50 0.52 0.46 0.49

F-statistic and p-valueof test of different slope coefficients from the

full-sample results

0.89 p=0.556

2.61 p=0.002

3.67 p=0.000

Notes: Standard errors in parentheses. OLS regression coefficients reported. The dependent variable is 6-point intensity of choice

described in section II.B.a, with a higher value indicating a higher probability of choosing option 2. The dependent variables are the

7-point responses described in section II.B.b,c, with a higher number indicating a higher predicted value of that aspect under option 2.

The first column reports the regression pooling all scenarios and including scenario fixed effects. The second column restricts the

sample to non-local decision scenarios. The third column restricts the sample to minor decision scenarios. The fourth column restricts

the sample to scenarios generated from surveys of important decisions, which generated scenarios we believe to be representative of

common decisions. * p < 0.05,

** p < 0.01,

*** p < 0.00

37

Appendix: Scenarios Presented in Surveys

1) Say you have to decide between two new jobs. The jobs are exactly the same in almost every way, but

have different work hours and pay different amounts.

Option 1: A job paying $80,000 per year. The hours for this job are reasonable, and you would be able to get

about 7.5 hours of sleep on the average work night.

Option 2: A job paying $140,000 per year. However, this job requires you to go to work at unusual hours, and

you would only be able to sleep around 6 hours on the average work night.

2) Suppose you promised a close friend that you would attend his or her 50st [―21st‖ in student samples]

birthday dinner. However, at the last minute you find out that you have won front row seats to see your favorite

musician, and the concert is at the same time as the dinner. This is the musician’s last night in town. You face

two options:

Option 1: Skip your friend’s birthday dinner to attend the concert.

Option 2: Attend your friend’s birthday dinner and miss the concert.

3) Suppose you are considering a new job, and have offers from two companies. Even though all aspects of

the two jobs are identical, employees’ salaries are different across the two companies due to arbitrary timing of

when salary benchmarks happened to be set. Everyone in each company knows the other employees’ salaries.

You must choose one of the two companies, which means you must decide between the following two options:

Option 1: Your yearly income is $105,000, while on average others at your level earn $120,000.

Option 2: Your yearly income is $100,000, while on average others at your level earn $85,000.

38

4) (Phrasing in study 1): Suppose you are a skilled artist, and you have to decide between two career

paths for your life.

Option 1: You devote yourself to your own style of painting. This would require a number of sacrifices, such as

having less time for friends and family, and making less money. For example, you expect that selling your

paintings will give you an income of $40,000 a year. If you choose this path, you don’t expect that your work

will be appreciated in your lifetime, but posthumously you will make an impact on the history of art, achieve

fame, and be remembered in your work.

Option 2: You become a graphic designer at an advertising company. This would give you more money and

more time with friends and family than Option 1. The company is offering you a salary of $60,000 a year,

which will afford you a much more comfortable lifestyle, but you will have no impact and leave no legacy to be

remembered.

(Phrasing in studies 2 and 4) : Suppose you are a skilled artist, and you have to decide between two

career paths for your life. There are two styles of painting that you consider to be your own style, and you enjoy

both equally. Style 1 happens to be much less popular than Style 2 today, but you know it will be an important

style in the future.

Option 1: You devote yourself to Style 1. You expect that selling your paintings will give you an income of

$40,000 a year. If you choose this path, you don’t expect that your work will be appreciated in your lifetime, but

posthumously you will make an impact on the history of art, achieve fame, and be remembered in your work.

Option 2: You devote yourself to Style 2. You expect that selling your paintings will give you an income of

$60,000 a year, but you will have no memorable impact. [In Study 2, each subject saw this question three times,

with different salaries in option 2. This number took a value of either $42,000, $60,000, $80,000, or $100,000.]

5) Suppose you are checking out a new supermarket that just opened near where you live. As you walk by

the fresh fruit display, you are offered your choice of a free snack:

Option 1: A freshly sliced apple.

39

Option 2: A freshly sliced orange.

6) Suppose that due to budget cuts, the school implements a ―student activities fee‖ of $15 dollars a week

to help pay for maintenance of facilities used for extracurricular student activities. However, the school allows

you to not pay the fee if instead you put in 2 hours of service a week shelving books at the library. You face two

options:

Option 1: Spend 2 hours a week shelving books.

Option 2: Pay $15 a week.

7) Say you are hanging out with a group of friends at your friend’s room. You are having a really good

time, but it is getting to be late at night. You have to decide between two options.

Option 1: Stay up another hour. It is likely you will feel tired all day tomorrow, but this particular evening you

are having an especially fun time.

Option 2: Excuse yourself from the group, and go to bed. You will be disappointed to miss the fun, but you

know you will feel better the next day and be more productive at paying attention in class and doing your

homework.

8) Imagine that for the first time in three years, your parents (or if your parents are gone, your closest

relatives who are older than you) have arranged for a special family gathering that will happen the day after

Thanksgiving, with everyone also invited to Thanksgiving dinner. You face two options. Would you choose to

go to the family gathering the day after Thanksgiving (and maybe to Thanksgiving dinner) if getting there

required a $500 roundtrip plane ticket for plane flights that were 5 hours each way?

Option 1: Go to the thanksgiving gathering, which requires a $500 round trip plane ticket.

Option 2: Miss the thanksgiving gathering, but save the money.

40

9) Suppose you have decided to leave Cornell, and are transferring to a new school. You have been

accepted to two schools, and are deciding where to go. The first school is extremely selective and high quality,

but is in a small town out in the country with a less active social scene. The second school is in a major city

with a great social scene, but is slightly less renowned. Which would you choose?

Option 1: Highly selective school, isolated socially and geographically.

Option 2: Less selective school, socially active and in a major city.

10) Suppose you are considering two summer internships. One is extremely interesting and involves work

you are passionate about, but does not advance your career. The other will likely be boring, but will help you

get a job in the future. Which would you choose?

Option 1: Interesting internship which does not advance career.

Option 2: Boring internship which will help you get a job.

11) Say you are driving by yourself to see your favorite musician in concert on their last day in town. You

are five minutes away, and the concert starts in ten minutes. On the drive, you witness a truck hit a parked car,

causing roughly $500 in damages, and then drive away without leaving their information. You notice the truck’s

license plate, and you are the only witness. You face two options:

Option 1: Keep driving and get to the concert on time.

Option 2: Call the police, in which case you will have to wait around the parked car to give a testimony. This

would take about half an hour. You would have trouble finding a seat and might miss the whole concert.

41

12) (Phrasing in study 1): Say you are moving to a new town. You are trying to decide between two similar

apartments which you could rent. The two apartments are identical in almost everything – including floor plan,

amenities, neighborhood character, schools, safety, etc. However, they have different rents and are located at

different distances from your work.

Option 1: An apartment which requires a 45-minute drive to work. The rent is about 20% of your monthly

income.

Option 2: A similar apartment, with only a 10-minute drive. The rent is about 40% of your monthly income.

(Phrasing in study 2): Say you are moving to a new town. The new town is known for its terrible traffic

jams, and driving there is widely considered to be unpleasant. You are trying to decide between two similar

apartments which you could rent. The two apartments are identical in almost everything – including floor plan,

amenities, neighborhood character, schools, safety, etc. However, they have different rents and are located at

different distances from your work.

Option 1: An apartment which requires a 45-minute drive each way to work. The commute has heavy traffic

almost the whole way. The rent is about 20% of your monthly income.

Option 2: A similar apartment which requires a 10-minute drive each way to work. The commute has heavy

traffic almost the whole way. The rent is about 40% of your monthly income.

13) Say you have been reassigned at your job, and will be moved to a new location. There are two offices

where you could request to work. One office is in a city where many of your friends happen to live, and pays

20% less than your current salary. The other office is in a city where you don’t know anyone, and pays 10%

more than your current salary. Your job will be exactly the same at either office. You must decide between the

following two options:

Option 1: Make 20% less than your current salary and move to the city with your friends.

Option 2: Make 10% more than your current salary and move to a city where you do not know anyone.