DISCUSSION PAPER SERIES

IZA DP No. 16200

Social Preferences:
Fundamental Characteristics and
Economic Consequences

Ernst Fehr
Gary Charness

MAY 2023
DISCUSSION PAPER SERIES

IZA DP No. 16200

Social Preferences:
Fundamental Characteristics and
Economic Consequences

Ernst Fehr
Zurich University and IZA
Gary Charness
University of California and IZA

MAY 2023

Any opinions expressed in this paper are those of the author(s) and not those of IZA. Research published in this series may
include views on policy, but IZA takes no institutional policy positions. The IZA research network is committed to the IZA
Guiding Principles of Research Integrity.
The IZA Institute of Labor Economics is an independent economic research institute that conducts research in labor economics
and offers evidence-based policy advice on labor market issues. Supported by the Deutsche Post Foundation, IZA runs the
world’s largest network of economists, whose research aims to provide answers to the global labor market challenges of our
time. Our key objective is to build bridges between academic research, policymakers and society.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper
should account for its provisional character. A revised version may be available directly from the author.

ISSN: 2365-9793

IZA – Institute of Labor Economics

Schaumburg-Lippe-Straße 5–9 Phone: +49-228-3894-0
53113 Bonn, Germany Email: [email protected] www.iza.org
IZA DP No. 16200 MAY 2023

ABSTRACT
Social Preferences:
Fundamental Characteristics and
Economic Consequences*
We review the vast literature on social preferences by assessing what is known about
their fundamental properties, their distribution in the broader population, and their
consequences for important economic and political behaviors. We provide, in particular,
an overview of the empirically identified characteristics of distributional preferences and
how they are affected by merit, luck, and risk considerations as well as by concerns for
equality of opportunity. In addition, we identify what is known about belief-dependent
social preferences such as reciprocity and guilt aversion. The evidence indicates that the
big majority of individuals have some sort of social preference while purely self- interested
subjects are a minority. Our review also shows how the findings from laboratory experiments
involving social preferences provide a deeper understanding of important field phenomena
such as the consequences of wage inequality on work morale, employees’ resistance to
wage cuts, individuals’ self-selection into occupations and sectors that are more or less
prone to morally problematic behaviors, as well as issues of distributive politics. However,
although a lot has been learned in recent decades about social preferences, there are still
many important, unresolved, yet exciting, questions waiting to be tackled.

JEL Classification: D0, D2, D9, H0, J0, P0

Keywords: social preferences, altruism, inequality aversion, image
concerns, reciprocity

Corresponding author:
Ernst Fehr
Department of Economics
University of Zurich
Blümlisalpstrasse 10
8006 Zurich
Switzerland
E-mail: [email protected]

* We thank Peter Andre, Alain Cohn, Martin Dufwenberg, David Levine, Johanna Mollerström, Bertil Tungodden,
and Joel van der Weele for useful comments on an earlier version of this paper.
 Table of Contents

1. Introduction
2. Social Preferences and Self-Interest in Paradigmatic
Economic Games
3. Fundamental Properties of Social Preferences
4. Economic and Political Consequences of Social Preferences
5. Summary and Outlook

1
 Detailed Table of Contents
1. Introduction

2. Social Preferences and Self-Interest in Paradigmatic Economic Games

3. Fundamental Properties of Social Preferences

3.1 Social Preferences over Payoff Distributions
3.1.1 Models of Distributional Preferences
3.1.2 Empirical Frequency of Different Distributional Preferences
- Concern for Equality or Concern for the Total Payoff?
- How Prevalent are Inequality Aversion and Spite?
- How Many Endogenous Preference Clusters?

3.2 The Role of Merit, Luck and Risk in Distributional Preferences

3.2.1 Equity and Entitlements
3.2.2 Modelling Entitlement Effects
3.2.3 Who are the Meritocrats?
3.2.4 Cultural Differences and the Relative Importance of Fairness & Efficiency Concerns
3.2.5 Shallow Meritocracy?
3.2.6 The Role of Risk in Distributional Preferences

3.3 Belief-Dependent Social Preferences

3.3.1 Reciprocity
- Models of Reciprocity
- Field Evidence on Reciprocity
- Laboratory Evidence on Reciprocity
- The Relative Importance of Distributional and Reciprocity Concerns
- Summary
3.3.2 The Role of Guilt Aversion in Social Preferences
- Conceptual and Intuitive Foundations of Guilt Aversion
- Early Experimental Evidence
- Overcoming Problems in Identifying Guilt Aversion
- Summary

3.4 The Role of Self-Image and Social Image in Social Preferences

- Theoretical Concepts
- Avoidance Behavior and Self-Image Concerns
- Social Image Concerns
- Summary

4. Economic and Political Consequences of Social Preferences

4.1 Implications for Labor Relations and Macroeconomics
4.1.1 Fairness Concerns, Wage Inequality, and Job Satisfaction
4.1.2 Fairness Concerns and Resistance to Wage Cuts
4.1.3 Screening, Selection and the Stability of Social Preferences

4.2 Implications for Contracts, Institutions and Incentives

4.2.1 The Effects on Contract Enforcement and Financial Incentives
4.2.2 Social Preferences as a Behavioral Foundation for Employment Contracts
4.2.3 Social Preferences, Contractual Incompleteness, and Property Rights

4.3 The Role of Social Preferences in Politics

5. Summary and Outlook

0
 1. Introduction
The topic of social preferences has received a great deal of attention in economics and in the
broader social-science literature in the past three decades. A large, and still growing,
literature has documented the existence of social preferences, their social and economic
implications, and the various conditions under which they influence the equilibria and
outcomes of human interactions. 1 The key characteristic of social preferences is that
individuals are willing to sacrifice money or other material resources to help or hurt other
people, to establish fairness and justice, or to increase groups’ joint payoff. In other words,
people with social preferences do not maximize their own material payoff but are other-
regarding – which has potentially far-reaching consequences for both theory and practical
applications.

In this paper, we will summarize what has been learned in this line of research and we
will identify what needs to be known and done to make further progress. First, we will
describe the evidence for the existence of social preferences gathered in canonical
(“paradigmatic”) experiments and suggested by behavioral patterns observed in the field in
Section 2. We will then review the various theoretical models and the empirical relevance of
the different motives that have been proposed as a foundation for social preferences in
Section 3. In this context, we will discuss preferences over the distribution of material
payoffs and the impact of merit, luck, and risk on these preferences. We will then review the
evidence on belief-dependent preferences such as reciprocity and guilt aversion. Moreover,
this section also discusses the role of self-image and social image in other-regarding
behaviors.

Because a large part of the empirical research on social preferences has used laboratory
experiments, it is important to point out that there is a broad consensus regarding the
relevance of qualitative findings from economic experiments (Levitt and List 2007b; Falk
and Heckman 2009; Camerer 2015; Kessler and Vesterlund 2015). While researchers may
disagree with regard to the generalizability of the quantitative findings from experiments –
such as the magnitude of discount rates – there is widespread agreement regarding the

1The influence of this literature can be illustrated, for example, by the citation rates of the most prominent models
of social preferences (Rabin (1993); Fehr and Schmidt (1999); Bolton & Ockenfels (2000); Charness and Rabin
(2002); Dufwenberg and Kirchsteiger (2004); Falk and Fischbacher (2006)). These six papers alone have received
a total of 40,000 Google Scholar citations and more than 6,500 citations in the Social Science Citation (SCCI)
Index (situation July 2022), but the overall impact of the literature on social preferences goes well beyond these
papers.

1
relevance of qualitative findings. 2 Applied to the domain of social preferences, this means
that while the strength of different social preferences may vary across different (lab or field)
environments, or between lab and field environments, the very existence and the directional
effects of social preferences such as inequality aversion, a concern for the total surplus
(“efficiency”), reciprocity, or guilt aversion can be reliable documented with laboratory
experiments. 3 In Section 4, we will discuss important economic consequences of social
preferences for labor markets and the macroeconomy, for the functioning of incentives and
contracts, for the foundation of incomplete (employment) contracts, and for the demand for
politically enforced redistribution. The subsections on labor markets and political
redistribution indicate, in particular, a strong congruence between lab and field findings.
Finally, we discuss what we have learned so far and point out important unresolved, yet
exciting, questions in Section 5.

There are a few other older review papers on social preferences such as those of Fehr
and Schmidt (2006) and Cooper and Kagel (2016), who provided selective surveys of aspects
of social preferences. Our paper differs by being more encompassing, by exploiting the
enormous accumulation of recent evidence about the distribution of social preferences in
broad population samples, by discussing the role of merit, luck, risk and image concerns in
other-regarding behaviors, as well as the deeper knowledge that has accrued with regard to
the identification of reciprocity and guilt aversion. Finally, our review paper also benefits
from the recent accumulation of exciting insights from field evidence that indicates the
relevance of social preferences for nominal wage rigidity, the morale effects of wage
inequality. or the demand for redistribution.

2
Individuals’ experimentally measured discount rates have, for example, been shown to be good predictors of
their administratively measured relative wealth levels (controlling for a host of relevant individual characteristics)
although the magnitude of the lab measured discount rate is often too high to be considered a meaningful
quantitative estimate (e.g., Epper et al. (2020)). Note, however, that it is surely not impossible that quantitative
experimental findings generalize to the field, but this needs careful empirical scrutiny for the involved behavioral
domains. The quantitative estimates of peer effects (Herbst and Mas 2015) provide one example of generalizable
quantitative findings. Benz and Meier (2008) also show a strong quantitative similarity of charitable donation of
students in a laboratory experiment and their donations in the field.
3
The first version of this paper included two additional sections. The first presented evidence on the predictive
validity of laboratory social preferences measures for field behaviors, while the second presented evidence on the
relationship between identity or parochial issues and social preferences. Due to space limitations, these sections
have been outsourced to two separate papers (Charness 2023; Fehr 2023).

2
2. Social Preferences and Self-Interest in Paradigmatic
Economic Games
The role of social preferences can, in principle, manifest itself in the myriads of human
interactions that take place every day in families, neighborhoods, markets, organizations, or
the political sphere. They can inhibit or enhance the effectiveness of economic incentives or
the functioning of markets and organizations; they can shape collective action, affect
contractual arrangements and institutions, influence how people vote, and undermine or
enhance political regimes. In fact, social preferences appear to at least partly drive many real-
world phenomena. They range from helping friends and neighbors to support for co-workers,
from charitable donations to participation in collective actions against oppressive
dictatorships, from affluent individuals’ political support for redistributing income towards
the poor to workers’ collective stance against wage cuts.

However, although social preferences of some sort are plausible drivers of behavior in
all these real-world examples, many other factors – such as the expectation of future material
benefits or reputational concerns – can influence individuals’ behavior in these situations.
Therefore, a skeptic who observes a seemingly other-regarding behavior in the field may not
be convinced that it is due to social preferences because one can always argue that there may
have been hidden, unobservable financial or reputational rewards. This is why a considerable
part of the research on social preferences is based on laboratory experiments where one can
rule out these hidden rewards with certainty by ensuring that the experiments are one-shot
and that the parties interact anonymously with each other. 4

There are several paradigmatic experiments that – when played one-shot and under
anonymity between the players involved – document the widespread existence of social
preferences. We call these experiments “paradigmatic” because they all capture an essential
feature of an important economic or social situation. The games are described in more detail
in Box 1 on “Paradigmatic Economic Games” (see below). They capture (i) situations that
involve the unilateral determination of a payoff distribution between parties (dictator game),
(ii) bilateral bargaining situations (ultimatum game), (iii) economic exchanges under

4 The implementation of anonymous one-shot experiments is thus a frequently employed conservative strategy
that makes sense, but it is also important to keep in mind that important aspects of social relations, and their
influence on social preferences, may be lost in these experiments (Frohlich, Oppenheimer and Kurki 2004). It is,
however, possible to introduce many real-life features (e.g., lack of anonymity, reputation formation, the role of
merit and entitlements, etc.) into the lab and to study them in a controlled manner.

3
incomplete contracts (gift exchange game, trust game), (iv) public goods contributions in the
absence of supporting institutions and material incentives for their provision (public goods
game) and (v) the private sanctioning by third parties of greedy, unfair, or norm-violating
behaviors (third party punishment game). All of these games have been replicated dozens,
and sometimes hundreds of times and show robust, replicable behavioral patterns (see
Camerer (2003) for a review). The critical player in these games (i.e., the person for whom
one can measure social preferences) always has a simple and transparent money maximizing
choice, while systematic (i.e., non-random) deviations from that choice indicate some form of
social preference.

For example, many responders in the ultimatum game from societies across the globe
indicate a social preference by rejecting uneven offers and earning nothing as a consequence
(Guth, Schmittberger and Schwarze 1982; Henrich et al. 2001) . The proposer often
anticipates this responder behavior, inducing him or her to make relatively fair offers so that
the responder appropriates on average roughly 40% of the pie even though the self-interest
model predicts that he or she should get 0%. In the dictator game, many subjects in the role of
dictators make positive unilateral transfers – on average typically around 20% of the
available money – to the recipient (Forsythe et al. 1994; Camerer 2003). Subjects in the role
of workers in the gift exchange game (Fehr, Kirchsteiger and Riedl 1993) respond to higher
wages with higher costly effort levels, although the minimal effort would always be the
money-maximizing choice. In the trust game (Berg, Dickhaut and McCabe 1995), the trustees
respond to trustors’ positive transfers with positive back-transfers, although the money
maximizing choice would be to transfer back nothing. In social dilemma games (Dawes,
Mctavish and Shaklee 1977; Dawes 1980) and public good games with (complete) defection
as a dominant strategy (Andreoni 1988), many subjects nevertheless cooperate and make
positive contributions to the public good. And third parties in third party punishment games
(Fehr and Fischbacher 2004) who observe that other individuals have been treated unfairly or
are the “victims” of a social norm violation punish the perpetrators at a cost to themselves.

The robust documentation of other-regarding behaviors in experiments has led many

observers to assume that in addition to self-interest, people also care for other people’s
payoffs (Thibaut and Kelley 1959; Messick and Mcclintock 1968). However, one
fundamental question that has bothered the field for a long time is how one can reconcile the
existence of social preferences with evidence from other paradigmatic games that appear to
indicate that people are largely selfish. How can we explain this without arbitrarily assuming

4
that individuals’ preferences are different across games? For example, in a market game with
responder competition (see Box 1 on Paradigmatic Economic Games), there is not just one
responder as in the bilateral ultimatum game but several, so that the responders compete with
each other for the share of the surplus the proposer offers. In this game, the parties behave in
a much more self-interested manner – compared to the bilateral ultimatum game – because
the proposers make much more unequal offers, and the rejection of these uneven offers
decreases dramatically, allowing the proposer to appropriate the lion’s share of the surplus
(Fischbacher, Fong and Fehr 2009).

In a market game with proposer competition, one adds several competing proposers to a
bilateral ultimatum game. Here again, the players behave much more in line with the self-
interest prediction because the competition among the proposers drives up their offers such
that the responder appropriates almost the whole surplus, i.e., responders accept a distribution
of payoffs (Roth et al. 1991; Fischbacher, Fong and Fehr 2009) that appears very unfair.
These facts are also consistent with the observations from competitive double auctions and
competitive posted offer markets that indicate that the observed prices and quantities traded
in these markets tend to quickly converge to the competitive equilibrium derived from selfish
preferences (Smith 1982). Similarly, cooperation rates are often extremely low in the final
period of finitely repeated public goods games, seemingly indicating that social preferences
may play no role when sufficient learning has been possible.

How is the claim that many people have social preferences compatible with these facts?
Does a little bit of learning or competition wipe out social preferences, i.e., are they simply
inexistent under competitive pressure, or even worse, is behavior in these paradigmatic games
that document social preferences just a strange behavioral aberration 5 or can models of social
preferences also explain the conditions under which other-regarding individuals behave as if
they were completely self-interested? We will address this problem when we discuss the
different models of distributional preferences in Section 3.1.

5
In the early days of social preference research, it was not infrequent to meet economists who labelled subjects
exhibiting behaviors indicating social preferences as “crazy” individuals or “crazy” types.

5
 Box 1: Paradigmatic Economic Games
In a dictator game one player – the dictator – is given a sum of money that she can unilaterally allocate
between herself and a passive recipient who cannot make a decision. Self-interest predicts zero transfers to
the recipient. In an ultimatum game, a proposer can make a single proposal of how to distribute a given a
sum of money between herself and a responder. The responder can accept or reject the proposed allocation.
If she rejects, both players receive nothing. If she accepts, the proposed allocation is implemented. Self-
interest predicts that the proposer makes the lowest possible money offer, which the responder will then
accept. In a third party punishment game, two players, the dictator A and the recipient B, participate in a
dictator game. A third player, the potential punisher C, observes how much A gives to B; then C can spend
a proportion of his endowment on punishing A. A third party with a punishment option can, in principle, be
added to any constituent base game such as, for example, the prisoners’ dilemma game. Selfish third
parties will never punish.

In a trust game, two players, A and B, each have an identical initial endowment. First, A decides whether
to keep his endowment or to send some or all of it to B. Then B observes A’s action and decides whether to
keep the amount she receives or share some if it with A. The experimenter triples A’s transfer, so that both
players are better off collectively if A transfers money and B sends back a sufficient amount. Self-interest
predicts that B sends back zero money and, therefore, A also sends no money. In a gift exchange game, a
subject in the role of an employer can offer a fixed wage to a subject in the role of a worker. After
observing the wage, the worker chooses a costly effort level that raises the overall surplus that can be
distributed among the parties. Self-interest predicts the lowest possible wage offer and the lowest possible
effort level. In a generic linear public goods game, players have a token endowment they can invest in any
proportion in a private project in their own favor or a public project to be shared equally among the group,
where the experimenter increases any donation to the public project. Therefore, investment into the public
project maximizes the group’s aggregate earnings, but each individual has a dominant money-maximizing
strategy to invest the whole endowment into the private rather than the public project.

In the market game with responder competition, the proposer decides how to split a given sum of money
between herself and one of the competing responders. All n>1 responders have to decide simultaneously
whether to accept or reject the proposal. If all reject, all parties receive zero payoff; if some responders
accept, one of them is randomly chosen to receive the proposed amount, and all other responders receive
zero. In the market game with proposer competition, all proposers simultaneously make a proposal of how
to split a given sum of money with a single responder who can accept one of the proposals or reject all of
them. The accepted proposal is implemented, while all other proposers earn zero. If all proposals are
rejected, all players earn zero payoff.

The predictions for the experimental games mentioned above are based on the assumption that the games
are played one-shot and interactions are anonymous.

6
 A second question is how widespread are social preferences, i.e., what is the
approximate share of selfish people, and which proportion deviates significantly from
selfishness. The strategic games discussed above are not well suited for answering these
questions because one needs several observations of the same individual at sufficiently
different costs of the deviation from selfishness in order to measure an individual’s (social)
preference. For example, if the cost of giving a dollar to another individual is two dollars,
significantly fewer individuals are likely to make this transfer compared to when the cost is
10 cents. In other words, whether an individual has selfish or social preferences is not a
matter of a choice in one particular situation, but rather a matter of a structural preference
parameter that indicates the limit of an individuals’ willingness to give in terms of the costs
of giving. 6 For this reason, we will discuss in Section 3.1 the results of papers that have
undertaken more “structural” measurements of social preferences based on many individual
choices.

Third, many of the paradigmatic experiments mentioned above were first conducted
with student populations from Western societies that anthropologists (Henrich, Heine and
Norenzayan 2010) have characterized as WEIRD populations because they seem to represent
an outlier compared to the universe of historically observed and existing human populations. 7
The frequent use of student subjects from WEIRD societies raises the question whether the
findings can be generalized to the broader population in WEIRD societies and to other
societies, a question we tackle in Sections 3.1. and 3.2.

A fourth important question concerns the properties of social preferences. Responders

in an ultimatum game, for example, may reject low, uneven offers for various reasons. They
could simply be spiteful (i.e., value the other’s payoff always negatively), or they could
prefer equity instead of an unfair distribution of income, or they may want to punish the
proposer for making an unfair proposal. Likewise, the precise motivation that underlies the
reciprocal behaviors of second movers in the trust or gift exchange game is unclear. Second-
movers may reward the fair intentions of first movers in these games, or they may aim for a
more equal distribution of income, or they may avoid feeling guilty if they fail to reciprocate.
Thus, while the above-mentioned paradigmatic games capture key aspects of important
human social interactions they are, in general, not ideal for the robust identification of the

6 Strictly speaking, the only individuals who are entirely selfish are those who are not even willing to pay a
penny for giving another person a dollar.
7 WEIRD stands for Western, Educated, Industrialized, Rich, and Democratic societies.

7
relative importance of different properties of social preferences. To clarify the motives
underlying social preferences more precisely, we must design experiments that can cleanly
separate the motives for the different behaviors, a question that we also address in Sections
3.1. – 3.4.

Fifth, social preferences are different from risk and time preferences because the latter
are completely independent from other people’s payoff and actions, while the very notion of
a social preference is based on the idea that others’ payoffs and/or actions matter. This raises
the question who the other people – whose payoff matters – are and which psychological,
social, and economic forces determine who matters. Social preferences may importantly
depend on the kind of social relationships individuals have with others and may, thus, be
characterized by social boundaries and parochial instincts. Due to space limitations, we deal
with this question in Appendix 9.

3. Fundamental Properties of Social Preferences

3.1. Social Preferences over Payoff Distributions

3.1.1. Models of Distributional Preferences

A number of social preference models (see Box 2) are based on the assumption that people
care about the distribution of payoffs between themselves and a set of relevant reference
agents (Fehr and Schmidt 1999; Bolton and Ockenfels 2000; Andreoni and Miller 2002;
Charness and Rabin 2002; Fisman, Kariv and Markovits 2007). These models are a natural
starting point for the modelling of social preferences because the behavioral patterns
observed in the paradigmatic games discussed above strongly suggest that some sort of
interdependent preferences are at play. Another main motivation for the construction of some
of these models (Fehr and Schmidt 1999; Bolton and Ockenfels 2000) was the existence of
important facts from (i) market games with proposer and responder competition and final
periods of public good games that appeared to contradict the existence – or at least the
widespread relevance – of social preferences, and (ii) the observation that the same people
are willing to engage in seemingly contradictory behaviors by increasing the other’s payoff at
a cost to themselves in some situations, while decreasing the other’s payoff in other
situations. For example, positive transfers in the dictator game indicate that the dictators
value the recipients’ payoff positively, while rejections in the ultimatum game reduce the

8
proposer’s payoff and thus indicate a negative evaluation of the proposer’s payoff. Likewise,
a rise in the effort level in response to a higher wage leads to an increase in the employer’s
payoff in the gift exchange game, while the third party’s sanctions in the third-party
punishment game reduces the dictator’s payoff in that game.
Thus, the question is whether social preference models can account for these seemingly
contradictory facts. Obviously, a simple model of altruism that assumes that other-regarding
individuals value others’ payoffs positively cannot account for these facts nor can a model of
spiteful/envious preferences, where individuals value others' payoff negatively, do so. Since
simple models of altruism and spite cannot explain these facts, the concept of a social value
orientation (SVO), that was developed by social psychologists (Liebrand 1984; Liebrand and
Mcclintock 1988; Van Lange et al. 1997) is not capable of doing so because the SVO concept
views an other-regarding individual as either altruistic/cooperative or as envious.
Surprisingly, however, relatively simple models of other-regarding preferences over
payoff-distributions – such as those by Fehr and Schmidt (1999) and Bolton-Ockenfels
(2000) – go a long way towards reconciling these facts. While this does not mean that these
models are necessarily empirically correct – because this requires further empirical testing
(see below) – it means that widespread seemingly selfish behaviors in certain strategic games
can be perfectly consistent with the existence of widespread social preferences.
In the two-player case, a simple, linearized, version of player i’s distributional social
preferences 𝑈𝑈𝑖𝑖 (𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 ) can be written as
𝑈𝑈𝑖𝑖 = (1 − 𝛼𝛼)𝜋𝜋𝑖𝑖 + 𝛼𝛼𝛼𝛼𝑗𝑗 𝑖𝑖𝑖𝑖 𝜋𝜋𝑖𝑖 < 𝜋𝜋𝑗𝑗𝑗𝑗 (1a)
𝑈𝑈𝑖𝑖 = (1 − 𝛽𝛽)𝜋𝜋𝑖𝑖 + 𝛽𝛽𝛽𝛽𝑗𝑗 𝑖𝑖𝑖𝑖 𝜋𝜋𝑖𝑖 ≥ 𝜋𝜋𝑗𝑗 , (1b)

where both 𝛼𝛼 and 𝛽𝛽 are from the interval (-1, +1), 𝜋𝜋𝑖𝑖 and 𝜋𝜋𝑗𝑗 represent the material payoffs of
players i and j, respectively, 𝛼𝛼 is the weight on player j’s payoff when i is behind (𝜋𝜋𝑖𝑖 < 𝜋𝜋𝑗𝑗 ),
while 𝛽𝛽 is the weight when i is ahead (𝜋𝜋𝑖𝑖 > 𝜋𝜋𝑗𝑗 ). For simplicity, we drop the subscript i for the
weights that player i assigns to the other player’s payoff. Depending on the parameters 𝛼𝛼 and
𝛽𝛽, this model captures a number of distributional social preference motives discussed in the
literature. If 𝛼𝛼 = 𝛽𝛽 = 0, players are selfish and their indifference curves in the �𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 �-space
are vertical (see Figure 1a). If both 𝛼𝛼 and 𝛽𝛽 are negative, a player has competitive or spiteful
social preferences because he or she always values the other player’s payoff negatively. The
indifference curves of spiteful players in �𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 �-space are illustrated in Figure 1b. Equations
(1a) and (1b) can also capture the inequality averse preferences described in Fehr and

9
Schmidt (1999), where subjects dislike both being ahead and being behind. This preference
implies that subjects value the other player’s payoff positively if ahead but negatively when
behind, which can be captured in terms of the parameters in equations (1a) and (1b) by 𝛼𝛼 < 0
and 𝛽𝛽 > 0. 8 The indifference curves of inequality averse players are illustrated in Figure 1c
below. The key property of these indifference curves is that individuals are willing to
sacrifice resources to reduce the other player’s payoff in order to diminish disadvantageous
inequality. In the two-player case, the social preferences Bolton-Ockenfels (2000) assume are
quite similar to those of Fehr and Schmidt, i.e., the players basically dislike being behind as
well as ahead. 9 Therefore, if a player is behind, she values the other player’s preferences
negatively, while she values them positively when ahead.

It is easy to see how social preferences like those in Fehr and Schmidt (1999) and
Bolton-Ockenfels (2000) can explain that people sometimes increase and sometimes decrease
other agents’ payoffs. In the Fehr-Schmidt model, the criterion for when individuals switch
from being benevolent to malevolent is whether they are ahead or behind the other player’s
payoff. As Fehr and Schmidt (1999) point out, this criterion may not always be the
empirically relevant one – because equity is not always identical to equality – but it may
nevertheless be a useful criterion in many experimental games and real-life situations. In the
Bolton-Ockenfels model, the criterion for switching from benevolent to malevolent is
whether the player receives an equal share of the overall surplus.

8
To see this more explicitly, recall that in the two-player case inequality aversion is defined by Fehr and Schmidt
as 𝑈𝑈𝑖𝑖 = 𝜋𝜋𝑖𝑖 − 𝛼𝛼�𝑖𝑖 �𝜋𝜋𝑗𝑗 − 𝜋𝜋𝑖𝑖 � 𝑖𝑖𝑖𝑖 𝜋𝜋𝑗𝑗 > 𝜋𝜋𝑖𝑖 and 𝑈𝑈𝑖𝑖 = 𝜋𝜋𝑖𝑖 − 𝛽𝛽𝑖𝑖̅ �𝜋𝜋𝑖𝑖 − 𝜋𝜋𝑗𝑗 � 𝑖𝑖𝑖𝑖 𝜋𝜋𝑖𝑖 > 𝜋𝜋𝑗𝑗 with both 𝛼𝛼�𝑖𝑖 > 0 and 0 < 𝛽𝛽𝑖𝑖̅ <
1, implying that players dislike inequality. These utilities can be rewritten as 𝑈𝑈𝑖𝑖 = (1 + 𝛼𝛼�𝑖𝑖 )𝜋𝜋𝑖𝑖 − 𝛼𝛼�𝑖𝑖 𝜋𝜋𝑗𝑗 if 𝜋𝜋𝑗𝑗 > 𝜋𝜋𝑖𝑖
and 𝑈𝑈𝑖𝑖 = �1 − 𝛽𝛽𝑖𝑖̅ �𝜋𝜋𝑖𝑖 + 𝛽𝛽𝑖𝑖̅ 𝜋𝜋𝑗𝑗 𝑖𝑖𝑖𝑖 𝜋𝜋𝑖𝑖 > 𝜋𝜋𝑗𝑗 . Define 𝛼𝛼 = − 𝛼𝛼�𝑖𝑖 and 𝛽𝛽 ≡ 𝛽𝛽𝑖𝑖̅ to arrive at equations (1a) and (1b).
9
In the two-player case, a version of the Bolton and Ockenfels preferences can be written as 𝑈𝑈𝑖𝑖 = 𝜋𝜋𝑖𝑖 + 𝑓𝑓(𝜎𝜎)
where 𝜎𝜎 = 𝜋𝜋𝑖𝑖 /(𝜋𝜋𝑖𝑖 + 𝜋𝜋𝑗𝑗 ) measures A’s relative payoff. 𝑓𝑓(𝜎𝜎) reflects the other-regarding part of an individual’s
utility function; it is increasing in 𝜎𝜎 for 𝜎𝜎 < ½ (i.e., 𝜋𝜋𝑖𝑖 < 𝜋𝜋𝑗𝑗 ) and decreasing for 𝜎𝜎 > ½ (i.e., 𝜋𝜋𝑖𝑖 > 𝜋𝜋𝑗𝑗 ). Note, that
the preferences illustrated in Figure 1c represent a piece-wise linear approximation of Bolton-Ockenfels
preferences.

10
 Box 2: Models of Distributional Preferences
Let 𝜋𝜋𝑖𝑖 and 𝜋𝜋𝑗𝑗 represent the material payoffs of players i and j, respectively. In the Fehr-Schmidt
(1999) model, individuals are assumed to derive disutility from inequitable outcomes. For simplicity,
inequity is formalized as inequality in the experimental games under consideration. This led to the following
utility function:
�
𝛼𝛼 �𝑖𝑖
𝛽𝛽
𝑖𝑖
𝑈𝑈𝑖𝑖 = 𝜋𝜋𝑖𝑖 − 𝑛𝑛−1 ∑𝑗𝑗≠𝑖𝑖 max�𝜋𝜋𝑗𝑗 − 𝜋𝜋𝑖𝑖 , 0� − ∑ max�𝜋𝜋𝑖𝑖
𝑛𝑛−1 𝑗𝑗≠𝑖𝑖
− 𝜋𝜋𝑗𝑗 , 0�, (2)
where 𝛼𝛼�𝑖𝑖 > 0 measures the disutility from the average disadvantageous inequality while 𝛽𝛽𝑖𝑖̅ > 0 measures the
disutility from the average advantageous inequality. The Fehr-Schmidt model implies that players care
positively for others’ payoff if they are better off than the other player (advantageous inequality) and
negatively if they are worse off than the other player (disadvantageous inequality).
The model by Bolton-Ockenfels (2000) stipulates a utility function
𝑈𝑈𝑖𝑖 = 𝑓𝑓𝑖𝑖 (𝜋𝜋𝑖𝑖 , 𝜎𝜎𝑖𝑖 ), (3)
i. e., a player’s utility is a function 𝑓𝑓𝑖𝑖 that depends positively on the player’s own material payoff and on the
relative share 𝜎𝜎𝑖𝑖 that the player receives in the game. If the total payoff in the game, Π = ∑𝑛𝑛𝑗𝑗=1 𝜋𝜋𝑗𝑗 , is positive,
the relative share is given by 𝜎𝜎𝑖𝑖 = 𝜋𝜋𝑖𝑖 /Π and if the total payoff is zero, 𝜎𝜎𝑖𝑖 is assumed to be equal to 1/n. The
model assumes that individuals derive additional utility from a higher 𝜎𝜎𝑖𝑖 if 𝜎𝜎𝑖𝑖 < 1/𝑛𝑛 while if 𝜎𝜎𝑖𝑖 > 1/𝑛𝑛, a
higher 𝜎𝜎𝑖𝑖 is utility-decreasing. In the Bolton-Ockenfels model, players do not care about the payoffs of
specific other players as long as their “own” relative share 𝜎𝜎𝑖𝑖 is unaffected. In the two-player case,
individuals care positively for the others payoff if they are better off and negatively if they are worse off than
the other player.
The distributional preferences in Charness and Rabin (2002) are assumed to be given by
𝑈𝑈𝑖𝑖 = (1 − λ𝑖𝑖 )𝜋𝜋𝑖𝑖 + λ𝑖𝑖 𝑊𝑊(𝜋𝜋1 , 𝜋𝜋2 , … , 𝜋𝜋𝑛𝑛 )
= (1 − λ𝑖𝑖 )𝜋𝜋𝑖𝑖 + λ𝑖𝑖 [δ𝑖𝑖 min(𝜋𝜋1 , 𝜋𝜋2 , … , 𝜋𝜋𝑛𝑛 ) + (1 − δ𝑖𝑖 ) Π] (4)
Here, subjects value a social welfare function 𝑊𝑊(𝜋𝜋1 , 𝜋𝜋2 , … , 𝜋𝜋𝑛𝑛 ) positively with weight λ𝑖𝑖 ∈[0,1], while a
subject’s own payoff 𝜋𝜋𝑖𝑖 has weight (1-λ𝑖𝑖 ). λ𝑖𝑖 > 0 indicates “other-regardingness”, i.e., that player i cares in
some way for other’s payoff. The social welfare function has two arguments: the payoff of the least well-off
player enters with weight δ𝑖𝑖 ∈(0,1), and the sum of the payoffs Π enters 𝑊𝑊 with weight (1-δ𝑖𝑖 ). Thus, if player
i cares for others’ welfare (λ𝑖𝑖 > 1) and player i cares for the total payoff (0 < δ𝑖𝑖 < 1), player i values the
payoff of each other player positively regardless of whether player i has a higher or a lower payoff than the
other player. The Charness-Rabin model therefore predicts altruistic behavior if player i cares for the total
payoff, and this tendency is particularly pronounced towards the player with the smallest payoff.
Altruistic CES preferences, as in Andreoni and Miller (2002) and Fisman, Kariv and Markovits (2007),
take the form
𝜌𝜌 𝜌𝜌
𝑈𝑈𝑖𝑖 �𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 � = �(1 − 𝛼𝛼′)𝜋𝜋𝑖𝑖 + 𝛼𝛼′𝜋𝜋𝑗𝑗 �1/ρ (5)

where 𝛼𝛼′∈[0,1] represents the weight on the other player’s payoff, while ρ, which obeys -∞ < 𝜌𝜌 ≤ 1,
captures the trade-off between equity and efficiency. Altruistic CES preferences and Charness-Rabin type
distributional preferences share the property that the payoff of other individuals is never valued negatively.

11
 Figure 1: Indifference curves for different types of distributional preferences
other’s payoff

other’s payoff
Figure 1a Figure 1b
selfish envious

self payoff self payoff

other’s payoff
other’s payoff

Figure 1c Figure 1d
inequality
altruistic
averse

self payoff self payoff

Note: The figure shows the indifference curves of different types of distributional preferences
captured by equations (1a) and (1b) above. a. selfish preferences (𝛼𝛼 = 𝛽𝛽 = 0). b. spiteful
preferences (−1 < 𝛼𝛼 < 0, −1 < 𝛽𝛽 < 0). c. inequality averse preferences (−1 < 𝛼𝛼 < 0, 0 <
𝛽𝛽 < 1). d. altruistic preferences (0 < 𝛼𝛼 < 1, 0 < 𝛽𝛽 < 1). The slope of the indifference curve
in the domain of disadvantageous payoffs is given by ((𝛼𝛼 − 1)/𝛼𝛼) while the slope in the
domain of advantageous payoffs is given by ((𝛽𝛽 − 1)/𝛽𝛽).

How can these theories explain why other-regarding individuals behave very selfishly
in competitive market games? A simple example may suffice to illustrate this. Consider a
population of players consisting only of people with a relatively strong aversion against
unequal payoffs in the sense of Fehr-Schmidt (e.g., they reject every offer in a bilateral
ultimatum game that is below 40% of the available pie). Now put these people into the
responder position of a market game with two competing responders who face a very unfair
offer of, say, 3% of the pie. The rules of the game are such that if one responder accepts and
the other rejects, the accepting responder gets the 3% and the proposer gets 97%. If both
accept, the responder who receives the 3% will be randomly drawn and if both reject, all

12
three players receive nothing. If our fair-minded responders believe that the other responder
will accept, they both believe that they will be unable to prevent a very unequal distribution
of income. Given their beliefs, they essentially face two very unattractive options in which
the proposer always gets 97% of the pie and one of the responders gets 3%. Therefore, even
strongly inequality averse responders will accept the 3% offer (Proposition 3 in Fehr-Schmidt
1999). Fischbacher, Fong and Fehr (2009) provide empirical evidence that these belief
mechanisms are a key driver of responder behavior in games with responder competition.

The example above illustrates that the mere belief that other players behave selfishly
might sometimes induce fair-minded players to behave selfishly. More generally, perhaps the
most important message from papers such as those of Fehr and Schmidt and Bolton and
Ockenfels is that the heterogeneity of social preferences is key because – depending on the
game played - the existence of selfish players can induce fair-minded players to behave as if
they were selfish and the existence of fair-minded players can induce selfish players to
behave as if they were fair-minded. 10

The simple linear social preference model described in equations (1) can also capture the
distributional preferences assumed in Charness and Rabin (2002) that are described in more
detail in Box 2. 11 In the two-player case, the weights given to the other player’s payoff in the
disadvantageous and the advantageous domain, respectively, is given by 12
λ(1−δ)
𝛼𝛼 = 1 + λ(1−δ) (6a)
λ
β = 1 + λ(1−δ) (6b)

Thus, if i cares about welfare (λ > 0) and about the total payoff (0 < δ < 1) she values the
other’s payoff positively even in the disadvantageous payoff domain, and the valuation is
even higher in the advantageous domain. Note, however, that the only reason why a

10
Selfish proposers in the ultimatum game have an incentive to make relatively fair offers if fair-minded
responders reject low offers. Selfish “employers” in the gift-exchange game have a reason to make fair wage
offers if fair-minded workers respond to fair offers with higher effort. The selfish players in public goods with
punishment games have an incentive to cooperate if cooperative individuals punish defectors. In all of these cases,
the existence of players with social preferences can induce fundamental changes in the selfish players’ incentives.
11
In this section, we discuss only the distributional preferences discussed in Charness and Rabin (2002). They
also extend their distributional preferences model with negative reciprocity to account for behaviors where players
incur cost to reduce the payoffs of “misbehaving” players. We deal with preferences for reciprocity in Section
3.3.
12
In case that i earns more than j, utility is given by 𝑈𝑈𝑖𝑖 = (1 − λ)𝜋𝜋𝑖𝑖 + λ[δ 𝜋𝜋𝑗𝑗 +(1 − δ)( 𝜋𝜋𝑖𝑖 + 𝜋𝜋𝑗𝑗 )] while in
case that j earns more than i, it is given by 𝑈𝑈𝑖𝑖 = (1 − λ)𝜋𝜋𝑖𝑖 + λ[δ 𝜋𝜋𝑖𝑖 +(1 − δ)( 𝜋𝜋𝑖𝑖 + 𝜋𝜋𝑗𝑗 )]. Reformulating the
first function in terms of the weight 𝛼𝛼 on the other player’s payoff and the second function in terms of the weight
β on the other’s payoff yields the expressions in the text.

13
Charness-Rabin player cares for the other player’s payoff in the disadvantageous domain is a
concern for the total payoff. Figure 1d illustrates the indifference curves this model implies.

From a behavioral viewpoint, the Charness-Rabin preferences are qualitatively quite

similar in the two-person case to non-linear other-regarding preferences as captured by a CES
utility function (see Box 2). Andreoni and Miller (2002) and Fisman, Kariv and Markovits
(2007) have assumed CES preferences. However, while the CES utility function is capable of
capturing very similar behaviors, the Charness-Rabin model provides a psychological
foundation for altruistic behaviors in terms of a motive to increase the total surplus and to
help the worst-off player. Knowing that these motives may underlie other-regarding
behaviors can provide a deeper understanding of these behaviors.

As in Charness and Rabin (2002), subjects with these CES preferences always value others’
payoffs positively and preferences are linear and given by 𝑈𝑈𝑖𝑖 �𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 � = (1 – 𝛼𝛼)𝜋𝜋𝑖𝑖 + 𝛼𝛼𝜋𝜋𝑗𝑗 in
the special case of ρ = 1. It is well known that preferences are Cobb-Douglas for the case of ρ
= 0, while social preferences take the Rawlsian form 𝑈𝑈𝑖𝑖 �𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 � = min(𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 ) for 𝜌𝜌 → −∞. In
the latter case, indifference curves in the �𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 �-space become horizontal for 𝜋𝜋𝑖𝑖 > 𝜋𝜋𝑗𝑗 and
vertical for 𝜋𝜋𝑖𝑖 < 𝜋𝜋𝑗𝑗 with a kink at the 45o line, implying that subjects’ social preferences are
strictly egalitarian. These egalitarian preferences arise in Charness and Rabin for the two
player case if subjects completely disregard their own payoffs (i.e., λ = 1) and the sum of
payoffs (i.e., δ = 1), but care only for the payoff of the least well-off player.

As in Charness and Rabin, CES preferences can thus capture a high willingness to pay
to increase the other’s payoff when ahead and a very low willingness to pay when behind.
However, similar to Charness and Rabin’s distributional preferences, CES preferences cannot
capture subjects’ willingness to reduce other’s payoff at a cost to themselves because the
other player’s payoff is never valued negatively.

3.1.2. Empirical Frequency of Different Distributional Preferences

In the following, we will discuss the empirical properties of distributional preferences that
emerged from more systematic attempts to identify the relevant parameters. Fehr-Schmidt
(1999) and Bolton-Ockenfels (2000) were inspired by the evidence from the paradigmatic
experiments. Their papers provided a unifying account for important, and seemingly
contradictory, empirical regularities but they did not involve a systematic attempt to identify

14
the model parameters empirically. To achieve identification of the parameters of
distributional preferences like those in equations (1), one needs information about subjects’
behavior on many budget lines with different slopes in the �𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 �-space. In other words, one
needs a large set of generalized dictator games in which one subject can unilaterally
determine the distribution of payoffs between herself and another subject at various costs of
redistributing payoffs. 13 By systematically varying the slope of negatively sloped and
positively sloped budget lines, it is possible to identify individuals’ indifference curves in the
domain of advantageous and disadvantageous inequality and to determine which of the
fundamentally difference distributional preference types illustrated in Figure 1 describes the
behavior of (sub)groups or individuals best. 14

Concern for equality or concern for the total payoff?

An important step towards answering this question was undertaken by Charness and Rabin
(2002) who recruited roughly 220 students for their dictator games. Their structural estimate
reveals that subjects on average put a positive weight of about 𝛽𝛽 ≈ 0.4 on the other player’s
payoff when ahead, but the 𝛼𝛼 parameter, which captures distributional preferences when
behind, fails to be significantly different from zero although it is positive with 𝛼𝛼 = 0.23 (t =
1.1). Overall, the authors interpret this as fairly strong evidence for social welfare motives as
formalized in their model and against inequality aversion motives, since a non-negative value
for 𝛼𝛼 means that subjects are not willing to reduce the other player’s payoff when behind, and
a positive value for 𝛼𝛼 even means that subjects positively care for the total payoff.

For example, when player A faced the choice between (200 for self, 800 for B) versus
(0, 0), none of their student subjects was willing to pay 200 to implement the (0, 0)

13 To identify more complex social preferences like reciprocity (Rabin 1993), guilt aversion (Battigalli and

Dufwenberg 2007), or Kantian Morality models (Alger and Weibull 2016), one needs to move beyond
generalized dictator games and introduce some strategic component into the game.
14 There is a considerable literature on social value orientation (SVO) in psychology that uses the so-called

ring measure of SVO (Liebrand 1984; Liebrand and Mcclintock 1988) and/or the triple-dominance measure
of SVO (Van Lange et al. 1997; Van Lange 1999) or the slider task (Murphy, Ackermann and Handgraaf
2011; Murphy and Ackermann 2014). These measures are based on generalized dictator games, but they
do not lend themselves easily to the estimation of utility functions. For example, they do not allow us to
differentiate between the four qualitatively different types of indifference curves displayed in Figure 1.
Instead, they use measures such as the ratio between the total payoff given to the other player and the total
payoff assigned to “self” (across all dictator games) to assign individuals to predefined SVO types such as
“cooperative” (= desire to maximize joint gains), “altruistic” (= desire to maximize the other player’s
payoff), and “competitive” (= desire to maximize the payoff difference). In particular, the SVO measures
cannot identify inequality aversion as defined in Fehr and Schmidt (1999) or Bolton-Ockenfels (2000). For
this reason, we do not use them in the following review of distributional preferences.

15
allocation, consistent with the claim that inequality aversion is absent (i.e., 𝛼𝛼 is non-
negative). In addition, roughly half of their subjects preferred the allocation (400, 400) over
(375, 750), i.e., they were not willing to sacrifice 25 to generate a gain of 350 for the other
subject. Thus, these subjects showed limited altruistic tendencies when their choices would
generate inequality, which illustrates why 𝛼𝛼 is not significantly different from zero. To see
why 𝛽𝛽 is clearly positive, consider the choice between (700, 200) and (600, 600); 73% of the
subjects preferred the (600, 600) allocation in this case.

Charness and Rabin (2002) primarily used a between-subjects approach, and therefore
did not have enough observations per subject to estimate individual utility functions. They
estimate a representative individual’s utility function, which necessarily neglects
heterogeneity. For example, while 50% of the subjects refused to pay 25 to implement (375,
750) instead of (400, 400), there was also the 50% who did so. In addition, they find that 33%
of their subjects preferred the allocation (500, 700) over (600, 300), meaning that these
subjects sacrifice 100 to give the other party 400 although this generates inequality to their
disadvantage. 15 These examples show that estimating an “average” subject’s preferences
hides important heterogeneity: although the average value of 𝛼𝛼 is not significantly different
from zero, a substantial share of the students in Charness and Rabin displays behavior
consistent with a concern for the total payoff.

Engelmann and Strobel (2004) report strong evidence for a motive to maximize the
group’s payoff (“efficiency motive”). In their experiments, undergraduate economics and
business administration students had to choose between different three-person allocations that
enabled the authors to disentangle the motive (i) to maximize the groups’ joint payoff, (ii) to
reduce overall inequality as predicted by the Fehr-Schmidt or the Bolton-Ockenfels model, or
(iii) to maximize the payoff of the worst-off individual in the three-person group (max-min
motive). They find that motive (i) and (iii) play a substantial role, while motive (ii) appears to
play no significant role in three-person dictator games in which the decision-maker is
indifferent (or close to indifferent) between the available options.

The generality of this conclusion has, however been challenged by showing that non-
economists put a substantially higher weight on equality – compared to the maximization of
the group’s overall payoff – than economists do (Fehr, Naef and Schmidt 2006). While a
majority of economists and business administration students preferred total payoff

15
Note the comparison to the (700, 200) and (600, 600) case; many people will sacrifice for equality, but few
will do so when this means coming out behind.

16
maximization over equality, various groups of non-economists, ranging from students of
various other disciplines to low-level employees of banks and financial institutions, showed
the opposite pattern. In addition, the strength of the total payoff motive may be weakened if
the decision-maker’s self-interest is at stake because she is not indifferent between the
available options. Bolton-Ockenfels (2006) show results indicating that a considerably larger
share of subjects is willing to deviate from their self-interest when the deviation generates a
more equal outcome compared to when it generates a higher group payoff. 16 The debate
suggests the importance of moving beyond (i) traditional student subject pools and (ii)
considering a larger class of games to arrive at firmer conclusions.

Many other studies have, in the meantime, estimated the structural parameters α and β
of model (1) above based on observations from strategic and non-strategic games. A recent
meta-analysis by Nunnari and Pozzi (2022) uses data from 42 laboratory studies with a total
of 289 estimates of the different populations’ average α and β values. The authors use state-
of-the-art methods to estimate the overall distribution of parameters across populations and
find weighted mean values for α = - 0.469 and β = 0.331, both of which are significant at the
p < 0.001 level. 17 This indicates that the observed behaviors are consistent with inequality
aversion on average. A similar qualitative conclusion is reached if one restricts attention to
the non-strategic games, which arguably offer the cleanest preference interpretation of the
estimated parameters: the mean values of α and β are then given by α = - 0.239 and β =
0.369 (both significant at p < 0.001), again indicating that subjects display inequality aversion
on average. In addition, these results show a considerable change in parameters between
strategic and non-strategic games, indicating that disadvantageous inequality aversion
appears much more important in strategic games, while aversion towards advantageous
inequality appears less important in strategic games.

The results discussed above suggest that there is substantial heterogeneity between
individuals and different subject pools. Measuring this heterogeneity is therefore important.
One possibility for characterizing a group’s heterogeneous social preferences is to estimate a
utility function for each individual. This approach has been applied to student subject pools in

16
In the three-player games of Engelmann and Strobel (2004), the subjects decide behind a veil of ignorance with
regard to their role in the game. Iriberri and Rey-Biel (2011) show that this role uncertainty favors social welfare
maximizing preferences (i.e., efficiency plus max-min motive) and underestimates selfish and inequity averse
preferences. In their study, the share of social welfare maximizers drops from 74% with role uncertainty to 21%
without role uncertainty.
17
In their state-of-the art in meta-analytic studies, Nunnari and Pozzi (2021) weigh the estimates of individual
studies according to their standard errors such that more precise estimates receive a higher weight.

17
seminal papers by Andreoni and Miller (2002) and Fisman, Kariv and Markovits (2007), who
estimated individuals’ CES utility functions (see Box 2). Their results indeed show enormous
heterogeneity in subjects’ distributional preferences. In addition, the Fisman-Jakiela-Kariv-
Markovits group shows in a series of papers striking differences between students and the
general population’s distributional preferences: the general population is much more other-
regarding (i. e., has a much higher 𝛼𝛼′ in equation (5)) and puts a much higher weight on
equality (i.e., displays 𝜌𝜌 < 0 in equation (5)) compared to students. This also holds, if one
controls for age. 18 Because of space limitations and because the CES approach neglects
important classes of distributional preferences –inequality aversion and envy – we describe
the results of these studies in more detail in Appendix 1.

How prevalent are inequality aversion and spite?

Kerschbamer (2015) developed a systematic approach – the Equality Equivalence Test (EET)
– that enables the identification of all four distributional preference types displayed in Figure
1. For this purpose, subjects are presented with choice lists in the domain of disadvantageous
inequality (DA-lists) and the domain of advantageous inequality (A-lists). In any given list,
the subjects face a series of binary choices where the equal payoff distribution E is always
paired with an alternative allocation (See Figure A2 in Appendix 2). A list essentially
confronts subjects with a series of positively and negatively sloped budget lines in the “self-
payoff / other payoff” space with two discrete options on each budget line.

Table 1 below shows the results of a relatively large number of EETs in student
samples (Kerschbamer (2015); Balafoutas, Kerschbamer & Sutter (2012); Paetzel,
Sausgruber & Traub (2014) ; Balafoutas et al. (2014); Krawczyk and Lee (2021)) and in
nationally representative population samples (Chapman et al. 2018; Kerschbamer and Muller
2020; Hedegaard et al. 2021).

18 The large differences between student samples and the broader population are consistent with research

reported in Snowberg and Yariv (2021) and Cappelen et al. (2015).

18
 Table 1 Empirical Frequency of Different Distributional Preference Types
in the Equality Equivalence Test

Preference types

Subject Pool Altruistic Inequality Envy/ Selfish

Study
(𝛼𝛼 ≥ 0; Averse Spite
(𝛼𝛼 ≈ 0;
𝛽𝛽 ≥ 0) (𝛼𝛼 < 0; (𝛼𝛼 =≤ 0;
𝛽𝛽 > 0) 𝛽𝛽 ≈ 0)
𝛽𝛽 ≤ 0)

Kerschbamer N = 92 33.7% 11.9 % 3.2% 48.9%

(2015) Univ. Innsbruck
Austria

Krawczyk & N = 101 48.5% 11.9% 9.8% 28.7%

Lee (2021) Univ. Warsaw
Poland
Balafoutas et al.
N = 132 28% 8.3% 5.3% 58.3%
(2012)
Student Univ. Innsbruck
Austria
Samples
Balafoutas et al.
N = 195 49.7% 6.7% 7.2% 33.8%
(2014)
Univ. Innsbruck
Paetzel et al.
N = 280 29.3% 9.3% 1.8% 58.3%
(2014)
Univ. Bremen
Germany

Hedegard et al.
N = 885 47.1% 23.2% 8.6% 20.0%
(2021)
Denmark

Kerschbamer & N = 2794 13.4 % 64.8% 14.0% 5.0%

Müller (2020) Germany
Nationally
represent-
Chapman et al. tative N = 1000 27.5% 41.9% 8.3% 16.6%
(2018) samples USA

2nd Wave of N = 2583 11.5% 67.8% 11.0% 7.9%

German Internet
Germany
Panel 19

Note: The equality equivalence test (EET) measures individuals’ distributional preferences by
identifying the slope of their indifference curves in �𝜋𝜋𝑖𝑖 , 𝜋𝜋𝑗𝑗 � space both in the domain of
disadvantageous inequality (providing a measure of the sign of 𝛼𝛼) and the domain of advantageous
inequality (providing a measure of the sign of 𝛽𝛽). To be classified as altruistic, either 𝛼𝛼 or 𝛽𝛽 or both
have to be strictly positive. To be classified as envious/spiteful, either 𝛼𝛼 or 𝛽𝛽 or both have to be strictly
negative. Individuals are classified as inequality averse if 𝛼𝛼 < 0 and 𝛽𝛽 > 0 holds.

19The results in the Kerschbamer and Müller (2020) paper are based on a first application of the EET to the
German Internet Panel (GIP) in 2016. In the meantime, the EET was performed with this sample a second time in
2018. We therefore also include the results of this second wave of data collection to illustrate the stability of the
preference classification over time in Table 1.

19
 Several striking facts emerge from these studies. First, the share of selfish individuals is
much larger in the student samples, where it varies between 29% and 58%, while it varies
only between 5% and 20% in the nationally representative samples. Second, the much larger
share of other-regarding subjects in the nationally representative samples is primarily due to a
much larger share of inequality averse individuals in these samples. While the share of
inequality averse individuals in student samples varies between 7% and 12%, this share is
between 23% and 68% in the nationally representative samples. Third, the share of envious
subjects is also slightly higher in the nationally representative samples, where it varies
between 8% and 14%, while in the student samples it varies between 2% and 10%.

Finally, the share of individuals with altruistic preferences is considerably larger in

student samples compared to nationally representative samples. The average share of
altruistic subjects over all student samples amounts to 37%, while the average share of
altruists is only 19% in the nationally representative samples. Thus, the general population
seem much more other-regarding and this higher share of other-regarding individuals reflects
a much larger share of inequality averse subjects and a somewhat larger share of envious
subjects. Interestingly, however, altruism is more prevalent in student samples.

The existence of large differences between the student samples and the general
population samples are corroborated by two earlier studies by Bellemare, Kröger and van
Soest (2008; 2011). Using a representative sample of the Dutch population that played
ultimatum and dictator games, they estimated a non-linear version of the Fehr-Schmidt model
and find that the young and educated individuals in their sample display a substantially lower
aversion against disadvantageous inequality than the rest of the sample.

How many endogenous preference clusters?

The EET assigns subjects to pre-defined preference categories based on the prevailing
theories of distributional preferences. However, these predefined categories might not
appropriately capture the empirical distribution of social preferences. Perhaps the empirical
preference clusters defy these predefined categories by, e. g., clustering around several
distinct altruistic types or several distinct inequality averse types. If that were the case,
lumping all altruistic or all inequality averse types together may overlook important aspects
of the broader distribution of preferences.

20
 Recently, Fehr®Epper®Senn (2022) applied a non-parametric Bayesian clustering
mechanism (“Dirichlet Process Means, DPM) to the data of four broadly representative Swiss
samples from which they had collected measures of distributional preferences. In each trial of
the preference elicitation task, the subjects in these samples faced seven discrete payoff
allocations located on a linear budget line that is either positively or negatively sloped (see
Figure A3 in Appendix 3).

An important aspect of the DPM approach is that it enables identification of preference

clusters without committing to a pre-specified number of preference types. The algorithm
allows for all possible type partitions of the data spanning from a representative agent (i.e., a
single data-generating process) up to as many types as there are individuals in the population
(i.e. n data-generating processes), i.e., it determines the number of preferences types
endogenously. In addition, it assigns each individual to one of the types, and the behavioral
(preference) properties of the types emerge endogenously.

Which preference types emerge from the Bayesian clustering algorithm? The DPM
approach reveals the existence of three different preference clusters. Figure 2 below provides
a graphical illustration of the observed clusters in all four Swiss data sets. Each dot captures
(i) one individual’s modal choice on the positively sloped budget line (horizontal axes) and
(ii) that individual’s modal choice on the negatively sloped budget line (vertical axes). The
figure clearly shows that the individuals form three clusters in each of the four data sets. In
the inequality averse cluster, the modal choice on the positively sloped budget lines and on
the negatively sloped budget lines is the equal payoff allocation. In the altruistic cluster,
individuals’ modal choice is the equal payoff allocation on the negatively sloped budget lines
and the self-payoff maximizing choice on the positively sloped budget lines. And in the
predominantly selfish cluster, the modal choice on negatively and positively sloped budget
lines is the self-payoff maximizing choice. Quantitatively, the distribution of individuals to
the three preference types is as follows (see Appendix Table A3): (i) Between 30 and 40
percent of the subjects are assigned to an altruistic cluster. (ii) Between 45 and 53 percent are
assigned to an inequality averse cluster and (iii) between 10 and 24 percent are assigned to
predominantly selfish cluster.

21
 Figure 2: Preference types in four broad samples of the Swiss population

Note: The figures depict subjects’ modal choices z among negatively sloped budget lines and among
positively sloped budget lines. The figures look almost identical if we depict subjects’ median choices.
Each dot represents one individual. Dots are jittered to make identical choices of individuals visible.
For each budget line, z = 6 indicates an own-payoff maximizing choice, z = 0 indicates an own-payoff
minimizing choice, and z = 3 indicates a payoff-equalizing choice. Note that z = 6 simultaneously
maximizes own payoff and the other individual’s payoff on positively sloped budget lines, while z = 6
maximizes the own payoff and simultaneously minimizes the other’s payoff on negatively sloped
budget lines. Panel (a) is constructed using panel subjects and their 2017 choices. Panel (b) is
constructed using panel subjects and their 2020 choices. Panel (c) is constructed using the choices of
individuals who only participated in the 2017 study. Panel (d) is constructed using the choices of
individuals who only participated in the 2020 study. The data are taken from Fehr®Epper®Senn (2022).

22
To what extent is the existence of the three qualitatively distinct clusters displayed in Figure
2 a peculiarity of the Swiss data? To answer this question, we analyzed the data from a large
Danish sample of N = 3691 adult individuals, taken from Epper et al. (2020). The Danish
data show a strikingly similar pattern to the four samples of the Swiss data presented in
Figure 2, making it immediately apparent that most individuals cluster into three distinct
behavioral types – altruistic, inequality averse, and predominantly selfish (see Figure A4 in
Appendix 3). Table A3 reports the results of the Bayesian clustering algorithm for the Danish
data set which formally confirms the clustering into the above-mentioned three types. And
again, the inequality averse type is the most frequent one with 37.3% of the subjects, while
32.5% are predominantly selfish and 30.2% are altruistic.

Taken together, the data from broad population samples indicate the existence of three
important distributional preference types: a large cluster of inequality averse subjects, another
large cluster of altruistic subjects, and a third cluster of predominantly selfish subjects. Often,
the predominantly selfish cluster comprises the lowest share of individuals. The frequency of
envious/spiteful subjects is generally too small and their location in preference space is too
scattered (see Figure 2), meaning that they do not constitute a distinct cluster but are
subsumed either under the selfish and the inequality averse cluster.

The data from the broad population samples differ strongly from the behavioral patterns
observed in student samples. In the latter, the share of inequality averse subjects is much
smaller and sometimes almost nonexistent, while the share of selfish subjects is much larger.
In the distributional preference domain, data from student subject pools thus clearly deviate
from the preferences of the broader population in important ways.

Finally, why does the selfish type typically only constitute a (nonnegligible) minority in
broad population samples? In our view, this is because if one offers subjects many different
choice situations with sufficiently small costs of engaging in generous or inequality reducing
behaviors, one can identify purely selfish types and individuals who are very close to purely
selfish types (who are never willing to pay anything or only extremely low amounts to help or
hurt others) with great precision. Apparently, many individuals in the broader population are
at least willing to pay some measurable positive amount for increasing and/or decreasing
others’ payoffs.

23
3.2. The Role of Merit, Luck, and Risk in Distributional Preferences

In the standard dictator game and other pure distribution games, the experimenter provides
the income that can be divided among the parties exogenously. Moreover, the role of the
powerful player, who can make a unilateral allocation decision, is not based on some sort of
merit but assigned randomly, and the subjects typically interact anonymously with each other
and have no information about the other players’ income, wealth, and social background. It
seems natural that no player in such an environment has a priori a greater normative claim on
the available resources compared to the other players, i.e., equality is a natural reference point
for judging the fairness and equitability of outcomes. It is, however, also clear that social
background, the saliency and status of particular individuals, the social proximity among
individuals, as well as their effort and contribution to the available resources, including the
risks and conditions under which they had to produce these resources, can play a role in what
is considered an equitable distributional claim. The potential importance of these factors is
widely acknowledged and will often drive a wedge between equality and equity (fairness)
and offers ample opportunities for examining these factors in a controlled way. 20

In this context, the notions of merit, luck, and effort play a key role. In many modern
societies. meritocracy appears to be a widely held normative ideal. According to a prominent
theory of equality of opportunity (Roemer and Trannoy 2015), individuals should be held
responsible for factors under their own control, while they should not be held responsible for
factors beyond their control. Based on this approach, there is no normative reason to
redistribute resources that are entirely generated by an individual’s effort, while earnings that
accrue at least partly from non-controllable external factors (“luck”) should be subject to
redistribution.

3.2.1. Equity and Entitlements

Psychologists and sociologists (Homans 1961; Adams 1963; Adams 1965) have
developed positive theories of justice that incorporate a widely applicable merit principle.

20 The very fact that equity and equality may diverge induced Fehr and Schmidt to first call the preferences
they examine as “inequity aversion”, i.e., “a general dislike for outcomes that are perceived as inequitable”.
However, under the special circumstances in the laboratory experiments they examined, where subjects
entered the laboratory as equals, knew nothing about each other’s background, and were randomly
allocated to different roles, equality appeared to be a more reasonable reference point.

24
According to this view, individuals perceive inequity if their outcome/input ratios diverge
from the ratio of their relevant comparison agent. The outcomes and inputs that enter the
equity calculus have typically been interpreted very broadly, i.e., any kind of reward that the
individuals experience can represent an outcome and any kind of perceived contribution to
the outcome can represent an input.

Equity theory thus allows for many subjective influences that may make its predictions
rather malleable and imprecise. However, if restricted to more easily measurable outcomes
(e.g., received material payoffs) and inputs (e.g., time spent on a task or material payoff
contributed to the group payoff), it makes sharper predictions. Mikula (1973) conducted, for
example, an experiment with recruits from the Austrian Army to examine whether subjects
follow the equity principle or the equality principle. Two matched parties produced a joint
monetary payoff, and then one of the individuals could unilaterally allocate this payoff to the
two parties. He found that the lower performing parties assigned themselves more than the
proportional monetary payoff equity theory predicted but less than the equal monetary
payoff. Likewise, the higher performing parties requested on average more than the equal
split but less than the reward predicted by equity theory. 21 More generally, the data confirm
the qualitative equity theory prediction that higher performing parties request (if they have
the power to decide) and are conceded (if the partner has the power to decide) a higher share
of the joint payoff. Likewise, lower performing parties allocate themselves and are allocated
a lower share of the joint payoff (Leventhal and Michaels 1969; Leventhal and Anderson
1970; Leventhal and Lane 1970; Lane and Messe 1971; Leventhal and Michaels 1971). Thus,
according to this literature, subjects behave as if they feel entitled to a larger (smaller) share
if they contribute more (less) to the joint payoff than their experimental counterpart.

One drawback in these experiments was that participants were typically deceived in
various ways. Often, there was no actual working partner who contributed to a joint surplus
or the performance ratios to which the parties were randomly allocated did not reflect the
subjects’ actual performances, or individuals had to make hypothetical choices without real
economic consequences. These practices may have generated doubts about the credibility of
the implemented procedures or might have affected subjects’ behavior in other ways. For this
reason, it makes sense to ask whether the notion of “earned entitlements” or “earned property

21 For example, when the performance ratio between the lower performing and the higher performing
party was between 62.5% and 37.5%, the higher performing parties requested on average 54.5% of the
joint payoff, while the lower performing parties requested 42.7%.

25
rights” implied by equity theory is indeed robust to the methods used in experimental
economics which rule out deception while implementing designs with transparent economic
consequences for the involved parties.

The experimental economics literature (e.g., Hoffman et al. (1994), Ruffle (1998), Fahr
and Irlenbusch (2000) Konow (2000); Cherry, Frykblom and Shogren (2002); Frohlich,
Oppenheimer and Kurk (2004); Cappelen et al. (2007); Krawczyk (2010); Lefgren et al.
(2016)) also strongly suggests that “earned entitlement” effects exist. For example, Hoffman
et al. (1994) and Cherry et al. (2002) show that the dictators take a higher share of the pie
when they earned the role of the dictator in a quiz or when they generated the pie that can be
distributed themselves. 22 While Hoffman et al. (1994) and Cherry et al. (2002) show that the
dictators behave more selfishly if they have acquired an earned entitlement, Konow (2000)
shows that the dictators also respect the recipients’ earned entitlement, i.e., the recipients’
earned entitlements also constrain the dictators’ selfish behavior. Subjects in Konow’s
experiment jointly produce the pie to be distributed in the subsequent dictator game in a
(letter production) task that enables the exact measurement of the matched subjects’ relative
contribution (i.e., the number of letters). He hypothesizes that fairness considerations will
induce the dictators to tilt the allocation given to the recipients towards their relative
contribution in the production task – a finding that his data nicely corroborates.

In a different design, Ruffle (1998) also shows that the recipients’ earned entitlements
constrain the dictators. He implemented a general knowledge quiz that was used to establish a
contest among the recipients that determines the pie size available in a subsequent dictator
game with a paired dictator (who did not participate in the contest). If dictators are paired
with a winning recipient (i.e., one with a high number of correctly answered questions), the
pie size is set at $10, while the pie size is $4 for a losing recipient. In a control treatment, the
outcome of the competition among recipients is determined randomly, i.e., by luck. Ruffle
finds that the dictators in the “quiz condition” allocate significantly higher incomes to the
winning recipients than in the “luck condition”, thus honoring the recipient’s contribution to

22Hoffman et al. (1994) also claim that if one not only introduces anonymity between the subjects but also
anonymity between the subjects and the experimenter, the dictators behave more selfishly and take a
higher share. However, subsequent research (Frohlich, Oppenheimer and Kurk 2004) suggests that the way
Hoffman et al. implemented experimenter-subject anonymity also generated doubts among the subjects
about whether the recipient was a real person, which may well induce selfish behavior. Other research on
double anonymity (e.g., Bolton, Katok and Zwick 1998; Barmettler, Fehr and Zehnder 2011) indicates that
lack of subject-experimenter anonymity has only minor, insignificant effects in dictator, ultimatum, and
trust games.

26
the available pie in the quiz condition. Fahr and Irlenbusch (2000) also provide strong
evidence indicating that trustees in a trust game strongly respect earned property rights.

3.2.2. Modelling Entitlement Effects

How should we model the motivational forces underlying the earned entitlement effect? To
set the stage for this discussion, it is useful to rely on a concrete example that is taken from
the seminal paper of Cappelen et al. (2007). In their experiment, subjects participate first in
two production tasks where they can finance an investment qi that generates an output xi =
aiqi from an identical endowment E. Each participant was randomly assigned to a high (ai =
4) or a low (ai = 2) rate of return on investment.

After the two production tasks, each subject was matched twice randomly with a
different partner and played a bilateral dictator game with each of the two partners. The pie
size in each dictator game was given by the paired players’ jointly produced total output X =
a1q1 + a2q2. Note that some pairs in this experiment face a total output produced with
identical returns on investment (a1 = a2) and identical investments (q1 = q2), while others can
distribute an output that is produced with different rates of return (a1 ≠ a2) and/or different
investment levels (q1 ≠ q2).

Which allocation would a sufficiently inequality averse dictator choose in this game?
This type of dictator would implement the allocation that equalizes the two players’ incomes.
If we denote s1 (s2) as player 1’s (player 2’s) share of the total payoff X, income equalization
Implies E – q1 + s1X = E – q2 + s2X. From this follows that player 1’s share is given by
𝑞𝑞1 −𝑞𝑞2
𝑠𝑠1 = 𝑠𝑠2 + . (7)
𝑋𝑋

Thus, inequality aversion is not identical to strict egalitarianism that would distribute the
earnings equally regardless of how they were generated, but it incorporates a meritocratic
aspect because it implies that differences in subjects’ investment levels are honored. An
individual receives a higher share if he or she contributed to higher output levels via higher
investments. This follows simply from the fact that the principle of equality is applied to all
material payoffs, and effort or investment costs are certainly part of this.

Another interesting aspect of inequality aversion is that it does not honor those
contributions to earnings that have nothing to do with the costs incurred to produce the
earnings. In the Cappelen et al. experiment, this means that inequality aversion does not

27
honor the luck of those participants who were randomly assigned a higher rate of return ai. In
other words, if two subjects invested the same amounts but faced different rates of return and
thus produced vastly different contributions to the overall earnings, inequality aversion
assigns both individuals the same share of the total earnings.

Which allocation would a dictator who is motivated by the forces stipulated by equity
theory implement in the Cappelen et al. experiment? He or she would equalize the
output/input ratios between the players. While a plausible interpretation of the “output”
seems to be the earnings received, there is more ambiguity when considering the inputs. This
depends crucially on whether one considers the investment levels qi or the contributions to
the total earnings, xi = aiqi, as an “input”. However, if we view equity theory as embodying a
merit principle, the appropriate input measure appears to be investment levels, implying that
equity is established once s1X/q1 = s2X/q2 or s1/q1 = s2/q2 holds.

If we denote the players’ ratios of investments by θ = q1/q2 and take into account that s1
+ s2 = 1, the players’ equitable payoff shares are given by
𝜃𝜃 𝑞𝑞1 1 𝑞𝑞2
𝑠𝑠1𝐸𝐸 = 1+ 𝜃𝜃
= and 𝑠𝑠2𝐸𝐸 = = . (8)
𝑞𝑞1 + 𝑞𝑞2 1+ 𝜃𝜃 𝑞𝑞1 + 𝑞𝑞2

Thus, the equitable payoff shares are given by the players’ relative investment shares. This
coincides with the notion of fairness advocated by Konow (2000) and with the notion of
liberal egalitarianism in Cappelen et al. (2007). Note that the notion of equity described in (8)
also implies that liberal egalitarians do not honor luck – in the form of randomly assigned
investment returns – while honoring merit – in the form of players’ relative investment
shares.

What kind of utility function rationalizes the behavior of players who are motivated
according to equity theory? A plausible function could look similar to the Fehr and Schmidt
utility function where the reference point is no longer the other players’ payoff, but the
dictator’s own equitable payoff as defined by 𝑠𝑠𝑖𝑖𝐸𝐸 , and income is denoted by 𝑦𝑦𝑖𝑖 :

𝑈𝑈𝑖𝑖 = 𝑦𝑦𝑖𝑖 − 𝛼𝛼 𝑚𝑚𝑚𝑚𝑚𝑚[𝑠𝑠𝑖𝑖𝐸𝐸 𝑋𝑋 − 𝑦𝑦𝑖𝑖 , 0] − 𝛽𝛽 𝑚𝑚𝑚𝑚𝑚𝑚[𝑦𝑦𝑖𝑖 − 𝑠𝑠𝑖𝑖𝐸𝐸 𝑋𝑋, 0] (9)

A dictator with this utility function would never give herself less than 𝑠𝑠𝑖𝑖𝐸𝐸 𝑋𝑋 and would allocate
herself exactly 𝑠𝑠𝑖𝑖𝐸𝐸 𝑋𝑋 if 𝛽𝛽 > 1. In contrast, Cappelen et al. (2007) assumed that deviations from
the equitable payoff impose symmetric (quadradic) costs on subjects, which led them to
postulate the utility function

28
 2
�𝑦𝑦𝑖𝑖 − 𝑠𝑠𝑖𝑖𝐸𝐸 𝑋𝑋�
𝑈𝑈𝑖𝑖 = 𝑦𝑦𝑖𝑖 − 𝛾𝛾 , (10)
2𝑋𝑋

The two utility functions above imply that there is a nonpecuniary disutility from receiving an
inequitably large share in case of 𝑦𝑦𝑖𝑖 > 𝑠𝑠𝑖𝑖𝐸𝐸 𝑋𝑋. Thus, because the players’ income shares add up
to one, subjects in this situation are willing to reduce their own income and behave
generously towards the other player. Likewise, the associated nonpecuniary disutility from
receiving a too low share generates a willingness to pay to behave in an envious manner
towards the other player in case of 𝑦𝑦𝑖𝑖 < 𝑠𝑠𝑖𝑖𝐸𝐸 𝑋𝑋.

Because social preferences are typically characterized by substantial heterogeneity, it is

unlikely that all individuals will behave in an inequality averse manner (i.e., obey (4)) or
according to utility functions (9) or (10). Therefore, to account for potential heterogeneity,
Cappelen et al. (2007) introduced two additional “fairness ideals” that deviate from the
meritocratic ideal of liberal egalitarians. The call them the libertarian ideal (𝑠𝑠𝑖𝑖𝐿𝐿 ) and the
strictly egalitarian ideal (𝑠𝑠𝑖𝑖𝑆𝑆𝑆𝑆 ). According to a strictly egalitarian ideal, the earnings should be
distributed between the players regardless of their rates or return and their investment levels,
i.e., 𝑠𝑠𝑖𝑖𝑆𝑆𝑆𝑆 = 1/2. In contrast, each player in the libertarian ideal is entitled to the earnings he or
she produced regardless of any luck or effort considerations, i.e., 𝑠𝑠𝑖𝑖𝐿𝐿 = aiqi/X.

To what extent are the different fairness ideals capable of capturing earned entitlement
effects? Liberal egalitarians who behave according to utility functions (9) or (10) will exhibit
earned entitlement effects if they contributed more effort or investments to the production
process. Inequality averse individuals, who behave according to (7), will display a similar
earned entitlement effect. In contrast, libertarians with a fairness ideal of 𝑠𝑠𝑖𝑖𝐿𝐿 = aiqi/X will
show an earned entitlement effect even if they provide higher output merely because of luck
(i.e., a higher ai).

The introduction of heterogenous fairness ideals by Cappelen et al (2007) represents an

interesting innovation; it implies that if people do not share in one particular situation, then
this may not mean that they do not care about fairness (i.e., put no weight on fairness in (9) or
(10)) but that they consider sharing unfair. Many interpretations of experiments overlook this
distinction and too often interpret people not sharing as selfishness. The distinction between
the weight put on fairness and the concrete fairness ideals also opens an avenue towards
studying the impact of institutions on distributional preferences.

29
 It has been argued, for example, that studying economics makes people more selfish
(possibly true, but the evidence is not entirely clear); but this approach misses the second
(perhaps even more important) dimension, that studying economics may change people’s
view of fairness. More generally, fairness ideals are likely to be shaped by educational and
political institutions as well as by the general moral infrastructure of societies.

Finally, what are the empirical results of the Cappelen et al. (2007) study? Based on the
assumption that there are three types of fairness ideals – 𝑠𝑠𝑖𝑖𝐸𝐸 , 𝑠𝑠𝑖𝑖𝑆𝑆𝑆𝑆 and 𝑠𝑠𝑖𝑖𝐿𝐿 – and that the
subjects behave according to utility function (9), the authors estimate a mixture model that
provides the relative share of individuals who are assigned to the different fairness ideals and
also an estimate of the distribution of the strength of fairness preferences as captured by 𝛾𝛾.
They find that the share of strict egalitarians is 43.5%, the share of liberal egalitarians is
38.1% and the share of libertarians is 18.4%. 23

The Cappelen et al. study deserves credit for introducing and estimating the share of
subjects assigned to heterogenous fairness ideals that are based on different normative
recognitions of “luck” and “effort”. At the same time, it is clear that the assumed existence of
three exogenously given fairness ideals is a strong assumption, and that it would be desirable
to develop experimental designs and econometric methods that make it possible to infer
individual subjects’ normative reference points from the data, instead of exogenously
assuming them.

Likewise, while assuming that different reference points can capture the distinction
between luck and effort, it is also possible that individuals’ fairness ideals directly affect their
interpretation of an individual’s deservingness. For example, an individual with a
meritocratic fairness ideal who faces a “lazy” individual may put a different weight (i.e., a
different 𝛼𝛼, 𝛽𝛽 or 𝛾𝛾 parameter) on being fair towards this individual compared to a situation
where she faces a hard-working individual. Thus, we believe it is an important task for future
research to find ways to identify individuals’ normative reference points and to examine how

23The high share of strict egalitarians is surprising in view of the fact that the subjects in this
study are students from a Norwegian business school. On the other hand, Norway is known for
a relatively strong egalitarian culture which may perhaps explain these results. In another study
with a broader sample of Norwegian and German students (Cappelen et al. 2013), the allocation
of individuals to the different fairness ideals is quite different, however, as only 22.5% of
individuals are strict egalitarians, while 42.5% are liberal egalitarians, and 35% are libertarians.

30
the strength of fairness preferences is affected by environments differentially characterized
by luck and effort. 24

3.2.3. Who are the Meritocrats?

The experimental literature on fairness ideals largely relies on an experimental design

involving an impartial third party whose own material payoff is not directly involved in the
decision to redistribute resources between the two other parties. This makes it difficult to
relate the findings of the literature on fairness ideals to the literature on altruism, inequality
aversion and envy discussed in section 3.1. In particular, we do not know whether and to
what extent altruistic or inequality averse subjects are meritocratic. Or are subjects who
display primarily selfish behaviors in the distribution tasks that involve no obvious
entitlements – such as the tasks used by Fisman, Kariv and Markovits (2007)
Fehr®Epper®Senn (2022) – behaving according to meritocratic principles when entitlement
effects exist?

To answer these questions, Epper, Fehr and Senn (2023) designed an experiment in
which the subjects participated in two types of allocation tasks. In the first part of the
experiment, the subjects allocated payoffs between themselves and another (anonymous)
party in multiple trials where they faced different negatively or positively sloped budget lines
in self-other payoff space across different trials. This part of the experiment enables the
authors to estimate subjects’ distributional preferences in the absence of any entitlement
effects. By applying the Bayesian clustering algorithm (described in Section 3.1.2) to the
subjects’ choices, they identify an inequality averse, an altruistic, and a predominantly selfish
cluster of individuals.

After the payoff allocation task, the subjects participated in several unrelated “filler”
tasks and a performance task that involved real effort and produced a measurable
performance output. Then they participated in another distribution task where they faced
exactly one budget line with a negative slope – but this time the task involved two different
entitlement manipulations involving luck or merit. In the two merit conditions, the subject

24 Cabeza (2021) conducted a study that varied the deservingness of players along the “effort” and “luck”
dimensions. She reports that changes in deservingness are associated with changes in the strength of
fairness preferences in both the domain of advantageous and disadvantageous inequality. If subjects
become “more deserving”, the decision-makers’ willingness to behave altruistically in the domain of
advantageous inequality increases while the willingness to behave enviously decreases.

31
was (i) matched with a partner who performed better in the performance task and (ii) with a
partner who performed worse. Thus, the subject had to make two payoff allocation decisions,
one when she was performing better and one when she was performing worse than the
partner. When the subject had performed better, the initial allocation – which was highlighted
on the decision screen (see Appendix Figure A 5) – implied a payoff advantage for the
subject. In contrast, when the subject had performed worse, the initial allocation implied a
payoff disadvantage for the subject. However, the decision-maker in this task was still free to
choose any feasible allocation on the budget line. The question is just whether the subjects
honor the fact that they face a partner who had performed better or worse in a preceding
performance task in their allocation decisions.

In the two luck conditions, everything was identical to the two merit conditions except
that the initial allocation on the budget line was based on a random draw. The subjects in both
the merit and luck conditions knew how the initial allocation had been determined.

Which allocations do the inequality averse, altruistic, and selfish subjects choose on the
budget line in the two luck conditions? It turns out that the selfish subjects predominantly
chose an allocation that is exactly or close to the own-payoff maximizing allocation – and
this behavior prevailed regardless of whether the subject was a lucky winner or an unlucky
loser. In sharp contrast, the inequality averse and the altruistic subjects overwhelmingly chose
the equal-payoff allocation in the two luck conditions, regardless of whether they had been a
winner or a loser.

Which allocations did the three distributional preference types choose in the two merit
conditions? Interestingly, the selfish types again predominantly chose exactly or close to the
own-payoff maximizing allocation, implying that they have little meritocratic concern. In
contrast, the altruistic and inequality averse subjects again behave very differently. Roughly
30 percent of the altruists still preferred the equal-payoff allocation in the two merit
conditions, but the large majority of the altruists showed substantial meritocratic concern. In
particular, when they had performed better in the performance task, they chose an allocation
that gave them more than the equal payoff and when they had performed worse, they chose
one that gave them less than the equal payoff.

The behavior of the inequality averse subjects had strong resemblance with those of the
altruistic subjects, except that a slightly higher share of the inequality averse subjects
(roughly 40 percent) preferred the equal payoff allocation. But a majority of the inequality

32
averse individuals felt entitled to take more than the equal payoff in case of a higher
performance and gave the partner more than the equal payoff in case of a lower performance.

These results indicate that the meritocrats largely consist of subjects who have other-
regarding distributional preferences, while the selfish individuals care little for meritocracy.

3.2.4. Cultural Differences and the Relative Importance of Fairness and Efficiency
Concerns

Almost all experiments documenting entitlement effects have been conducted with Western
student populations. These subjects are part of an educational and employment environment
that permanently evaluates their performance and requires high effort levels to pass frequent
examinations, provide satisfactory work results, or move up the career ladder. Meritocratic
and libertarian ideas may well flourish in this environment, raising the question about the
prevalence of difference fairness ideals in the broader population and in other cultures.

Jakiela (2015) studies entitlement effects in a younger population from rural villages in
Kenya and compares them with those observed in a US student population. She implemented
a luck treatment, where the roll of a die determines the size of the overall budget, and an
effort treatments, where either the dictator or the recipient produced the budget in a real effort
task. Jakiela’s experiment is motivated by observations that these entitlement effects may not
exist in poor rural communities with strong traditions of solidarity and mutual assistance
(Platteau 2000).

In Jakiela’ design, potential entitlement effects are strengthened with an additional

feature – the so-called “Giving” and “Taking” treatment. In the Giving treatments, the budget
is provisionally allocated to the dictators who also rolled the die in the luck condition and
performed the real effort task in the effort condition. In the Taking treatment, the budget is
provisionally allocated to the recipients who also rolled the die and performed the real effort
task. Jakiela finds that the Kenyan dictators behave very similarly in the luck and the effort
treatments. This contrasts sharply with the behavior of US students who showed strong
behavioral differences between the Luck and Effort treatments: dictators allocated
significantly higher income shares to those parties who had generated the budget through real
effort rather than the role of a die.

Almas, Cappelen and Tungodden (2020), henceforth ACT, were the first to study the
relevance of the distinction between “luck” and “effort/merit” in two broad population

33
samples – a Norwegian and a US sample consisting of 1000 decision makers in each of the
two countries. Moreover, more recently, Almas et al. (2021; 2022) conducted a large study
with 65’800 participants that comprise representative country samples taken from 60
different countries around the globe. In each country, at least 1000 individuals participated as
impartial spectators in an experiment like that in ACT (2020). This study enables the
examination of the potential universality of the role of merit vs. luck vs. efficiency costs on
preferences for redistribution in a highly controlled experimental set-up that is kept constant
across all countries.

In these experiments, the decision makers were put in the role of an “impartial
spectator” who had the option to redistribute income between two subjects (“workers”).
Initially, before redistribution, one of the workers had an income equivalent to US $6 while
the other had US $0. In the luck treatment, a lottery draw (from an earlier experiment on
Mechanical Turk in which the workers participated) generated the inequality between the
subjects. In contrast, the productivity differences between the two workers in a real effort task
determined the initial inequality in the merit treatment. The impartial spectator could
redistribute the $6 in $1 units without any cost in both the luck and the merit treatments.

In addition to the luck and merit treatments, the authors also conducted a so-called
efficiency treatment. This treatment is identical to the luck treatment except that
redistribution is associated with large cost: for every $1 given to the poorer workers, the
income of the worker who was initially richer is reduced by $2. With such a large
redistribution cost, one would expect a substantial increase in inequality acceptance.

Several findings of these studies stand out (see Figure 3 below). First, there are huge
cultural differences in the extent to which the merit treatment causes an increase in inequality
acceptance relative to the luck treatment (Figure 3a). The merit treatment causes basically no
increase in inequality in countries like India and China, presumably because the implemented
inequality is already very high in the luck treatment. However, in the overall sample, the Gini
coefficient in the merit treatment is 26 percentage points higher than in the luck treatment,
and the implemented Gini coefficient increases by 50-60 percentage points in the merit vs.
the luck treatment in countries like Canada, Australia, or Portugal.

34
 Figure 3: The relative importance of merit and luck for distributional preferences

Note: The figure shows how the introduction of performance-dependent inequality in the merit treatment
and costly redistribution in the efficiency treatment increases the implemented inequality (delta Gini)
relative to the luck treatment for each country. The treatment effects are estimated with a country specific
regression controlling for pre-specified background characteristics. The figure shows much larger
treatment effects for the merit compared to the efficiency treatment.

Second, the effect of the efficiency treatment – when compared to the luck treatment – is
much lower than the effect of the merit treatment in almost all countries (Figure 3b and 3c).

35
On average, the Gini coefficient in the efficiency treatment is only 5 percentage points higher
than in the luck treatment, and the treatment effect is zero or even negative in many countries.
The efficiency treatment increases inequality acceptance at most by (slightly less than) 20
percentage points. Almas et al. (2021, 2022) thus document the near universality of the
important role of merit for redistributive preferences, including the near universality of the
greater role of fairness considerations relative to efficiency considerations for redistribution.

Third, the study finds vast differences in fairness ideals across countries. While almost
75% of the subjects display a libertarian ideal in countries like India and China, this share is
in the range of 10-15% in countries like Canada, Australia, or Norway. Conversely, the share
of meritocratic ideals is vanishingly small in India and China, while it is close to 50% in
Canada, Australia, and Norway. Interestingly, the cultural/country differences in fairness
ideals primarily show up in the different shares of libertarians versus meritocrats, while the
share of strict egalitarians is roughly between 20 and 30% in most countries.

3.2.5. Shallow Meritocracy?

The distinction between earnings generated through individuals’ choices, as well as earnings
that accrue to individuals because of lucky circumstances, is at the heart of the meritocratic
fairness ideal. A key issue in this concept of fairness is, however, that choices are
endogenous. In other words, individuals’ choices are themselves affected by circumstances.
For example, an individual who faces discrimination will generally have weaker incentives to
exert high effort because the reward for effort is lower. There are myriads of external
circumstances – such as gender norms, socio-economic background, ethnicity, or race – that
are associated with unequal opportunities that generate differences in individuals’ choices
and earnings. Moreover, when people make merit judgements based on earnings or effort
information, they often do not know the external circumstances under which earnings and
effort choices took place. In this context, the question then arises to what extent individuals’
merit judgements take these unequal opportunities into account.

In a recent paper entitled “Shallow Meritocracy”, Andre (2022) examines this question
in a series of experiments involving a broadly representative sample of 4000 US respondents.
Andre introduces an interesting twist into an effort treatment where two workers could
generate earnings in a piece-rate environment, and an impartial spectator then had the
opportunity to redistribute the earnings between the workers. The piece rate was either $0.50

36
or $0.10, each with a 50% chance, per correctly solved task, and the spectators knew that a
lottery determined the piece rate. In the control treatment, the workers did not know the
realized piece rates during the production phase, i.e., they faced an identical incentive
environment which the spectators knew about. In the treatment condition, the workers knew
the realized piece rates before they started working in the production phase. Therefore, the
workers faced different incentives to perform, and again the spectators knew this. When
spectators made their redistributive decision for a pair of workers they knew (i) each worker’s
effort share (i.e., relative share of the pair’s solved tasks), (ii) each worker’s relative
contribution to the total earnings of the pair, and (iii) the piece-rate condition under which
each worker had to perform the task.

The question then is whether spectators’ redistributive choices consider that workers in
the treatment condition face very different incentives depending on the assigned piece rate.
The answer is that they do not. Spectators condition their redistributive choices in the control
and the treatment conditions in the same way on the disadvantaged workers’ effort share or
the disadvantaged workers’ earnings contribution. Spectators redistribute income to the
disadvantaged workers, but they do not redistribute more to them in the treatment condition.

It is possible that spectators lack perspective-taking abilities, i.e., when they make their
decisions, it does not come to mind that disadvantaged workers had a good reason to solve
fewer tasks. To test this, the author explicitly informs the spectators that “the piece-rates
strongly influence the number of tasks a worker completes” and spectators are also informed
how large, on average, this incentive effect is. Yet again, the spectators’ redistribution to
disadvantaged workers is identical in the control and the treatment conditions.

The previous result means that explicit knowledge of the average incentive effect of
low piece rates does not induce spectators to assign more merit to disadvantaged workers.
But perhaps spectators would assign more merit if they knew with certainty that the specific
disadvantaged worker whom they face would provide high effort levels under high piece
rates. To test this hypothesis, Andre (2020) collected further data from a large number of
workers under both high and low piece rates. This enabled him to inform the spectator in a
subsequent experiment about the counterfactual effort share of the disadvantaged worker, i.e.,
although in the concrete redistributive situation of this experiment the disadvantaged workers
had a low effort share, the spectator had credible information that this worker would work
hard if had he faced a high piece rate. Interestingly, when this specific counterfactual

37
information was provided, the spectators redistributed significantly more money to
disadvantaged workers compared to a situation where they lacked counterfactual information.

Taken together, the study by Andre (2022) suggests that individuals are held
responsible for lower effort, dedication, and perseverance even though the disadvantageous
environment is the source of their lower performance. The reason for this is that many
spectators (37%) are “actual choice meritocrats”, i.e., they condition their merit judgements
on the actual behavior of the disadvantaged individuals, while only 26% of the spectators (the
“comparable choice meritocrats”) condition their choices on the specific counterfactual
behavior of the disadvantaged. However, because specific counterfactual information is often
unavailable in practice, the two types of meritocrats may often behave very similarly.

This disregard of unequal opportunities in the assessment of merit has also been documented
in several other papers. Dong, Huang and Lien (2022) and Preuss et al (2022) conduct
experiments with related designs, and document similar findings to Andre (2022). A final
related experiment is found in Bhattacharya and Mollerstrom (2022), where the experimental
conditions do not only differ in how (dis)advantageous it is for different agents to exert effort.
Instead, the opportunity to work is randomly assigned to the workers in the experiment.
Bhattacharya and Mollerstrom show that even if luck fully determines the opportunity to
work, spectators reward work status – not only with higher earnings but also higher implied
utility. Taken together, this growing literature shows that the disregard of unequal
opportunities in merit judgements is robust and arises in different settings and cultures.

3.3. Belief-Dependent Social Preferences

In distributional models of social preferences, the players care only about material payoffs:
not only their own, but also those of others. The choice between actions is unaffected by the
path leading to these choices or the perceived intentions behind the action of others. In this
sense, these models are consequentalist. But the path and the perceived intentions may very
well matter. A small example makes this point: Two brothers are dividing a pie. One cuts
it into two pieces and offers the choice of pieces to his brother. The second brother takes
the larger piece, and the first brother complains that he would have taken the smaller piece.
The second brother says: “You got what you wanted. What is the problem?” Certainly, the
path and the perceived (un)kindness of the players’ actions are likely to matter in this

38
case. 25

In view of the potential importance of beliefs, this section discusses prominent

classes of belief-dependent theories of social preferences – reciprocity theories, guilt
aversion theory, and theories that include emotions in social preference research.

3.3.1. Reciprocity

Models of Reciprocity 26

Rabin (1993) is the path-breaking paper in economics on intentions-based reciprocity,

based on the tools of psychological game theory (Geanakoplos, Pierce and Stachetti 1989).
The key idea in this research is that people have beliefs about others’ beliefs and that these
higher-order beliefs can affect their utility. Reciprocity models (e.g., Rabin 1993,
Dufwenberg and Kirchsteiger 2004) assume that the desire to raise or lower others’ payoffs
depends on the perceived fairness or unfairness of their behavior: kind intentions are met
with kind responses, while unkind intentions are met with unkind responses. In Rabin’s
model, player i’s utility includes both his material payoff 𝜋𝜋𝑖𝑖 and a non-pecuniary fairness
payoff that is the product of player i’s belief about player j’s kindness 𝑓𝑓̃𝑗𝑗 and player i’s
kindness 𝑓𝑓𝑖𝑖 as follows:

𝑈𝑈𝑖𝑖 ≡ 𝜋𝜋𝑖𝑖 + α ∙ 𝑓𝑓̃𝑗𝑗 ∙ 𝑓𝑓𝑖𝑖 . (12)

The fairness payoff is given by 𝑓𝑓̃𝑗𝑗 ∙ 𝑓𝑓𝑖𝑖 and α ≥ 0 measures the weight given to the fairness
payoff. Rabin defines the kindness terms in such a way that if j is believed to be kind to i,
then 𝑓𝑓̃𝑗𝑗 > 0, while if j is believed to be unkind to i, then 𝑓𝑓̃𝑗𝑗 < 0. Likewise, if i is kind to j,
𝑓𝑓𝑖𝑖 > 0 and if i is unkind to j, then 𝑓𝑓𝑖𝑖 < 0. From this follows that player i can increase his
utility by responding to kindness (i.e., 𝑓𝑓̃𝑗𝑗 > 0) with kindness (𝑓𝑓𝑖𝑖 > 0), and to unkindness
(𝑓𝑓̃𝑗𝑗 < 0) with unkindness (𝑓𝑓𝑖𝑖 < 0). For given beliefs of i about j’s strategy, player i is kind to j
if she gives player j his fair payoff. The fair material payoff is again defined for given beliefs
of player i about j’s strategy, and is given by the average of the lowest and the highest
material payoff on the pareto frontier that i can give to j. Thus, the fair payoff as well as

25 The second brother may, for example, infer unkind intentions from the fact that the first brother split
the cake unequally.
26 We focus our attention on intentions-based models of reciprocity because they have led to a

considerable empirical literature. Other models of reciprocity exist, however, such as that by Levine
(1998), that also have interesting implications.

39
player i’s kindness are entirely independent of player i’s own material payoff, i.e., fairness
does not involve any interpersonal payoff comparisons or notion of fairly distributing the
potentially available surplus.

To illustrate the logic underlying his model, let us consider the prisoners’ dilemma
game in the figure below. Suppose player 1 believes that player 2 believes that player 1 will
cooperate. Suppose further that player 1 also believes that player 2 will cooperate. This
means that player 1 believes that player 2 is kind to him because player 2 is believed to be
giving him a payoff of 4 instead of zero. Therefore, if player 1 puts enough weight on
fairness payoffs he will have an incentive to respond to 2’s perceived kindness with
cooperation. Since the game is symmetric, the same logic applies to player 2’s choices, i.e.,
the strategy (cooperate, cooperate) is a fairness equilibrium for large enough values of α.

Player 2

Cooperate Defect

Player Cooperate 4, 4 0, 6
1
Defect 6, 0 1, 1

The Rabin (1993) model considers only simultaneously played two-player games.
Retaining the kindness feature, Dufwenberg and Kirchsteiger (2004) formally extend and
develop the notion of reciprocity by making it applicable to sequential and three-person
games. This extension permits application to the more standard experimental games that are
played in sequence. Dufwenberg and Kirchsteiger also slightly change the definition of
kindness such that the 2nd mover’s conditional cooperation in the sequential Prisoner’s
Dilemma can be part of a fairness equilibrium. Falk and Fischbacher (2006) combine the
notion of inequity aversion with reciprocity considerations in a hybrid model where a
person is less bothered by another’s refusal to come out on the short end of a split than by a
refusal to share equally. Like Rabin (1993) and Dufwenberg and Kirchsteiger (2004), they
also use a kindness function.

A key feature is the assumption that harmful behavior by the other player (meaning that you
come out behind) is resented less when the other player had no “reasonable” alternative

40
choices. For example, in Falk, Fehr, and Fischbacher (2003), when the 1st mover in a binary
ultimatum game can propose (800 for self, 200 for other) or (500, 500), the rejection rate of
the (800, 200) offer is 44%, while if the alternative offer to (800, 200) is (200, 800), the
rejection rate of (800, 200) offer is only 27%. In the latter case, it is somewhat unreasonable
to expect the first mover to propose (200, 800). When the equal-payoff alternative
(500,500) is replaced by (800,200), the first mover clearly has no choice whatsoever and
the rejection rate drops to 18%. Thus, depending on the alternative to the (800, 200) offer,
the responder can make different inferences about the proposer’s intention. When (500,
500) is the alternative, a proposer’s unfairness intention when offering (800, 200) becomes
very visible while if (800, 200) is the alternative to itself, the proposer has no meaningful
choice and thus no unkind intention can be inferred.

Charness and Rabin (2002) also combine distributional preferences with reciprocity.
As mentioned earlier, the key innovation of this model is that people care about the total
payoffs for the reference group. The full model is given as follows:

𝑈𝑈𝑖𝑖 (𝑠𝑠, 𝑑𝑑) ≡ (1 − 𝜆𝜆) ∙ 𝜋𝜋𝑖𝑖 + 𝜆𝜆 ∙ [𝛿𝛿 ∙ min[𝜋𝜋𝑖𝑖 , 𝑚𝑚𝑚𝑚𝑚𝑚𝑚𝑚≠𝑖𝑖 {𝜋𝜋𝑚𝑚 + 𝑏𝑏𝑑𝑑𝑚𝑚 }]

+(1 − 𝛿𝛿) ∙ (𝜋𝜋𝑖𝑖 + ∑𝑚𝑚≠𝑖𝑖 max [1 − 𝑘𝑘𝑑𝑑𝑚𝑚 , 0]𝜋𝜋𝑚𝑚 ) − 𝑓𝑓 ∑𝑚𝑚≠𝑖𝑖 𝑑𝑑𝑚𝑚 ∙ 𝜋𝜋𝑚𝑚 ] (13)

Unpacking this array of symbols, i’s utility is a weighted average of his own financial
payoff (with weight 1 − 𝜆𝜆) and a social-welfare function (with weight 𝜆𝜆). This social-
welfare function is itself a weighted average of concern for the minimum payoff (with
weight 𝛿𝛿) of anyone in the reference group and concern for the total social payoff (with
weight 1 − 𝛿𝛿). Of note is the “demerit function” 𝑑𝑑𝑚𝑚 , which measures what is termed
“misbehavior”, and which can be interpreted roughly as a measure of how much Player
𝑘𝑘 deserves. The higher the value of 𝑑𝑑𝑘𝑘 , the less player i thinks Player 𝑘𝑘 deserves.
Misbehavior is defined by the group’s standards. Negative reciprocity takes two forms: 1)
concern withdrawal and 2) willingness to sacrifice own payoffs. Concern withdrawal
means that one’s concern for the payoff of another player is reduced by adding 𝑏𝑏𝑑𝑑𝑚𝑚 to that
player's payoff and subtracting 𝑘𝑘𝑑𝑑𝑚𝑚 from that player's contribution to the total social payoff
(b and k are non-negative numbers). One might also be willing to sacrifice money to hurt
another player (reflected in the f term near the end), being so upset by the other player’s
misbehavior that they are willing to destroy surplus. This logic can explain rejections in the
ultimatum game.

The last model we discuss in this section is that of Cox, Friedman, and Gjerstad

41
(2007), who present a non-equilibrium approach that combines a form of distributional
preferences with reciprocity considerations and is more tractable. In this approach, both
status (relative position) and reciprocity affect one’s emotional state, which in turn affects
the choices that are made by a utility-maximizing agent. They introduce a parametric model
of social preferences in which one’s emotional state determines the marginal rate of
substitution between own and others’ payoffs, and thus one’s subsequent choices. In turn,
one’s emotional state responds to relative status and to the kindness or unkindness of
others’ choices. They find that structural estimations of this model with six existing data
sets demonstrate that social preferences depend on status, reciprocity, and perceived
property rights.

One objection to reciprocity models and, in particular, to that proposed by Charness

and Rabin (2002), is that they are considerably more complex compared to purely
distributional preference models. One might thus wonder whether this complexity is
needed. However, it seems fair to say that there are many papers that provide data that
distributional models cannot explain without considering reciprocity. Below we discuss
several studies that corroborate this view. 27

Before we discuss the empirical evidence for reciprocity one point is important. Kind
or unkind intentions are assumed to play a key role in intentions-based models of
reciprocity. One would, therefore, expect empirical researchers to have invested a lot of
effort in the identification of subjects’ kindness perceptions. However, with a few
exceptions (Offerman 2002; Dhaene and Bouckaert 2010), the empirical literature has often
ignored this point, unfortunately. Instead of identifying the subjects’ (un)kindness
perceptions, the experimenters (and the reviewers) have typically been the judges of what
behavior is kind or unkind. Yet, what really matters is what the participants in the
experiment consider to be kind or unkind. Thus, if an experimenter assumes, for example,
that a particular behavior is kind but the subjects themselves do not perceive it this way, and

27Another objection to reciprocity models is that they can be associated with the existence of
rather unintuitive equilibria even in very simple games. Consider, e. g., a binary ultimatum
game where the first mover can either propose (500, 500) or (800, 200). Here, the strategy pair
(800, 200) for the proposer and “accept (500, 500) but reject (800, 200)” for the responder can
be a fairness equilibrium (involving mutual hostility) in Rabin’s model. However, why should
a proposer who knows that the responder rejects (800, 200) with certainty ever make that
proposal? Likewise, why should a proposer interpret the rejection of the unfair (800, 200) offer
as a hostile act rather than as an understandable response to an unfair offer?

42
hence do not reciprocate, one may erroneously conclude that positive reciprocity is absent
when in fact the theory predicts the absence of reciprocal responses.

Field Evidence on Reciprocity

There is evidence from field settings that suggest negative reciprocity is often present.
Workers have been known to engage in sabotage or increased theft rates after a pay cut or
other actions perceived to be unfair (see for example Greenberg (1990), Shminki,
Cropanzano, and Rupp (2002)), particularly when procedural justice in the organization is
low (Skarlicki and Folger 1997). And the studies by Krueger and Mas (2004) and Mas
(2006; 2008) present results suggesting retribution with real firms and workers. The case for
positive reciprocity in the field is weaker; cases involving tipping when on the road or
higher response rates to mailed surveys that include small gifts may instead reflect guilt
aversion, discussed below. 28

Field evidence from uncontrolled natural environments rarely allows researchers to

unambiguously pin down motives because reputational and repeated game issues may play a
role. Field experiments are, however, capable of controlling these factors. Gneezy and List
(2006) conducted such a field experiment that examined the effect of paying more than the
participants were led to expect. In experiments in two different locations, they advertised to
the public a job with a wage of $10 (or $12) per hour; however they then told the people who
showed up that they would instead pay them $20 per hour in one treatment. The job was six
hours of work, split into morning and afternoon sessions. The research question is whether
this surprise overpayment leads to greater effort.

In fact, there was significantly higher productivity in the morning but no effect after
lunch. Breaking production into four 90-minute segments, the average production in the
control treatment was 40.7, 40.5, 41.2, and 39.6; this compares to 51.7, 44.9, 41.7, and 40.3
in the gift treatment. So, there is a large (27%) increase in productivity during the first 90
minutes, which declines precipitously to only 11% during the second 90 minutes, and then

28
However, we do have one clean example of positive reciprocity from the field: One of the co-authors was
approaching the toll plaza for the Richmond-San Rafael bridge around 1993. A driver had somehow gotten out of
the line of waiting cars and she was trying to get back in. She kept trying, but other drivers weren’t letting her in.
The co-author’s turn came along, and he waved her into the space in front of him; she waved back profusely.
When he arrived at the toll booth, he had his dollar out for the toll-taker, but he was informed that the car in front
of him had paid for him. Unless the other driver did this habitually for strangers (or she thought it was expected
and so not paying would have led her to feel guilty,) this is positive reciprocity in the field.

43
completely disappears. This corresponds with an attitude of “What have you done for me
lately?”. Does positive reciprocity stem from a form of warm glow that dissipates with time?
The authors argue that this shows that reciprocity is a weak and ephemeral phenomenon and
that paying higher wages may not lead to higher net profits. While we do not dispute the
behavioral findings, we note that the authors did not measure the workers’ kindness
perceptions. Thus, it remains unclear whether the higher productivity in the morning reflects
intention-based reciprocity. Likewise, it remains unclear whether the absence of higher
productivity in the afternoon occurred even though workers saw the $20 wage as kind, i.e.,
whether reciprocity is indeed a very short-lived response.

In this context, a key question is how the recruited subjects interpreted the surprise
wage increase. Is it, e.g., possible that subjects inferred from the higher wage that the
employer is much wealthier than they thought, or that the employer made an announcement
mistake when the subjects were initially invited for $12? If true, then the initial effort
response to the higher wage may not be a result of an increase in fairness perceptions but it
may have resulted from short-run effects on subjects’ mood. 29 Bellemare and Shearer (2009)
conducted a field experiment on the influence of wage gifts on worker productivity that
circumvents this problem by providing a natural explanation for the wage increase. In
addition, they increased the statistical power for identifying a wage effect with a within-
subject design. Bellemare and Shearer found that the wage gift significantly increased
workers’ productivity.

At the time Gneezy and List conducted their studies (2004), $10 or $12 was a very
fair wage for this type of work (prevailing rates were roughly $8.50), i.e., workers were
already substantially “overpaid” at the baseline wage. Perhaps, this made it difficult to further
increase the kindness perception in the group that received $20 per hour. This interpretation
would be consistent with the “fair wage – effort” hypothesis put forward by Akerlof and
Yellen (1990). This hypothesis stipulates that workers respond to wage levels with their
effort only if they are underpaid relative to a fair reference wage. Amounts paid above this
fair wage are not predicted to have positive effects on effort. Thus, the surprise may be not
that positive higher productivity vanishes over time, but that it was found in the first place.

29The importance of fairness perceptions in response to a wage increase has been documented in Cohn,
Fehr and Goette (2015), who elicited independent measures of fairness perceptions and reciprocity
preferences. They show that workers respond to a wage increase with higher effort if two conditions are
met: (i) workers perceive the wage increase as an increase in fairness and (ii) workers have reciprocal
preferences.

44
 Kube, Maréchal, and Puppe (2012) consider how the form of payment or reward
affects behavior. Their experiment consisted of a 3-hour job with an advertised pay of €12.
In the baseline treatment, this was done as advertised. In two other treatments, there was a
surprise bonus announced; in one case this was €7, and in the other it was a thermos bottle
worth €7. There were also treatments where it was clear that there was more effort put into
the gift (e.g,. the additional money was folded into a complex origami). The figure below,
taken from their paper, shows the results.

Throughout the whole experiment, people are much more productive in the bottle treatment
(and in other treatments where a non-monetary gift was made) than in the baseline, a 25
percentage-point increase on average. The monetary bonus had a modest increase of 5% over
the baseline. While there is no significant positive reciprocity in response to a higher wage,
there is a great deal of positive reciprocity when the worker is given a gift. Although this
paper did not measure subjects’ kindness perceptions, a plausible interpretation of this
difference is that giving a gift is much more likely to be perceived as kind. It is also quite
interesting that, in a treatment where people were given a choice between the bottle and the
cash, more than 80% chose the cash gift! It seems that, at least to some extent, it is (the
perception of) the thought that counts. In any event, the article suggests that the “currency of
reciprocity” may indeed not be surprise money that is given without further explanation.

Figure 4: The effect of monetary and non-monetary gifts on productivity

Kube, Maréchal, and Puppe (2013) implemented a wage cut treatment to study the
role of negative reciprocity. Since internal review boards might not be happy with a
researcher promising a wage and then reneging, they chose their words carefully when they

45
advertised a job. The announcement stated: “The hourly wage is projected to be €15” (the
exact German wording was “Ihr Stundenlohn beträgt voraussichtlich €15”), leaving some
room for later wage changes. There were three treatments: No change in pay (€15), pay
reduction (€10), and pay increase (€20). In the pay increase condition, productivity is on
average very similar to the no change in pay condition, while there is an immediate and
sustained decline in productivity by 21% in the pay cut condition, indicating the considerable
strength of negative reciprocity.

Why do subjects respond much more strongly to wage cuts compared to wage
increases? In addition to the ambiguities in interpreting the intentions underlying a wage
increase, loss aversion could also play a role. In the field experiments discussed above, the
experimenter invites subjects to the experiment by generating an explicit expectation about
the wage which constitutes a natural reference point. Thus, wage cuts will be interpreted as
losses, while wage increases will be coded as gains. If losses loom psychologically larger
than gains, then wage cuts are more likely to generate perceptions of unkindness while the
kindness inferences associated with wage increases are more muted.

Laboratory Evidence on Reciprocity

Most laboratory studies also did not explicitly measure subjects’ kindness perceptions, but
the lab enables treatment variations that render it more likely for (un)kindness perceptions to
be induced. Most of the papers discussed below used a control treatment in which a random
device determines the first-mover’s choice in a sequentially played game. Thus, the first
movers in the control treatment cannot indicate any intentions, which maximizes the chances
of treatment differences in intentions.

Charness (1996; 2004) is the first paper to simultaneously test for positive and negative
reciprocity while controlling for distributional preferences. He considers gift-exchange in a
bilateral setting, varying whether a self-interested firm or a random device generated the
wage. In both cases, the firm benefits from the worker’s chosen effort. There is a strong
positive relationship between wage and effort in both treatments. However, the effort level
with low wages is lower when a self-interested firm chose the wage than when it was
generated exogenously, suggesting the presence of negative reciprocity. On the other hand,
there was little difference across treatments in the effort level with high wages. Here,
intentionally receiving a low wage is very likely to be seen as unkind because the worker
knows that higher wages are possible, and frequently even experiences that higher wages are

46
paid (as the worker sequentially faces offers from 10 different employers). In contrast, it
appears less obvious that higher wages are kind because they may be offered for strategic
reasons to elicit a higher effort level. This may help to explain why we observe that positive
reciprocation is less prevalent than negative reciprocation.

Offerman (2002) also uses the random versus intentional first-mover choice approach
of Blount (1995) and of Charness (1996) to study positive and negative reciprocity. He
considers players’ responses to an unambiguous helpful or hurtful choice. The helpful choice
generates a positive payoff for the responder, while the hurtful choice causes a negative
payoff. The responders could sacrifice one unit to either increase or decrease the first mover’s
payoff by four units. 75% of the responders reciprocate intentional helpful choices, while
50% of the responders reciprocate random and unintentional helpful choices. The difference
of 25 percentage points is not significant but this most likely reflects the limited number of
observations (12). In contrast, the effect of negative intentionality is quite strong:
Responders reciprocate 83.3% of the intentional versus 16.7% of the unintentional hurtful
choices. This difference is significant at p < 0.01.

To what extent are these large differences in reciprocation patterns a result of subjects’
kindness perceptions? Offerman collected proxy measures of responders’ (un)kindness
experiences by measuring their positive and negative emotions. He finds that intentional
hurtful choices generate much stronger negative emotions than unintentional hurtful choices
do. In contrast, intentional helpful choices generate about the same positive emotions as
unintentional helpful choices do. Moreover, negative (positive) emotions after a hurtful
(helpful) first-mover choice are significantly correlated with punishing (rewarding)
responses. Taken together, these results suggest that the differences in reciprocation patterns
are driven by differences in the extent to which intentional hurtful and helpful choices trigger
(un)kindness perceptions. The results are thus nicely in line with reciprocity theory but also
suggest that inducing kindness experiences through intentional choices can be more difficult
than inducing unkindness experiences. This differential ease with which one can induce
kindness versus unkindness may well have been a factor in the results reported in Charness
(1996, 2004).

Several other papers (Brandts and Sola 2001; Brandts and Charness 2003; Falk, Fehr
and Fischbacher 2003; Charness and Levine 2007; Falk, Fehr and Fischbacher 2008; Dhaene
and Bouckaert 2010) support the conclusion that intentions matter for reciprocal responses.

47
The evidence in these papers, as well as those discussed previously in this section, typically
suggests that in addition to distributional fairness concerns, intentions are likely to play a key
role for deviations from self-interested behavior.

While several of these papers do not measure kindness perceptions explicitly, they
nevertheless implement plausible manipulations of intentions via random versus intentional
first-mover choice designs or via other means. Brandts and Charness (2003) design an
experiment, for example, where a particular outcome can be reached in two different ways
that indicate very different intentions. There are three stages in their game. First, player A
sends a message about her intended play to Player B. Second, the players then simultaneously
make choices in the game depicted below. Third, the players are informed about their choices
at the second stage; if B chooses B2, she is then given an opportunity to punish or reward
player A at a cost to herself.

In this game, A prefers the outcome (A2, B2) while B prefers (A1, B2). The outcome (A2,
B2) can be reached in two ways. A can send a message that he will play A1 but then he
chooses A2. Alternatively, the outcome (A2, B2) can also be reached after a truthful message
by A. The authors found that the responder was twice as likely to punish player A after a
choice of A2 if A had previously lied about his play than if he had told the truth.

Player B

B1 B2

Player A A1 2, 2 6, 9

A2 2, 2 12, 3

Not all laboratory evidence appears to support the relevance of intentions-based

reciprocity. Bolton, Brandts, and Ockenfels (1998) find strong evidence for distributional
preferences, no evidence for positive reciprocity, and only weak evidence for negative
reciprocity. We believe, however, that their findings do not show the absence of reciprocity
but are rather the result of an experimental design that very likely failed to elicit kindness and
unkindness perceptions. Participants made choices in 2x6 matrix games via the strategy
method. The payoff matrices used for their different treatments are shown in Appendix 5. In
our view, the presentation of payoffs in this way makes it very difficult to induce any
kindness or unkindness judgements for at least two reasons. First, the 2x6 game matrix is so

48
complex that even a researcher can use a guiding hand to understand the (un)kindness
interpretations built into the matrix. Second, the game is a simultaneous move game and
although the authors use the strategy method, the kindness judgements only emerge through a
complex reasoning chain. This is rather demanding and presupposes a “theory of mind”
capacity that many people are unlikely to have naturally.

The Relative Importance of Distributional and Reciprocity Concerns

There is evidence from student samples that reciprocity concerns can outweigh distributional
concerns if the experimental design makes it easy to infer intentions. For example, in Falk,
Fehr and Fischbacher (2008) distributional concerns play almost no role when a random
device transparently determines the first mover’s behavior, while strong positive and negative
reciprocity prevails when first-movers themselves make the choices. There are, however, also
situations where distributional concerns can minimize reciprocal behavior. Xiao and
Bicchieri (2010) consider the interplay between reciprocity and equality concerns with a
variant of the Berg, Dickhaut, and McCabe (1995) investment (“trust”) game, comparing in
particular the trustees’ willingness to reciprocate when reciprocation makes them worse off
than the investor. Their results show that “the proportion of non-reciprocating decisions is
twice as large when reciprocity promotes inequality” and that the first movers expect this.

A thorough analysis of the relative importance of reciprocity and inequality aversion is

provided by Bellemare, Kröger and van Soest (2011). They simultaneously estimated the
structural parameters measuring inequality aversion (α, β) and intentions-based reciprocity (ι
for negative reciprocity, κ for positive reciprocity) on the basis of a representative sample of
the Dutch population that participated in ultimatum games with randomly determined versus
intentional offers. They find that disadvantageous inequality aversion is much higher than
aversion to advantageous inequality (|α| = 0.796 < β = 0.183) in the general population, and
that “inequity aversion tends to be more important than perceived intentions in the population
as a whole”. However, the aversion to disadvantageous inequality is much lower for the
highly educated and young subjects below the age of 35 than in the general population.
Moreover, negative reciprocity quantitatively dominates aversion to disadvantageous
inequality (|α| = 0.248 < ι = 0.474) among the young and educated.

49
Summary

What have we learned about reciprocity in this section? First, there are widespread and
numerous examples of negative reciprocity in the lab (and in the field, although instrumental
concern for the future may color the choices made). On the other hand, positive reciprocity is
found in some but not all cases. Why this is the case?

A first important reason could be loss aversion, i.e., the notion that losses loom larger
than gains. Recall that reciprocity theory defines (un)kindness always relative to a reference
point. Kindness is naturally related to receiving more than the reference point while
unkindness means receiving less. The finding of Offerman (2002) that negative emotions
after an unkind choice are much stronger than positive emotions are after a kind choice is
consistent with this view. A second reason is that certain actions – like surprisingly paying
higher wages without providing a plausible explanation for the wage increase – can be
interpreted in different ways and may thus not be viewed as kind. This interpretation is in line
with the findings in Kube, Marechal and Puppe (2012), which indicate that clearly
interpretable gifts (payoff-equivalent to a wage increase) trigger strong reciprocal effort
responses, while mere wage increases show insignificant effects. When employers want to
induce reciprocal effort responses, they must carefully design their gifts, and not just offer
more money. We typically do not give money to our spouses at Christmas but think carefully
about the gifts for them.

A third potential reason for the difficulties in inducing positive reciprocity is that a
certain minimal level of kindness is taken for granted in many modern societies: people
usually hold the door for each other. Typically, if one expects kind or favorable treatment and
receives it, there is no strong emotional jolt; on the other hand, if one expects kindness and
receives unkind or hurtful treatment, the emotional response is much stronger. This means
that the threshold for inducing further kindness perceptions is higher than the threshold for
inducing unkindness perceptions. Consistent with this conjecture, Khalmetski, Ockenfels, and
Werner (2015) find that surprising gifts are quite effective in leading to behavior that
resembles positive reciprocity.

A fourth potential reason is related to the fact that reciprocity is a cognitively

demanding concept because it requires reading other people’s intentions. This is much harder
than just noticing that one received less than another individual. In this context, it is
interesting that there appear to be large cultural differences with regard to the extent to which

50
intent is taken into account in moral reasoning. This has been shown by anthropologists who
assessed how people judge the badness and punish-worthiness of bad outcomes that either
occur accidentally or intentionally (Barrett et al. 2016; Curtin et al. 2020). While intentions
played an important role for Western populations, they played a more minor role for many
other cultures. These findings suggest that it may be interesting to study the role of intentions
versus outcomes in social preferences across different cultures. In addition, except for the
Bellemare, Kröger, van Soest (2011) study, very little appears to be known about the
distribution of reciprocity in broad population samples.

3.3.2. The Role of Guilt Aversion in Social Preferences

Guilt aversion is based on the idea that people prefer to avoid feeling guilty. The basic idea
is that decision makers experience guilt if they believe they disappointed others who
depended on them. The avoidance of guilt may thus induce prosocial behaviors. However,
unlike reciprocity and inequality aversion, guilt aversion cannot explain behaviors that
reduce other people’s payoff. Guilt aversion leads to a concept of utility in which a player's
preferences over strategies depend on her beliefs about the beliefs of others, even if there is
no strategic uncertainty. 30

Conceptual and Intuitive Foundations of Guilt Aversion

Guilt aversion has its roots in social psychology: Baumeister, Stillwell, and Heatherton
(1994; 1995) advance the notion that people suffer from guilt if they inflict harm on others.
One way to inflict harm is to let others down.

To illustrate the intuition behind guilt, Dufwenberg (2002) introduces the marital
investment game, presented in the game tree in Figure 5 below 31: A wife can provide

30 The role of other emotional factors such as anger, frustration or anxiety is shortly discussed in
Appendix 7.
31
Huang and Wu (1994) present the first applied theoretical work in economics to incorporate guilt. Two
different forms of guilt aversion have been described. What we describe above is simple guilt aversion, where
one experiences guilt for disappointing another person’s expectations. A second form of guilt aversion is guilt-
from-blame whereby one experiences guilt only to the extent that one expects to be blamed for a bad outcome
(Battigalli and Dufwenberg 2007; Battigalli and Dufwenberg 2009). A key difference between simple guilt and
guilt-from-blame is the effect of observation. With simple guilt, B feels guilt regardless of observability; only
guilt sensitivity and second-order beliefs matter. With guilt from blame, B only feels guilt if A observes her
effort. In the field, it may well be that observability of actions is a key driver for guilt-influenced behavior. Here,
we limit our attention to simple guilt.

51
support for the husband to pursue a profitable education. If she refuses to do so, each
person receives one monetary payoff unit. If she agrees to support the husband, then she
forgoes the chance to invest in herself. In this case, her personal earnings are zero, but her
husband’s earnings are quadrupled. The husband can then choose to stay in the marriage
and share the earnings or choose to divorce and keep the rewards. Since the husband
promised not to do this, the wife will be quite upset, and the husband might therefore feel
guilty.

Figure 5: The Marital investment game with simple guilt

In this game, guilt aversion naturally enters the scene as follows: “When a husband
suddenly divorces his wife … the stronger the wife’s belief that her husband would stay,
the more disappointed she is. … The husband may be averse to letting a trusting wife
down, and the stronger he believes that she believes that he will stay the more guilty he
feels by forcing divorce.”

Dufwenberg models guilt-averse preferences here by presuming some individual

sensitivity to guilt (𝛾𝛾 > 0) for the husband, and then multiplying this sensitivity by the
husband’s second-order beliefs 𝜏𝜏 ′′ concerning what the wife expects. Thus, the stronger the
sensitivity to guilt and the higher the second-order beliefs, the greater the chance that the
husband stays in the marriage.

52
Early Experimental evidence

The first experiment to examine psychological game theory is Dufwenberg and Gneezy
(2000). 32 They study the lost-wallet game where player 1 (the finder) finds a wallet
containing a money amount x. The finder can take the money or ensure that the wallet is
given back to the owner. Once the owner has received the wallet, he can reward the finder
with an amount y from a given endowment of 20. For example, if the owner gives the
finder y = 5, then the parties payoffs are (5 for the finder, 15 for the owner). The authors
used x as a treatment variable set at 4, 7, 10, 13, and 16 in the different treatments. The key
issue here is whether the owner’s choice of the reward y is positively correlated with her
expectation of the finder’s expectation of y conditional on 1 choosing Leave.

The game was played only once in a session. First, the finder was given an
envelope with x ∈ [4, 7, 10, 13, 16]; then the finder chose whether to leave it or to return it;
in the latter case, the owner chose y. Beliefs were then elicited. Incentivized beliefs about
y (for the known x) were elicited from the finders. The owners were asked to make an
incentivized guess of the average guess of the finders who chose to leave the money in the
envelope. The results show a significant positive correlation between the reward y and the
owner’s expectation of the finder’s expectation of y. This positive correlation is the very
first evidence of what the authors call letdown aversion, a term changed to guilt aversion in
later work (as suggested by the younger co-author of this article). Thus, the psychological
mechanism hypothesized in guilt aversion theory appears to be operative here.

Guilt aversion can potentially also play an important role in economic interactions
under incomplete contracting. Charness and Dufwenberg (2006) show this by studying the
game below where player 1 can enter a trade that involves the sequential exchange of
goods. If 1 forgoes the trading opportunity, each party receives a payoff of 5. In case player
1 trusts by sending his goods to 2, player 2 can send her goods to player 1, and if the batch
arrives the payoffs are (12, 10). However, the sent goods may not arrive with probability
1/6 in which case player 1 receives nothing. Note that in this game player 1 cannot
distinguish whether bad luck or bad behavior by player 2 generated a payoff of zero.

32
Due to space limitations, we are not able to provide an exhaustive discussion of the evidence on the role of
guilt. For a more exhaustive presentation, see Cartwright (2019).

53
 Figure 6: Sequential Exchange with Moral Hazard

In this game a self-interested player 1 who expects to face a self-interested player 2 will
never trust. If player 2 is, however, guilt averse then it may be rational for player 1 to trust.
From the viewpoint of guilt aversion theory, the key question is whether player 2’s belief
about player 1’s expectation about 2 choosing share is positively correlated with player 2’s
actual probability of choosing share. Higher beliefs by player 2 about 1’s expectation will
increase the probability that a guilt averse player 2 will actually choose share.

Charness and Dufwenberg (2006) conjectured that giving player 2 the option to
communicate with player 1 may be a particularly powerful way of strengthening player 2’s
second order belief because player 2 has a strong incentive to convince player 1 that 2 will
share. Intuitively, if player 2 tries to persuade player 1 that he will choose to share, then
player 2 can hardly avoid the inference that in case of trust player 1 had a high belief that 2
will share. Thus, guilt aversion provides a mechanism that can explain why communication
between parties often enhances cooperative behavior.

The authors consider the effect of free-form communication in a one-way message

treatment where player 2 could send a hand-printed message of any nature to 1 before the
game commenced. Particular attention was paid to promises (statements of intent) since
these might be particularly effective in affecting beliefs.

The experimental results clearly show the effectiveness of communication. In

addition, they reveal behavioral patterns consistent with guilt aversion as an underlying
mechanism. Examining subjects’ verbatim messages is interesting and instructive (e.g., “I
swear upon my mother’s grave that I will share.”), and shows that promises were
particularly effective: Player 1 chose trust 79% of the time and 2’s chose share 79% of the

54
time when a promise was made; this compares to 37% and 33%, respectively, when no
promise was made. Of particular interest for guilt aversion, there is a strong relationship
between 2’s belief about 1’s belief that 2 will choose share and 2’s actual choice of whether
to share. These beliefs were significantly higher for 2’s who chose share compared to 2’s
who chose grab (66.7% versus 42.7%).

A comparison between the above discussed paper and Charness and Dufwenberg
(2010) reveals that the effectiveness of communication is very much context dependent. In
the latter paper, a second mover can only circle a message saying “I promise to choose
Roll” or send a blank sheet. This bare promise is almost completely ineffective relative to
the treatment with no communication. Richer messages tend to move beliefs more and it
may well be that self-generated messages are particularly effective (in the same way that
personal gifts were more effective than money in Kube, Marechal and Puppe (2012). One
open topic of interest is whether self-generated messages are more effective than if these
messages were known to have been chosen from a menu of messages.

Andreoni and Rao (2011) investigate the effect of one-way and two-way
communication on giving in a dictator game. They used an anonymously played one-shot
$10 dictator game with free-form communication, varying who in the pair could send
messages. Although they did not measure beliefs, their results are suggestive for guilt
aversion as a factor underlying communication-driven generosity.

In their main experiment, there is a Baseline treatment without communication, a

treatment (A for “ask”) in which the recipient sent a written message containing a
numerical transfer request. In treatment E (for “explain”), only the dictator could send a
written message (along with the pass request). There are also two combination treatments:
In “Ask then Explain” (AE), the receiver first sent a numerical request and the dictator’s
pass choice was accompanied by an explanation. In “Explain then Ask” (EA) the dictator
first indicated a (non-binding) split, and then the recipient responded with a numerical
request. The dictator then chose an allocation.

Andreoni and Rao document dramatic differences in the mean transfer according to
whether it is possible to make a numerical request. In AE, for example, dictators give 30%,
which is four times more than in treatment E. The two treatments with bilateral
communication have the highest mean transfer and, perhaps surprisingly, there is little
difference in the mean transfer across these. Transfers in treatment A are also significantly

55
higher than in the baseline and treatment E, and statistically indistinguishable from AE and
EA. It seems that treatment E could be considered an “Excuse” treatment, since the mean
transfer is much lower than even the Baseline.

There are several possible explanations for the power of an ask. The authors
themselves consider self-image maintenance and guilt aversion but social norm compliance
is also a possibility. Perhaps, self-image maintenance plays a role in treatment E, where
saying “sorry” to a recipient who is given little may help subjects to convince themselves
that their selfishness is not too bad. Guilt aversion appears, however, a much more plausible
source of the power of the ask. It appears quite likely that when a recipient asks for an
amount, the dictator believes that the recipient is expecting a similar amount to be
transferred; failing to do so could lead to guilt. 33 It is also possible, however, that the
equality/fairness norm becomes more salient when recipients can ask, as they typically
invoke fairness concerns in their asks. Because the role of dictator and recipient have been
randomly assigned, there is little normative legitimacy for the dictator to deviate from
equality, and the recipients’ asks may make this salient.

Vanberg (2008) argued that player 2 may keep her promise not because of guilt
aversion (i.e., the aversion felt because she disappoints player 1’s payoff expectation) but
because she experiences an obligation to keep her promise. He developed a clever
experimental design that distinguishes between the guilt aversion and the obligation
hypothesis: Each pair of player plays the same subgame with probability ½ that player 2
faces in Figure 6 where she has to choose between grab and share. If player 1 becomes the
decision-maker (“dictator”), player 2 is assigned the role of the recipient and vice versa.
Both players are given the opportunity to send two messages before the players know
which role they are assigned. Thus, when the roles are not yet assigned, the players have an
incentive to informally provide mutual insurance against exploitation by the other player,
and mutually promising to play share might achieve this. In fact, the vast majority of
players indeed promised to play share during pre-play communication.

To identify whether guilt aversion or feeling obliged to keep a promise is the motive

33Andreoni and Rao (2011) argue that guilt aversion has trouble explaining the similar behavior in EA and
AE, suggesting that perhaps a strategically biased perception of the other party’s beliefs is present and that
maintaining distortions is more difficult with communication. Yet, it is not so clear that the dictators’ second
order beliefs should differ between EA and AE, so we do feel that guilt aversion is a very plausible driver
behind the “power of asking”, at least in part.

56
underlying the play of share, Vanberg randomly rematched half of the recipients with
dictators in one treatment (“switch condition”). Critically, while dictators are informed
about whether such a switch occurred, the recipients are not. In the case of a switch, the
dictator can see the conversation that had occurred with the switched-out prior dictator, i.e.,
they learn whether the recipient has previously received a promise from a different dictator.

The key to this design is that recipients cannot know whether there was a switch, so
their first-order beliefs can depend only on whether a promise was received, not on whether
they are finally matched with the person who has made that promise. Since dictator subjects
know this, their second-order beliefs should be equally affected by promises made by others
(as in the switch condition) as by promises that they themselves made (as in the no switch
condition). The data support this hypothesis. Thus, the guilt aversion explanation for
promise keeping predicts that dictators should be more likely to share whenever the partner
has received a promise, in both the switch and no-switch condition. In contrast, the
obligation hypothesis predicts that promises made in the no-switch condition have higher
commitment power (i.e., lead to more play of share) because the dictator in the switch
condition did not make a promise to the partner she faces.

The main result in Vanberg (2008) suggests that feeling obliged to keep promises
drives the behavior rather than second-order beliefs. Promises led to significantly higher
share rates in the switch compared to the no-switch condition (73% versus 52%), although
they are associated with the same second-order beliefs in the two conditions. Moreover, the
second-order beliefs in the switch condition are higher when the recipient received a
promise from her previous partner compared to when he received no promise. Note that the
recipient’s previous partner generated this variation in second-order beliefs and the partner
should thus, according to guilt aversion, provide higher share rates after the promise than
after no promise. Yet, in fact the share rates are almost identical (54% versus 52%).

Overcoming Problems in Identifying Guilt Aversion

A key issue in identifying guilt aversion by measuring players’ second order beliefs is that
these beliefs are endogenous. In the sequential exchange game in Figure 6, player 2’s belief
about 1’s expectation could easily be a result of 2’s choice instead of driving it. For
example, if 2 believes that player 1 correctly anticipates her choice, then 2’s second order
belief perfectly coincides with 2’s action. In addition, because players’ second order beliefs

57
are typically elicited after the players have made their choices, the beliefs may simply
reflect an ex-post rationalization of the choices made.

Several approaches have been put forward to solve this identification problem, such
as the disclosure approach, the menu approach, and approaches that experimentally
manipulate second-order beliefs. Applied to a dictator game, e.g., the disclosure approach
elicits the recipient’s expectations about the dictator’s transfer and informs the dictator
about this expectation. Thus, there is no need to elicit the dictator’s second order belief
because the experimenter knows that belief directly. The disclosure approach has been
implemented in several papers and led to mixed evidence for guilt aversion. Some papers
found no evidence for a positive correlation between second order beliefs and the players’
prosocial choices (Ellingsen et al. 2010; Bellemare, Sebald and Suetens 2017), while others
found a significant correlation (Reuben, Sapienza and Zingales 2009; Bellemare, Sebald
and Strobel 2011). Bellemare, Sebald and Strobel (2011) also document that directly
eliciting subjects’ second-order beliefs leads to a strong overestimation of guilt aversion. In
their disclosure treatment, the estimated willingness to pay (WTP) for avoiding guilt is only
roughly 50% of the WTP in their baseline treatment that uses uncorrected second order
beliefs to measure guilt aversion.

The disclosure approach is not without drawbacks, however. While it circumvents

the endogeneity problem associated with the direct elicitation of second order beliefs, it
introduces the possibility of other confounds. For example, providing information about a
partner’s expectation may also signal something about the prevailing social norm, implying
that guilt aversion may be confounded with social norm compliance. In addition, if players
are informed about the other players’ expectations, they may wonder why the experimenter
does this, whether the other players know this, and if so, whether other players have biased
their expectations strategically. Ellingsen et al. intended to mitigate these concerns by
telling the dictators that the recipients are not told of their expectations. But the second
movers then might have wondered what was not being revealed to them. Doubt about the
environment might have reduced the sense of guilt for those who felt uninformed.

Several papers have therefore tried to solve the endogeneity problem by

implementing the so-called menu-approach. Here, the dictators or second movers stipulate a
choice conditional on different, exogenously given, beliefs of their partners (Khalmetski,
Ockenfels and Werner 2015; Bellemare, Sebald and Suetens 2017; Bellemare, Sebald and

58
Suetens 2018; Attanasi, Rimbaud and Villeval 2019). This approach generally leads to
results that support the existence of guilt aversion, but the supporting evidence needs to be
taken with a grain of salt because asking subjects to state their choices for different levels of
the other party’s first-order beliefs clearly suggests to them that they could or should make
their choices dependent on these beliefs. Guilt aversion theory claims, however, that
subjects naturally do so to avoid feeling guilty.

Is there a way to overcome the difficulties in reliably identifying guilt aversion

through the exogenous variation of second-order beliefs? The papers by Ederer and
Stremnitzer (2017) and Khalmetski (2016) indeed solve this problem. Ederer and
Stremnitzer use a variation of the sequential exchange game displayed in Figure 6. Like in
Charness and Dufwenberg (2006), player 2 can send free form messages to player 1 before
player 1 decides whether to trust. 34

Ederer and Stremnitzer successfully implement an exogenous variation in beliefs by

introducing a chance move that generates either a reliable or an unreliable trading
technology in case that player 1 plays trust. If the trading technology is reliable, there is a
5/6 probability that player 2 has the option to trade (i.e., the option to play share), while in
case of an unreliable technology the option to choose share only materializes with
probability (1/6). 35 Importantly, once the type of trading technology is determined, both
players are informed about it, which will obviously affect the players’ expectations. In
particular, player 1 will expect a higher payoff in case of a reliable technology than in an
unreliable trading technology because player 2 has the option of choosing share with a high
probability in the former case. Moreover, player 2 will know this and will thus have higher
2nd order expectations which, according to guilt aversion theory, should trigger a higher rate
of share choices in the presence of a reliable technology.

Ederer and Stremnitzer (2017) verify that the reliable trading technology indeed
induces higher second order beliefs regarding the probability of playing share. Moreover,

34Khalmetski (2016) used a sender-receiver game where the sender has an incentive to lie to
the receiver which hurts the latter. Based on a clever design, he exogenously shifts first and
second order beliefs without changing the sender’s material incentive to lie. His results indicate
that senders are more likely to tell the truth when the senders have higher second order beliefs
about receivers’ payoff expectations. The effect prevails, however, only when incentives for
lying are small, but not when they are large.
35 Ederer and Stremnitzer also allow for different degrees of sharing, but we abstract from this to keep the

exposition simple.

59
this exogenous increase in second order beliefs holds regardless of whether player 2 made a
promise, sent no message, or merely engaged in empty talk. Guilt aversion thus predicts
that this exogenous increase in beliefs will be associated with a higher percentage of share
choices under a reliable technology regardless of the message player 2 sends. Interestingly,
however, the percentage of sharing choices is only higher under the reliable technology in
case of a promise. When player 2 made no promise or simply engaged in empty talk, the
exogenous shift in second order expectations was not accompanied with an increase in
choosing share.

These results may help us understand the conditions under which guilt aversion
theory applies. It seems that moving second order expectations is not always enough to
move behavior in the direction of guilt aversion, but that additional psychological
conditions – such as a promise – need to be present to induce guilt averse behavior. The
special status of promises is also supported by evidence in Bracht and Regner (2013) who
find a significant effect of promises even after controlling for second-order beliefs.

Summary

The theory of guilt aversion provides a simple and intuitively powerful account of prosocial
behavior. There is considerable behavioral evidence that is consistent with guilt aversion
and behavioral measures of individuals’ guilt aversion have been shown to be correlated
with psychological measures of guilt proneness (Bellemare, Sebald and Suetens 2019).
However, the rigorous identification of behavioral guilt aversion is non-trivial and involves
difficult identification problems. Guilt aversion also does not appear to manifest itself in all
environments. The setting must give a sense of fairness, otherwise guilt sensitivity might
diminish. Messages must be credible and persuasive to move beliefs, which in turn dictate
choices. In addition, once beliefs change, the content of message matters for this change to
become behaviorally relevant. We also feel that there is something to the commitment-
based story of why promises are so effective. We do not view this idea as being in
opposition to guilt aversion but rather as complementary. While we do not have a full
characterization of when such emotions come into play, we expect that further research will
help to clarify these connections.

60
3.4. The Role of Self-Image and Social Image Concerns for Prosocial Behavior
One’s image can be important for a great variety of reasons. Just like looking in a mirror and
adjusting physical appearance, people will take actions to appear more favorably. This
applies to both self-image and social image, which to some extent seem inextricably
intertwined: improving one’s self-image or self-esteem may well have positive spillovers
socially, and one’s social image can lead to a better self-image. This topic was first
considered in the field of psychology, where impression management, self-concept, and self-
presentation are terms that reflect the care taken to appear more favorably to self and others.
The Tesser (1988) self-evaluation maintenance model proposes that people are motivated to
maintain both positive self-views as well as their perception about how other individuals
view them. Overall, there is considerable evidence indicating that social comparisons affect
self-esteem (Gastorf and Suls 1978; Molleman, Pruyn and Van Knippenberg 1986).

In this section, we discuss how concerns about one’s image – whether self or social –
affects people’s prosocial behaviors. While many of the prosocial behaviors we discussed in
the previous sections appear to be driven by genuinely prosocial motivations such as altruism,
equity-seeking, or positive reciprocity, it makes intuitive sense that the desire to appear
prosocial to oneself or to others might also drive some prosocial behaviors.

We begin this section with a short presentation of related theory, after which we will
discuss the evidence for self-image and social image driven prosociality. In this context, we
will also pay particular attention to the effect size of image-driven prosociality relative to
other, more genuine, forms of prosociality.

Theoretical concepts

Self-perception theory (Bem 1973) stipulates that people observe their own behavior to infer
what they are thinking and how they are feeling. People without initial attitudes or emotional
responses develop them by observing their own behavior to assess what their attitudes and
motives must have been to induce that behavior. While one might presume that one’s
personality and attitudes drive one’s actions, self-perception theory suggests that this is not
always the case. Instead, “we are what we do”. One way of formalizing self-signaling is to
assume that it is an attempt to influence the beliefs of a future self, who cannot remember the
original motivation for the behavior. Bodner and Prelec (2003) were the first who developed

61
a dual-self signaling model, an approach that was later also applied in Benabou and Tirole
(2006; 2011), Grossman (2015), and Grossman van der Weele (2017). 36

Consider, for example, a simplified version of Benabou and Tirole’s (2006) model.
Here, an agent has an intrinsic valuations va for contributing a to a charity that generates a
costs C(a), so the direct benefit of choosing a is vaa - C(a). Because the agent cares about her
social image, there are also reputational costs and benefits 𝑅𝑅 that depend on what others infer
on average about the agent’s intrinsic prosociality, 𝐸𝐸(𝑣𝑣𝑎𝑎 |𝑎𝑎), given the agent’s observable
prosociality a:

𝑅𝑅(𝑎𝑎) ≡ 𝑥𝑥[𝛾𝛾𝑎𝑎 𝐸𝐸(𝑣𝑣𝑎𝑎 |𝑎𝑎)], with 𝛾𝛾𝑎𝑎 ≥ 0.

The sign of 𝛾𝛾𝑎𝑎 indicates that people would like to appear prosocial and 𝑥𝑥 > 0 is the
visibility of their actions. Note that R(𝑎𝑎) could represent an instrumental or an affective value
of one’s social image. Letting 𝜇𝜇𝑎𝑎 ≡ 𝑥𝑥𝛾𝛾𝑎𝑎 , an agent with reputational concerns chooses a to
satisfy:

max𝑎𝑎∈𝐴𝐴 [𝑣𝑣𝑎𝑎 𝑎𝑎 − 𝐶𝐶(𝑎𝑎) + 𝜇𝜇𝑎𝑎 𝐸𝐸(𝑣𝑣𝑎𝑎 |𝑎𝑎)]

In a nutshell, this model assumes that individuals care for the intrinsic value of
prosociality 𝑣𝑣𝑎𝑎 𝑎𝑎, the material cost involved in behaving prosocially 𝐶𝐶(𝑎𝑎), and their social
image 𝜇𝜇𝑎𝑎 𝐸𝐸(𝑣𝑣𝑎𝑎 |𝑎𝑎). They choose 𝑎𝑎 to optimally balance these components. The model can
also be applied to self-image concerns by assuming that there is some chance that one might
later fail to remember or (remember in a self-serving manner) the reasons for making a
choice. If observable actions are easier to remember than imagined motives, signaling our
type to our later self with our current actions could make sense. Perhaps the exact feelings or
signals at that time are obscured with some probability proportional to x and the agent later
cares about “who she is”.

Bénabou and Tirole (2006) use a slightly more complicated model to derive the
image-spoiling and crowding effects of rewards as well as several other interesting social
phenomena (social stigma, social norms, etc.). Grossman and van der Weele (2017) apply a
self-signaling model in the spirit of Benabou and Tirole directly to explain prosocial

36Dufwenberg and Battigalli (2022) make a strong case that image concerns imply that the decision
maker’s belief about other’s beliefs enters the utility function. In other words, proper modelling of image
concerns requires the tools of psychological game theory. In the case of self-image concerns, beliefs about
other self’s beliefs, and in case of social image concerns, beliefs about other people’s beliefs enter the utility
function. To ease exposition, we use a simplified version of the Benabou-Tirole (2006) model.

62
behaviors – such as information avoidance in altruistic choice (Dana, Weber and Kuang
2007) – that models distributional preferences cannot explain.

Avoidance behavior and self-image concerns

Self-image concerns may play a role if people cross the street to avoid passing by a poor
beggar. There are other plausible explanations for this behavior, however. Perhaps I give
money to charities that I know and trust, so I have no self-image problem regarding my
prosociality. Nevertheless, if I pass a beggar in need and give nothing, I frustrate my
empathic concern and feel bad. But I don’t feel bad because I have a bad self-image, I feel
bad because I empathize with the beggar. In the same way, I feel bad when I see a starving
child on TV or when I learn about a famine. It may have nothing to do with my self-image.
Similarly, social pressure to give to a charity that randomly shows up on one’s door or on the
street can be aversive per se and thus avoiding such situations may have nothing to do with
image concerns. 37

Laboratory experiments provide a more precise tool to identify these concerns. Dana,
Kuang, and Weber (henceforth DKW) (2007) conducted a paradigmatic and influential study
providing support for self-image concerns in altruistic choice. They implemented a binary
dictator game in a baseline treatment, a hidden information treatment, and a plausible
deniability treatment. In the baseline treatment, the allocator can choose between option A,
yielding allocation (6,1), and option B with allocation (5,5). In the hidden information
treatment, the chooser has again two options (A and B) with corresponding payoffs of 6 or 5
for herself but unknown payoffs for the other party. There are two possible states of the
world. In one state the players’ preferences over A and B are aligned because choosing A
yields payoffs of (6,5) while choosing B leads to (5,1). In the state with a conflict of interest,
the payoffs are the same as in the baseline treatment. In the hidden information treatment, the
chooser does not know which state of the world prevails, but she has the option to reveal the
true state at no cost by clicking a box on the computer screen. In the plausible deniability
treatment, the options were identical to the baseline treatment and the choosers were given

37 This argument applies, in our view, to the study of DellaVigna, List and Malmendier (2012) who
convincingly show that some people dislike the social pressure associated in door-to-door funding
campaigns. In their experiment, flyers posted on their front doors told some households that an
individual will come to solicit donations for a charity at a particular time. In the control treatment, no
flyers were distributed. They find that flyers significantly reduced the share of households opening the
door and they also reduced overall donations.

63
10s to decide (as in the other treatments), but the computer could cut off the choosers at
random time points within this 10 s interval and implement (with 50% probability) either
(6,1) or (5,5). Moreover, only the choosers were informed about a cut off, while the receivers
never knew whether the chooser or the computer made the decision. Thus, the chooser can
hide her responsibility for the for an unfair outcome behind a random mechanism in this
treatment, which may facilitate an “other-deceptive motive”. However, if they determine the
selfish outcome on time (i.e., without being cut off), they are still responsible for that
outcome. Delaying decision-making, in contrast, allows them to increase the likelihood of an
actual responsibility delegation which may facilitate a “self-deceptive motive”.

74% of the dictators chose the fair (5,5) option in the baseline treatment, while only
37.5% of the dictators who were randomly assigned to the conflict state chose the fair option
in the hidden information treatment. Thus, despite the fact that dictators in the conflict state
of the hidden information treatment faced identical cost and benefits from a fair choice
compared to the baseline treatment, they were much less likely to behave fairly. The key
reason for this is that only 50% of the dictators who were assigned to the conflict state
actively chose to seek information about the true state. All of those who remained
uninformed chose the selfish option, while the informed dictators chose the fair option in
75% of the cases.

In the plausible deniability treatment, dictators who were not cut off chose the fair
option in 45% of the cases, significantly less than in the baseline. This is consistent with the
view that dictators exploit receivers’ incomplete information about who determined the
outcome, but it is also consistent with guilt aversion. Since receivers knew that the software
could cut off the dictator’s decision, they may have had lower expectations, dictators who
anticipated this simply have felt less guilty when making the selfish choice. 24% of the
dictators delayed their decisions long enough for the software to make the choice. The cut
offs occurred after much more time than what subjects typically needed for deciding in the
baseline treatment, consistent with the view that subjects were willingly cut off.

To what extent can self-image concerns really explain willful ignorance (and willful
delay), i.e., to what extent can we go beyond mere intuitions about the potential role of self-
image? A key paper in this regard is Grossman and van der Weele (2017), who developed a
multiple-self model where a decision-maker (DM)-self manages her image vis-à-vis an
observer-self. One can illustrate the basic intuition of their paper with the help of the notation

64
used when describing the simplified Benabou and Tirole (2006) model above where the
decision-maker’s utility is given by

𝑣𝑣𝑎𝑎 𝑎𝑎 − 𝐶𝐶(𝑎𝑎) + 𝜇𝜇𝑎𝑎 𝐸𝐸(𝑣𝑣𝑎𝑎 |𝑎𝑎, 𝜎𝜎).

The DM self knows her preferences and derives altruistic utility 𝑣𝑣𝑎𝑎 𝑎𝑎 from behaving
prosocially (a = 1), and also puts a positive weight 𝜇𝜇𝑎𝑎 on her self-image 𝐸𝐸(𝑣𝑣𝑎𝑎 |𝑎𝑎, 𝜎𝜎). The
observer self lacks introspective knowledge of the DM’s altruism parameter 𝑣𝑣𝑎𝑎 but can infer
it from the DM’s actions and knowledge σ about the state of the world. The state of the world
σ is either unknown, the conflict state, or the aligned state. The observer’s inference about the
DM’s altruism is given by 𝐸𝐸(𝑣𝑣𝑎𝑎 |𝑎𝑎, 𝜎𝜎) which is the basis for the DM’s self-image.

The key insight derived from this model is based on the existence of a “willful
ignorance equilibrium” in which the selfish individuals (𝑣𝑣𝑎𝑎 = 0) neither behave prosocially (a
= 0) in the baseline treatment nor in the hidden information treatment. Altruistic individuals,
however, can be sorted into two groups: those with a strong preference for altruism (𝑣𝑣𝑎𝑎 ≥ 𝑣𝑣*
> 0) and those with a weak preference for altruism (𝑣𝑣𝑎𝑎 < 𝑣𝑣*). The strong altruists (i) choose
the prosocial option in the baseline treatment and (ii) acquire information and choose the
prosocial option in the conflict state of the hidden information treatment.

The third category of players, the weak altruists, are necessary to explain willful
ignorance in the DKW paper. In the baseline treatment, the DM and the observer know that
the conflict state prevails. Choosing option A with payoffs (6,1) thus provides a clear signal
to the observer that the DM’s altruism is low, which hurts the DM’s self-image. In the hidden
information treatment, however, neither the DM nor the observer know the state of the world.
Therefore, maintaining the uncertainty and choosing A has a much lower signaling value
regarding the DM’s altruism because choosing A was not an unfair choice with 50%
probability. But the critical assumption here is that the observer does not infer something
negative about the DM’s altruism from the mere fact that the DM decides to remain ignorant.
In this regard, self-image models are also self-deception models because individuals can fool
themselves into believing that they are more prosocial than they in fact are. 38

38 Carlson et al. (2020) report evidence that a non-negligible share of individuals recalls being more
generous in the past than they actually were – an effect that occurred mainly in those individuals who
violated their own fairness standards. Motivated misremembering of selfish behaviors may be one source
of self-deception and maintaining a positive self-image.

65
 Thus, the essence of the model is that the hidden information treatment enables the
DM to reduce the negative signaling value of behaving unfairly by obfuscating the observer’s
inferences about her altruism in case of an unknown state of the world. For weak altruists, the
total value of an altruistic act – which consists of intrinsic altruistic utility plus the signaling
value – thus declines below the cost of the altruistic act.

To what extent is self-image driven willful ignorance a robust phenomenon and how
large are its quantitative effects on altruistic behavior? A recent meta-analysis by Vu et al.
(2023) takes a large number of studies comprising roughly 6500 individuals and 56 treatment
effects into account. The meta-analysis shows that across all studies, 57% of the individuals
make an altruistic choice in the full information treatment and that this share is 15.6% lower
in the conflict state of the hidden information treatment. Thus, the share of strong altruists
appears to be roughly 2.5 times larger than the share of weak altruists.

Grossman (2014) examined the robustness of willful ignorance to small, seemingly

innocuous changes in the experimental design. He changes the default for learning the true
state in the hidden information treatment. In their “Default Non-Revealed (NR)” treatment,
which is similar to DWK, subjects need to click a box to receive information; they remain
uninformed otherwise. In “Active Choice” treatment, participants must click on a box to
either learn the state or to not learn the state. In the “Default Revealed (R)” treatment, the
default is reversed so that participants learn the state unless they click on a box to say they
don’t want the information. Figure 7, taken from Grossman (2014) , shows the results.

Figure 7: The Percentage of Dictators Choosing Ignorance

The basic result in DWK is confirmed. About 45% of the participants in “Default NR”
maintained their ignorance and then almost invariably chose allocation (6, 1). Yet matters
change dramatically when one must make an active choice of one of the two boxes, with the

66
non-reveal rate dropping significantly to 25%. The most compelling evidence, though,
comes from reversing the default. When one is required to affirmatively state that one does
not wish to learn the true state, only 3% of the participants opt not to learn it. This is a
remarkable departure from the baseline rate, and it illustrates the sensitivity of preference for
ignorance to the details of the environment. When remaining ignorant, the damage to one’s
self-image seems considerably more severe when one can only remain ignorant by
deliberately electing to do so, indicating limits to self-deception.

Carpenter and Robbett (2022) conducted another robustness test. They implemented
the plausible deniability treatment of DKW (2007) but instead of only one binary dictator
game, subjects participated in 45 different binary dictator games. The reason for the large
number of games was that the authors were interested in estimating subjects’ distributional
preferences parameters. The 45 games included a large number of different situations that
involved many different costs of altruistic choices on negatively sloped “budget lines” (with
two options). In some cases, the cost of the altruistic act was as low as in DWK’s conflict
treatment, i.e., the altruistic choice increased the total payoff, while in other cases the costs
were high enough so that the total payoff decreased. In addition, subjects also faced many
positively sloped “budget lines”. Note that this set-up is likely to decrease demand effects
associated with standard single shot dictator games because the large variation across games
provides many justifications for selfish behaviors (e.g., high costs of altruism on negatively
sloped budget lines or joint payoff maximization on positively sloped budget lines).
Carpenter and Robbett show that plausible deniability does not lead to a significant change in
the estimated distributional preference parameters, i.e., moral wiggle room does not affect
subjects’ preference for altruism in the domain of advantageous and disadvantageous
inequality. This finding is important because it indicates that moral wiggle room may play
little role in the typical experiments (Fisman, Kariv and Markovits 2007; Fisman et al. 2015;
Fehr, Epper and Senn 2022) used to estimate distributional preference parameters.

Willful ignorance is one important example of avoidance behavior, but the literature
has also considered other forms. Dana, Cain, and Dawes (2006), for example, give subjects a
choice between playing a $10 dictator game or taking $9 with an exit option. The recipient
received nothing with this exit option, but in this case the receiver never knew that there had
been an option to make an award in a dictator game. About a third of the dictators chose to
exit, hiding behind the receivers’ ignorance. The authors also conducted a control treatment –
the private dictator game – where the receivers never knew about the existence of the dictator

67
game or the source of any money paid out. Interestingly, very few dictators chose to exit the
private dictator game. Thus, when exiting cannot affect the receivers’ expectations, the exit
option is rarely chosen, suggesting that dictators want to avoid violating the receivers’
expectations. The results are therefore consistent with guilt aversion, but self-image concerns
could also have played a role because the very fact that another anonymous individual knows
that the dictator behaved selfishly may be associated with higher self-image cost. Knowing
that the recipient knows that I behaved selfishly is perhaps like looking into the mirror. 39

The literature on information avoidance in moral wiggle room experiments has

undoubtedly increased our knowledge about the intricacies and driving forces underlying
prosocial behavior. There are clearly relevant conditions under which self-image appears to
play a role. However, the literature frequently also uses a rhetoric suggesting that preferences
for fairness and altruism are merely “illusory” (Larson and Capra 2009) or that “the
underlying motivation driving much fair behavior might be self-interest, coupled with a
desire to maintain the illusion of not being selfish” (Dana, Weber and Kuang 2007).
Likewise, the question is asked “if good deeds stem from altruistic tastes, why do the same
people often seize upon the thinnest of veils to revert to selfishness?” (Benabou and Tirole
2011).

One may ask whether this rhetoric is justified in view of the overall evidence. The
meta-analysis of Vu et al (2023) shows that the share of subjects who behave consistently
altruistic in the baseline and the hidden information condition is 2.5 times larger than the
share of subjects who engage in information avoidance and behave selfishly when the state is
unknown. In addition, we must keep in mind that moral wiggle room experiments in the spirit
of DKW do not prove the absence of truly altruistic components among those who avoid
information. Based on the insights of the Grossman and van der Weele model, the weak
altruists may well put positive value on prosocial acts; the evidence only allows us to say that
their altruistic utility is not sufficiently high to behave consistently across both treatment
conditions. Also, while it is true that seemingly innocuous design features (like the need to
click a box to get informed about the state of the world) induces a minority of individuals to

39Lazear, Malmendier, and Weber (2012) also consider avoidance behavior in a set-up that triggers a
psychological mechanism similar to that in Dana, Cain and Dawes (2006). They compare giving behavior in
a standard $10 dictator game with a treatment where the dictators can exit the dictator game silently
without any cost. In case of exit, the recipient never learns that he was potentially part of a dictator game.
Thus, by exiting the dictator can manipulate the recipient’s expectations, which reduces self-image cost
and/or guilt.

68
revert to selfish behavior, equally innocuous design features (like the need to click a box to
remain uninformed) basically remove this effect (Grossman 2014). And finally, if dictators
make not just one but a large number of choices, moral wiggle room (plausible deniability)
may have little effect on the estimated distributional preference parameters (Carpenter and
Robbett 2022).

Social Image Concerns

The formal framework in Benabou-Tirole (2006) can be applied to self-image concerns and
to social image concerns. In the latter case, the multiple-self interpretation can be discarded
and the term 𝜇𝜇𝑎𝑎 𝐸𝐸(𝑣𝑣𝑎𝑎 |𝑎𝑎, 𝜎𝜎) represents the decision makers’ intrinsic valuation of being
viewed as a prosocial actor by relevant third parties. Note, however, that willful ignorance in
experiments like DKW (2007) can hardly be interpreted in terms of social signaling because
the experiments are one shot and subjects remain anonymous. In addition, the receivers do
not know whether the dictator acquired information. Therefore, the social image
interpretation has little explanatory power in this setting.

There are, however, powerful demonstrations of the relevance of social image

concerns on prosocial behaviors, and many authors have provided models of such concerns
(Hollander 1990; Benabou and Tirole 2006; Ellingsen and Johannesson 2007; Ellingsen and
Johannesson 2008). Social image concerns are, in a sense, just another term for the
preference for social approval. Rege and Telle (2004), e.g., conducted a simultaneously
played one shot public goods experiment where complete free-riding is the dominant
strategy. They implemented a private condition and a public condition where all group
members saw how much an individual had contributed. In both conditions, each individual
first privately committed to a contribution level by deciding how much of her monetary
endowment to put into an envelope dedicated to the public good. Thereafter, each subject had
to stand up and put her “public envelope” into a box in front of a room. In the private
condition, the public envelopes remained sealed, but each subject had to open her public
envelope, count the money in the envelope, and write the amount contributed on a blackboard
in the public condition. Obviously, the public condition generates strong social
image/approval incentives such that the average contribution rate increased from 34% in the
private condition to 68 percent in the public condition.

69
 Ariely, Bracha, and Meier (2009) also exploit the distinction between a private setting
(where choices remained private), and a public setting (where choices were made public to
the other participants in the study) to examine social image effects on prosocial behavior. In
their lab study, participants performed a real-effort task (pressing two keys sequentially) that
led to donations on their behalf to the American Red Cross. In the public setting, the subjects
worked much harder to generate donations to the Red Cross. In addition, the authors tested an
interesting prediction that follows from several social signaling models: that monetary
incentives crowd out prosocial behaviors driven by social image concerns because public
observers will attribute at least a part of subjects’ effort to the monetary incentive rather than
to their prosociality. Ariely, Bracha and Meier confirm this prediction neatly.

Andreoni and Bernheim (2009) consider social signaling in a $20 dictator game with a
stochastic element to the decision. Specifically, there is some probability p that Nature
transfers a fixed value x0 to the recipient, i.e., Nature overrides the dictator’s decision. To
heighten social image concerns, participants were informed that all participants and outcomes
would be publicly identified at the end of the session. Note that only the outcomes for the two
players but not the dictators’ choices are made public. This means that selfish dictators can
hide their choices in those dictator games in which Nature chooses a low transfer with
positive probability, and thus avoid damaging their social image by making the same transfer
that Nature would have made.

Andreoni and Bernheim elicit dictators’ choices for four values of p (0, 0.25, 0.50, and
0.75) and two values (0 and 1) for Nature’s fixed transfer. Their findings provide strong
support for the idea that people care about their social image. Between 57% (in case of x0 =0)
and 69% (in case of x0 = 1) of the dictators chose the equal split when p = 0. However, the
frequency of equal splits dramatically declines with increasing p, and more and more
dictators hide behind Nature’s fixed value and choose exactly x0. For example, for p = 0.5,
and x0 = 0, 72% of the dictators choose exactly 0 and only 28% the equal split. Likewise, for
p = 0.5, and x0 = 1, roughly 45% of the dictators choose x0 = 1 and the frequency of equal
split is slightly below 40%. Thus, taken together, the evidence suggests that social image
concerns (being viewed as fair) constitutes a strong motive. However, we need to consider
that whenever an individual’s social image is at stake, the individual’s self-image is also
likely to be involved. This is so because what could be a stronger hit to one’s self-image than
losing one’s social image? Therefore, the strong effects observed in the experiments
discussed above may be the result of a combined effect of social and self-image concerns.

70
Summary

The evidence suggests that both self and social image concerns play a role in behaviors that
have a clear prosocial meaning. In contrast to other-regarding preferences like altruism,
inequity aversion, reciprocity or guilt aversion, self and social image concerns share a more
self-regarding flavor. This does not mean, however, that truly other-regarding preferences
and image concerns are mutually exclusive. In fact, it appears quite plausible that individuals
simultaneously hold various other-regarding and image-based motives. In terms of
quantitative importance, a recent meta-analysis finds that other-regarding motives appear to
be considerably more important compared than self-image concerns (Vu et al. 2023).
Moreover, both self and social image concerns do not seem to lend themselves to easily
explain the rejection of low offers in ultimatum games or the payoff-reducing behaviors
observed in allocation tasks with positively sloped budget lines because these behaviors lack
a clear prosocial meaning.

Image concerns imply that situational and institutional factors, that are completely
irrelevant and innocuous in a world without image concerns, can become highly relevant.
Self-image concerns can be mobilized for prosocial behaviors by removing uncertainties
about the prosocial effects of these behaviors, by reducing opportunities that enable plausible
deniability, or by making decisions to remain uninformed public knowledge. All these
precautions reduce the likelihood that players can deceive themselves into believing they are
more altruistic than they in fact are. Likewise, making decisions transparent and publicly
known to relevant others activates social image concerns.

The delegation of decision-making rights to agents, combined with ignorance about the
agents’ decisions, is an institutional set-up that appears particularly prone for mitigating
responsibility for problematic behaviors. There is, in fact, a considerable experimental
literature that documents the responsibility-alleviating and self-interest enhancing capacities
of delegating decision-making rights (Charness 2000; Fershtman and Gneezy 2001; Charness
and Jackson 2009; Hamman, Loewenstein and Weber 2010; Bartling and Fischbacher 2012),
and image concerns might also drive the behaviors observed in these experiments.

Authoritarian rulers, emperors, and mafia bosses have long known about such
responsibility-alleviating effects as they often let others do the dirty work. Experimental
evidence also suggests that those who delegate and hope that their agents implement selfish

71
allocations on their behalf indeed receive less punishment compared to those who directly
enact the selfish act. Although the literature on delegation and responsibility is exciting,
space constraints induce us to defer it to Appendix 9.

4. Economic and Political Consequences of Social Preferences

Like time and risk preferences, social preferences have broad implications for a wide variety
of domains. In Section 4 we review evidence how social preferences (i) affect employees’
responses to wage inequality and wage cuts, (ii) the extent to which this affects firms’
employment decisions, (iii) how the affect the allocation of workers with varying prosociality
levels to different industries, (iv) how incentives and contract are influenced by them, and (v)
how they affect the political demand for redistribution. These topics do not exhaust the
consequences of social preferences, but space constraints force us to limit consideration on
the above-mentioned issues. For this reason, we mention here only briefly two other
important topics for which they are relevant: Cooperation and normative public economics
(“welfare economics”).

There is a large theoretical and empirical literature on the role of social preferences for
cooperation and collusion, including several review papers (e.g., van Lange et al. (2014);
Fehr and Schurtenberger (2018); Balliet, Mulder and van Lange (2011)). Theory suggests that
altruistic preferences tend to facilitate cooperation while the role of behindness aversion or
negative reciprocity is more nuanced and depends on the possibility, and specific features, of
peer punishment opportunities. If such opportunities are provided in public goods
experiments, behindness aversion and negative reciprocity are motivational forces that
facilitate the punishment of free-riders which helps to maintain cooperation levels (Fehr and
Schmidt 1999, Fehr and Gächter (2000), Falk and Fischbacher 2006). However, if such
opportunities are absent, these forces may have detrimental effects on cooperation and induce
players to cease cooperating.

The relevance of peer pressure and peer punishment has also been established in recent
field studies of large-scale cooperation. Breza, Kaur and Krishnaswamy (2019) studied
whether large groups of decentralized workers cooperate to prevent downward pressure on
wages. They implemented a field experiment in 183 local labor markets in rural India and
show that only 1.8% of agricultural workers are willing to accept jobs below the prevailing
wage when this choice is observable by other workers. In addition, they document that this

72
unwillingness to accept low wages is due to workers willingness to sanction those who accept
wage cuts, and if acceptance of low wages is not observable (and thus cannot be sanctioned
by their peers) the willingness to undercut the going wage increases substantially. Moreover,
consistent with the aggregate implications of downward rigidity, Breza et al. also show that
measures of social cohesion in local markets correlate with downward wage rigidity and its
employment effects across India.

Another powerful real-world example of large-scale cooperation supported by peer

punishment is provided by the hugely successful recruitment of soldiers for the British Army
at the beginning of World War I when the army relied entirely on the voluntary recruitment
of soldiers. Between August 1914 (when Britain declared war on Germany) and September
1914 roughly 479 000 volunteers were recruited and until December 1915 roughly 2.5
million men joined the British Army voluntarily. Those who did not join faced the contempt
of their community members who attached big red patches to the free-riders’ front doors at
night, so that everybody could see that the person living there was a dodger (Simkins (1988)).
Recently, Becker (2022) documented how young women publicly shamed young men who
refused to join the army. In many towns and cities, the women handed out white feathers to
men in civilian clothes, marking them out as cowards. The young women often took
substantial risk when doing so because the affected men retaliated. Becker collects evidence
from local newspaper articles and exploits the gradual spread of the movement to show that
during the 10 days after the first mention of White Feather Girls in the news, volunteering
surged by a third.

Cooperation is not always a good for the overall society. The cooperation between
companies for the purpose of maintaining high prices, or the cooperation within criminal
organizations is an example. Another example is vote-buying, a frequent practice in many
countries with weak democratic institutions. In a fascination study, Finan and Schechter
(2012) document how social preferences for reciprocity facilitate vote-buying in municipal
elections in Paraguay. Vote-buying constitutes a serious puzzle in a secret voting
environment with selfish voters because they cannot commit to vote for the candidate who
tries to buy their vote. A selfish voter would just take the bribe and merely claim that he or
she voted for the bribing politician. However, if voters have an intrinsic preference for
reciprocity this commitment problem can be overcome.

In Paraguay, politicians hire respected community leaders in each village to interact

with voters and offer them money and other forms of aid for the promise of their vote. Finan

73
and Schechter (2012) show that these community leaders have a very good knowledge of
individual voters’ preferences for reciprocity and preferentially target reciprocal voters for
vote-buying. A one standard deviation increase in reciprocity increases the likelihood of
being targeted for vote-buying by 44% - a finding that is robust to a large set of controls
including other social preferences and voters’ network relationships in the village.

Another domain where social preferences matter is the role of individual heterogeneity
in public goods provision. The theory of inequality aversion predicts that groups that are
more heterogeneous – in terms of their wealth (endowments) or in terms of the benefits they
derive from public goods – are less likely to achieve and maintain successful cooperation, a
prediction that is supported by a large experimental literature (Chan et al. 1996; Chan et al.
1999; Anderson, Mellor and Milyo 2008) and consistent with field observations (Mayer
2001; Fajnzylber, Lederman and Loayza 2002).

The influence of prosocial preferences on groups’ abilities to maintain high levels of

cooperation has been documented in Gächter and Thöni (2005). They measured individuals’
willingness to cooperate in a one-shot social dilemma game and subsequently formed three
types of homogeneous groups: (i) groups comprising individuals with a high prosocial
preference, (ii) intermediate groups and (iii) groups comprising selfish individuals. Gächter
and Thöni show that aggregate group cooperation in public good games with a dominant free-
riding strategy is close to maximal in groups of type (i), intermediate levels of cooperation
are achieved in groups of type (ii) and the lowest cooperation levels prevail in groups with
predominantly selfish individuals.

These lab findings are nicely echoed in field evidence. Rustagi and Kosfeld (2015)
measured prosocial and antisocial tendencies of village leaders in Ethiopa in a third-party
punishment game. These leaders were responsible for monitoring and sanctioning of free-
riders communities that are strongly relying on the successful management of forest
commons. Rustagi and Kosfeld show that villages with prosocial leaders have significantly
better forest outcomes. These results continue to hold after careful consideration of reverse
causality issues and omitted variable bias. Similar results of the effects of prosocial
preferences on cooperation have been reported in Rustagi, Engel and Kosfeld (2010) and
Carpenter and Seki (2011).

Social preferences are also important for welfare economics. If individuals display
altruistic or inequality averse distributional preferences, it may not make much sense to

74
compute optimal policies on the basis of social welfare functions that assume that every
individual only cares for his or her own consumption. Isn’t economics, after all, built on a
deep commitment to respect individuals’ preferences? Likewise, if people care also for
equality of opportunity, it appears of paramount importance to incorporate that notion into
modern welfare economics rather than computing optimal policies on the basis of a standard
utilitarian welfare function that assumes that individuals only care for their own consumption.

With regard to the incorporation of distributional preferences into welfare economics

there has been some progress recently. In a recent AEA Distinguished Lecture, Emmanuel
Saez (2021) emphasized the importance of concerns about inequality, poverty and relative
position for positive and normative public economics. Likewise, Aronsson and Johansson-
Stenman (2020a) have studied optimal income taxation in the presence of externalities and
inequality averse individuals, and the same authors have derived optimal second-best taxation
conditions when individuals have social preferences (Aronsson and Johansson-Stenman
2020b). And more recently, Eden and Piacquadio (2023) discussed the normative content of
other-regarding preferences. However, despite these recent studies, it is probably fair to say
that the bulk of normative economics neglects other-regarding preferences.

4.1 Implications of Social Preferences for Labor Relations and Macroeconomics

4.1.1 Fairness Concerns, Wage Inequality, and Job Satisfaction

If people care for equity and reciprocity, it is likely that wage inequalities that violate their
equity standards will have detrimental effects on performance and satisfaction. This
viewpoint is expressed by many practitioners in Human Resource Management. Bewley
(1999) who interviewed several hundred personnel managers about pay-related issues
concludes, for example: “The main function of internal structure is to ensure internal pay
equity, which is critical for good morale” (p. 82). However, are the views managers express
in surveys and interviews also backed by behavioral data from laboratory experiments and
field studies? As we will see below, the behavioral data generally provide a strong
endorsement for the role of fairness and equity concerns in the assessment of wage
inequalities and indicate the conditions under which pay inequalities have detrimental effects
on performance.

The notion of inequity aversion implies that employees who work under identical
conditions and provide identical effort should be paid identically. If, instead, they are offered

75
unequal wages, the prediction is that workers who dislike disadvantageous inequality and
receive a lower wage will reduce their effort. Gächter and Thöni (2010) conducted a
laboratory experiment in a repeated one-shot gift exchange setting (i.e., an environment with
noncontractible effort) in which one experimental employer faces two workers and makes flat
wage offers, wi and wj, to each of the two workers. The workers are then informed about wi
and wj after which they choose their effort levels ei and ej which are associated with effort
costs of c(ei) and c(ej). Gächter and Thöni find indeed that disadvantageous wage
discrimination for worker i (i.e., increasing wj for a given level of wi) reduces her effort level,
while advantageous wage discrimination (i.e., decreasing wj for a given level of wi) leaves ei
unaffected.

In subsequent studies (Gächter, Nosenzo and Sefton 2012; Gächter, Nosenzo and
Sefton 2013), the authors also showed that social preferences affect workers’ effort behavior
even in the absence of wage inequality. Their design is similar to that of Gächter and Thöni
(2010), but after the workers observed the employer’s wage offer, they chose their effort
levels sequentially. Thus, the effort level of employee 1, who chose first, could affect the
effort level of employee 2, who chose second. It turns out that if both workers received
generous wage offers, the effort level of employee 2 is strongly positively correlated with the
effort level of employee 1 – a finding that inequality averse preferences predict.

The lab-based papers discussed above implemented a situation where all parties knew
wages, effort levels, and the workers’ output at different effort levels. In this set-up, the
involved parties have clean data that enables them to compare their outcomes. However, what
happens if, e.g., there are large productivity differences between the workers, but they do not
know the exact differences and are only informed that their productivity differs? In this case,
social comparison processes are necessarily based on less precise information which –
depending on workers’ beliefs about the co-worker’s productivity – may strengthen or
weaken the impact of differential wage payments on effort choices. For example, if workers
believe that productivity differences are minor, then wage inequality may have negative
effects on effort, while they may consider wage differences to be more justified if workers
believe that there are large productivity differences. Charness and Kuhn (2007), who were the
first to implement three-player (one-shot) gift exchange games, implemented a design like the
one described above. They report that co-workers’ wages do not affect workers’ effort
choices. A possible interpretation of this finding is that the subjects in their experiments do
not have social preferences. This is, however, unlikely to be the case because workers

76
responded strongly with their effort choices to generous own wages. An alternative
interpretation is that workers believed that there are large productivity differences that
justified differential wage payments.

To what extent are the findings above about the negative effort spillovers of higher co-
workers’ wages generalizable to the field? Cohn et al. (2014) implemented a wage cut in a
field experiment that offered a one-time job opportunity to workers who performed the job in
teams of two. The firm placed a job advertisement stipulating an hourly wage of about €10 on
an online search platform. The task for both workers was identical and consisted of selling
promotional cards that permitted entrance to specific nightclubs. The experiment had two
phases that were spread over two subsequent weekends with 6 hours of work per weekend.

In phase one (the first weekend) both workers received the same the same hourly wage
of €12 while in phase two (the second weekend) either (i) none of the workers or (ii) only of
the two workers or (iii) both of the workers received a wage cut of €3 relative to the first
weekend. Note that even the workers who received a wage cut on the second weekend earned
a higher total income for the overall job (6x12 + 6x9 = 126) than that which they initially
could have expected with the hourly wage of about €10 initially announced (12x10 = 120).

The authors find that a unilateral wage cut leads to a 34% reduction in the performance
of the worker whose wage is cut relative to the no-wage cut group while the performance of
the worker whose wage is not cut remains unchanged. 40 Thus, similar to the lab experiments
discussed above, disadvantageous wage inequality is associated with a large negative effect
on effort, while advantageous wage inequality leaves effort unchanged. One might therefore
expect that a multilateral wage cut would lead to a smaller effort reduction compared to a
unilateral wage cut, which is indeed what Cohn et al observed. A multilateral wage cut to
both workers reduced their performance “only” by 15% relative to the no-wage cut group,
and this difference between the unilateral and the multilateral cut is highly significant. These
findings are in line with the predictions of a model of inequality aversion.

Breza, Kaur and Shamdasani, henceforth BKR, (2018) implemented a month-long field
experiment with Indian manufacturing workers who worked for a daily wage. Their study
provides a rich collection of facts regarding the relevance of wage inequality for workers’

40To interpret this finding, it is useful to understand that while the two workers in a “team” worked
during the same shift and in the same environment (e.g. at a well frequented subway station) there was
no interdependence in their task and their interactions during a shift were minimal. In particular, they
had little information about their co-workers’ effort during the shift.

77
daily labor supply (i.e., showing up for work), their effort during work, and the impact of
wage disparities on the subsequent ability to cooperate in other tasks.

BKR assembled production units consisting of three workers who sit together in a
separate physical space during work and lunch breaks. The workers in a production unit thus
form a natural reference group. They randomized workers into (i) a pay disparity condition
where the daily wage reflected workers’ baseline productivity that was assessed during an
initial training period, and into (ii) a pay compression condition in which all three members
of a unit are paid the same wage. By randomizing workers with different baseline
productivities to production units, they introduced variation in the extent to which pay
differences overstate productivity differences in the disparity condition. And by randomizing
production units to different tasks that differ with regard to the observability of co-workers’
output, the authors can examine the impact of output observability on effort during work. In
addition, the authors also measured workers’ attendance on the job (they are only paid when
they appear on the job) and their knowledge about co-workers’ wages within and across
production units. BKR show that workers within a unit are well aware of their co-workers’
wages, while little wage information travels across production units.

BKR find that for workers with a similar baseline productivity and a given absolute pay
level, a worker’s output declines by 0.33 standard deviations (22 percent) on average when
the worker is paid less than his two co-workers. Moreover, holding the level of absolute pay
constant, there is no evidence that receiving a higher wage than one’s peers increases output.
In fact, pay inequality in the presence of similar baseline productivities across workers
appears to cause a general dissatisfaction with the job situation in the sense that both overpaid
and underpaid workers reduce their attendance at the job compared to the compressed pay
condition. This means that the workers in the pay disparity group give up valuable earnings –
on average by 9.3% - by substantially reducing attendance.

Note that these facts about the effect of over and underpayment on effort/output nicely
coincide with the lab findings of Gächter and Thöni (2010) and the field finings of Cohn et al
(2014). BKR also show that in tasks where individuals’ productivity differences are easily
observable, paying different wages has no negative impact on workers’ output. This suggests
that wage inequality has no negative effect when productivity differences justify pay
differences – a finding that is consistent with our interpretation of the lab evidence in
Charness and Kuhn (2007).

78
 BKR also report a remarkable finding about the detrimental impact of unjustified pay
disparities on the subsequent ability of workers to cooperate even when it is in their self-
interest to cooperate. On the last day of their job, the workers participated in two cooperative
games in each of which they could earn money on the basis of group piece rates for
performance. The outcome of these games did not affect the firm’s payoff. In the first game,
the members of a unit had to build towers of raw materials; the higher the tower, the more
each worker in the unit earned. In pay disparity units with little or no baseline productivity
differences, the workers built towers that were 17% shorter on average compared to the
compressed pay units. In contrast, when pay differences were justified – based on baseline
productivity or task observability – pay disparity units performed equally well as compressed
pay units do.

Finally, their endline survey reveals that workers from pay disparity units show a
significantly lower social cohesion – in terms of their willingness to borrow or lend, seek or
give advice or visit one another’s homes – compared to workers in the compressed pay group.

More recently, Cullen and Perez-Truglia (2022) conducted a field experiment on the
effects of salary comparisons with a sample of 2060 employees from a large corporation in
Southeast Asia. They document substantial misperceptions of managers’ and co-workers’
wages and identify the causal impact of changes in salary perceptions with the help of an
information provision experiment. They find that a higher perceived peer salary has a large
negative effect on employee’s own effort. A 10% increase in employee’s perception of peer
salaries significantly reduces the number of hours they work by 9.4%, the number emails they
send by 4.3% and the sales performance by 7.3%. The authors also collected survey evidence
that is consistent with the view that social preferences are the mechanism underlying these
peer comparison effects, as they also find that higher perceived peer salaries have negative
effects on pay and job satisfaction.

These negative effects of peer salaries on job satisfaction are consistent with the
findings of Card et al. (2012), who conducted a field experiment with “University of
California employees” by providing them with easy access to information about peer salaries.
Employees who had salaries below the peer median subsequently displayed a reduced job
satisfaction and an increased intention to switch jobs, while those with salaries above the
median showed no changes in job satisfaction and no intention to switch. These findings are
consistent with the view that employees have a considerable aversion against
disadvantageous inequality.

79
 D’Ambrosio, Clark and Bazzaretta (2018) provide further evidence for the role of fair
reference wages on quitting behavior using data from the German Socio-Economic Panel
(SOEP). The SOEP contains panel data on which income people consider as fair for their
current job. They show that individuals’ fair income gap (the difference between what they
earn and what they consider fair for their current job) is not only significantly associated with
individuals’ life and job satisfaction, but it also influences workers’ emotional states such as
the frequency of feeling happy or feeling sad and a strong influence on the frequency of
experiencing anger. Finally, consistent with these findings on subjective assessments of well-
being, the fair income gap also predicts the probability of quitting within the next year.

The evidence provided by Dube, Giuliano and Leonhard (2019) further provides strong
support for the view that fairness concerns involving comparisons with (higher) peer wages
have a substantial impact on workers’ quitting behavior. The authors estimated the own-wage
and the peer-wage elasticities of employees’ job quitting behavior at a large US retailer with
hundreds of stores nationwide. They exploited a regression discontinuity that resulted from
the firm’s response to the federal minimum wage increases in 1996 and 1997 to identify the
causal effect of own and peer wages on quitting behavior. The results show that job
separations are extremely sensitive to rising peer wages with peer wage elasticities of 20, 9
and 3 for three, six and nine months after the raise. This result contrasts sharply with the
rather low own-wage elasticities they found. Their estimates suggest that, holding the gap
between own and peer wage constant, a uniform raise in wages has no impact on quitting
behavior. Thus, the overall effect of wages on separations is mostly driven by peer
comparisons.

Finally, Dube et al. show that the peer wage effects are asymmetric because they are
driven by comparisons with higher paid peers, which again suggests that aversion against
disadvantageous wage inequality is an important driver of labor market behavior. The overall
findings from surveys, lab experiments, and field evidence suggest that aversion against
disadvantageous wage inequality that cannot be justified by effort or productivity differences
generates strong behavioral effects in terms of a reduced willingness to perform, an increased
willingness to quit, a lower job satisfaction, a reduced social cohesion and lower willingness
to cooperate.

80
4.1.2 Fairness Concerns and Resistance to Wage Cuts

There is a considerable literature indicating the importance of fairness concerns for the
presence of downward wage rigidity. Surveys conducted by Kaufmann (1984), Blinder and
Choi (1990), Agell and Lundborg (1995; 2003) and Bewley (1995; 1998; 2002) all point in
the direction that workers strongly resist (nominal) wage cuts for fairness reasons even in
recessions, and that personnel managers are keenly aware of this resistance. While these
surveys do not pin down the concrete social preferences underlying workers’ fairness
concerns, a plausible interpretation of the evidence suggests that preferences such as negative
reciprocity or inequality aversion (with suitable reference points) may provide the
motivational raw material for these concerns.

Because fairness concerns induce workers to resist wage cuts and personnel managers
anticipate this resistance, the survey evidence suggests that firms will be very reluctant to cut
wages, which in turn mitigates labor market adjustments to exogenous shocks. There is
indeed substantial laboratory and field evidence of downward wage rigidity and often the
data point towards the existence of fairness concerns as the underlying mechanism.

Regarding the field evidence related to whole labor markets, Dickens et al. (2007)
document wage rigidity in many countries. Fehr and Goette (2005) show downward nominal
wage rigidity for Switzerland, while Grigsby, Hurst & Yildirmaz (2021) document it for the
US. In addition, Fehr and Goette (2005) show that downward wage rigidity is negatively
related to employment and Kaur (2019) also finds that it is associated with employment
distortions. Kaur studies downwards rigidity in the context village labor markets in India and
documents strong rigidity. In addition, she reports that Indian village workers consider
nominal wage cuts to be very unfair, suggesting that fairness related resistance to wage cuts is
driving rigidity. Interestingly, workers do not consider real wage cuts that arise from
avoiding nominal pay rises in response to inflation to be unfair – a finding that is consistent
with data reported in Kahneman, Knetsch and Thaler (1986).

Firm-level evidence is provided by Greenberg (1990) who reports that workers

responded to a temporary wage cut that was triggered by a negative demand shock, with an
increase in employee theft during the period for which pay was cut. Krueger and Mas (2004)
document that workers at Bridgestone/Firestone’s Decatur, Illinois, plant responded to the
firm’s attempt to cut wages and hire replacement workers with the provision of lower quality
tires. Their monthly data show that defective tires were produced primarily during those

81
months in which the firm demanded wage reductions and incumbents worked side by side
with replacement workers.

Coviello, Deserranno and Persico (2022) report evidence on workers’ responses to a

wage cut in a sales call center in the US. This company paid its sales representatives on the
basis of two performance indicators: commissions based on net sales (gross sales minus
refunds due to dissatisfied customers) and conversion rates (percentage of calls resulting in
positive gross sales). When the company raised the required conversion rates, which was
associated with a 13% earnings reduction at a given performance, many sales representatives
responded by keeping gross sales constant but increasing customer refunds by intentionally
selling suboptimal items to the customers. Note that this behavior not only hurt the company
but also the workers themselves, indicating that workers were willing to take costly actions to
punish the firm for cutting their wages.

Labor relations are often long-term. Therefore, if workers respond to wage cuts with
reduced effort, sabotage, or higher theft rates, one may interpret this as a rational punishment
for employers in a repeated game, i.e., the workers’ responses may not necessarily result from
their social preferences. For this reason, it is useful to study responses to wage cuts in more
short-term employment situations in lab and field experiments where there is no prospect for
future employment.

The laboratory evidence on downward wage rigidity comes from experimental labor
markets (e.g., Fehr, Kirchsteiger and Riedl (1993); Fehr and Falk (1999); Charness (2004);
Charness and Brandts (2004); Brown, Falk and Fehr (2004)) that are designed in such a way
that in the absence of social preferences, employers have an incentive to pay competitive
wage levels that are rather low. However, if effort is non-contractible and workers respond to
the low competitive wage levels with lower effort (due to social preferences such as
reciprocity or inequality aversion), even selfish employers have a pecuniary incentive to pay
high, non-competitive wages. The evidence from these experiments indeed indicates that
employers are reluctant to cut wages to competitive levels because they will then receive low
effort levels from their workers. The experiments also show that even in the presence of a
large excess supply of workers, the experimental employers shy away from cutting wages to
low, competitive levels because of anticipated detrimental effects on workers’ performance.

Do the negative effort responses triggered by wage cuts in the lab generalize to field
experimental settings that credibly rule out repeated game effects? There are studies

82
indicating that the answer is “yes” (Kube, Marechal and Puppe 2013; Cohn et al. 2014). We
discussed already the negative productivity effects of wage cuts in Cohn et. al. in the previous
section. Likewise, the evidence in Kube, Marechal and Puppe (2013) indicates large negative
productivity effect of 20% from a wage cut relative to a no-wage-cut treatment.

Another interesting study documenting the employment effects of downwards wage

rigidity is Breza, Kaur and Shamdasani, BKR, (2021). These authors implemented hiring
shocks in local Indian labor markets by giving jobs to an average of 24 percent of the labor
force of casual male workers in external jobsites for two to four weeks – a shock that
substantially reduced how many workers remained in the local economy.

Their approach exploits the strong seasonality in labor demand in these local labor
markets. The hiring shock led to immediate and strong rises in wages and a fall in
employment in the local markets during the peak season, when demand for casual labor is
generally high. However, in the lean season, when the demand for casual labor is generally
low, the hiring shock had basically no impact on employment and wages in the local market.
This is a remarkable finding since there were apparently enough unemployed workers who
filled the gap generated by removing 24 percent of the available labor force, indicating severe
rationing of labor supply – a finding that could not have happened in a competitive labor
market with flexible wages. If the local labor market during the lean season had been cleared
before the hiring shock, then the shock also should have led to large wage increases and a fall
in employment. However, apparently there were enough unemployed workers before the
hiring shock who were willing to work at the going wage but could not find employment at
that wage. This made it possible for employers to find enough workers without needing to
raise wages despite the considerable reduction in the local labor force. Breza, Kaur and
Shamdasani also provide evidence indicating that moral hazard or nutrition efficiency wage
models cannot explain their data, while a model that relies on workers’ resistance to wage
cuts can.

Quach (2020) provides evidence for downward wage rigidity by exploiting the
following natural experiment from the US. In May 2016, the federal Department of Labor
announced that starting December 1, 2016, salaried workers earning less than $913 per week
would be entitled to overtime compensation if they work more than 40 hours in a week. In
response to this announcement, many employers promised raises to their employees in
anticipation of the new rule. However, one week before the rule became effective, a federal

83
court ordered an injunction on the new policy, implying that the employers would not face
any legal obstacles if they wanted to refrain from the promised pay rises.

Quach (2020) shows that employers nevertheless increased wages. For the median
worker, for example, wages rose by 5.8%, suggesting the employers shied away from cutting
nominal wages to pre-announcement levels The pay rises took the form of bunching many
employees at $913 per week and reclassifying some workers from salaried pay to hourly pay.
Quach also shows that workers who received pay rises through bunching experienced the
same wage growth compared to workers slightly above the bunching threshold, suggesting
that firms did not lower the future wage growth of workers whose wages exhibited rigidity.
Moreover, the paper shows that even a year after the proposed overtime policy was nullified,
the employers continued to bunch the salaries of new hires at the $913 threshold, indicating
that wage nominal wage rigidity also affected the new hires.

The evidence from Quach (2020) suggests that even the mere promise of a pay rise
based on a temporary legal requirement makes it hard for employers to subsequently lower
wages. This finding is also consistent with the laboratory evidence documented in Fehr, Falk
and Zehnder (2006) where the temporary implementation of a legal minimum wage led to
lasting effects on wages that prevailed even long after the legal minimum wage was removed.

4.1.3 Screening and Selection based on Social Preferences

If social preferences are a relatively stable attribute of individuals, employers may want to
attract workers with particular social preferences and avoid workers with others. Workers
with altruistic preferences, e.g., may be valuable for employers because they generate
positive spillover effects on other workers in interdependent production processes.
Conversely, employers may shy away from workers with envious or spiteful social
preferences because they may have detrimental effects on cooperation among employees and
between the envious employee and the employer. Likewise, employers might avoid workers
who are negatively reciprocal because they may have a strong tendency to engage in
counterproductive activities when they are aggrieved. 41

41
Selection and sorting only make sense if individuals’ social preferences or their assignment to a particular social
preference type exhibits a reasonable degree of stability over time. In Appendix 9, we discuss evidence suggesting
that this is the case.

84
 With regard to self-selection of employees, there is a relatively large literature
suggesting that people with more prosocial inclinations tend to self-select themselves to a
higher degree into the public sector in countries with a high degree of trust into the public
sector (e.g., Dur & Zoutenbier (2014)). Most of these studies are based on self-reported data
about motivation or self-reported prosocial actions. However, evidence based on revealed
preference data also exists. Buurman et al. (2012) show that early career public sector
workers are more likely to donate to a charity compared to observationally equivalent private
sector workers. Gregg et al. (2011) study British Household Panel Data and show that
workers who are more prosocial – in terms of providing unpaid overtime work – are more
likely to sort into the non-profit sector. They also find that this effect is strongest for
industries with “caring characteristics” such as health, education, and social care.

Prosocial individuals may not only prefer working in companies and sectors with caring
or helping characteristics, but they may also shy away from sectors or companies involved in
immoral business practices such as the intentional sale of toxic financial assets, the marketing
of tobacco products to underage smokers, or the aggressive marketing of opioids by the
pharmaceutical industry. Schneider, Brun and Weber (2020) used administrative, laboratory,
and survey data to study the hypothesis that the least prosocial (i.e., most immoral) people are
most likely to work in jobs perceived to involve (or actually involving) immoral activities.
Moreover, if working for a company/industry that is perceived to be involved in immoral
activities is emotionally aversive, standard economic theory would predict compensating
wage differentials. In other words, labor market competition would induce
companies/industries perceived to be more immoral to pay, ceteris paribus, higher wages.

To examine the compensating wage differentials hypothesis, they collected survey data
from the Swiss population on the perceived morality/immorality of different industries in
Switzerland. Then they regressed the gross hourly wages across the industries on the
industries’ perceived immorality, controlling for observable industry and workers’
characteristics. The results indicate a strong positive correlation between the perceived
immorality and the gross hourly wages, with industries such as tobacco and weapons
manufacturing paying the highest wages and construction and sports facilities being among
the lowest paying industries.

Because the correlational evidence from administrative data is, of course, not yet fully
convincing, they also collected a measure of prosociality (immorality) where subjects had the
chance to earn more money by giving wrong advice that reduced both (i) another individual’s

85
earnings and (ii) charitable donations to UNICEF. In addition, the subjects also participated
in a separate competitive experimental labor market with two different treatment conditions –
a neutral work condition and an immoral work condition. In the immoral work treatment, the
subjects were competing for jobs that required them to give wrong advice to another
individual that reduced that individual’s earnings and the charitable donations to UNICEF. In
the neutral treatment, the job involved giving advice that increased another individual’s
earnings and donations to UNICEF.

The striking result of this experiment is that the reservation wages, and thus the
competitive equilibrium wages, for the immoral job are much higher than for the neutral job.
Moreover, subjects with less prosocial preferences have a much higher frequency of
employment in the market involving the immoral task. Thus, the lab experiments provide
causal evidence for compensating wage differentials for immoral jobs and for selective
sorting of more immoral individuals into these jobs. Finally, the authors also show with the
help of survey evidence that subjects who are less prosocial are more willing to work for
industries perceived to be more immoral.

Dohmen et al. (2009) provide further evidence on the sorting/selection hypothesis of

social preferences. Their results suggest that workers’ attitudes towards positive and negative
reciprocity can have quite far-reaching effects on their workplace behavior and their earnings.
They exploit an interesting survey measure of reciprocal preferences developed by Perugini
et al. (2003). This personality questionnaire was developed and psychologically validated
with the intention to identify positive and negative reciprocity as independent preference
characteristics, i.e., the questionnaire is deliberately designed to minimize the correlation
between positive and negative reciprocity. This survey measure of intrinsic reciprocity was
included in the German Socio-Economic Panel (SOEP) in 2005, and Dohmen et al. (2009)
used it to estimate the association between workers’ willingness to voluntarily provide effort
(in the form of overtime work) and their positive and negative reciprocity.

Controlling for a large number of individual characteristics, Dohmen et al. show that
positive reciprocity measured in 2005 is significantly associated with workers’ actual
overtime work in the years 2005 as well as in the years 2006 and 2007. Moreover, the
coefficient on positive reciprocity is almost twice as large for workers who perceived their
current wage as fair, while if workers perceive their current wage as unfair, the association
between overtime work and positive reciprocity is zero. These results are consistent with the

86
view – derived from theories of inequity aversion and reciprocity – that wages perceived as
fair induce reciprocal workers to increase their work effort.

Dohmen et al. also find that negatively reciprocal workers are less willing to perform
overtime work. In addition, they show that positively reciprocal workers are less likely to be
absent from the workplace, while negatively reciprocal workers are more absent. Likewise,
positively reciprocal workers “consume” fewer days for paid sick leave while negatively
reciprocal workers are more days on paid sick leave.

Based on these results, one would expect that positively reciprocal workers are more
valuable employees, i.e., that the labor market will reward them with higher wages, while
negatively reciprocal workers are less valuable. Dohmen et al. estimate Mincer-type wage
equations and find indeed, that positively reciprocal workers earn higher monthly and annual
labor incomes. They do not find a negative impact of negative reciprocity on wages but
instead they show that negatively reciprocal workers have a higher probability of being
unemployed while positive reciprocity reduces the probability of being unemployed.

Barr and Serneels (2009) conducted trust games with several hundred employees from
20 manufacturing companies in Ghana; 164 of them were in the role of the second-mover,
which provides a (noisy) measure of the workers’ willingness to display reciprocal behavior.
Note that the first-mover in this game reaps a positive rate of return if she gets back more
than what she transferred. Barr and Serneels categorize a worker as highly reciprocal if his
back-transfer yields a rate of return for the first-mover of more than 50%. 42 The authors show
that the output per worker across companies is strongly positively correlated with the share of
high reciprocators among employees. This correlation persists when controlling for capital
inputs and sector fixed effects. Moreover, a Mincer-type earnings regression that includes a
dummy for highly reciprocal workers indicates that these workers earn a wage premium.

Although the papers by Dohmen et al (2009) and Barr and Serneels (2009) do not
establish a causal relationship between workers’ willingness to reciprocate and their
workplace behaviors and earnings, they nevertheless constitute suggestive correlations that
deserve further scrutiny. Their findings are consistent with what one would theoretically
expect based on knowledge about the behavioral properties of the involved social preferences
and they are also consistent with the literature on the impact of early childhood characteristics

42The parameters of the trust game are such that at a 100% return for the first mover, the payoffs between
the two parties are equal, while the second-mover reaps all the surplus generated from the first-mover’s
transfer at a zero rate of return.

87
on later life outcomes. Verdunst et al. (2019) show, e.g., that teachers’ ratings of kindergarten
boys’ prosociality are positively associated with the boys’ earnings in adulthood after
controlling for a large set of covariates.

The findings in Dohmen et al. and Barr and Serneels are also in line with laboratory
evidence of Bartling, Fehr and Schmidt (2012) on the role of employer’s screening in
experimental labor markets. In these experiments, employers in some treatments can
condition their job offers – in terms of wages, rent-sharing, and employees’ opportunities for
effort discretion – on information about employees’ past performance levels in other firms.
The experimental employers make ample use of this information and offer completely
different compensation packages to the workers depending on their past performance.
Workers with high past effort levels – generally based on workers’ willingness to reciprocate
to generous job offers – receive generous current job packages with high wages, a high share
of the overall surplus, and broad opportunities for effort discretion. In contrast, workers with
low past performance received mediocre job packages with low wages, no rent-sharing, and
tightly controlled effort opportunities. By exogenously varying (i) the employers’ screening
opportunities and (ii) the degree of labor market competition, the authors are also able to
show the causal impact of these factors on the extent to which employers reward workers’
reciprocal behavior (i.e., their past performance). These findings suggest that, under the
realistic assumption that employers can acquire information about their employees’ effort
attitudes, positively reciprocally motivated workers are rewarded with better job packages.

4.2 The Role of Social Preferences for Contracts, Institutions and Incentives
4.2.1 The Effects on Contract Enforcement and Financial Incentives

In the past 20 years several authors illustrated the advantages and disadvantages of different
incentive schemes in the light of fairness concerns. For example, it has been shown
theoretically (e.g. Sliwka (2007)) and experimentally (e.g. Fehr and Gächter (2002); Fehr and
List (2004); Falk and Kosfeld (2006)) that explicit incentive contracts may undermine
voluntary cooperation that is based on social preferences. As a consequence, explicit incentive
contracts may be less efficient than implicit alternatives based on trust, informal bonuses or
informal sanctions.
Fehr and Gächter (1997; 1998), for example, tested the impact of trust and reciprocity on
contract enforcement in a standard two-stage gift-exchange game where principals commit to

88
pay a wage and state a desired effort level, and workers respond to the offer with an effort
choice. In an additional treatment they added a third stage in which the experimental firms can
pay to reward or punish the worker for her effort choice. There was a positive wage-effort
relation in both treatments but in the three-stage treatment the average effort level was much
higher than in the two-stage treatment. Thus, workers apparently anticipated that firms reward
high effort choices and punish low ones. Note that in the absence of social preferences we
would neither observe a positive wage-effort relation nor would we see that firms reward and
punish at the third stage because the worker-firm interactions were one-shot.
Fehr, Klein and Schmidt (2007) provided evidence suggesting that social preferences
may affect the contracts principals offer to their agents. They consider the following three types
of contracts: a trust contract in a two-stage gift exchange environment like the one described
above, an incentive contract that introduced an explicit incentive into the trust contract and a
bonus contract that introduced the option of informally rewarding agents in a third stage. The
incentive contract is based on a verification technology that enables the principal to fine the
agent in case of verified shirking. The verification technology is imperfect and the fine is
limited, implying that the highest effort level that can be implemented is positive but falls short
of the efficient effort level. The informal bonus contract contains no explicit incentives but
gives the principal the opportunity to reward agents ex-post, i. e., after effort is observed. The
bonus contract does not rely on effort verification and enforcement by third parties. Instead,
the principal promises a nonbinding, voluntary bonus payment if the agent’s effort is
satisfactory. This bonus contract is an implicit contract because third parties do not enforce the
principal’s promise.
If all actors were completely selfish, the incentive contract is the only viable contract, and
the trust and bonus contracts would be equally bad. However, in fact the incentive contract
dominates the trust contract, but the bonus contract turns out to be much more efficient than
the incentive contract. How is it possible that social preferences are not strong enough to render
the trust contract more efficient than the incentive contract but strong enough to make the bonus
contract the most efficient one? Fehr, Klein and Schmidt (2007) show that inequality aversion
preferences can explain this puzzle.

4.2.2 Social Preferences as a Behavioral Foundation for Employment Contracts

The existence of simple employment contracts that pay a fixed wage and give employers the
right to tell the employee what to do (i.e., to exert authority) is a long-standing puzzle in

89
economics. Why should the trading parties ever agree to such a seemingly inefficient
contractual arrangement? Why do they agree on a fixed wage that may prevent trade (i.e.,
employment) in certain states of the world? Why do they not continuously renegotiate the
contract terms to achieve ex-post (i.e., when conditions changed) efficient outcomes, as
suggested by Alchian and Demsetz (1972)? And if continuous and efficient ex-post
renegotiation is always possible, isn’t the characterization of the employment contract as an
authority relation thoroughly misguided? A negotiation, after all, means that task assignments
are subject to both parties’ agreement.
Hart and Moore (2008) tackle these and related questions by dropping the assumption
that “ex post trade is perfectly contractible” and that “renegotiation always leads to ex post
efficiency” (p. 3). 43 In the absence of perfect ex post contractibility we are in the world of
incomplete contracts with gift exchanges and informal relationships where social preferences
are typically playing a key role. Moreover, social preferences and their interactions with
contractual arrangements are deeply affecting ex post inefficiencies (Fehr, Gachter and
Kirchsteiger 1997; Fehr, Klein and Schmidt 2007), and thus also which contracts are most
efficient. This raises the question whether social preferences could also render employment
contracts with rigid wages more efficient compared to contracts that allow for the flexible
adjustment of wages to the prevailing state of the world?
Fehr, Hart and Zehnder (2011) indeed show experimentally that rigid contracts can be
superior to flexible contracts. In the experiment, there is ex ante uncertainty whether a good
state of the world (e.g., high output prices) or a bad state of the world prevails. In addition, the
parties’ values and costs in the different states of the world are not verifiable so that state-
contingent contracts cannot be written. The advantage of a flexible contract, that fixes only a
wage range but not the wage level, is that it allows adjusting the wage w such that trade between
an employer and an employe is also possible when output prices are low. In contrast, a contract
with wages that are rigidly fixed ex ante (i.e., before the state of the world is known) may
prevent trade in this situation. The flexible contract may, however, also have a disadvantage
because it provides scope for diverging expectations regarding the wage that will be paid ex
post. In other words, while the rigid contract pins down wage expectations ex ante and thus
avoids ex post disappointments, under flexible wages workers may feel entitled to higher ex
post wages in a good state of the world which provides scope for ex post disappointments. In

43 Hart and Moore (2008, p. 3) ask the following fundamental question: If the relevant parties can always
sit down together ex post (i.e., after the state of the world is revealed) and bargain to an efficient outcome,
why should “authority, hierarchy, delegation, or indeed anything apart from asset ownership matter”?

90
the presence of (i) non-contractible effort levels and (ii) fairness concerns (social preferences)
workers may thus shirk more under flexible contracts than under rigid contracts. Moreover, the
lower effort levels under flexible contracts may even render that contract less profitable than
the rigid one.
The experimental results confirm the above conjectures. The drawback of the rigid contract
is that it prevents trade in the bad state of the world, but this is often over-compensated by the
fact that rigid contracts elicit considerably higher effort levels in the good state of the world.
To prevent disappointments, and low effort levels, in the good state of the world employers
pay much higher ex post wages under flexible contracts but still observe a non-negligible
amount of shirking. In contrast, under rigid contracts the wage is competitively fixed at very
low levels by market competition 44, and despite these low wages low effort levels rarely occur.
Overall, this renders fixed wage contracts more profitable than flexible contracts. Note that
under selfish actors this result could not occur because the flexible contract would always
dominate the rigid one.
Fehr, Hart and Zehnder (2015) show that the trade-off between trading frequency and effort
provision that prevails under flexible contracts is also present when the parties can engage in
informal agreements or ex-post renegotiation. Brandts, Ellman and Charness (2016) provide a
robust replication of the results of Fehr, Hart and Zehnder 2011). In addition, they conduct a
treatment that allows for free-form communication (and thus continuous bargaining) between
the trading parties at any point in time. In this treatment, the flexible contract becomes more
efficient and is chosen more often than the rigid one. This treatment may be interpreted as
implementing an approximation of efficient ex-post contracting. In other words, the absence of
such efficient ex-post contracting opportunities, which is a key assumption in the Hart and
Moore (2008) approach, is indeed important for the superiority of the rigid contract. 45

4.2.3 Social Preferences, Contractual Incompleteness, and Property Rights

Many investments are relationship-specific, meaning that they are valuable only within a
particular relationship. In the presence of incomplete contracts, such investments bear the risk
of being exploited ex-post – the so-called hold-up problem. Rational parties anticipate being

44Under flexible contracts, competition only determines the lower bound on wages.
45In practice, companies have rather rigid rules about who is allowed to approach whom in the hierarchy,
and for what purpose, Thus, while easily implementable in the lab with a small number of involved parties,
continuous free-form communication between the relevant parties is typically not possible in companies.

91
held up ex post, and thus underinvest in relation-specific assets. The property rights literature
shows that the appropriate allocation of asset ownership can mitigate the underinvestment
incentive. Incomplete contracting and the associated hold-up problem have thus provided an
important economic rationale for the allocation of asset ownership to those parties that are most
vulnerable to exploitation (Grossman and Hart 1986; Hart and Moore 1990).
The key assumption behind the property rights approach is that contracts are incomplete,
which is justified by assuming that payoff-relevant information is observable to the involved
parties but not verifiable by a third-party enforcer. There are many other applications of the
“observable but not verifiable assumption” in economics and the assumption has therefore
become one of the most important cornerstones of modern institutional economics. 46 However,
all these applications of the incomplete contracting approach are subject to a fundamental
criticism that has been raised by Maskin and Tirole (1999). They show that if parties commonly
observe payoff-relevant information, one can construct an extensive form mechanism that leads
to truthful revelation of the relevant information in the unique subgame perfect equilibrium of
the game implied by the mechanism.
Thus, in principle, one could design contracts that embody this extensive form
mechanism, and if the mechanism works as predicted by theory, all commonly observable
information could be turned into truthfully reported verifiable information. This means that the
second-best institutional arrangements derived under incomplete contracting would become
superfluous because a superior contractual arrangement exists. The question, however, is
whether the above-mentioned extensive form mechanism, which is based on the work of Moore
and Repullo (1988), indeed works as predicted.
Fehr, Powell and Wilkening (2021) examine this question experimentally, and show that
negative reciprocity thoroughly undermines the functioning of Maskin-Tirole-type
mechanisms. Most parties are unwilling to enter a contract that incorporates the mechanism,
and if they enter, these contracts typically perform worse than contracts without the
mechanism. Intuitively, a key reason for the failure of the mechanisms is that they are based
on large fines for the trading parties if they “misbehave”, but the threat and execution of large
fines is also likely to induce extreme hostility (i.e., negative reciprocity) between the parties.

46 The assumption has, for example, been used to understand property rights and firm boundaries, the

optimal scope of governments, problems of privatization, the control of insiders by outsiders through
voting rights, financial contracts, and patterns of international trade and technology adoption.

92
Adding the mechanism to a usual hold-up problem is like handing out guns at a fist fight. The
guns are unlikely to make the fight more peaceful.
Thus, social preferences in the form of negative reciprocity undermine the criticism of
the theoretical foundations of incomplete contracting models. Ironically, to the extent to which
social preferences are a force that contributes to contractual incompleteness, they help sustain
their own behavioral importance. Why? Because incomplete contracts provide the terrain – gift
exchanges, informal sanctions and rewards, informal agreements – under which social
preferences can play an important role.
Overall, social preferences may contribute to the prevalence of incomplete contracting in
two ways. First, they may render institutions like the above discussed mechanisms, that render
contracts more complete, dysfunctional. Second, they may mitigate the contracting problems
that arise under incomplete contracts. One example of this is the relatively high efficiency of
incomplete bonus contracts in Fehr, Klein and Schmidt (2007), which we discussed above.
Another example is provided by the large experimental literature on behavior under the hold-
up problem (see Yang (2021) for a review). This literature shows that the underinvestment
problem is typically considerably less severe than predicted by the self-interest model (see, e.
g., Gantner, Güth and Königstein (2001); Ellingsen and Johannesson (2004a; 2004b);
Dufwenberg, Smith and van Essen (2013)). A key reason for this is that fairness concerns
induce the parties to take their ex-ante investments in the ex-post bargaining process partially
into account. This means that the investing parties experience less exploitation than predicted
under self-interest which weakens underinvestment. Negative reciprocity and disadvantageous
inequality aversion appear to be important forces in this context because they provide a
preference-based commitment to credibly reject very unfair offers. Parties who can hold-up
their counterparts thus often face the threat of complete disagreement (like in the simple
ultimatum game), and therefore they shy away from fully exploiting their ex-post bargaining
power.

4.3 The Role of Social Preferences in Politics

Individuals with non-selfish distributional preferences care not only about their own payoffs
but also about payoff distributions. Their preferences should thus also affect their desire to
re-distribute payoffs, which should affect their willingness to vote for redistributive policy
proposals. Suppose, for example, an economic environment like that in the famous model of

93
Meltzer and Richard (1981), but assume that people are inequality averse. In this
environment, individuals face the decision to vote on a proportional tax rate τ to be levied on
all individuals in the population and redistributed equally as a lump sum. An individual with
gross income 𝑦𝑦𝑖𝑖 who receives a lump sum transfer T will thus have a consumption level of
1
𝑐𝑐𝑖𝑖 = (1 − 𝜏𝜏)𝑦𝑦𝑖𝑖 + 𝑇𝑇. Assuming that there are quadratic costs of taxation of (2)𝜏𝜏 2 per tax
1
dollar, the government budget is balanced if 𝑇𝑇 = �𝜏𝜏 − (2)𝜏𝜏 2 � 𝑦𝑦�, where 𝑦𝑦� represents the

average gross income in the population. If one assumes that individuals have Fehr-Schmidt
preferences, the preferred tax rate 𝜏𝜏 ∗ is given by

where 𝛼𝛼𝑖𝑖 captures the aversion against disadvantageous inequality (“envy”) and 𝛽𝛽𝑖𝑖 the
aversion against advantageous inequality (“empathy”). The above solution for 𝜏𝜏𝑖𝑖∗ suggests
that the preferred tax rates for a selfish and an inequality averse individual may look like
those in Figure 8 below.

Figure 8: Preferred tax rate as a function of gross income and social preferences

preferred
tax rate

inequality
averse

selfish

gross income

More precisely, the model yields the following predictions and implications for empirical
research: (i) Low-income individuals obviously have a selfish reason for choosing
redistributive taxation, i.e., even in case of 𝛼𝛼𝑖𝑖 = 𝛽𝛽𝑖𝑖 = 0 they favor a high tax rate.

94
(ii) Selfish individuals (𝛼𝛼𝑖𝑖 = 𝛽𝛽𝑖𝑖 = 0) generally demand less redistribution, i.e., have a lower
preferred tax rate. However, at low incomes this non-pecuniary driver of redistribution may
be difficult to identify empirically because selfish reasons already demand a high tax rate.

(iii) A higher gross income 𝑦𝑦𝑖𝑖 will generally lower the demand for redistribution but this
effect will be mitigated for inequality averse individuals. In fact, the term that multiplies
(1/𝑦𝑦�) may become close to zero for very inequality averse individuals, implying that their
demand for redistribution does not decline with gross income.

(iv) For individuals with a relatively low income, disadvantageous inequality aversion 𝛼𝛼𝑖𝑖 is
the main non-pecuniary driver of the demand for redistribution because the social
comparisons involve many individuals who earn more than the low-income individual.

(v) For individuals with a relatively high income, 𝛽𝛽𝑖𝑖 is the main non-pecuniary driver of the
demand for redistribution because the social comparisons involve many individuals who earn
less than the high-income individual.

To what extent do individuals in laboratory experiments vote for redistribution? Sausgruber

and Tyran (2009) recruited students for a voting experiment in which purely redistributive
taxation on the basis of the majority rule was possible. Because low income recipients have a
selfish reason to vote for redistribution, they show that even only a little bit of inequality
aversion can generate a majority vote in favor of redistribution. Assuming the same
distribution of disadvantageous and advantageous inequality aversion in their subject
population that assumed in Fehr and Schmidt (1999), they show that the inequality aversion
model predicts voting outcomes quite accurately.

Durante, Putterman and van der Weele (2014) conducted perhaps the most extensive
lab study on preferences for redistribution in a tax policy setting. Undergraduate students
were assembled in groups of 21 subjects. Pre-tax incomes were calibrated to proportionally
reproduce the actual US pre-tax income distribution. Each subject made a decision on the
preferred proportional tax rate t ∈ {0.1, 0.2, …, 0.9, 1} that generated tax revenue that was
equally redistributed as a lump sum to all 21 members of the group. The final group outcome
was not determined by voting but by the decision of one randomly chosen subject from the
group. This has the advantage that every subject’s decision had the same probability of being
decisive, i.e., incentive compatibility also held for subjects with extreme preferences.

Durante et al. varied (i) the source of pre-tax income (earned or randomly assigned), (ii)
whether the preferred tax had to be chosen before or after pre-tax income is known and (iii)

95
the direct costs and deadweight losses of taxation. They find that (i) higher direct costs of
taxation and higher pre-tax income reduce the demand for redistribution, (ii) subjects more
confident about their performance in the task used to determine earned pre-tax incomes also
demand lower redistribution, (iii) subjects take deadweight losses of taxation into account
when deciding about tax rates, and (iv) most subjects are willing to pay to reduce income
inequality among others. Durante et al. also estimate the parameters of a Charness and Rabin
model and find that the weight given to increasing the income of the worst-off player in the
group is about 3 times higher than the weight given to aggregate earnings.

To what extent are individuals’ social preferences predictive of their political behaviors
outside the laboratory, i.e., their preferences for left versus right wing parties and their
demand for redistribution in the broader society? One problem that arises here is that
objective information about individuals’ voting behavior is typically not available because
individual votes cast are secret. Researchers must therefore rely on non-incentivized surveys.
However, as shown below, there are ways to validate individuals’ survey answers.

Kerschbamer and Müller (2020) measured the social preferences of a large

representative sample of the German population (the German Internet Panel, GIP, see Table
1) that also contained various questions indicating individuals’ views on redistribution spread
over several survey waves. These are questions like “Should the government mitigate income
differences?” or “Should people, who work more and consequently earn more, pay more or
less taxes than they currently do?”. Kerschbamer and Müller show that, compared to selfish
subjects, inequality averse and altruistic subjects have (i) a higher propensity to vote for left-
wing parties, (ii) self-report that they more left-leaning, and (iii) are more in favor of
redistribution as measured by the first principal component of the bundle of redistribution
questions in the GIP. These results also hold when the authors control for age, gender,
income, education, risk aversion and patience.

Yet there could be many other potential reasons why people might be for or against
redistribution such as their expected future income, their history of misfortunes (i.e.,
unemployment or negative health shocks), their (false) beliefs about the prevailing inequality,
their beliefs about their relative incomes, or their beliefs about the role of luck and effort for
economic success in life. All of these reasons have been intensely discussed and examined in
the political economy literature on redistribution, which raises the question whether
distributional preferences are also predictive of people’s demand for redistribution if one
controls for these motives. In addition, there is the question of the extent to which answers to

96
non-incentivized survey questions such as whether the government should mitigate income
differences validly capture the demand for redistribution.

Fehr, Epper and Senn (2021) tackle these problems by measuring social preferences in
a broad sample of the Swiss population that is representative in terms of age, education,
income, and gender. In addition, they elicit measures of the motives for redistribution
mentioned in the previous paragraph that allows them to control for these motives. Moreover,
they exploit the fact that many strongly redistributive referenda were put to vote under the
rules of Swiss direct democracy during the last 10-12 years. Several of these proposals such
as the “fair taxes initiative” would have involved substantially higher taxation of high
incomes if they had been implemented. Because these redistributive proposals were put to
vote in a national referendum, they were broadly discussed in the population and the media.
This has the advantage that the people have some memory about the discussed benefits and
costs of the proposals and that the survey results on people’s attitudes towards the proposals
could be validated with the actual voting results by examining the geographic and
sociodemographic distribution of votes with the distribution of survey answers. In addition,
Fehr, Epper and Senn (2021) validated the survey results with people’s actual donations to
organizations that support or oppose redistributive proposals.

Fehr, Epper and Senn document that their population is characterized by three distinct
types of social preferences – a relative majority of inequality averse individuals, a smaller but
still large group of altruistic individuals, and a relative minority of predominantly selfish
individuals (see panels a and c in Figure 2). They report the following findings: (i) At low
incomes, differences in the support for redistribution across the different preference groups is
very small. (ii) The support for redistribution declines sharply with increasing income for
selfish individuals. (iii) As depicted in Figure 8, social preferences strongly mitigate the
decline in support for redistribution. (iv) As a consequence of (ii) and (iii), inequality averse
individuals with incomes above the median have a much higher demand for redistribution
(0.57 standard deviations higher) than selfish individuals. (v) Inequality averse individuals
generally also have a higher demand for redistribution compared to altruistic individuals, but
this difference is not significant. However, altruistic individuals demand significantly more
redistribution than the selfish ones do. All these results hold when controlling for a host of
socio-demographic variables and the other motives for redistribution mentioned above. Taken
together, the results thus suggest that social preferences play an important role in the demand
for redistribution.

97
 The results discussed above do, however, not yet exhaust the role of fairness concerns
and social preferences for the demand for redistribution. The reason for this is that both the
Kerschbamer & Müller paper and the Fehr, Epper and Senn paper used a distributional
measure of social preferences that does not account for people’s concern for meritocracy. In
section 3.2 on “who are the meritocrats?”, we reported data that indicate that selfish people
show little concern for meritocracy, while altruistic and inequality averse people displayed a
relatively strong respect for meritocracy. Thus, when discussing the influence of social
preferences on redistribution, it is also necessary to acknowledge the influence of the
meritocratic dimension of social preferences.

Starting with Fong (2001), there is a sizeable literature that shows that people’s beliefs
about the role of effort and luck in economic success in life is a key factor in their demand for
redistribution (Alesina and La Ferrara 2005; Alesina and Giuliano 2011). In particular,
individuals who believe that effort (and not luck) is a primary driver for success in life often
attribute low income to a lack of effort, i.e., people with low income are considered
responsible for their situation and do not deserve help through redistributive legislation. The
combination of beliefs about the important role of effort with meritocratic concerns thus
reduces the demand for redistribution. Moreover, because selfish individuals (see section 3.2)
show few meritocratic concerns, a belief about the role of effort should predominantly inhibit
the demand for redistribution by altruistic and inequality averse individuals. This is exactly
what Fehr, Epper and Senn (2021) found. In other words, other-regarding individuals who
believe that effort is a primary driver of economic success behave as if their 𝛼𝛼𝑖𝑖 and 𝛽𝛽𝑖𝑖
parameters are smaller than those who believe that luck is a primary driver. Thus, taking the
meritocratic dimension into account leads to a more nuanced view about the role of social
preferences in redistributive politics. The widespread existence of meritocratic other-
regarding preferences has been documented not only in the laboratory but also in several
survey experiments conducted with general population samples (Almas, Cappelen and
Tungodden 2020; Cappelen et al. 2022), and can, in particular, also explain the popularity of
workfare programs (Fong, Bowles and Gintis (2005); Drenik & Perez-Truglia (2018)).
Survey experiments have also documented how social preferences and demand for
redistribution in general, and meritocratic preferences in particular, are sensitive to the
(mis)perceptions a person holds about society, and her own place (e.g., rank) in it (Cruces,
Perez-Truglia and Tetaz 2013; Karadja, Mollerstrom and Seim 2017; Fehr, Mollerstrom and
Perez-Truglia 2022).

98
 5. Summary and Outlook

Over the previous two to three decades a lot has been learned about the properties, the
prevalence and the consequences of social preferences but there is also still a lot that needs to
be learned. In the following, we outline several open questions that offer exciting research
opportunities.

Perhaps, the most fundamental question is related to the determinants of social

preferences. There is already initial evidence suggesting that they are formed in childhood
through different role models or different early childhood education practices (Van Lange et
al. 1997; Cappelen et al. 2020; Kosse et al. 2020) but the set of societal determinants is
probably much larger. For example, is it possible for companies and other organizations to
shape the social preferences of their employees by structuring rewards, incentives, and the
overall company culture in different ways? What is the effect of detrimental health and
income shocks on social preference? How does the break-up of marriages or other events that
disrupt or improve the relation between family members or members of a community affect
social preferences? A recent paper (Cassar et al. 2022) suggests, for example, that
allomaternal care increases prosocial preferences in a community. And how do a society’s
governance institutions shape individuals’ social preferences.

Very little is known about these determinants although Rustagi (2022) recently has
made some advances in understanding how a history of self-government and democratic
interactions tends to favor preferences for cooperation. He exploits a natural experiment in
Switzerland, where during the middle-ages, the absence of an heir resulted in the extinction
of a prominent noble dynasty, which enabled some Swiss municipalities to become self-
governing whereas others remained under feudalism for another 600 years. Rustagi shows
that individuals from self-governing communities display stronger preferences for conditional
cooperation (measured in a behavioral experiment) as well as higher voter turnout and higher
charitable donations. 47 These findings are also consistent with those of Guiso, Sapienza and
Zingales (2016) who show that Northern Italian cities that experienced a period of
independence in the Middle Ages have significantly higher prosocial behaviors in terms of

47
Because Switzerland tracks every family’s place of origin in registration data, Rustagi can identify the “cultural
origin” of individuals and document the persistence of cultural transmission at the individual level in a context of
historically low migration rates.

99
organ/blood donations, the frequency of cheating in national exams taken by children in each
Italian town and the number of non-profit organizations.

When discussing the potential determinants of social preferences, the relationship

between intrinsic social preferences and social norms may also become important. We define
a social norm as a commonly known standard of behavior that is based on a widely shared
view how individual group members ought to behave in a given situation. Thus, in contrast to
preferences, which are a property of individuals, social norms are a property of whole groups
of people. They constitute an external normative constraint on individuals’ behavior that
arises from the fact that the normative standard is widely shared and deviations from the
standard are met with disapproval, ridicule, and other forms of sanctioning. However, over
time external normative constraints may be internalized which turns them into preferences
but very little is known conceptually and empirically about these internalization processes,
the factors that shape them, and ways to model them (although see Enke (2019); Schulz et al.
(2019); Ellingsen and Mohlin (2022)). In addition, there is very little empirical research that
simultaneously elicits and measures social norm driven and social preference driven
behaviors (for an exception see Carpenter and Robbett (2022)).

Another important unresolved question concerns the determinants of individuals

reference points for their fairness and equity judgements. In the absence of reliable empirical
knowledge, models like those of Fehr and Schmidt (1999) have pragmatically assumed that
(at least in experiments) equality between the involved parties is a good first-order
approximation of individuals actual reference point. As the section on the role of merit, luck
and risk in social preferences has made clear, however, there are many situations in which
equality may be the wrong reference point for many individuals. It is therefore important to
develop methods that enable the reliable empirical identification of individuals’ reference
agents and reference outcomes. An interesting step in this direction has recently been
undertaken by Hvidberg, Thustrup-Kreiner and Stantcheva (Forthcoming) and Xu et al.
(2023). 48

Finally, as already pointed out at the beginning of the section on economic

consequences, it would be desirable to study the deeper implications of heterogenous social

48There exists also an older literature in labor economics that discussed reference points such as one’s own past
wages, peer wages in the company, the company’s ability to pay, workers’ perceived contributions, etc. as
potential reference points (e.g., Levine (1993)). It is, however, probably fair to say that no firm conclusions have
been reached by this literature.

100
preferences for normative (public) economics. What are optimal institutions, incentives and
tax-transfer schedules when the population is characterized by heterogenous social
preferences or if individuals care for equality of opportunity? To answer these questions it is
important to move beyond measuring social preferences in bilateral settings and consider how
these preferences change when there are many recipients. An interesting step in this direction
has recently been made by Charite, Fisman and Kuziemko (2021) who show that people
exclusively care of the very poor (positively), the very rich (negatively) and their local
“income neighbors” directly above them (negatively). Overall, the role of social preference
research for normative and positive public economics may be substantial and, perhaps,
change what economists recommend to policy makers.

101
 References
Adams, J. S. "Inequity in Social-Exchange." Advances in Experimental Social Psychology 2,
no. 4 (1965): 267-99.
———. "Toward an Understanding of Inequity." Journal of Abnormal Psychology 67, no. 5
(1963): 422-&.
Agell, J., and P. Lundborg. "Survey Evidence on Wage Rigidity and Unemployment: Sweden
in the 1990s." Scandinavian Journal of Economics 105, no. 1 (2003): 15-29.
———. "Theories of Pay and Unemployment - Survey Evidence from Swedish
Manufacturing Firms." Scandinavian Journal of Economics 97, no. 2 (1995): 295-
307.
Akerlof, G. A., and Y. L. Yellen. "The Fair Wage-Effort Hypothesis and Unemployment."
Quarterly Journal of Economics 105, no. 2 (1990): 255-83.
Alchian, A. A., and H. Demsetz. "Production, Information Costs, and Economic
Organization." American Economic Review 62, no. 5 (1972): 777-95.
Alesina, A., and P. Giuliano. "Preferences for Redistribution." In Handbook of Social
Economics, edited by J. Benhabib, A. Bisin and M.O. Kackson, 93-131: Elsevier,
2011.
Alesina, A., and E. La Ferrara. "Preferences for Redistribution in the Land of Opportunities."
Journal of Public Economics 89, no. 5-6 (2005): 897-931.
Alger, I., and J. W. Weibull. "Evolution and Kantian Morality." Games and Economic
Behavior 98 (2016): 56-67.
Almas, I., A. W. Cappelen, E. O. Sorensen, and B. Tungodden. "Attitudes to Inequality:
Preferences and Beliefs." In IFS Deaton Review of Inequalities: Institute for Fiscal
Studies, 2022.
———. "Fairness across the World: Preferences and Beliefs." Federal Reserve Bank of New
York, 2021.
Almas, I., A. W. Cappelen, and B. Tungodden. "Cutthroat Capitalism Versus Cuddly
Socialism: Are Americans More Meritocratic and Efficiency-Seeking Than
Scandinavians?" Journal of Political Economy 128, no. 5 (2020): 1753-88.
Anderson, L.R., J.M. Mellor, and J. Milyo. "Inequality and Public Good Provision: An
Experimental Analysis." The Journal of Socio-Economics 37, no. 3 (2008): 1010-28.
Andre, P. "Shallow Meritocracy." University of Bonn, 2022.
Andreoni, J. "Why Free Ride - Strategies and Learning in Public-Goods Experiments."
Journal of Public Economics 37, no. 3 (1988): 291-304.
Andreoni, J., and B. D. Bernheim. "Social Image and the 50-50 Norm: A Theoretical and
Experimental Analysis of Audience Effects." Econometrica 77, no. 5 (2009): 1607-
36.
Andreoni, J., and J. Miller. "Giving According to Garp: An Experimental Test of the
Consistency of Preferences for Altruism." Econometrica 70, no. 2 (2002): 737-53.
Andreoni, J., and J. M. Rao. "The Power of Asking: How Communication Affects
Selfishness, Empathy, and Altruism." Journal of Public Economics 95, no. 7-8
(2011): 513-20.
Ariely, D., A. Bracha, and S. Meier. "Doing Good or Doing Well? Image Motivation and
Monetary Incentives in Behaving Prosocially." American Economic Review 99, no. 1
(2009): 544-55.
Aronsson, T., and O. Johansson-Stenman. "Inequality Aversion, Externalities, and Pareto-
Efficient Income Taxation." University of Gothenburg, 2020a.
———. "Optimal Second-Best Taxation When Individuals Have Social Preferences." Umea
School of Business: Department of Economics, 2020b.
Attanasi, G., C. Rimbaud, and M. C. Villeval. "Embezzlement and Guilt Aversion." Journal
of Economic Behavior & Organization 167 (2019): 409-29.
1
Balafoutas, L., R. Kerschbamer, M. Kocher, and M. Sutter. "Revealed Distributional
Preferences: Individuals Vs. Team." Journal of Economic Behavior & Organization
108 (2014): 319-30.
Balafoutas, L., R. Kerschbamer, and M. Sutter. "Distributional Preferences and Competitive
Behavior." Journal of Economic Behavior & Organization 83, no. 1 (2012): 125-35.
Balliet, D., L. B. Mulder, and P. A. M. Van Lange. "Reward, Punishment, and Cooperation:
A Meta-Analysis." Psychological Bulletin 137, no. 4 (2011): 594-615.
Barr, A., and P. Serneels. "Reciprocity in the Workplace." Experimental Economics 12
(2009): 99-112.
Barrett, H. C., A. Bolyanatz, A. N. Crittenden, D. M. T. Fessler, S. Fitzpatrick, M. Gurven, J.
Henrich, M. Kanovsky, G. Kushnick, A. Pisor, B. A. Scelza, S. Stich, C. von Rueden,
W. Y. Zhao, and S. Laurence. "Small-Scale Societies Exhibit Fundamental Variation
in the Role of Intentions in Moral Judgment." Proceedings of the National Academy
of Sciences of the United States of America 113, no. 17 (2016): 4688-93.
Bartling, B., E. Fehr, and K. M. Schmidt. "Screening, Competition, and Job Design:
Economic Origins of Good Jobs." American Economic Review 102, no. 2 (2012): 834-
64.
Bartling, B., and U. Fischbacher. "Shifting the Blame: On Delegation and Responsibility."
Review of Economic Studies 79, no. 1 (2012): 67-87.
Battigalli, P., and M. Dufwenberg. "Belief-Dependent Motivations and Psychological Game
Theory." Journal of Economic Literature 60, no. 3 (2022): 833-82.
———. "Dynamic Psychological Games." Journal of Economic Theory 144, no. 1 (2009): 1-
35.
———. "Guilt in Games." American Economic Review 97, no. 2 (2007): 170-76.
Baumeister, R. F., A. M. Stillwell, and T. F. Heatherton. "Guilt - an Interpersonal Approach."
Psychological Bulletin 115, no. 2 (1994): 243-67.
———. "Personal Narratives About Guilt - Role in Action Control and Interpersonal
Relationships." Basic and Applied Social Psychology 17, no. 1-2 (1995): 173-98.
Becker, A. "Shamed to Death: Social Image Concerns and War Participation." University
College London, 2022.
Bellemare, C., S. Kroeger, and A. Van Soest. "Measuring Inequity Aversion in a
Heterogeneous Population Using Experimental Decisions and Subjective
Probabilities." Econometrica 76, no. 4 (2008): 815-39.
Bellemare, C., S. Kroger, and A. van Soest. "Preferences, Intentions, and Expectation
Violations: A Large-Scale Experiment with a Representative Subject Pool." Journal
of Economic Behavior & Organization 78, no. 3 (2011): 349-65.
Bellemare, C., A. Sebald, and M. Strobel. "Measuring the Willingness to Pay to Avoid Guilt:
Estimation Using Equilibrium and Stated Belief Models." Journal of Applied
Econometrics 26, no. 3 (2011): 437-53.
Bellemare, C., A. Sebald, and S. Suetens. "Guilt Aversion in Economics and Psychology."
Journal of Economic Psychology 73 (2019): 52-59.
———. "Heterogeneous Guilt Sensitivities and Incentive Effects." Experimental Economics
21, no. 2 (2018): 316-36.
———. "A Note on Testing Guilt Aversion." Games and Economic Behavior 102 (2017):
233-39.
Bellemare, C., and B. Shearer. "Gift Giving and Worker Productivity: Evidence from a Firm-
Level Experiment." Games and Economic Behavior 67, no. 1 (2009): 233-44.
Bem, D. "Self-Perception Theory." Advances in Experimental Social Psychology 6, no. 1
(1973): 1-62.
Benabou, R., and J. Tirole. "Identity, Morals, and Taboos: Beliefs as Assets." Quarterly
Journal of Economics 126, no. 2 (2011): 805-55.

2
———. "Incentives and Prosocial Behavior." American Economic Review 96, no. 5 (2006):
1652-78.
Benz, M., and S. Meier. "Do People Behave in Experiments as in the Field? Evidence from
Donations." Experimental Economics 11, no. 3 (2008): 268-81.
Berg, J, J Dickhaut, and K McCabe. "Trust, Reciprocity and Social History." Games and
Economic Behavior 10, no. 1 (1995): 122-42.
Bewley, T. F. "A Depressed Labor-Market as Explained by Participants." American
Economic Review 85, no. 2 (1995): 250-54.
———. "Fairness, Reciprocity, and Wage Rigidity." SSRN, 2002.
———. "Why Not Cut Pay?" European Economic Review 42, no. 3-5 (1998): 459-90.
Bewley, T.F. . Why Wages Don't Fall During a Recession. Cambridge MA: Harvard
University Press, 1999.
Bhattacharya, P., and J. Mollerstrom. "Lucky to Work." Unpublished Working Paper (2022).
Blinder, A. S., and D. H. Choi. "A Shred of Evidence on Theories of Wage Stickiness."
Quarterly Journal of Economics 105, no. 4 (1990): 1003-15.
Blount, S. "When Social Outcomes Arent Fair - the Effect of Causal Attributions on
Preferences." Organizational Behavior and Human Decision Processes 63, no. 2
(1995): 131-44.
Bolton, G. E., and A. Ockenfels. "Erc: A Theory of Equity, Reciprocity, and Competition."
American Economic Review 90, no. 1 (2000): 166-93.
———. "Inequality Aversion, Efficiency, and Maximin Preferences in Simple Distribution
Experiments: Comment." American Economic Review 96, no. 5 (2006): 1906-11.
Bolton, G.E, J. Brandts, and A. Ockenfels. "Measuring Motivations for the Reciprocal
Responses Observed in a Simple Dilemma Game." Experimental Economics 1 (1998):
207-19.
Bracht, J., and T. Regner. "Moral Emotions and Partnership." Journal of Economic
Psychology 39, no. C (2013): 313-26.
Brandts, J., and G. Charness. "Do Labour Market Conditions Affect Gift Exchange? Some
Experimental Evidence." Economic Journal 114, no. 497 (2004): 684-708.
———. "Truth or Consequences: An Experiment." Management Science 49, no. 1 (2003):
116-30.
Brandts, J., M. Ellman, and G. Charness. "Let's Talk: How Communication Affects Contract
Design." Journal of the European Economic Association 14, no. 4 (2016): 943-74.
Brandts, J., and C. Sola. "Reference Points and Negative Reciprocity in Simple Sequential
Games." Games and Economic Behavior 36, no. 2 (2001): 138-57.
Breza, E., S. Kaur, and N. Krishnaswamy. "Scabs: The Social Suppression of Labor Supply."
NBER Working Paper 25880 (2019).
Breza, E., S. Kaur, and Y. Shamdasani. "Labor Rationing." American Economic Review 111,
no. 10 (2021): 3184-224.
———. "The Morale Effects of Pay Inequality." Quarterly Journal of Economics 133, no. 2
(2018): 611-63.
Brown, M., A. Falk, and E. Fehr. "Relational Contracts and the Nature of Market
Interactions." Econometrica 72, no. 3 (2004): 747-80.
Buurman, M., J. Delfgaauw, R. Dur, and S. Van den Bossche. "Public Sector Employees:
Risk Averse and Altruistic?" Journal of Economic Behavior & Organization 83, no. 3
(2012): 279-91.
Cabeza, B. "Desert Concerns and Distributional Preferences." 2021.
Camerer, C.F. "The Promise and Success of Lab–Field Generalizability in Experimental
Economics: A Critical Reply to Levitt and List." In Handbook of Experimental
Economic Methodology, edited by G. Frechette and A. Schotter, 249-95: Oxford
Academic, 2015.

3
Camerer, Colin F. Behavioral Game Theory - Experiments in Strategic Interaction.
Princeton, New Jersey: Princeton University Press, 2003.
Cappelen, A., J. List, A. Samek, and B. Tungodden. "The Effect of Early-Childhood
Education on Social Preferences." Journal of Political Economy 128, no. 7 (2020):
2739-58.
Cappelen, A. W., A. D. Hole, E. O. Sorensen, and B. Tungodden. "The Pluralism of Fairness
Ideals: An Experimental Approach." American Economic Review 97, no. 3 (2007):
818-27.
Cappelen, A. W., K. O. Moene, E. O. Sorensen, and B. Tungodden. "Needs Versus
Entitlementsan International Fairness Experiment." Journal of the European
Economic Association 11, no. 3 (2013): 574-98.
Cappelen, A. W., J. Mollerstrom, B. A. Reme, and B. Tungodden. "A Meritocratic Origin of
Egalitarian Behaviour." Economic Journal 132, no. 646 (2022): 2101-17.
Cappelen, A. W., K. Nygaard, E. O. Sorensen, and B. Tungodden. "Social Preferences in the
Lab: A Comparison of Students and a Representative Population." Scandinavian
Journal of Economics 117, no. 4 (2015): 1306-26.
Card, D., A. Mas, E. Moretti, and E. Saez. "Inequality at Work: The Effect of Peer Salaries
on Job Satisfaction." American Economic Review 102, no. 6 (2012): 2981-3003.
Carlson, R. W., M. A. Marechal, B. Oud, E. Fehr, and M. J. Crockett. "Motivated
Misremembering of Selfish Decisions." Nature Communications 11, no. 1 (2020).
Carpenter, J., and A Robbett. "Measuring Socially Appropriate Social Preferences." IZA
Working Paper No. 15590 (2022).
Carpenter, J., and E. Seki. "Do Social Preferences Increase Productivity? Field Experimental
Evidence from Fishermen in Toyama Bay." Economic Inquiry 49, no. 2 (2011): 612-
30.
Cartwright, E. "A Survey of Belief-Based Guilt Aversion in Trust and Dictator Games."
Journal of Economic Behavior & Organization 167 (2019): 430-44.
Cassar, A., A. Cristia, P. Grosjean, and S. Walker. "It Makes a Village: Allomaternal Care
and Prosociality." Working Paper, Department of Economics, UNSW Sidney (2022).
Chan, K. S., S. Mestelman, R. Moir, and R. A. Muller. "Heterogeneity and the Voluntary
Provision of Public Goods." Experimental Economics 2, no. 1 (1999): 5-30.
———. "The Voluntary Provision of Public Goods under Varying Income Distributions."
Canadian Journal of Economics-Revue Canadienne D Economique 29, no. 1 (1996):
54-69.
Chapman, J., M. Dean, Ortoleva P., E. Snowberg, and C. F. Camerer. "Econographics."
NBER Working Paper 24931 (2018).
Charness, G. "Attribution and Reciprocity in a Simulated Labor Market: An Experimental
Investigation." Barcelona: Universitat Pompeu Fabra, 1996.
———. "Attribution and Reciprocity in an Experimental Labor Market." Journal of Labor
Economics 22, no. 3 (2004): 665-88.
———. "Parochial Social Preferences " Working Paper, Department of Economics, UC
Santa Barbara (2023).
———. "Responsibility and Effort in an Experimental Labor Market." Journal of Economic
Behavior & Organization 42, no. 3 (2000): 375-84.
Charness, G., and M. Dufwenberg. "Bare Promises: An Experiment." Economics Letters 107,
no. 2 (2010): 281-83.
———. "Promises and Partnership." Econometrica 74, no. 6 (2006): 1579-601.
Charness, G., and M. O. Jackson. "The Role of Responsibility in Strategic Risk-Taking."
Journal of Economic Behavior & Organization 69, no. 3 (2009): 241-47.
Charness, G., and P. Kuhn. "Does Pay Inequality Affect Worker Effort? Experimental
Evidence." Journal of Labor Economics 25, no. 4 (2007): 693-723.

4
Charness, G., and D. I. Levine. "Intention and Stochastic Outcomes: An Experimental
Study." Economic Journal 117, no. 522 (2007): 1051-72.
Charness, G., and M. Rabin. "Understanding Social Preferences with Simple Tests."
Quarterly Journal of Economics 117, no. 3 (2002): 817-69.
Cherry, T. L., P. Frykblom, and J. F. Shogren. "Hardnose the Dictator." American Economic
Review 92, no. 4 (2002): 1218-21.
Cohn, A., E. Fehr, and L. Goette. "Fair Wages and Effort Provision: Combining Evidence
from a Choice Experiment and a Field Experiment." Management Science 61, no. 8
(2015): 1777-94.
Cohn, A., E. Fehr, B. Herrmann, and F. Schneider. "Social Comparison and Effort Provision:
Evidence from a Field Experiment." Journal of the European Economic Association
12, no. 4 (2014): 877-98.
Cooper, D. J., and J. H. Kagel. "Other-Regarding Preferences: A Selective Survey of
Experimental Results." In The Handbook of Experimental Economics, edited by J. H.
Kagel and A. E. Roth, 217 - 89. Princeton, New Jersey: Princeton University Press,
2016.
Coviello, D., E. Deserranno, and N. Persico. "Counterproductive Worker Behavior after a
Pay Cut." Journal of the European Economic Association 20, no. 1 (2022): 222-63.
Cox, J. C., D. Friedman, and S. Gjerstad. "A Tractable Model of Reciprocity and Fairness."
Games and Economic Behavior 59, no. 1 (2007): 17-45.
Cruces, G., R. Perez-Truglia, and M. Tetaz. "Biased Perceptions of Income Distribution and
Preferences for Redistribution: Evidence from a Survey Experiment." Journal of
Public Economics 98 (2013): 100-12.
Cullen, Z., and R. Perez-Truglia. "How Much Does Your Boss Make? The Effects of Salary
Comparisons." Journal of Political Economy (2022).
Curtin, C. M., H. C. Barrett, A. Bolyanatz, A. N. Crittenden, D. M. T. Fessler, S. Fitzpatrick,
M. Gurven, M. Kanovsky, G. Kushnick, S. Laurence, A. Pisor, B. Scelza, S. Stich, C.
von Rueden, and J. Henrich. "Kinship Intensity and the Use of Mental States in Moral
Judgment across Societies." Evolution and Human Behavior 41, no. 5 (2020): 415-29.
D'Ambrosio, C., A. E. Clark, and M. Barazzetta. "Unfairness at Work: Well-Being and
Quits." Labour Economics 51 (2018): 307-16.
Dana, J., D. M. Cain, and R. M. Dawes. "What You Don't Know Won't Hurt Me: Costly (but
Quiet) Exit in Dictator Games." Organizational Behavior and Human Decision
Processes 100, no. 2 (2006): 193-201.
Dana, J., R. A. Weber, and J. X. Kuang. "Exploiting Moral Wiggle Room: Experiments
Demonstrating an Illusory Preference for Fairness." Economic Theory 33, no. 1
(2007): 67-80.
Dawes, R. M. "Social Dilemmas." Annual Review of Psychology 31 (1980): 169-93.
Dawes, R. M., J. Mctavish, and H. Shaklee. "Behavior, Communication, and Assumptions
About Other Peoples Behavior in a Commons Dilemma Situation." Journal of
Personality and Social Psychology 35, no. 1 (1977): 1-11.
DellaVigna, S., J. A. List, and U. Malmendier. "Testing for Altruism and Social Pressure in
Charitable Giving." Quarterly Journal of Economics 127, no. 1 (2012): 1-56.
Dhaene, G., and J. Bouckaert. "Sequential Reciprocity in Two-Player, Two-Stage Games: An
Experimental Analysis." Games and Economic Behavior 70, no. 2 (2010): 289-303.
Dickens, W. T., L. Goette, E. L. Groshen, S. Holden, J. Messina, M. E. Schweitzer, J.
Turunen, and M. E. Ward. "How Wages Change: Micro Evidence from the
International Wage Flexibility Project." Journal of Economic Perspectives 21, no. 2
(2007): 195-214.
Dohmen, T., A. Falk, D. Huffman, and U. Sunde. "Homo Reciprocans: Survey Evidence on
Behavioural Outcomes." Economic Journal 119, no. 536 (2009): 592-612.

5
Dong, L., L. Huang, and J. W. Lien. "They Never Had a Chance: Unequal Opportunities and
Fair Redistribution." Unpublished Working Paper (2022).
Drenik, A., and R. Perez-Truglia. "Sympathy for the Diligent and the Demand for Workfare."
Journal of Economic Behavior & Organization 153 (2018): 77-102.
Dube, A., L. Giuliano, and J. Leonard. "Fairness and Frictions: The Impact of Unequal Raises
on Quit Behavior." American Economic Review 109, no. 2 (2019): 620-63.
Dufwenberg, M. "Marital Investments, Time Consistency and Emotions." Journal of
Economic Behavior & Organization 48, no. 1 (2002): 57-69.
Dufwenberg, M., and U. Gneezy. "Measuring Beliefs in an Experimental Lost Wallet Game."
Games and Economic Behavior 30, no. 2 (2000): 163-82.
Dufwenberg, M., and G. Kirchsteiger. "A Theory of Sequential Reciprocity." Games and
Economic Behavior 47, no. 2 (2004): 268-98.
Dufwenberg, M., A. Smith, and M. Van Essen. "Hold-Up: With a Vengeance." Economic
Inquiry 51, no. 1 (2013): 896-908.
Dur, R., and R. Zoutenbier. "Working for a Good Cause." Public Administration Review 74,
no. 2 (2014): 144-55.
Durante, R., L. Putterman, and J. van der Weele. "Preferences for Redistribution and
Perception of Fairness: An Experimental Study." Journal of the European Economic
Association 12, no. 4 (2014): 1059-86.
Eden, M., and P. G. Piacquadio. "The Normative Content of Other-Regarding Preferences."
Working Paper, Department of Economics, Brandeis University (2023).
Ederer, F., and A. Stremitzer. "Promises and Expectations." Games and Economic Behavior
106 (2017): 161-78.
Ellingsen, T., and M. Johannesson. "Is There a Hold-up Problem?" Scandinavian Journal of
Economics 106, no. 3 (2004a): 475-94.
———. "Paying Respect." Journal of Economic Perspectives 21, no. 4 (2007): 135-49.
———. "Pride and Prejudice: The Human Side of Incentive Theory." American Economic
Review 98, no. 3 (2008): 990-1008.
———. "Promises, Threats and Fairness." Economic Journal 114, no. 495 (2004b): 397-420.
Ellingsen, T., M. Johannesson, S. Tjotta, and G. Torsvik. "Testing Guilt Aversion." Games
and Economic Behavior 68, no. 1 (2010): 95-107.
Ellingsen, T., and E. Mohlin. "A Model of Social Duties." Working Paper, Department of
Economics, Stockholm School of Economics (2022).
Engelmann, D., and M. Strobel. "Inequality Aversion, Efficiency, and Maximin Preferences
in Simple Distribution Experiments." American Economic Review 94, no. 4 (2004):
857-69.
Enke, B. "Kinship, Cooperation, and the Evolution of Moral Systems." Quarterly Journal of
Economics 134, no. 2 (2019): 953-1019.
Epper, T., E. Fehr, H. Fehr-Duda, C. T. Kreiner, D. D. Lassen, S. Leth-Petersen, and G. N.
Rasmussen. "Time Discounting and Wealth Inequality." American Economic Review
110, no. 4 (2020): 1177-205.
Epper, T., E. Fehr, and J. Senn. "Who Are the Meritocrats?" Working Paper, Department of
Economics, University of Zurich (2023).
Fahr, R., and B. Irlenbusch. "Fairness as a Constraint on Trust in Reciprocity: Earned
Property Rights in a Reciprocal Exchange Experiment." Economics Letters 66, no. 3
(2000): 275-82.
Fajnzylber, P., D. Lederman, and N. Loayza. "Inequality and Violent Crime." Journal of Law
& Economics 45, no. 1 (2002): 1-40.
Falk, A., E. Fehr, and U. Fischbacher. "On the Nature of Fair Behavior." Economic Inquiry
41, no. 1 (2003): 20-26.
———. "Testing Theories of Fairness - Intentions Matter." Games and Economic Behavior
62, no. 1 (2008): 287-303.

6
Falk, A., E. Fehr, and C. Zehnder. "Fairness Perceptions and Reservation Wages - the
Behavioral Effects of Minimum Wage Laws." Quarterly Journal of Economics 121,
no. 4 (2006): 1347-81.
Falk, A., and U. Fischbacher. "A Theory of Reciprocity." Games and Economic Behavior 54,
no. 2 (2006): 293-315.
Falk, A., and J. J. Heckman. "Lab Experiments Are a Major Source of Knowledge in the
Social Sciences." Science 326, no. 5952 (2009): 535-38.
Falk, A., and M. Kosfeld. "The Hidden Costs of Control." American Economic Review 96,
no. 5 (2006): 1611-30.
Fehr, D., J. Mollerstrom, and R. Perez-Truglia. "Your Place in the World: Relative Income
and Global Inequality." American Economic Journal-Economic Policy 14, no. 4
(2022): 232-68.
Fehr, E. "The External Validity of Social Preference Measures." Working Paper, Unviversity
of Zurich (2023).
Fehr, E., T. Epper, and J. Senn. "The Fundamental Properties, Stability and Predictive Ability
of Social Preferences." Working Paper, Department of Economics, University of
Zurich (2022).
———. "Other-Regarding Preferences and Redistributive Politics." Working Paper,
Department of Economics, University of Zurich (2021).
Fehr, E., and A. Falk. "Wage Rigidity in a Competitive Incomplete Contract Market."
Journal of Political Economy 107, no. 1 (1999): 106-34.
Fehr, E., and U. Fischbacher. "Third-Party Punishment and Social Norms." Evolution and
Human Behavior 25, no. 2 (2004): 63-87.
Fehr, E., and S. Gachter. "Cooperation and Punishment in Public Goods Experiments."
American Economic Review 90, no. 4 (2000): 980-94.
———. "Do Incentive Contracts Undermine Voluntary Cooperation7." University of Zurich,
2002.
———. "How Effective Are Trust- and Reciprocity-Based Incentives?" In Economics,
Values, and Organization, edited by A. Ben-Ner and L. Putterman, 337-63.
Cambridge: Cambridge University Press, 1998.
Fehr, E., S. Gachter, and G. Kirchsteiger. "Reciprocity as a Contract Enforcement Device:
Experimental Evidence." Econometrica 65, no. 4 (1997): 833-60.
Fehr, E., and L. Goette. "Robustness and Real Consequences of Nominal Wage Rigidity."
Journal of Monetary Economics 52, no. 4 (2005): 779-804.
Fehr, E., O. Hart, and C. Zehnder. "Contracts as Reference Points-Experimental Evidence."
American Economic Review 101, no. 2 (2011): 493-525.
———. "How Do Informal Agreements and Revision Shape Contractual Reference Points?"
Journal of the European Economic Association 13, no. 1 (2015): 1-28.
Fehr, E., G. Kirchsteiger, and A. Riedl. "Does Fairness Prevent Market Clearing - an
Experimental Investigation." Quarterly Journal of Economics 108, no. 2 (1993): 437-
59.
Fehr, E., A. Klein, and K. M. Schmidt. "Fairness and Contract Design." Econometrica 75, no.
1 (2007): 121-54.
Fehr, E., and J. A. List. "The Hidden Costs and Returns of Incentives: Trust and
Trustworthiness among Ceos." Journal of the European Economic Association 2, no.
5 (2004).
Fehr, E., M. Naef, and K. M. Schmidt. "Inequality Aversion, Efficiency, and Maximin
Preferences in Simple Distribution Experiments: Comment." American Economic
Review 96, no. 5 (2006): 1912-17.
Fehr, E., M. Powell, and T. Wilkening. "Behavioral Constraints on the Design of Subgame-
Perfect Implementation Mechanisms." American Economic Review 111, no. 4 (2021):
1055-91.

7
Fehr, E., and K. M. Schmidt. The Economics of Fairness, Reciprocity and Altruism -
Experimental Evidence and New Theories. Amsterdam - Boston: Elsevier, 2006.
———. "A Theory of Fairness, Competition, and Cooperation." Quarterly Journal of
Economics 114, no. 3 (1999): 817-68.
Fehr, E., and I. Schurtenberger. "Normative Foundations of Human Cooperation." Nature
Human Behaviour 2, no. 7 (2018): 458-68.
Fershtman, C., and U. Gneezy. "Strategic Delegation: An Experiment." Rand Journal of
Economics 32, no. 2 (2001): 352-68.
Finan, F., and L. Schechter. "Vote-Buying and Reciprocity." Econometrica 80, no. 2 (2012):
863-81.
Fischbacher, U., C. M. Fong, and E. Fehr. "Fairness, Errors and the Power of Competition."
Journal of Economic Behavior & Organization 72, no. 1 (2009): 527-45.
Fisman, R., P. Jakiela, S. Kariv, and D. Markovits. "The Distributional Preferences of an
Elite." Science 349, no. 6254 (2015).
Fisman, R., S. Kariv, and D. Markovits. "Individual Preferences for Giving." American
Economic Review 97, no. 5 (2007): 1858-76.
Fisman, R., I. Kuziemko, and S. Vannutelli. "Distributional Preferences in Larger Groups:
Keeping up with the Joneses and Keeping Track of the Tails." Journal of the
European Economic Association 19, no. 2 (2021): 1407-38.
Fong, C. "Social Preferences, Self-Interest, and the Demand for Redistribution." Journal of
Public Economics 82, no. 2 (2001): 225-46.
Fong, C.M., S. Bowles, and H. Gintis. "Behavioural Motives for Income Redistribution." The
Australian Economic Review 38, no. 3 (2005): 285-97.
Forsythe, R., J. L. Horowitz, N. E. Savin, and M. Sefton. "Fairness in Simple Bargaining
Experiments." Games and Economic Behavior 6, no. 3 (1994): 347-69.
Frohlich, N., J. Oppenheimer, and A. Kurki. "Modeling Other-Regarding Preferences and an
Experimental Test." Public Choice 119, no. 1-2 (2004): 91-117.
Gächter, S., D. Nosenzo, and M. Sefton. "The Impact of Social Comparisons on Reciprocity."
Scandinavian Journal of Economics 114, no. 4 (2012): 1346-67.
———. "Peer Effects in Pro-Social Behavior: Social Norms or Social Preferences?" Journal
of the European Economic Association 11, no. 3 (2013): 548-73.
Gachter, S., and C. Thoni. "Social Comparison and Performance: Experimental Evidence on
the Fair Wage-Effort Hypothesis." Journal of Economic Behavior & Organization 76,
no. 3 (2010): 531-43.
———. "Social Learning and Voluntary Cooperation among Like-Minded People." Journal
of the European Economic Association 3, no. 2-3 (2005): 303-14.
Gantner, A., W. Guth, and M. Konigstein. "Equitable Choices in Bargaining Games with
Joint Production." Journal of Economic Behavior & Organization 46, no. 2 (2001):
209-25.
Gastorf, J. W., and J. Suls. "Performance Evaluation Via Social-Comparison - Performance
Similarity Versus Related-Attribute Similarity." Social Psychology 41, no. 4 (1978):
297-305.
Geanakoplos, J., D. Pierce, and E. Stachetti. "Psychological Games and Sequential
Rationality." Games and Economic Behavior 1, no. 1 (1989): 60-79.
Gneezy, U., and J. A. List. "Putting Behavioral Economics to Work: Testing for Gift
Exchange in Labor Markets Using Field Experiments." Econometrica 74, no. 5
(2006): 1365-84.
Greenberg, J. "Employee Theft as a Reaction to Underpayment Inequity - the Hidden Cost of
Pay Cuts." Journal of Applied Psychology 75, no. 5 (1990): 561-68.
Gregg, P., P. A. Grout, A. Ratcliffe, S. Smith, and F. Windmeijer. "How Important Is Pro-
Social Behaviour in the Delivery of Public Services?" Journal of Public Economics
95, no. 7-8 (2011): 758-66.

8
Grigsby, J., E. Hurst, and A. Yildirmaz. "Aggregate Nominal Wage Adjustments: New
Evidence from Administrative Payroll Data." American Economic Review 111, no. 2
(2021): 428-71.
Grossman, S. J., and O. D. Hart. "The Costs and Benefits of Ownership - a Theory of Vertical
and Lateral Integration." Journal of Political Economy 94, no. 4 (1986): 691-719.
Grossman, Z. "Self-Signaling and Social-Signaling in Giving." Journal of Economic
Behavior & Organization 117 (2015): 26-39.
———. "Strategic Ignorance and the Robustness of Social Preferences." Management
Science 60, no. 11 (2014): 2659-65.
Grossman, Z., and J. J. van der Weele. "Self-Image and Willful Ignorance in Social
Decisions." Journal of the European Economic Association 15, no. 1 (2017): 173-217.
Guiso, L., P. Sapienza, and L. Zingales. "Long-Term Persistence." Journal of the European
Economic Association 14, no. 6 (2016): 1401-36.
Guth, W., R. Schmittberger, and B. Schwarze. "An Experimental-Analysis of Ultimatum
Bargaining." Journal of Economic Behavior & Organization 3, no. 4 (1982): 367-88.
Hamman, J. R., G. Loewenstein, and R. A. Weber. "Self-Interest through Delegation: An
Additional Rationale for the Principal-Agent Relationship." American Economic
Review 100, no. 4 (2010): 1826-46.
Hart, O., and J. Moore. "Contracts as Reference Points." Quarterly Journal of Economics
123, no. 1 (2008): 1-48.
———. "Property-Rights and the Nature of the Firm." Journal of Political Economy 98, no.
6 (1990): 1119-58.
Hedegaard, M., R. Kerschbamer, D. Muller, and J. R. Tyran. "Distributional Preferences
Explain Individual Behavior across Games and Time." Games and Economic
Behavior 128 (2021): 231-55.
Henrich, J., R. Boyd, S. Bowles, C. Camerer, E. Fehr, H. Gintis, and R. McElreath. "In
Search of Homo Economicus: Behavioral Experiments in 15 Small-Scale Societies."
American Economic Review 91, no. 2 (2001): 73-78.
Henrich, J., S. J. Heine, and A. Norenzayan. "The Weirdest People in the World?" Behavioral
and Brain Sciences 33, no. 2-3 (2010): 61-+.
Herbst, D., and A. Mas. "Peer Effects on Worker Output in the Laboratory Generalize to the
Field." Science 350, no. 6260 (2015): 545-49.
Hoffman, E., K. Mccabe, K. Shachat, and V. Smith. "Preferences, Property-Rights, and
Anonymity in Bargaining Games." Games and Economic Behavior 7, no. 3 (1994):
346-80.
Hollander, H. "A Social-Exchange Approach to Voluntary Cooperation." American
Economic Review 80, no. 5 (1990): 1157-67.
Homans, C. G. Social Behavior: Its Elementary Forms. New York: Harcourt, Brace &
World, 1961.
Hvidberg, K.B., S. Stantcheva, and C. Thustrup-Kreiner. "Social Position and Fairness Views
on Inequality." Review of Economic Studies (Forthcoming).
Iriberri, N., and P. Rey-Biel. "The Role of Role Uncertainty in Modified Dictator Games."
Experimental Economics 14, no. 2 (2011): 160-80.
Jakiela, P. "How Fair Shares Compare: Experimental Evidence from Two Cultures." Journal
of Economic Behavior & Organization 118 (2015): 40-54.
Karadja, M., J. Mollerstrom, and D. Seim. "Richer (and Holier) Than Thou? The Effect of
Relative Income Improvements on Demand for Redistribution." Review of Economics
and Statistics 99, no. 2 (2017): 201-12.
Kaufmann, R.T. "On Wage Stickiness in Britain's Competitive Sector." British Journal of
Industrial Relations 22 (1984): 101-12.
Kaur, S. "Nominal Wage Rigidity in Village Labor Markets." American Economic Review
109, no. 10 (2019): 3585-616.

9
Kerschbamer, R. "The Geometry of Distributional Preferences and a Non-Parametric
Identification Approach: The Equality Equivalence Test." European Economic
Review 76 (2015): 85-103.
Kerschbamer, R., and D. Muller. "Social Preferences and Political Attitudes: An Online
Experiment on a Large Heterogeneous Sample." Journal of Public Economics 182
(2020).
Kessler, J.B., and L. Vesterlund. "The External Validity of Laboratory Experiments: The
Misleading Emphasis on Quantitative Effects." In Handbook of Experimental
Economic Methodology, edited by G. Frechette and A. Schotter, 391-406: Oxford
Academic, 2015.
Khalmetski, K. "Testing Guilt Aversion with an Exogenous Shift in Beliefs." Games and
Economic Behavior 97 (2016): 110-19.
Khalmetski, K., A. Ockenfels, and P. Werner. "Surprising Gifts: Theory and Laboratory
Evidence." Journal of Economic Theory 159 (2015): 163-208.
Konow, J. "Fair Shares: Accountability and Cognitive Dissonance in Allocation Decisions."
American Economic Review 90, no. 4 (2000): 1072-91.
Kosfeld, M., and D. Rustagi. "Leader Punishment and Cooperation in Groups: Experimental
Field Evidence from Commons Management in Ethiopia." American Economic
Review 105, no. 2 (2015): 747-83.
Kosse, F., T. Deckers, P. Pinger, H. Schildberg-Horisch, and A. Falk. "The Formation of
Prosociality: Causal Evidence on the Role of Social Environment." Journal of
Political Economy 128, no. 2 (2020): 434-67.
Krawczyk, M. "A Glimpse through the Veil of Ignorance: Equality of Opportunity and
Support for Redistribution." Journal of Public Economics 94, no. 1-2 (2010): 131-41.
Krawczyk, M., and F. Le Lec. "How to Elicit Distributional Preferences: A Stress -Test of the
Equality Equivalence Test." Journal of Economic Behavior & Organization 182
(2021): 13-28.
Krueger, A. B., and A. Mas. "Strikes, Scabs, and Tread Separations: Labor Strife and the
Production of Defective Bridgestone/Firestone Tires." Journal of Political Economy
112, no. 2 (2004): 253-89.
Kube, S., M. A. Marechal, and C. Puppe. "The Currency of Reciprocity: Gift Exchange in the
Workplace." American Economic Review 102, no. 4 (2012): 1644-62.
———. "Do Wage Cuts Damage Work Morale? Evidence from a Natural Field Experiment."
Journal of the European Economic Association 11, no. 4 (2013): 853-70.
Lane, I.M., and L.A. Messe. "Equity and the Distribution of Rewards." Journal of
Personality and Social Psychology 20, no. 1 (1971): 1-17.
Larson, T., and C. M. Capra. "Exploiting Moral Wiggle Room: Illusory Preference for
Fairness? A Comment." Judgment and Decision Making 4, no. 6 (2009): 467-74.
Lazear, E. P., U. Malmendier, and R. A. Weber. "Sorting in Experiments with Application to
Social Preferences." American Economic Journal-Applied Economics 4, no. 1 (2012):
136-63.
Lefgren, L. J., D. P. Sims, and O. B. Stoddard. "Effort, Luck, and Voting for Redistribution."
Journal of Public Economics 143 (2016): 89-97.
Leventhal, G. S., and D. Anderson. "Self-Interest and Maintenance of Equity." Journal of
Personality and Social Psychology 15, no. 1 (1970): 57-+.
Leventhal, G. S., and D.W. Lane. "Sex, Age, and Equity Behavior." Journal of Personality
and Social Psychology 13, no. 4 (1970): 312-16.
Leventhal, G. S., and J.W. Michaels. "Locus of Cause and Equity Motivation as
Determinants of Reward Allocation." Journal of Personality and Social Psychology
17, no. 3 (1971): 229-35.

10
Leventhal, G. S., and J.W. Michaels. "Extending the Equity Model: Perception of Inputs and
Allocation of Reward as a Function of Duration and Quantity of Performance."
Journal of Personality and Social Psychology 12, no. 3 (1969): 303-09.
Levine, D. I. "Fairness, Markets, and Ability to Pay - Evidence from Compensation
Executives." American Economic Review 83, no. 5 (1993): 1241-59.
Levine, D. K. "Modeling Altruism and Spitefulness in Experiments." Review of Economic
Dynamics 1, no. 3 (1998): 593-622.
Levitt, S. D., and J. A. List. "Viewpoint: On the Generalizability of Lab Behaviour to the
Field." Canadian Journal of Economics-Revue Canadienne D Economique 40, no. 2
(2007b): 347-70.
Liebrand, W. B. G. "The Effect of Social Motives, Communication and Group-Size on
Behavior in an N-Person Multi-Stage Mixed-Motive Game." European Journal of
Social Psychology 14, no. 3 (1984): 239-64.
Liebrand, W. B. G., and C. G. Mcclintock. "The Ring Measure of Social Values -
Computerized Procedure for Assessing Individual-Differences in Information-
Processing and Social Value Orientation." European Journal of Personality 2, no. 3
(1988): 217-30.
Mas, A. "Labour Unrest and the Quality of Production: Evidence from the Construction
Equipment Resale Market." Review of Economic Studies 75, no. 1 (2008): 229-58.
———. "Pay, Reference Points, and Police Performance." Quarterly Journal of Economics
121, no. 3 (2006): 783-821.
Maskin, E., and J. Tirole. "Unforeseen Contingencies and Incomplete Contracts." Review of
Economic Studies 66, no. 1 (1999): 83-114.
Mayer, S. E. "How Did the Increase in Economic Inequality between 1970 and 1990 Affect
Children's Educational Attainment?" American Journal of Sociology 107, no. 1
(2001): 1-32.
Meltzer, A. H., and S. F. Richard. "A Rational Theory of the Size of Government." Journal of
Political Economy 89, no. 5 (1981): 914-27.
Messick, D. M., and C. G. Mcclintock. "Motivational Bases of Choice in Experimental
Games." Journal of Experimental Social Psychology 4, no. 1 (1968): 1-25.
Mikula, G., and H. Uray. "Neglect of Individual Performances in Financial Profit with
Respect to Social Situations." Zeitschrift Fur Sozialpsychologie 4, no. 2 (1973): 136-
44.
Molleman, E., J. Pruyn, and A. Van Knippenberg. "Social Comparison Processes among
Cancer Patients." Britisch Journal of Social Psychology 25, no. 1 (1986): 1-13.
Moore, J., and R. Repullo. "Subgame Perfect Implementation." Econometrica 56, no. 5
(1988): 1191-220.
Murphy, R. O., and K. A. Ackermann. "Social Value Orientation: Theoretical and
Measurement Issues in the Study of Social Preferences." Personality and Social
Psychology Review 18, no. 1 (2014): 13-41.
Murphy, R. O., K. A. Ackermann, and M. J. J. Handgraaf. "Measuring Social Value
Orientation." Judgment and Decision Making 6, no. 8 (2011): 771-81.
Nunnari, S., and M. Pozzi. "Meta-Analysis of Social Preferences Estimates." Working Paper,
Bocconi University, Italy (2022).
Offerman, T. "Hurting Hurts More Than Helping Helps." European Economic Review 46, no.
8 (2002): 1423-37.
Paetzel, F., R. Sausgruber, and S. Traub. "Social Preferences and Voting on Reform: An
Experimental Study." European Economic Review 70 (2014): 36-55.
Platteau, J. P. "Does Africa Need Land Reform?" Evolving Land Rights, Policy and Tenure in
Africa (2000): 51-73.
Prelec, D., and R. Bodner. "Self-Signaling and Self-Control." In Time and Decision:
Economic and Psychological Perspectives on Intertemporal Choice, edited by G.

11
 Loewenstein, D. Read and R. F. Baumeister, 277–98. New York: Russell Sage
Foundation, 2003.
Preuss, M., G. Reyes, J. Somerville, and J. H. Wu. "Inequality of Opportunity and Income
Redistribution." Unpublsihed Working Paper (2022).
Quach, S. "The Extent of Downward Nominal Wage Rigidity: Evidence from a Natural
Experiment." Princeton, NJ: Princeton University 2020.
Rabin, M. "Incorporating Fairness into Game-Theory and Economics." American Economic
Review 83, no. 5 (1993): 1281-302.
Rege, M., and K. Telle. "The Impact of Social Approval and Framing on Cooperation in
Public Good Situations." Journal of Public Economics 88, no. 7-8 (2004): 1625-44.
Reuben, E., P. Sapienza, and L. Zingales. "Is Mistrust Self-Fulfilling?" Economics Letters
104, no. 2 (2009): 89-91.
Roemer, J., and A. Trannoy. "Equality of Opportunity." In Handbook of Income Distribution,
edited by Atkinson. A. and F. Bourguignon. Amsterdam - New York: Elsevier, 2015.
Roth, A. E., V. Prasnikar, M. Okunofujiwara, and S. Zamir. "Bargaining and Market
Behavior in Jerusalem, Ljubljana, Pittsburgh, and Tokyo - an Experimental-Study."
American Economic Review 81, no. 5 (1991): 1068-95.
Ruffle, B. J. "More Is Better, but Fair Is Fair: Tipping in Dictator and Ultimatum Games."
Games and Economic Behavior 23, no. 2 (1998): 247-65.
Rustagi, D. "Historical Self Governance and Norms of Cooperation." Working Paper,
University of Nottingham (2022).
Rustagi, D., S. Engel, and M. Kosfeld. "Conditional Cooperation and Costly Monitoring
Explain Success in Forest Commons Management." Science 330, no. 6006 (2010):
961-65.
Saez, E. "Public Economics and Inequality: Uncovering Our Social Nature." Aea Papers and
Proceedings 111 (2021): 1-26.
Schminke, M., R. Cropanzano, and D. E. Rupp. "Organization Structure and Fairness
Perceptions: The Moderating Effects of Organizational Level." Organizational
Behavior and Human Decision Processes 89, no. 1 (2002): 881-905.
Schneider, F.H., F. Brun, and R.A. Weber. "Sorting and Wage Premiums in Immoral Work."
Zurich, Switzerland: University of Zurich, 2020.
Schulz, J. F., D. Bahrami-Rad, J. P. Beauchamp, and J. Henrich. "The Church, Intensive
Kinship, and Global Psychological Variation." Science 366, no. 6466 (2019): 707-+.
Simkins, P. Kitchener’s Army: The Raising of the New Armies, 1914-1916. Manchester:
Manchester United Press, 1988.
Skarlicki, and Folger. "Retaliation in the Workplace: The Roles of Distributive, Procedural,
and Interactional Justice (Vol 82, Pg 434, 1997)." Journal of Applied Psychology 82,
no. 6 (1997): 888-88.
Sliwka, D. "Trust as a Signal of a Social Norm and the Hidden Costs of Incentive Schemes."
American Economic Review 97, no. 3 (2007): 999-1012.
Smith, V. L. "Microeconomic Systems as an Experimental Science." American Economic
Review 72, no. 5 (1982): 923-55.
Snowberg, E., and L. Yariv. "Testing the Waters: Behavior across Participant Pools."
American Economic Review 111, no. 2 (2021): 687-719.
Tesser, A. "Toward a Self-Evaluation Maintenance Model of Social Behavior." In Advances
in Experimental Social Psychology, edited by L. Berkowitz, 181-227: Academic
Press, 1988.
Thibaut, J. W., and H. H. Kelley. The Social Psychology of Groups. New York: Wiley, 1959.
Van Lange, P. A. M. "The Pursuit of Joint Outcomes and Equality in Outcomes: An
Integrative Model of Social Value Orientation." Journal of Personality and Social
Psychology 77, no. 2 (1999): 337-49.

12
Van Lange, P. A. M., D. P. Balliet, C. D. Parks, and M. Van Vugt. Social Dilemmas:
Understanding Human Cooperation. New York: Oxford University Press, 2014.
Van Lange, P. A. M., W. Otten, E. M. N. DeBruin, and J. A. Joireman. "Development of
Prosocial, Individualistic, and Competitive Orientations: Theory and Preliminary
Evidence." Journal of Personality and Social Psychology 73, no. 4 (1997): 733-46.
Vanberg, C. "Why Do People Keep Their Promises? An Experimental Test of Two
Explanations." Econometrica 76, no. 6 (2008): 1467-80.
Vu, L., I. Sorraperra, M. Leib, J. van der Weele, and S. Shalvi. "Willful Ignorance: A Meta-
Analytic Review." Psychological Bulletin forthcoming (2023).
Xiao, E., and C. Bicchieri. "When Equality Trumps Reciprocity." Journal of Economic
Psychology 31, no. 3 (2010): 456-70.
Xu, X., S. Metsälampi, M. Kichler, K. Kotakorpi, P. H. Matthews, and T. Miettinen. "Which
Income Comparisons Matter to People, and How? Evidence from a Large Field
Experiment." Working Paper, Hanken School of Economics, Finland (2023).
Yang, Y. "A Survey of the Hold-up Problem in the Experimental Economics Literature."
Journal of Economic Survey 35, no. 1 (2021): 227 - 49.

13
 Online Appendix

Online Appendix for

Social Preferences
Fundamental Characteristics and Economic Consequences

Ernst Fehr & Gary Charness

0
 Online Appendix

Table of Contents for Online Appendix

Appendix 1: Heterogeneity in Altruistic Distributional Preferences between

Individuals and Subject Pools

Appendix 2: The Equality-Equivalence Test

Appendix 3: Endogenous Distributional Preference Clusters

Appendix 4: Material for “Who are the Meritocrats”?

Appendix 5: Distributional Preferences under Risk

Appendix 6: Payoff Matrix used in Bolton, Brandts and Ockenfels (1998)

Appendix 7: Belief-Dependent Preferences and Emotions

Appendix 8: Responsibility and Delegation

Appendix 9: The Stability of Social Preferences

1
 Online Appendix

Appendix 1
Heterogeneity in Altruistic Distributional Preferences between Individuals
and Subject Pools

This appendix describes the characterization of individual heterogeneity in terms of

individuals’ estimated CES utility functions. Andreoni and Miller (2002) digged deeper into
individual heterogeneity by recruiting 176 student subjects who make between 8 – 11 choices
in dictator games with varying prices of giving, allowing them to check for violations of the
generalized axioms of revealed preferences (GARP).1 They find that less than 2% commit
GARP violations, meaning that the choices of the remaining 98% can be represented by a
quasi-concave utility function. They also classify individuals into one of three predefined
categories: selfish subjects, egalitarian subjects who maximize 𝑈 𝜋 , 𝜋 = min (𝜋 , 𝜋 ), and
utilitarians who maximize (0.5 𝜋 + 0.5 𝜋 ). While 43 percent of their subjects display
choices that perfectly fit these preference categories, the remaining 57 percent are allocated to
these categories by minimizing the distance from the three pre-specified utility functions.
Based on this procedure, they classify 47.2% of the 176 subjects as selfish, 30.4% as
egalitarian and 22.4% as utilitarian.

To what extent do the 57% of “impure” subjects actually fit the three predefined
preference categories? To answer this question, the authors estimate a representative CES
function (2) for the “impure” individuals in each category. The results indicate that the
estimated parameters deviate quite substantially from the parameters of the ideal types. For
example, the average 𝛼′ of the “impure” selfish subjects is 0.24, indicating a non-negligible
deviation from selfishness, and the average 𝜌 of the egalitarian types is –0.35 which is a long
way from −∞ which would indicate strict egalitarianism. While such deviations from the

1 There is a considerable literature on social value orientation (SVO) in psychology that uses the so-called
ring measure of SVO (Liebrand 1984; Liebrand and Mcclintock 1988) and/or the triple-dominance measure
of SVO (Van Lange et al. 1997; Van Lange 1999) or the slider task (Murphy, Ackermann and Handgraaf
2011; Murphy and Ackermann 2014). These measures are based on generalized dictator games but they
do not lend themselves easily to the estimation of utility functions. Instead, they assign individuals to
predefined SVO types such as “cooperative” (= desire to maximize joint gains), “altruistic” (= desire to
maximize the other player’s payoff), and “competitive” (= desire to maximize the payoff difference) or they
use the ratio between the total payoff given to the other player and the total payoff assigned to “self” (across
all dictator games) as an individual SVO measure. In addition, these SVO measures cannot identify
inequality aversion as defined in Fehr and Schmidt (1999) or Bolton and Ockenfels (2000) and they do not
capture the fundamental difference in the slope of indifference curves between the domains of
advantageous and disadvantageous inequality. For this reason, we do not include them in the following
review of distributional preferences.

2
 Online Appendix

pure types are inevitable when people are classified into subgroups it is important to keep
them in mind.

Two further observations related to Andreoni and Miller (2002) are worth mentioning.
First, even those individuals who perfectly fit the selfish preference assumption in their
choice data may not be perfectly selfish because the smallest relative price of giving was 0.25
– for every dollar given, the partner received $4. Thus, we do not know what would have
happened if the relative price had been lower.2 Second, 34 subjects in one of their sessions
also faced upwards sloping budget line in 𝜋 , 𝜋 -space that involved disadvantageous
inequality. Subjects could reduce inequality in these budget lines by decreasing both players’
payoffs, and 8 of the 34 subjects (23.5%) actually did so. Thus, they observed some evidence
in favor of inequality aversion when behind but no strong inferences can be made here given
the small sample size, and the CES utility function is not capable of capturing these
preferences.

The Fisman-Jakiela-Kariv-Markovits group undertook one of the most systematic

characterizations of individual heterogeneity in altruistic distributional preferences in a series
of papers (Fisman, Kariv and Markovits 2007; Fisman, Jakiela and Kariv 2015; Fisman et al.
2015; Li et al. 2022). Subjects in their experiments faced many different budget constraints in
the material payoff space, giving them substantial power to estimate the individual preference
parameters 𝛼′ and  of the CES utility function. In Fisman, Kariv and Markovits (2007) and
Fisman, Jakiela and Kariv (2015), they report the parameter estimates of 76 and 72 Berkely
undergraduates, respectively; moreover, they estimate the distributional preferences of 208
Yale Law School (YLS) students in Fisman et al. (2015) as well as of 503 US medical
students in Li et al. (2017). In Figures 2a and 2b we show the cumulative distribution of the
estimated 𝛼′ and  parameter for the Berkeley and the Yale Law School students and
Appendix Table A1 classifies the individuals into three categories: those close to selfishness
(’ > 0.95), intermediate altruists (0.55  ’  0.95) and egalitarian altruists (0.45 < ’ <
0.55). The figures and Table A1 illustrate that between 30 and 40 percent of the students put
literally a weight of zero or a weight close to zero on other individuals’ payoffs (𝛼′ > 0.95),
while only between 8 and 25 percent of them are egalitarian altruists. Moreover, the student

2
Some people may be inclined to discount situations in which the cost of altruistic acts is low, but social life is in
fact pervaded by situations in which low-cost favors can be given to other people. When a colleague in the
workplace asks for help, when a stranger in a city asks for directions, or when students help each other answer
questions, the costs involved are often very low, while the benefits for the receiving party are high.

3
 Online Appendix

subject pools appear to be more oriented towards efficiency compared to equality because
only between 30 and 37% of them reveal a  < 0.

Are these results from student samples generalizable to the general population? To answer
this question, the Fisman-Jakiela-Kariv-Markovits group also conducted experiments with a
large sample of roughly 1000 Adult Americans from the American Life Panel (ALP). The
ALP subjects are broadly comparable with the US population in terms of demographic and
socio-economic characteristics. To control for age, Fisman et al. (2015) use only the ALP
subjects under age 40 for the comparison with the student sample. The figures show that the
ALP sample under age 40 displays a much higher concern for the payoff of others (Figure 2a)
and a much higher concern for equity compared to efficiency (Figure 2b) than the student
sample. The following facts displayed in Table A1 are, in particular, noteworthy: (i) Among
the ALP subjects under age 40, the share of individuals that are close to selfishness is only
16.2% which is much smaller than the 30-40% among the students. (ii) The share of
egalitarian altruists is with 37.2% of ALP subjects under age 40 much larger than the 8-26%
among the students. (iii) The share of equality-oriented individuals ( < 0) is with 47% of
ALP subjects under 40 much larger than corresponding share among the students.3

These large differences between student samples and the broader population are
consistent with research reported in Snowberg and Yariv (2021) and Cappelen et al. (2015).
Snowberg and Yariv document that subjects from a representative sample of the US
population transfer a much higher share of income (39%) to recipients in simple dictator
games compared to the transfers given by a large sample of all Caltech undergraduate
students, who gave only 14%. Likewise, Cappelen et al. (2015) report that in a representative
sample of the Norwegian population the share transferred was 40.3% for men and 41.7% for
women, while male students only gave 22.6% and female students gave 32.2%.

3
The FJKM group also shows that the much higher degree of other-regardingness and the much higher equality
orientation of the broad population sample does not depend on socio-economic status. In other words, the
individuals with high education and income in the ALP sample (N = 152) display very similar parameters
compared to the rest of the ALP sample.

4
Online Appendix

Figure A1a
The estimated weight on self-payoff (’) among students and in a broad sample of the US
population under age 40. High ’ means a low concern for others’ payoff.
(based on data from FKM 2007, FJK 2015, FJKM 2015)4
Figure A1b
The estimated weight of efficiency relative to equality concerns () among students and a
broad sample of the US population under age 40. High  means a low concern for equality.
(based on data from FKM 2007, FJK 2015, FJKM 2015)
4 FKM (2007) indicates Fisman, Kariv and Markovits (2007), FJK (2015) indicates Fisman, Jakiela and
Kariv (2015) and FJKM indicates Fisman, Jakiela, Kariv and Markovits (2015). See also Table A1 for the
type classification that follows from the estimates displayed in Figures 2a and 2b.
5
 Online Appendix

Table A1: Empirical Properties of Altruistic Distributional Preferences

Study Subject Pool Egalitarian Intermediate Close to

altruism altruism Selfishness
<0
0.45 < ’ < 0.55  ’  ’ > 0.95
0.55 0.95

FKM 2007 N = 148 UC 8.1% 49.3% 39.9% 37.%

& FJK 2015 Berkeley students

FJKM 2015 N = 208 Yale 14.5% 53.9% 31.8% 20.3%

Law School
Students

JDK 2017 N = 503 Students 25.7% 41.5% 28.2% 29.2%

from US medical
schools

FJKM 2015 N = 309 Adult 37.2% 42.7% 16.2% 47.3%

Americans under
40 (ALP
subjects)

N = 693 Adult 27.7% 50.5% 16.0% 57.0%

Americans over
40 (ALP
subjects)

LDK 2017 N = 208 36.8% 42.8% 15.1% 48.3%

US Physicians

Note. The table shows key components of the distribution of individuals’ estimated weights (’) on other persons’
payoffs based on studies co-authored by D (Dow), F (Fisman), J (Jakiela), K (Kariv), L (LI) and M (Markovits).
Thus. FKM (2007) indicates the paper by Fisman, Kariv and Markovits (2007). The estimates are based on the
assumption that distributional preferences can be captured by a CES utility function like in equation (3) and on
each subjects’ distributional choices in 50 randomly chosen budget sets. The efficient frontier of the budget set
(i.e., the “budget line”) is always negatively sloped such that one cannot measure the willingness to pay to reduce
others’ income for the sake of equality (“inequality aversion”). However, the CES function enables the
identification of individuals’ preference for equality within the class of altruistic preferences with the parameters
’ and . ’ = 1/2 indicates that individuals put equal weight on others’ payoff, and  < 0 implies that the income
share spent on others’ payoff rises as the price of giving rises, i.e., subjects are equality-oriented ( < 0) and not
efficiency-oriented (0 <  < 1).

6
 Online Appendix

One noteworthy feature of the experimental design on which the data in Figure 2a and
2b and Table A1 are based is that the price of giving is randomly determined for every
subject, i.e., different subjects see different prices. This means that some subjects may have
seen a relatively large number of low prices for giving, which makes identification of purely
selfish subjects very precise, while other subjects may have seen only a few low prices of
giving, so that their assignment to the selfish versus intermediate category may be coarser.

Another important feature of the data collected by the Fisman-Jakiela-Kariv-Markovits

group is that the subjects do not face upwards sloping budget lines in (𝜋 , 𝜋 )-space. Thus,
by construction, the subjects do not face a situation in which they can decrease both players’
payoffs to reduce disadvantageous inequality. Given this restriction, the CES approach is a
powerful tool for identifying altruistic distributional preferences, but it cannot capture
spiteful, envious, or inequality averse preferences5.

5
In Fisman, Kariv, Markovits (2007), the authors had budget constraints with vertical and horizontal segments,
but their student subjects never made pareto-damaging choices on these segments, which led the authors to
believe that inequality aversion is not important.

7
 Online Appendix

Appendix 2
The Equality Equivalence Test

In the equality equivalence test subjects are presented choice lists in the domain of
advantageous payyofs (A-lists) and the domain of disadvantageous payoffs (DA-lists). In a
disadvantageous list (DA-list, see Figure A2 below), the equal payoff allocation E is always
paired with a list of alternative allocations in which the other subject’s payoff is kept constant
at a level of 𝜋 > 𝜋 , while 𝜋 systematically varies across alternative allocations. In an
advantageous list (see Figure A2), E is always paired with a list of alternative allocations in
which the other subject’s payoff is kept constant at a level of 𝜋 < 𝜋 while 𝜋 systematically
varies across alternative allocations.

Starting the binary choice list with the choice between the 𝜋 , 𝜋 -combination and E
where 𝜋 is lowest (and hence, below the egalitarian payoff, see Figure A1), the decision
maker is more benevolent towards the other subject (i.e., willing to pay to increase the other’s
payoff) in the DA domain, the earlier he or she moves from E towards an alternative
allocation 𝜋 , 𝜋 . In the advantageous domain, the decision maker is more benevolent if he
or she, starting the binary choice list with the choice between the 𝜋 , 𝜋 -combination and E
where 𝜋 is highest (and hence, above the egalitarian payoff), moves earlier to the equal
payoff allocation E. However, the EET can also identify inequality aversion in the DA
domain because some binary choice pairs essentially imply a choice on a positively sloped
“budget line”. Likewise, the EET can also identify positively sloped indifference curves in
the A domain (“spite”) because some binary choice pairs in this domain are located on
positively sloped “budget lines”.

A potential drawback of the EET is that the equal payoff allocation is part of every
binary choice the subjects face, which may render equality very salient and thus induce a
behavioral bias towards equality. However, a study by Krawczyk and Lee (2021) indicates
that the results are robust to the introduction of a reference allocation that does not involve
equality. In addition, the results of the EET by Kerschbamer (2015) indicates that 48.9% of
his student subjects reveal selfish preferences (see Table 1 below), which is even higher than
the 39.9% of selfish students in Fisman, Kariv and Markovits (2007) or the 31.8% of selfish
students in Fisman et al. (2015). Likewise, Table 1 presents the data from several other
studies with student samples that indicate a relatively high share of selfish subjects that
approaches 60% in some student samples. Moreover, among the student subjects with other-

8
 Online Appendix

regarding distributional preferences, those with altruistic preferences are far more prevalent
compared to inequality averse or envious preferences. The share of altruistic student subjects
varies between 28 and 48%, while the share of inequality averse subjects is between 7 and
12%. Typically, envious/spiteful subjects are the least frequent across the student data with 3-
10%.

Figure A2: Choice Alternatives in the Equality-Equivalence Test

The figure illustrates how the Equality Equivalence Test (EET) works by depicting the alternatives to
the equal payoff allocation which is at (10, 10). The figure is taken from Kerschbamer and Müller
(2021). It shows three binary choice lists in the disadvantageous domain (DA-lists) and three lists in the
advantageous domain (A-lists). The DA-lists enable the identification of the slope of a subject’s
indifference curve in the DA domain (𝛼), while the A-lists enable identification of the slope in the A
domain (𝛽).

9
 Online Appendix

Appendix 3: Endogenous Distributional Preference Clusters

Figure A3: Example Decision Screen and Budget Line

Figure A3a shows the decision screen in Fehr®Epper®Senn (2022). Figure A3b illustrates the
negatively sloped budget line that corresponds to the set of alternatives depicted on the decision
screen. Subjects faced many positively sloped and many negatively sloped budget lines.

10
 Online Appendix

Table A2: Empirical Frequency of Endogenous Distributional Preference Clusters

Endogenous Preference Clusters

Study Subject Pool Altruistic Inequality Predomin

Averse antly
selfish

Swiss Student Univ. of Zurich 60.1% 1.5% 36.4%

Sample N = 66

Swiss Panel 42% 48.1% 9.9%

Sample (2017)
N = 467

Swiss Panel 38.7% 45.2% 16.1%

Sample (2020)
Fehr®Epper Broadly N = 467
®Senn representative
(2022) Swiss Samples Swiss Sample 29.6% 53.7% 16.7%
(Only 2017)
N = 348

Swiss Sample 30.5% 45.5% 24.0%

(Only 2020)
N = 916

Epper et al. Broad Danish Sample (2017) 30.2% 37.3% 32.5%

(2020) N = 3691

Note: The data of the broad Swiss population samples is taken from Fehr®Epper®Senn (2022). Each of
the four data sets is broadly representative of the German and French language areas in Switzerland in
terms of age, gender, income, and education. The Danish data consist of a broad sample of Danish
individuals collected in 2017 by Epper et al. (2020). The roughly 3700 subjects in this study are from the
Copenhagen area and have an age between 32-42. The data from the Swiss student sample is taken from
unpublished work by Epper, Fehr and Senn. The table is based on choice problems in which subjects
could choose one of 7 payoff allocations that are equidistantly placed on linear budget lines with different
positive and negative slopes in 𝜋 , 𝜋 space. A non-parametric Bayesian clustering algorithm assigns
subjects to one of three preference types. Importantly, these preference types are not assumed a priori but
arise endogenously from the empirical properties of the data sets. Interestingly, qualitatively similar
preference types emerge in all of the data sets above.

11
 Online Appendix

Figure A4: Preference types in in a large Danish population sample (N = 3691)

Note: The figures depict subjects’ median choices z among negatively sloped budget lines and among
positively sloped budget lines. Each dot represents five individuals whose median choice is identical*.
Dots are jittered to make the median choice of the five-person groups visible. For each budget line, z =
1 indicates an own-payoff maximizing choice, z = 0 indicates an own-payoff minimizing choice, and z
= 0.5 indicates a payoff-equalizing choice. Thus, five-person groups whose z-value is close to 0.5 on
positively and negatively sloped budget lines display inequality averse behaviors, individuals whose z-
value is close to 1 on both types of budget lines display selfish behaviors, and individuals whose z-
value is close to 0.5 on negatively sloped budget lines but z = 1 on positively sloped budget lines display
altruistic behaviors. The data are taken from Epper et al. (2020).

12
 Online Appendix

Appendix 4: Material for the Section on “Who are the Meritrocrats?”

Figure A5: Example of a Decision Screen in the Merit and Luck condition of
Epper, Fehr and Senn (2023)

Note: Subjects in the merit and luck condition faced exactly the same negatively sloped budget lines
illustrated in the above screen. The initial allocation was highlighted but subjects were completely
free in choosing any of the seven available allocations.

13
 Online Appendix

Appendix 5: Distributional Preferences under Risk

When individuals care for others’ payoffs, a whole new set up of questions arises if outcomes
are risky. A key issue concerns the question whether people care for others’ expected payoffs
or for their realized payoffs. This also concerns the issue whether individuals care for
equality of opportunity, i.e., have a preference for lower inequality in ex-ante expected
payoffs or whether they have a preference for more equal ex-post realized payoffs. Another
issue when risk is present is how individuals’ own risk preferences and their beliefs about
others’ risk preferences affect their other-regarding behavior.

The problem of ex-ante expected payoffs versus ex-post realized payoffs comes into
sharp focus in a dictator game that involves the sharing of chances to win an indivisible
resource that has a value of R = 100 for both parties. The dictator chooses x, which
determines the probability with which the recipient wins R, while the dictator wins R with

probability (1 − ). Here, equality of opportunity implies the equalization of chances but

there will always be inequality ex-post. An individual with utility function 𝑈(𝜋 , 𝜋 ) that
obeys the plausible restriction 𝑈(𝑅, 0) > 𝑈(0, 𝑅) will always choose x = 0. Not only
inequality averse players, but players with Charness-Rabin preferences as well, may plausibly
obey the restriction 𝑈(𝑅, 0) > 𝑈(0, 𝑅) and thus choose x = 0.

This prediction contrasts, however, with the results of experiments showing that many
dictators are willing to transfer some chance of winning to the recipients (Krawczyk and Le
Lec 2010; Brock, Lange and Ozbay 2013). Models that are solely based on the realized ex-
post payoffs have a hard time explaining this fact, whereas models in which players also care
about the ex-ante expected payoffs of others can explain it.

Now suppose that the above-described game is slightly changed so that the payoff to
the two players is no longer exclusive, i.e., if the dictator transfers a chance x, then the
dictator wins R with probability (1 − ) and the recipient can simultaneously also win R

with , i.e., there are two independent draws. Note that there may not be any ex-post

inequality in this game because both players can end up with 0 or with R. Therefore,
inequality averse dictators have less reason to worry about inequality, implying that they are
more likely to be willing to share chances with the recipient. Krawcyk and Le Lec (2010)
indeed show that dictators transfer more chances in the dictator game with independent draws

14
 Online Appendix

compared to the game with exclusive payoffs. This result suggests that players also care
about ex-post payoffs.

Further evidence for the relevance of ex-post payoffs is provided by Brock, Lange and
Ozbay (2013), who designed six different dictator games where they systematically varied
the risk for the dictators and the recipients across games in such a way that if players’ cared
only about ex-ante expected payoffs, they would behave identically across all six games.
Their find treatment differences, however, that are indicative for the relevance of ex-post
payoff concerns. For example, subjects in the standard dictator game without any risk (and a
dictator endowment of 100) transfer a significantly higher x to the recipient compared to a
dictator game where the dictator’s payoff is still certain, but a transfer of x gives the recipient
a payoff of 100 with probability . Note that a positive transfer x in the game where the

recipient faces a risky payoff implies that the dictator may end up with a lower payoff than
the recipient. Inequality averse dictators, who care for ex-post inequality, will thus tend to
give less in the risky dictator game.6 Another key result documented in Brock, Lange and
Ozbay (2013) is that subjects’ giving in the standard dictator game is highly predictive for
their willingness to equalize ex-ante expected values in dictator games involving risks.

The question whether subjects care for equality of opportunity or for equality of ex-post
payoffs was also addressed in Cappelen et al. (2013). In their experiments, there was first a
risk-taking phase and then a distribution phase. Subjects made 4 decisions in the risk-taking
phase between the payoff y of a sure alternative ( 𝑦 ∈ {25, 200, 300, 400} and a 50:50 chance
of receiving nothing or 800 NOK. In the distribution phase, each subject was paired
sequentially with 8 different subjects who participated in the risk-taking phase, and one of the
four risk-taking problems was drawn randomly for each pair. Then, an “impartial” spectator,
who was informed about subjects’ choices and outcomes in the drawn risk-taking problem,
was asked to distribute the pair’s total earnings between the two subjects.

6 Alternatively, because the certainty equivalent of a given transfer x is less valuable for risk averse
recipients, dictators who care for the total payoff may give less in the risky dictator game. However, based
on this logic risk averse dictators should give more in a dictator game in which their own payoff is risky –
they receive a payoff of 100 with probability 1 − – while the recipient receives the transfer x with
certainty. The reason is that a transfer of x decreases the certainty equivalent of the dictator’s payoff by
less than x, i.e., giving is surplus-enhancing. The evidence strongly suggests the opposite, as dictators give
much less in this game compared to the standard dictator game (Freundt and Lange 2017). Moreover,
Freundt and Lange also find that the dictators who believe that recipients are risk averse do not give less
to the recipients.

15
 Online Appendix

Before presenting the results, it is important to emphasize that complete equality of

opportunity existed between the two paired subjects in the risk-taking phase. If spectators
redistribute ex-post from the richer to the poorer subject, they thus explicitly express a
preference for less ex-post inequality. Almost all of the spectators’ redistributive choices
involved redistribution from the poorer to the richer subject.7 If the pair consisted of two risk
takers where one was lucky while the other was unlucky, the spectators strongly redistributed
from the lucky to the unlucky one – they chose the equal split in more than 40% of the cases
and they did not redistribute at all in only roughly 30% of the cases. In contrast, if an unlucky
risk-taker was paired with an individual who chose the safe option, the unlucky risk-taker
received much fewer transfers and the equal split was only chosen in roughly 15% of the
cases. Spectators thus made the unlucky risk takers more responsible for their choices
compared to a situation where both were unlucky. Finally, there is also a substantial amount
of redistribution when a lucky risk-taker is paired with an individual who chose the safe
payoff, but the lucky risk-taker was nevertheless given a higher payoff in roughly 80% of the
cases.

Thus, taken together, the literature suggests that subjects on average care about both ex-
ante equality of opportunity and ex-post equality of outcomes but there is strong
heterogeneity in the weight that individual subjects put on the different conceptions of
equality. Cappelen et al. (2013) estimate a mixture model that enables them to assign
individuals to three different types – individuals who care only for ex-post equality (“ex-post
egalitarians”, EPs), individuals who do not care about ex-post equality (“ex-ante
egalitarians”, EAs), and individuals who care about ex-post equality among those who made
the same choice in the risk-taking task (“choice egalitarians”, CEs). Roughly 30% of their
subjects (students from the Norwegian School of Economics) are EPs, 27% are CEs and 43%
are EAs.

In this section we have so far mainly dealt with the question how social preferences are
affected by outcome risks. However, the perceived sources of inequality may also be subject
to risk and uncertainty. If individuals do not know whether a particular inequality is due to
luck or differential performance, how does this affect their willingness to redistribute
income? Cappelen et al (2022) study this situation, and document that this kind of uncertainty

7In case that a lucky risk-taker met a subject who chose the safe payoff it would have been possible to
redistribute from the poorer to the richer subject.

16
 Online Appendix

can push meritocrats towards behaving more egalitarian – with more risk averse spectators
exhibiting a stronger drive towards egalitarian behavior.

Finally, we deal with the question how to combine concerns for equality of opportunity
and equality of outcomes in theoretical modelling. Saito (2013) addresses this issue,
providing an axiomatic foundation for “expected inequality-averse” preferences. Individuals
with such preferences put a weight of  (0  𝛿  1) on preferences for equality of opportunity
and a weight (1-) on preferences for equal ex-post outcomes.8

To make things concrete, let 𝒙 = (𝑥 , 𝑥 , … , 𝑥 ) denote an allocation of material

payoffs to individuals. Assume that there are m different states of the world, each one of
which is obtained with probability 𝑝 , 𝑠 ∈ {1, … , 𝑚}, and denote the allocation obtained in
state s by 𝒙𝒔 = (𝑥 , 𝑥 , … , 𝑥 ), then the expected material payoff allocation is given by

𝐸(𝒙) = ∑ 𝑝 𝒙𝒔 = (∑ 𝑝𝑥 , ∑ 𝑝 𝑥 ,…,∑ 𝑝 𝑥 ),

where ∑ 𝑝 𝑥 denotes the expected material payoff of individual i across states. Likewise,
the allocation of expected utilities is given by

𝐸(𝑈(𝒙)) = ∑ 𝑝 𝑈(𝒙𝒔 )

Saito shows that if and only if a decision-maker obeys “his” axioms, the preferences of a
decision-maker are represented by the following preference function V:

𝑉 = 𝛿𝑈(𝐸(𝒙)) + (1 − 𝛿)𝐸(𝑈(𝒙), (11)

where 𝑈(𝒙) is given by the Fehr-Schmidt Utility function. Thus, the utility of an expected
inequality averse player is affected by the inequalities in the expected material payoffs with
weight 𝛿 and by the inequalities in realized ex-post payoffs with weight (1 − 𝛿). It is also
noteworthy that the preference function (8) also applies under further plausible assumptions if
𝑈(𝒙) is given by Charness-Rabin type preferences.

It is easy to see that an individual who puts a sufficiently high weight 𝛿 on equality of
opportunity is willing to share the chances of receiving an indivisible resource in a dictator
game although this creates chances for high ex-post inequality. Overall, however, the Saito
model has undergone very little empirical testing. For example, it would be interesting to
know to what extent the behavior of individual subjects in the six different treatment
conditions of Brocks, Lange and Ozbay (2013) are consistent with the Saito model and which

8Several other authors have also provided axiomatic foundations of inequality averse preferences (Neilson 2006;
Rohde 2010) but none of them involves preferences for equality of opportunity.

17
 Online Appendix

parameters (, , ) explain their behaviors.9 To our knowledge, there is no paper that jointly
estimated 𝛿 and the parameters in 𝑈(𝒙). One complication in applying (8) to data is that the
distributional preference models – such as Fehr-Schmidt or Charness-Rabin – assume risk
neutrality, but it is well known that risk aversion also exists at the typical experimental stake
levels. This means that behavior in distributional problems under risk is affected by a
complicated mix of risk aversion as well as by preferences for equality of opportunity and
other-regarding preferences for ex-post outcomes.10

9
Recall that in their experiments an individual with  = 1 would behave identically across all treatments. Thus,
behavioral variation across treatments may provide at least some qualitative insights with regard to the parameter
constellations that may explain their data.
10 Cettolin, Riedl and Tran (2017) and Freundt and Lange (2017) have independent measures of dictators’

and recipients risk aversion and can relate them to the dictators’ behavior in risk-involving dictator games.
Cettolin, Riedl and Tran (2017) show that dictators’ risk aversion strongly predicts lower transfers in both
dictator games that render the payoff of the recipients risky and in dictator games that render the payoff of
the dictators risky. Freundt and Lange (2017) also show that a rise in dictators’ risk aversion is associated
with a decline in generosity in games where the dictators’ payoff is subject to risk.

18
 Online Appendix

Appendix 6: Payoff Matrices used in Bolton, Brandts and Ockenfels (1998)

19
 Online Appendix

Appendix 7: Belief-Dependent Social Preferences and Emotions

Psychological game theory (Battigalli and Dufwenberg 2009; Battigalli and Dufwenberg
2022) is a powerful and flexible tool for modelling emotions and their effect on social
preferences. This appendix discusses a few papers that provide evidence for the role of
emotions in experimental games.
An article by Khalmetski, Ockenfels, and Werner (2015) considers “surprising
gifts”, a concept related to experiencing elation or joy and so seems to be the flip side to
guilt. Their model generalizes the guilt-aversion model to capture positive surprises, and an
extension of the model permits choosers to care about the intentions of others behind
surprises. The authors provide experimental tests using a series of dictator-game
experiments. Interestingly, their design featured the Ellingsen et al. (2008) notification
feature, but had different results. The responder made a guess about how much the dictator
will give; the recipient who guessed closest to the average received by all recipients
received a substantial bonus. In the meantime, the dictator completed a form stating how
much she will give for each (rounded) guess possibly made by the recipient. In the
PUBLIC treatment, dictators were told that recipients would later be informed that their
guesses were revealed to the dictators before they made their choices, while in the
PRIVATE treatment dictators were told that the recipients would not be so informed.

This study finds evidence that dictators do consider what recipients know about the
dictators’ intentions behind surprises. In support of their model, they find smaller amounts
given if the inference concerning a dictator’s intentions is unclear. In one experiment, the
average amount sent in PUBLIC was 16.8%, substantially higher than the 10.1% average in
PRIVATE. The share of dictators who exceeded the recipient’s expectation was more than
twice as high in the PUBLIC treatment, 28.4% versus 11.7%. The experimental results
suggest that guilt aversion can be an important motivation for giving in dictator games.
Their analysis highlights that many subjects also like to exceed others’ expectations,
perhaps a surprising result.

Battigalli, Dufwenberg, and Smith (2019) provide a model of frustration and anger.
Anger and aggression are certainly prevalent in our world, and it is often frustration that
leads to anger and aggression. This phenomenon has been useful for some “populist”
demagogues; the 2016 and 2020 U.S. presidential elections are cases in point. Battigalli,

20
 Online Appendix

Dufwenberg, and Smith (2019) assume that a frustrated player is inclined to hurt those
deemed blameworthy, formalizing types of blame. With simple anger, all others are
blamed equally regardless of the choices made. With anger-from-blame, others are blamed
to the extent that they could have avoided a player’s frustration with different choice
behavior. Regarding anger-from blame intentions, others are blamed to the extent that the
player believes they intended to cause her frustration.

An example taken from Battigalli and Dufwenberg (2022) illustrates this notion.

Battigalli and Dufwenberg (2022) example

In this game, if Penny’s anger sensitivity is sufficiently high, simple blame indicates
she would choose d, targeting the person that she can most efficiently punish. Here this
would be Don. With anger from blaming behavior, Penny would choose b with a
sufficiently high blame sensitivity, since Don (who had no choice) is not blameworthy.
Anger from blaming intentions is a bit trickier because intention must be inferred. Here
Penny would choose a; Ben is not blameworthy because he had no way to know the actual
chance move and so could not have bad intentions.

Battigalli and Dufwenberg (2022) discuss an array of other emotions from the
standpoint of psychological game theory and belief-dependent preferences. These include
elation, regret, and anticipatory feelings. Given space limitations, we only sketch these out
briefly. Elation is the opposite of disappointment. While elation is indeed rare in the
economics profession and the literature Gill and Prowse (2012) present evidence that
disappointment is stronger than elation, much as negative reciprocity seems stronger than
positive reciprocity), this can be readily modeled in the same manner as is disappointment
but with a negative term replaced by a positive one. With regret, one must assess what
would have happened if she had chosen differently. One may also have anticipatory

21
 Online Appendix

feelings that provide current positive or negative utility (Loewenstein et al. 2001). Anxiety
is one such anticipatory feeling. Since future outcomes depend on one’s own actions, as
well as those of others), this allows psychological utility to depend on one’s own plan.
Caplin and Leahy (2001) provide a nice example of a doctor who must choose whether to
reveal information to a patient, anticipating the patient’s anticipatory feelings.

22
 Online Appendix

Appendix 8 – Responsibility and Delegation

Feelings of responsibility may have substantial effects on social preferences. When

one has the responsibility for determining an outcome that may affect the well-being of
others, one may respond more positively than when this responsibility is borne by another
entity. The flip side of the coin is that being able to delegate a decision may weaken the
effect on one’s image, so that one might expect more selfish behavior.

To the best of our knowledge, the first research on this topic in the field of economics
is Charness (2000). The intuition is that one will be more honest (or pro-social) when one
bears the responsibility of making the determination: “The responsibility-alleviation effect
states that a shift of responsibility to an external authority dampens internal impulses towards
honesty, loyalty, or generosity.” The article uses gift-exchange evidence to demonstrate that
participants are more generous (provide higher effort) when wages are determined by draws
from a bingo cage than when assigned by a third party, so that even a modest shift in
perceived responsibility affects behavior. Charness and Jackson (2009) identify the same
principle in a Stag Hunt. They make a comparison to what players do unilaterally to what
they do when what they do when making a choice on behalf of another party as well. While
only one-third of the population choose differently in these two cases, almost 90 percent of
these people (p = 0.001) plays a less risky strategy (Hare) when choosing for a group than
when playing only for themselves. So, responsibility-alleviation applies to strategic
situations as well as to honesty. Being held responsible (either by yourself or by others) does
affect one’s social preferences.

A notion quite related to responsibility-alleviation is delegation. Here one formally

shifts at least the appearance of responsibility to another party, trying to evade being held
responsible for decisions by other people (and perhaps even to self-deceive about this
responsibility). The model makes the same predictions with delegation as with
responsibility-alleviation. To the extent that one is perceived to be less responsible by having
delegated an action or choice, one can be selfish at a lower cost.

Fershtman and Gneezy (2001) note that employing a messenger to deliver bad news is
quite common and considers strategic delegation in the ultimatum game. In the second half
of some sessions (the first half was the basic game), there were four delegation environments
varying whether the delegation was available to the dictator or to the responder and whether
the delegation contract was observable (the delegator can provide the delegate with an

23
 Online Appendix

incentive scheme and can, for example, specify the proposal to be made). Proposers’ profits
were higher is this case, perhaps because proposers were perceived to be less selfish.
However, there is no significant difference when the delegation contract was observable.
When responders could delegate, they benefited from doing so. Here, “the responders
induced the agents to be tough, and as a result the proposers indeed made more generous
offers”. Interestingly, the effect of responder delegation with an unobservable contract goes
in the opposite direction, with higher proposals and higher proposer profit. The authors
conclude that delegation impacts the outcome of the game.

Hamman, Loewenstein, and Weber (2010) also find that delegation can foster self-
interest, suggesting an additional benefit for the principal-agent relationship. They state:
“Through the use of agents, therefore, accountability for morally questionable behavior can
become vertically diffused, with no individual taking responsibility”. In one experiment, the
Baseline treatment was a standard $10 dictator game. In the Agent treatment (fixed payments
for agents), one of three agents made the choice. In the Agent/Choice treatment, the dictator
could choose whether to delegate the choice to an agent in periods 8-12 of the 12-period
session. Their Figure 1 presents the results.

In the Baseline, dictators allocated an average of $2.26 over the first eight rounds,
while principals in the Agent conditions shared significantly less ($1.70); this difference was
considerably larger in periods 5-8 ($2,32 versus $1.11). In the Agent treatments, dictators
demonstrated self-interest, showing a preference for agent’s who made small offers (the
history of each agent was given). When agents gave $0, they were retained by the dictator

24
 Online Appendix

93% of the time; in contrast, when agents gave $5, they were retained 39% of the time. In a
second experiment, agents sent messages to dictators concerning the amount they indicated
(cheap talk) they would give to recipients. Dictators were very sensitive to the
announcements when selecting an agent. Giving was 62% lower in this treatment than in the
Baseline.

In summary, “… acting through agents allows principals to maintain positive

impressions of their own behavior and role in determining outcomes. Agents serve this
function through a subtle interplay of psychological factors. Principals do not feel that they
are behaving unfairly because they do not directly take immoral actions; they simply hire
agents. They also do not feel responsible for the ultimate outcomes.” Thus, once again,
alleviating a sense of responsibility leads to more selfish actions, since one’s self-image
suffers less by using an intermediary. This is therefore a viable strategy in some
environments.

Bartling and Fischbacher (2013) discuss delegation and responsibility in the context
of being blamed. A perceptive quote from Machiavelli in mentioned: “Princes should
delegate to others the enactment of unpopular measures and keep in their own hands the
means of winning favors”. The article considers responsibility attributions in delegated
choices. Participants could choose could “either choose a fair allocation or an unfair
allocation or delegate the choice”, with punishment measuring responsibility attribution. A
key result is that one can effectively shift this attribution using delegation and that this can be
strong motivation for delegating decisions. It is also found that intention and responsibility
matter significantly.

The experimental design involves a dictator game with punishment and delegation
possibilities. In each group, there is a dictator (A), a possible delegee (B), and two recipients
(C). Either the dictator or the delegee chooses to allocate 20 tokens fairly amongst the four
parties (five each) or unfairly (nine each for the dictator and possible delegee and only one
for each of the recipients). When punishment was possible, one of the recipients is randomly
chosen and can assign punishment points by sacrificing one point to reduce the payoffs of
others by up to seven points.

There were three treatments that varied the punishment environment. In the random
treatment, the dictator can delegate the decision between the fair and the unfair allocation to
only a die roll. It is common information that there is a 40% chance that the die chooses the

25
 Online Appendix

unfair allocation matching the observed behavior in D&P). In the asymmetric treatment, the
dictator must either choose the fair allocation or delegate the choice. In the asymmetric
treatment, the dictator can choose the fair allocation or delegate, but he cannot choose the
unfair allocation.

Their Figures 2 and 3 show the punishment results in D&P (left panel) and in the two
treatments described above (right panel). In D&P, the punishment rate when A cannot
delegate and is unfair is slightly lower than when A could delegate and is instead unfair.
When A delegates and B is unfair, B gets the lion’s share of the punishment. The same
pattern holds when the punishment points assigned are instead considered.

There are two main points illustrated in Figure 3. First, A is successfully able to
diminish punishment by delegating. A is blamed when A is unfair, but the rate drops
considerably when A delegates. Compare the punishment rates: about 45% when A does not
delegate to about 16% (the average of the rates for dice fair + dice unfair) when A delegates.
Second, when A delegates and B is unfair, B receives the bulk of the punishment and the rate
for A is about 15%. These Figures make it clear that delegating shifts the blame to the
intermediary. Here, the image is primarily social, although the dictator might also feel better
by delegating than by being unfair.

26
 Online Appendix

Recipients do blame the dictator less for a delegated unfair outcome, but only in the
Choice treatment is the blame directed to the intermediary. Delegation (significantly)
reduces the average punishment by 25%. The results for the dictator are similar in the
Random treatment, where the dictator punishment by around 21%. Dictators hardly being
helpful by foregoing the equal split. Since they bypassed the punishment-minimizing equal
split, it is difficult to argue that they are being helpful, yet they are punished less. They
conclude that their results “illustrate just how easily a person may avoid negative judgment
by delegating.” Of course, this presumes that the dictator manages to escape the knowledge
that she was being selfish by delegating.

27
 Online Appendix

Appendix 9 – The Stability of Social Preferences

In this appendix, we review evidence that examines the extent to which social preferences are
relatively stable. Measuring the stability of social preferences over time appears
straightforward as long as the measurement tools indeed deliver a preference measure and not
merely a behavioral measure that is confounded by beliefs and other types of preferences (as
discussed in the section on external validity), and as long as the measurement tool at different
points in time is identical. In addition, the preference measure is ideally not just based on a
single behavioral measure like the choice of the transfer in a standard dictator game but
instead on many choice situations across which the costs and benefits of the transfer vary.
Otherwise, the recovered preferences contain a lot of measurement errors and noise, which
may generate spurious preference instability.
Measuring social preferences across contexts is trickier because the notion of stability
is theory-dependent. To illustrate this point, consider the behavior of responders in two
versions of the ultimatum game (Blount 1995). In version 1, a random mechanism determines
the first-mover’s offer exogenously while the first-mover herself makes the offer in version 2.
Suppose that the responders are negatively reciprocal but not inequality averse. Then
responders reject low offers in version 2 of the game but not in version 1 because a low offer
does not indicate an unkind intention in version 1 but it does so in version 2 of the game. If
one erroneously assumes that responders are inequality averse, one would conclude that the
responders’ inequality averse preferences are highly unstable because inequality averse
responders should reject low offers regardless of whether they are randomly determined or
volitionally chosen. However, if one correctly assumes that the responders are negatively
reciprocal, their change in behavior across the two games is exactly what a stable preference
for negative reciprocity predicts. Thus, the extent to which one can interpret changes in
behavior across different contexts as changes in preferences is strongly dependent on the
assumption about the underlying psychological mechanism. For this reason, care needs to be
exercised when preference stability is assessed by examining behaviors across contexts.
With the above caveats in mind, what does the evidence on the stability of social
preferences show? Bruhin et al. (2019) estimated the structural parameters twice for
advantageous (’) and disadvantageous (’) inequality aversion in a sample of N = 196
students three months apart with the same experimental paradigm. They found that the
intertemporal correlation of individuals’ ’ is 0.48 while the correlation for ’ is 0.56.

28
 Online Appendix

Fehr®Epper®Senn (2022) also measured individuals’ social preferences in a broad

Swiss sample (N = 415) at two points in time that were three years apart (in 2017 and 2020).
The subjects faced the exact same large set of budget lines which makes it possible to study
preference stability (i) at the level of choice for individual budget lines, (ii) at the level of
individuals’ estimated structural preference parameters and (iii) at the level of individuals’
assignments to different preference types. At the choice level, roughly 55% of the choices are
perfectly identical across time points and 67% of the choices are identical or coincide with
the closest neighboring allocation on the budget line. At the level of individuals’ structural
parameters, they find an intertemporal rank correlation of 0.458 for ’ and 0.428 for ’.
Finally, at the level of type assignment, they find that 68% of the individuals are assigned to
the same preference type (altruistic, inequality averse, selfish) across the two points in time,
and that among the individuals classified as other-regarding (altruistic or inequality averse) in
2017, 89% are again classified as other-regarding in 2020. Among the individuals classified
as selfish in 2017, 60% are again classified as selfish in 2020.
Moreover, two waves of the German Internet Panel implemented the same equality
equivalence Test (Kerschbamer and Muller 2020). In total N = 2583 individuals participated
twice in this test, 2 years apart (2016 and 2018). This permits an analysis of the stability of
individuals’ assignment to four pre-defined preference types (selfish, altruistic, inequality
averse, envious; see Table 2). This analysis shows that 60% of individuals remain assigned to
the same preference type across the two years, and that among the 76% of individuals who
were classified as altruistic or inequality averse in 2016, 84.5% were again assigned to these
two preference types.
Chuang and Schechter (2015) also report significantly positive intertemporal
correlations between 0.21 and 0.32 involving survey measures of negative reciprocity taken
in 2007, 2009 and 2010. Likewise, Carlsson, Johansson-Stenman and Nam report
significantly positive intertemporal correlations of social preference related behaviors
(voluntary money and labor contributions to a natural public good) at four different points in
time spread across six years.

Thus, taken together, the data suggest a reasonable degree of stability in social
preference when measured at the level of choices, structural parameters, or preference type
assignment. However, the data also suggests a non-negligible degree of noisiness and/or
measurement error. Nevertheless, the observed degree of stability appears sufficiently strong
to suggest that workers with different degrees of prosociality may self-select into different

29
 Online Appendix

sectors or to make it worthwhile for employers to screen potential employees based on certain
social preference characteristics.

30
 Online Appendix

References
Battigalli, P., and M. Dufwenberg. "Belief-Dependent Motivations and Psychological Game
Theory." Journal of Economic Literature 60, no. 3 (2022): 833-82.
———. "Dynamic Psychological Games." Journal of Economic Theory 144, no. 1 (2009): 1-
35.
Brock, J. M., A. Lange, and E. Y. Ozbay. "Dictating the Risk: Experimental Evidence on
Giving in Risky Environments." American Economic Review 103, no. 1 (2013): 415-
37.
Bruhin, A., E. Fehr, and D. Schunk. "The Many Faces of Human Sociality: Uncovering the
Distribution and Stability of Social Preferences." Journal of the European Economic
Association 17, no. 4 (2019): 1025-69.
Caplin, A., and J. Leahy. "Psychological Expected Utility Theory and Anticipatory Feelings."
Quarterly Journal of Economics 116, no. 1 (2001): 55-79.
Cappelen, A. W., J. Konow, E. O. Sorensen, and B. Tungodden. "Just Luck: An Experimental
Study of Risk-Taking and Fairness." American Economic Review 103, no. 4 (2013):
1398-413.
Cappelen, A. W., J. Mollerstrom, B. A. Reme, and B. Tungodden. "A Meritocratic Origin of
Egalitarian Behaviour." Economic Journal 132, no. 646 (2022): 2101-17.
Cappelen, A. W., K. Nygaard, E. O. Sorensen, and B. Tungodden. "Social Preferences in the
Lab: A Comparison of Students and a Representative Population." Scandinavian
Journal of Economics 117, no. 4 (2015): 1306-26.
Cettolin, E., A. Riedl, and G. Tran. "Giving in the Face of Risk." Journal of Risk and
Uncertainty 55, no. 2-3 (2017): 95-118.
Chuang, Y. T., and L. Schechter. "Stability of Experimental and Survey Measures of Risk,
Time, and Social Preferences: A Review and Some New Results." Journal of
Development Economics 117 (2015): 151-70.
Fisman, R., P. Jakiela, and S. Kariv. "How Did Distributional Preferences Change During the
Great Recession?" Journal of Public Economics 128 (2015): 84-95.
Fisman, R., P. Jakiela, S. Kariv, and D. Markovits. "The Distributional Preferences of an
Elite." Science 349, no. 6254 (2015).
Fisman, R., S. Kariv, and D. Markovits. "Individual Preferences for Giving." American
Economic Review 97, no. 5 (2007): 1858-76.
Freundt, J., and A. Lange. "On the Determinants of Giving under Risk." Journal of Economic
Behavior & Organization 142 (2017): 24-31.
Gill, D., and V. Prowse. "A Structural Analysis of Disappointment Aversion in a Real Effort
Competition." American Economic Review 102, no. 1 (2012): 469-503.
Kerschbamer, R., and D. Muller. "Social Preferences and Political Attitudes: An Online
Experiment on a Large Heterogeneous Sample." Journal of Public Economics 182
(2020).
Khalmetski, K., A. Ockenfels, and P. Werner. "Surprising Gifts: Theory and Laboratory
Evidence." Journal of Economic Theory 159 (2015): 163-208.
Krawczyk, M., and F. Le Lec. "'Give Me a Chance!' An Experiment in Social Decision under
Risk." Experimental Economics 13, no. 4 (2010): 500-11.
———. "How to Elicit Distributional Preferences: A Stress -Test of the Equality Equivalence
Test." Journal of Economic Behavior & Organization 182 (2021): 13-28.
Li, J., L.P. Casalino, R. Fisman, S. Kariv, and D. Markovits. "Experimental Evidence of
Physician Social Preferences." PNAS 119, no. 28 (2022).
Liebrand, W. B. G. "The Effect of Social Motives, Communication and Group-Size on
Behavior in an N-Person Multi-Stage Mixed-Motive Game." European Journal of
Social Psychology 14, no. 3 (1984): 239-64.

1
 Online Appendix

Liebrand, W. B. G., and C. G. Mcclintock. "The Ring Measure of Social Values -

Computerized Procedure for Assessing Individual-Differences in Information-
Processing and Social Value Orientation." European Journal of Personality 2, no. 3
(1988): 217-30.
Loewenstein, G. F., E. U. Weber, C. K. Hsee, and N. Welch. "Risk as Feelings."
Psychological Bulletin 127, no. 2 (2001): 267-86.
Murphy, R. O., and K. A. Ackermann. "Social Value Orientation: Theoretical and
Measurement Issues in the Study of Social Preferences." Personality and Social
Psychology Review 18, no. 1 (2014): 13-41.
Murphy, R. O., K. A. Ackermann, and M. J. J. Handgraaf. "Measuring Social Value
Orientation." Judgment and Decision Making 6, no. 8 (2011): 771-81.
Neilson, W. S. "Axiomatic Reference-Dependence in Behavior toward Others and toward
Risk." Economic Theory 28, no. 3 (2006): 681-92.
Rohde, K. I. M. "A Preference Foundation for Fehr and Schmidt's Model of Inequity
Aversion." Social Choice and Welfare 34, no. 4 (2010): 537-47.
Saito, K. "Social Preferences under Risk: Equality of Opportunity Versus Equality of
Outcome." American Economic Review 103, no. 7 (2013): 3084-101.
Snowberg, E., and L. Yariv. "Testing the Waters: Behavior across Participant Pools."
American Economic Review 111, no. 2 (2021): 687-719.
Van Lange, P. A. M. "The Pursuit of Joint Outcomes and Equality in Outcomes: An
Integrative Model of Social Value Orientation." Journal of Personality and Social
Psychology 77, no. 2 (1999): 337-49.
Van Lange, P. A. M., W. Otten, E. M. N. DeBruin, and J. A. Joireman. "Development of
Prosocial, Individualistic, and Competitive Orientations: Theory and Preliminary
Evidence." Journal of Personality and Social Psychology 73, no. 4 (1997): 733-46.