Identity Utility and Other-Regarding Preferences – an Experiment
Draft: February 2015
Michael Kurschilgen
Max Planck Institute for Research on Collective Goods, Bonn
kurschilgen@coll.mpg.de
I test the central idea of identity utility (Akerlof and Kranton 2000), i.e. that human choices
reflect a tradeoff between material selfishness and compliance with one’s normative ideal, in the
context of other-regarding preferences (Charness and Rabin 2002). This context is particularly
interesting since the empirically observed heterogeneity of other-regarding preferences, implies,
according to identity utility, that people’s normative ideals will differ even more than their
observed choices. As a consequence, increasing the moral cost of norm deviation should make
choices not only less selfish but also more heterogeneous. My experimental results largely
confirm the theoretical predictions. Interestingly, I find egalitarians to be more willing to deviate
from their stated normative ideal than welfarists. All results also hold both when participants are
uninformed and informed about others’ behavior.
Identity, Other-Regarding Preferences, Introspection, Social Information, Modified Dictator
Game, Experiment
C91, D03, A13
1
1. Introduction
It is a robust finding of behavioral and experimental economics that people not always maximize
their material self-interest. Typical examples include generous giving behavior in dictator games
(Engel 2011, Forsythe et al. 1994) and rejection behavior in ultimatum games (Güth et al. 1982,
Oosterbeek et al. 2004). Two of the most prominent approaches to explain the deviations from
material selfishness have been the concepts of other-regarding preferences (Andreoni and Miller
2002, Bolton and Ockenfels 2000, Charness and Rabin 2002, Fehr and Schmidt 1999) on the one
hand, and identity utility (Akerlof and Kranton 2000, Bénabou and Tirole 2011) on the other.
According to the former, non-selfish choices reflect people’s intrinsic concern for others’
payoffs. In contrast, the latter describes human behavior as a tradeoff between material
selfishness on the one hand and the wish to comply with a certain normative ideal on the other.
A person’s choice thus depends on the relative (moral) cost of deviating from one’s ideal self.
The present paper is the first to study how other-regarding preferences and identity utility
relate to each other. Specifically, I start from the robust empirical observation that otherregarding preferences are heterogeneous (Charness and Rabin 2002, Fisman et al. 2007), not only
in the extent to which they deviate from material selfishness but also in the nature of their
deviation. Whereas some people are willing to forego personal earnings for the sake of higher
social efficiency, others are willing to pay for higher equality. This empirically observed
heterogeneity represents an interesting test-bed for the concept of identity utility: if people’s
deviations from selfishness differ, identity utility implies that their normative ideals will differ
even more. As a consequence, identity utility predicts that increasing the relative cost of deviating
will make choices not only less selfish but also more heterogeneous, as people’s choices move
closer to their respective normative ideals. The present paper tests these predictions.
Experimentally, I elicit participants’ other-regarding preferences with the help of a
modified dictator game (MDG), in which participants are either decider or recipient. In addition,
I make use of the idea of self-reflection or introspection (Krupka and Weber 2009, Smith 1790)
in order to induce an increase of the relative cost of deviating from one’s normative ideal. In line
with the theoretical predictions, I find that choices become less selfish and more heterogeneous,
reflecting even more heterogeneous normative valuations of efficiency vs. equality. However,
egalitarians seem significantly more willing to deviate from their stated normative ideal than
welfarists. The results are robust to providing subjects with information about other people’s
choices.
2
The next section describes the theoretical framework and derives testable predictions.
Section three presents the experimental design and section four reports the experimental results.
Section five concludes the paper.
2. Theoretical Framework
I conceptualize other-regarding preferences following Charness and Rabin (2002):
⎧ ρπ + (1− ρ ) π if π ≥ π
j
i
i
j
⎪
Ui = ⎨
⎪⎩ σπ j + (1− σ ) π i if π i ≤ π j
(1)
The type space is two-dimensional, as depicted in Figure 1. Every person is characterized by her
concern for others when she is richer ( ρ ) and when she is poorer ( σ ). Selfish types, who are
only interested in their own material payoff, have ρ = σ = 0 . Social efficiency orientation is
described by ρ > 0 and σ > 0 whereas inequality aversion is captured by ρ > 0 and σ < 0 .
Other types are theoretically possible but empirically irrelevant.
The model of Fehr and Schmidt (1999) is functionally equivalent to Charness and Rabin
(2002) but places more restrictions on the range of plausible parameter values thus not allowing
for efficiency orientation, which empirically has been shown to be highly relevant. The same
shortcoming applies to Bolton and Ockenfels (2000), who have a slightly different
conceptualization of equality concerns that focuses on relative payoff differences instead of
absolute differences.
Andreoni and Miller (2002) have suggested a more general functional form for otherregarding preferences. Their CES-function not only allows for π i and π j to be perfect
substitutes, as both Fehr and Schmidt (1999) and
Charness and Rabin (2002) assume, but also
Leontief, Cobb-Douglas and many other. However, the increase in generality comes along with
an additional free parameter (i.e. the elasticity of substitution), which makes the empirical
elicitation disproportionately more involved without adding important substance for the purpose
of this paper.
Empirically, other-regarding preferences have been found to differ particularly along two
dimensions: (a) the extent to which a person deviates from material selfishness, i.e. the relative
price someone is willing to pay to attain a certain distributional goal, and (b) the nature of the
deviation, i.e. in which type of situation someone is willing to give up money. This second
empirical regularity makes other-regarding preferences an interesting test-bed for identity utility.
3
Figure 1: Identity utility and other-regarding preferences
Identity utility (Akerlof and Kranton 2000) describes human behavior as a tradeoff
between material selfishness and compliance with a normative ideal:
U ( a ) = π ( a ) − γ D ( a − a! )
(2)
According to this model, a person’s utility increases in her material payoffs π ( a ) but decreases
as her action a deviates from her normative ideal a! . The relative importance of normative
compliance is determined by γ ≥ 0 . The smaller (larger) γ
the closer a person’s utility
maximizing choice a** will be to her selfish optimum a* (to her normative ideal a! ).
In the context of other-regarding preferences, this has some interesting implications,
illustrated in Figure 1. If two players’ choices ( a1** and a2** ) in the two-dimensional parameter
space of Charness and Rabin (2002) differ, not only in their distance from the selfish optimum
a* but in their orientation, this implies that their normative ideals ( a!1 and a! 2 ) will differ even
more. As a consequence, increasing γ should make choices (a) move away from selfish
optimization, (b) move closer to people’s respective normative ideals, and thus (c) be more
heterogeneous along the dimension of normative dissent. Empirically, virtually all participants
display ρ ≥ 0 . In contrast, people have been found to differ substantially along the σ dimension, reflecting differing relative inclinations toward social welfare ( σ > 0 ) or equality (
σ < 0 ).
4
3. Experimental Design
a. Paradigm
Participants play a modified dictator game (MDG) similar to the one of Iriberri and Rey-Biel
(2011). There are two roles: one player is decider, the other is recipient. The decider is confronted
on her computer screen, successively, with four decision panels, each panel consisting of nine
decision tasks (see Figure 2), i.e. a total of 36 tasks. In every task, the decider may choose
between Option A and Option B. Option A is always profit maximizing. By choosing Option B a
decider can either create additional income for the recipient (panels 1 and 2) or destroy parts of it
(panels 3 and 4). In panels 1 and 3 the decider is richer than the recipient whereas in panels 2 and
4 she is poorer.
Ahead
Create
Option A
Option B
Option A
Option B
Task
πi
πj
πi
πj
πi
πj
πi
πj
1
2
3
4
5
6
7
8
9
170
170
170
170
170
170
170
170
170
70
70
70
70
70
70
70
70
70
160
160
160
160
160
160
160
160
160
82
84
88
94
102
112
124
138
154
110
110
110
110
110
110
110
110
110
120
120
120
120
120
120
120
120
120
100
100
100
100
100
100
100
100
100
132
134
138
144
152
162
174
188
204
A
A
Option A
Destroy
Behind
B
B
Option B
A
A
Option A
B
B
Option B
Task
πi
πj
πi
πj
πi
πj
πi
πj
1
2
3
4
5
6
7
8
9
140
140
140
140
140
140
140
140
140
130
130
130
130
130
130
130
130
130
130
130
130
130
130
130
130
130
130
118
116
112
106
98
88
76
62
46
90
90
90
90
90
90
90
90
90
180
180
180
180
180
180
180
180
180
80
80
80
80
80
80
80
80
80
168
166
162
156
148
138
126
112
96
A
A
B
B
A
A
B
B
Figure 2: Decision panels and tasks in the MDG
Player i is the decider and player j the recipient in the MDG. Each of the 4 decision panels consists of 9 tasks in
which the decider chooses between an option A and an option B.
The MDG serves as an instrument to elicit participants’ other-regarding preferences in
the two-dimensional parameter space of the Charness and Rabin (2002) utility function (see
5
Appendix 5). The design of the MDG aims for subjects to make deliberate, well-thought choices.
For that purpose, I deviate from the MDG of Iriberri and Rey-Biel (2011) in two respects: First, I
let deciders choose between two options (Option A: selfish, Option B: destroy or create) instead
of three (Option A: selfish, Option B: create, Option C: destroy). Second, instead of presenting
the tasks randomly, I classify them into four panels and sort them within every panel by the
relative price of creating/destroying.
Figure 3: Cartesian type space of the MDG
Deciders’ choice behavior in the MDG allows categorizing them in the Cartesian type
space depicted in Figure 3. Players that always choose the profit maximizing Option A are
categorized as selfish, which corresponds to the origin of the graph. Every time a decider chooses
Option B over Option A she pays a price of 10 tokens. The further East (West) of the origin a
dot is, the more money a decider is willing to give up in order to create (destroy) income when
she is richer than the recipient. The further North (South) of the origin a dot is, the more money a
decider gives up to create (destroy) income when she is poorer than the recipient. Hence, the
North-East (South-East) extreme of the graph represents the maximum amount of efficiency
(equality) a decider can choose. Dot x in Figure 3, for instance, corresponds to a decider who
6
being richer pays 80 tokens to create income for the recipient while being poorer pays 20 tokes to
destroy income of the recipient. Dot x thus deviates 80+20=100 tokens from Max Profits,
10+110=120 tokens from Max Efficiency, and 10+70=80 tokens from Max Equality.
In order to test the predictions stated in the previous section, I measure for each
treatment (a) the mean distance from Max Profits, (b) the mean distance from Max Efficiency
and Max Equality, and (c) the standard deviation of choices along the σ -dimension.
b. Treatments
I run two sets of treatment comparisons: (1) with uninformed deciders, (2) with informed
deciders. The uninformed set compares behavior in treatments BASE and REFLECT. The
BASE treatment consists simply of the MDG described above. In the REFLECT treatment, after
reading the instructions but before assigning the roles of decider and recipient, subjects are asked
for their moral judgments. Specifically, they have to state for each of the 36 tasks they will
subsequently be seeing in the MDG: “Which of the two Options (A or B) do you find morally
right?” The instructions on the computer screen make it clear that the answers to this question
are not payoff relevant and will not be revealed to other participants. Subsequently, subjects are
assigned their roles and play the payoff-relevant MDG. When playing the MDG, deciders are
reminded on their screens of their own, previously stated, moral judgments.
The idea of the REFLECT treatment is to strengthen participants’ awareness of their own
moral convictions and thus their relative weight, without imposing any specific normative
content, i.e. to increase γ without altering a! . This idea can in fact be traced back to Adam
Smiths concept of introspection, who called for strengthening one’s normative self by becoming
“the impartial spectator of one’s own character and conduct” (Smith 1790). A similar approach is
used by Krupka and Weber (2009), who have subjects deliberate about what others possibly said
one should do (“injuctive focus”). It contrasts for instance with the more intrusive approach of
Dal Bó and Dal Bó (2009), who provide participants with messages that define moral behavior,
and Bicchieri and Xiao (2009), who manipulate dictators’ normative expectations by telling them
what the majority thought should be done.
To test the effect of self-reflection on informed subjects, I run a new baseline, called
INFO. The only difference to BASE is that in INFO deciders are informed that the game has
been run before with more than 100 deciders. On their computer screens they then see for each
7
of the 36 decision tasks the percentage of deciders that chose the selfish Option A in previous
experiments.1
The REFLECT+INFO treatment combines the respective elements of REFLECT and
INFO. First, before knowing their role in the subsequent game, subjects are asked to state what
they believe to be “morally right”, just as in the REFLECT treatment. Thereupon they are
informed that the game has been run before with more than 100 deciders. When playing the
MDG they are reminded on their screens for each of the 36 tasks of both their own moral
judgments and which percentage of previous deciders chose Option A.
In the INFO+REFLECT treatment the order of the two elements is reversed. After
reading the instructions, subjects are informed that the game has been run before with more than
100 deciders. They are then asked to make their moral judgments while seeing on their screens
for each of the 36 tasks which percentage of previous deciders chose Option A. When playing
the MDG they are reminded on their screens for each of the 36 tasks of both their own moral
judgments and which percentage of deciders in the Baseline preferred Option A.
c.
Procedure
The experiment was conducted at the EconLab in Bonn, Germany. 736 subjects were recruited
via email from a pool of more than 5000 people, using the software ORSEE (Greiner 2004), 304
participants for the BASE treatment, 144 for INFO, 96 for REFLECT, 96 for
REFLECT+INFO, and 96 for INFO+REFLECT (between-subject design). Upon arriving at the
laboratory, participants were seated in visually completely isolated cubicles. Experimental
instructions (see Appendix 1) were identical in both treatments. They were handed out in paper
to the participants and read aloud by the experimenter. Participants were then asked to turn their
attention to the computer screens in front of them. The experiment was computerized in ztree
(Fischbacher 2007).
At the end of experiment, the computer randomly picked one decision task per panel for
payoff. The corresponding token amounts from those four decision tasks were added and
changed into Euros (100 tokens = 1 Euro). Participants earned on average € 6 (ca. US$ 8) for
approximately 20 minutes in the lab, which corresponds to about twice the typical student’s
hourly wage.
1
The information is taken from the real choice behavior of deciders in the BASE treatment, see Appendix 4.
8
4. Results
Figure 4 depicts deciders’ choices in BASE and REFLECT. In both treatments, the modal
behavior is pure selfishness, as illustrated by the large black dot in the origin of both graphs. In
BASE, 46 percent of deciders never pick Option B, in REFLECT the share of purely selfish
deciders decreases to 31 percent. On average, deciders in BASE only deviate 37 tokens from the
selfish maximum, nearly doubling the distance to 71 tokens in REFLECT (Mann-Whitney
ranksum test, two-sided, N=159, p=0.005). The largest part of the change away from selfishness
goes into the direction of efficiency, reducing the average distance to Max Efficiency from 152
tokens in BASE to 121 in REFLECT (MW, two-sided, N=159, p=0.010). The distance to Max
Equality is also reduced from 173 to 163 tokens but fails to reach significance (MW, two-sided,
N=159, p=0.160). Heterogeneity along the σ -dimension increases, as predicted, from 32 tokens
in BASE to 43 in REFLECT (Levene’s F-test for equality of variances, two-sided, N=159,
p=0.002).
0
80
0.
0.
0.
0.
.1
42
−2.50
29
60
19
−0.71
13
40
0
20
−0.31
7
−0.17
<−− to destroy
to create −−>
Price paid when behind
80
60
40
20
0
20
40
0
−0
rho
20
1
42
0.
29
0.
0.
0.
.1
−0
.3
−0
.7
.5
−0
−2
19
80
13
−2.50
0
60
7
−0.71
1
40
1
20
−0.31
0
−0.17
0.13
.3
0
40
−0
0
60
80
20
60
0.19
1
0.13
0.29
0
40
.7
60
0.19
80
.5
0.29
0.42
−0
80
−2
0.42
sigma
80
60
40
20
0
20
40
60
Price paid when ahead
<−− to destroy
to create −−>
<−− to destroy
to create −−>
Price paid when behind
sigma
80
Price paid when ahead
<−− to destroy
to create −−>
rho
Figure 4: Choices in BASE and REFLECT
The left (right) graph depicts individual choices in BASE (REFLECT).
For informed players the effect of self-reflection is very similar, see Figure 5. The
deviation from Max Profits increases from 32 tokens in INFO to 62 in REFLECT+INFO (MW,
two-sided, N=108, p=0.010) and 60 in INFO+REFLECT (MW, two-sided, N=107, p=0.006).
The deviation from Max Efficiency decreases from 150 in INFO to 124 in REFLECT+INFO
(MW, two-sided, N=108, p=0.018) and 127 in INFO+REFLECT (MW, two-sided, N=107,
p=0.029). Choices approach Max Equality descriptively in REFLECT+INFO, from a mean
distance of 182 tokens to 169 (MW, two-sided, N=108, p=0.233), and significantly in
9
INFO+REFLECT, from 182 to 163 (MW, two-sided, N=107, p=0.007). σ -heterogeneity
increases from 28 tokens in INFO to 39 in REFLECT+INFO (Levene test, two-sided, N=108,
p=0.012) and 36 in INFO+REFLECT (Levene test, two-sided, N=107, p=0.027).
40
0.13
20
0
0
42
0.
29
19
0.
0.
13
0.
.3
.1
−0
−0
.7
0
80
7
60
−2.50
1
40
−0.71
1
20
−0.31
0
−0.17
<−− to destroy
to create −−>
Price paid when behind
80
60
40
20
0
20
40
60
80
60
0.19
.5
0.
0.
0.
.1
0.
rho
80
0.29
−0
80
42
−2.50
29
60
19
−0.71
13
40
0
−0.31
7
20
1
−0.17
0.42
−2
0
sigma
0
<−− to destroy
to create −−>
Price paid when behind
80
60
40
20
0
20
40
20
.3
42
0.
0.
0.
0.
.3
.7
.5
.1
−0
−0
−0
−2
rho
29
80
19
−2.50
13
60
0
−0.71
7
40
1
−0.31
1
20
0
−0.17
60
80
0
0.13
−0
0
40
1
20
0.19
0
0.13
60
.7
40
80
0.29
−0
0.19
Price paid when ahead
<−− to destroy
to create −−>
0.42
.5
60
−0
80
0.29
−2
0.42
sigma
80
60
40
20
0
20
40
60
Price paid when ahead
<−− to destroy
to create −−>
<−− to destroy
to create −−>
Price paid when behind
sigma
80
Price paid when ahead
<−− to destroy
to create −−>
rho
Figure 5: Choices in INFO, REFLECT+INFO and INFO+REFLECT
The left (middle) [right] graph depicts individual choices in INFO (REFLECT+INFO) [INFO+REFLECT].
Looking at incentivized choice behavior in the MDG, I thus find strong evidence in favor
of self-reflection making choices less selfish and more heterogeneous. Moreover, choices move
both strongly in the direction efficiency and, somewhat less strongly, in the direction of equality.
The effect is virtually identical for informed as for uninformed deciders. My results thus support
the predictions derived from identity utility.
So far, however, I have only worked with two stylized normative ideals: Max Efficiency
and Max Equality. Figure 6 shows what deciders, behind the veil of ignorance, actually state as
“morally right behavior”.2 The large black dots in the North-East and South-East corner of the
graph suggests that the stylized approach may actually not be too far off. In addition, many
participants’ self-stated moral ideal seems to be a weighted average between maximizing
efficiency, equality, and to some extent even own profits. According to their moral judgments, 22
percent can be categorized as inequality averse as they state that is morally right to increase the
recipient’s income when one is richer but to reduce it when one is poorer, 48 percent self-declare
efficiency oriented, and 15 percent are exactly on the horizontal line that distinguishes efficiency
from equality orientation. Notably, 13 percent state it is morally right to be selfish.
2
I pool the moral judgments from REFLECT, REFLECT+INFO, and INFO+REFLECT. This seems conceivable
as there is no significant difference between informed and uninformed deciders, neither with respect to choices, nor
moral judgments.
10
60
80
20
40
0
40
20
0.42
80
0.29
60
0.19
40
0.13
20
0
0
−2.50
80
0.
0.
0.
0.
−2
42
60
29
−0.71
19
40
13
−0.31
0
20
.5
−0 0
.7
−0 1
.3
−0 1
.1
7
−0.17
<−− to destroy
to create −−>
Price paid when behind
sigma
80
60
Price paid when ahead
<−− to destroy
to create −−>
rho
Figure 6: Moral judgments behind the veil of ignorance
Deciders’ moral judgments in REFLECT, REFLECT+INFO, INFO+REFLECT are pooled since there is no
significant difference between them.
How do these moral judgments translate into actual choice behavior, once the veil of
ingnorance is lifted? Identity utility would predict them to become more selfish but not change
their general normative orientation. As Figure 7 illustrates, this is exactly what happens. Each
arrow starts at a player’s moral judgment and points to her actual choice in the MDG. The left
graph shows the behavior of self-declared equality-oriented players, hence all arrows start by
definition in the South-East quadrant. Strikingly, they almost never leave the quadrant. Faced
with a real incentivized choice, equality-oriented players become significantly more selfish. The
distance from Max Profits decreases from 134 to 57 tokens (Wilcoxon signrank test, two-sided,
N=28, p<0.001) just as the distance from Max Equality rises from 45 to 130 tokens (WSR, twosided, N=28, p<0.001). Notably, however, their distance to Max Efficiency stays virtually
unchanged, only minimally decreasing from 165 to 161 (WSR, two-sided, N=28, p=0.871).
The picture is even more striking for the self-declared efficiency-oriented players as
literally none of them leaves the North-East quadrant. Also their choices are significantly more
selfish than their moral judgments. The mean distance form Max Profits decreases from 137 to
94 tokens (WSR, two-sided, N=62, p<0.001) just as the distance from Max Efficiency increases
from 43 to 86 tokens (WSR, two-sided, N=62, p<0.001). Also these players do not revise their
11
general normative orientation as their distance from Max Equality shrinks only minimally from
178 to 174 tokens (WSR, two-sided, N=62, p=0.111).3
80
0.42
80
0.29
60
0.29
60
0.19
40
0.19
40
0.13
20
0.13
20
0
60
80
−2.50
80
rho
0.
0.
0.
42
−0.71
29
60
19
40
13
−0.31
0.
0.
0.
0.
20
0.
−2.50
0
−0.17
42
−0.71
29
40
19
−0.31
13
20
0
−0.17
0
0
0
sigma
0.42
<−− to destroy
to create −−>
Price paid when behind
80
60
40
20
0
80
60
40
20
Price paid when ahead
<−− to destroy
to create −−>
<−− to destroy
to create −−>
Price paid when behind
sigma
0
Price paid when ahead
<−− to destroy
to create −−>
rho
Figure 7: Moral judgments and actual choices
The left (right) graph shows the deciders that self-declared equality (efficiency) oriented. Every arrow starts at an
individual decider’s moral judgment and points at their actual choice.
Comparing Figure 6 with Figure 4 and Figure 5 it is striking how the rather extended type
space of moral judgments translates into a much narrower type space of actual choices. Just as
predicted by identity utility subjects have, independent of their individual conception of a
normative ideal, selfishness as a common denominator. As players trade off compliance with
their individual normative ideal against selfish profit maximization, their actual incentivized
choices become much more homogeneous than their moral judgments behind the veil of
ignorance. Interestingly, however, types differ quite a bit with respect to the price they are willing
to pay for normative compliance. Efficiency oriented players move from the judgment that giving
up 137 tokens would be morally right to actually giving up only 94 tokens. Equality oriented
players, however, declare it morally right to pay 135 tokens and end up paying only 57 tokens to
implement a more equal outcome. This is illustrated in Figure 7 by the fact that the arrows in the
left graph are substantially longer than those of the right graph. The difference between selfdeclared efficiency and equality types is significant (MW, two-sided, N=90, p=0.016).
3
The 15 percent mixed types display a similar pattern, their choices being significantly more selfish than their moral
judgments (WSR, two-sided, N=19, p=0.008) but without becoming more efficiency nor equality oriented. Finally,
of the 17 self-declared selfish deciders, 16 confirmed their moral judgment with an identical subsequent choice.
12
5. Conclusion
This paper has compared two of the most prominent conceptual frameworks to explain
deviations from material selfishness: other-regarding preferences and identity utility. The context
of other-regarding preferences is a particularly interesting test-bed for identity utility since the
empirically observed heterogeneity implies that if people’s actual choices differ, their normative
ideals should differ even more. Thus, increasing the relative cost of deviating should make
choices not only less selfish but also more heterogeneous, moving them closer to people’s
individual normative ideals.
Experimentally, I have elicited participants’ other-regarding preferences in a modified
dictator game (MDG). By means of self-reflection behind the veil of ignorance, I have
experimentally induced a higher moral cost of deviating from one’s normative ideal, both with
uninformed and with informed players.
The results largely confirm the theoretical predictions. Increasing the relative weight of
identity makes choices substantially less selfish and more heterogeneous. Choices also move
significantly closer to the maximum attainable level of efficiency in the game. In contrast, I find
choices to only weakly become more egalitarian even though many participants state strong
moral preferences for equality. The reason seems to be that self-declared egalitarians are more
willing to deviate from their stated normative ideal than welfarists.
My experimental results show that people’s moral judgments about the appropriate
tradeoff between efficiency and equality vary substantially more than their subsequent choices,
the reason being that virtually all participants, independent of their specific normative
convictions, are to some extent selfish. This has an interesting implication. An important part of
social science, most notably the literature on social dilemmas, is concerned with finding ways to
restrain people’s selfishness and encourage their moral responsibility. However, when people
differ with respect to their normative ideals, as participants do in the simple, non-strategic, twodimensional setting of this paper, appealing to morality might actually increase the difficulty of
finding common ground and thus the potential of conflict. Further research should examine
whether, in strategic settings characterized by heterogeneity of normative conceptions, human
selfishness may actually serve a social purpose as a useful coordination device.
13
References
Akerlof, George A. and Rachel E. Kranton. 2000. "Economics and Identity." The Quarterly
Journal of Economics, 115(3), 715-53.
Andreoni, James and John Miller. 2002. "Giving According to Garp. An Experimental Test of
the Consistency of Preferences for Altruism." Econometrica, 70, 737-53.
Bénabou, Roland and Jean Tirole. 2011. "Identity, Morals, and Taboos: Beliefs as Assets." The
Quarterly Journal of Economics, 126(2), 805-55.
Bicchieri, Cristina and Erte Xiao. 2009. "Do the Right Thing: But Only If Others Do So."
Journal of Behavioral Decision Making, 22, 191-208.
Bolton, Gary E. and Axel Ockenfels. 2000. "Erc: A Theory of Equity, Reciprocity and
Competition." American Economic Review, 90, 166-93.
Charness, Gary and Matthew Rabin. 2002. "Understanding Social Preferences with Simple
Tests." The Quarterly Journal of Economics, 117, 817-69.
Dal Bó, Ernesto and Pedro Dal Bó. 2009. "" Do the Right Thing." The Effects of Moral
Suasion on Cooperation," National Bureau of Economic Research,
Engel, Christoph. 2011. "Dictator Games. A Meta-Study." Experimental Economics, 14, 583-610.
Fehr, Ernst and Klaus M. Schmidt. 1999. "A Theory of Fairness, Competition, and
Cooperation." The Quarterly Journal of Economics, 114, 817-68.
Fischbacher, Urs. 2007. "Z-Tree. Zurich Toolbox for Ready-Made Economic Experiments."
Experimental Economics, 10, 171-78.
Fisman, Raymond, Shachar Kariv and Daniel Markovits. 2007. "Individual Preferences for
Giving." American Economic Review, 97, 1858-76.
Forsythe, Robert, Joel L. Horowitz, N.E. Savin and Martin Sefton. 1994. "Fairness in
Simple Bargaining Experiments." Games and Economic Behavior, 6, 347-69.
Greiner, Ben. 2004. "An Online Recruiting System for Economic Experiments," K. Kremer and
V. Macho, Forschung Und Wissenschaftliches Rechnen 2003. Göttingen: 79-93.
Güth, Werner, Rolf Schmittberger and Bernd Schwarze. 1982. "An Experimental Analysis of
Ultimatum Bargaining." Journal of Economic Behavior and Organization, 3, 367-88.
Iriberri, Nagore and Pedro Rey-Biel. 2011. "The Role of Role Uncertainty in Modified
Dictator Games." Experimental Economics, 14, 160-80.
Krupka, Erin L. and Roberto A. Weber. 2009. "The Focusing and Informational Effects of
Norms on Pro-Social Behavior." Journal of Economic Psychology, 30, 307-20.
Oosterbeek, Hessel, Randolph Sloof and Gijs van de Kuilen. 2004. "Cultural Differences in
Ultimatum Game Experiments. Evidence from a Meta-Analysis." Experimental Economics, 7, 17188.
Smith, Adam. 1790. The Theory of Moral Sentiments, or, an Essay Towards an Analysis of the Principles
by Which Men Naturally Judge Concerning the Conduct and Character, First of Their Neighbours, and
Afterwards of Themselves. To Which Is Added, a Dissertation on the Origin of Languages. London: Strahan.
14
Appendix 1: Experimental Instructions
General Information
Welcome to our experiment!
If you read the following explanations carefully, you will be able to earn a substantial sum of money, depending
on the decisions you make. It is therefore crucial that you read these explanations carefully.
During the experiment there shall be absolutely no communication between participants. Any violation of this
rule means you will be excluded from the experiment and from any payments. If you have any questions, please
raise your hand. We will then come over to you.
During the experiment we will not calculate in euro, but instead in tokens. Your total income is therefore initially
calculated in tokens. The total number of tokens you accumulate in the course of the experiment will be
transferred into Euro at the end, at a rate of
100 tokens = 1 Euro.
At the end you will receive from us the cash sum, in euro, based on the number of tokens you have earned.
The Experiment
In the experiment, there are two roles: decider and recipient.
At the beginning of the experiment you will be randomly allotted one of the two roles. One half of the
participants will be deciders, the other half will be recipients. During the entire experiment, you will remain in
the same role.
On your computer screen you will be shown 4 tables, one after the other. Every table consists of 9 decision
tasks.
A decision task could for example read as follows:
Option A
Option B
Decider (You)
12
10
Recipient
5
7
Your decision (A or B):
In every decision task the decider has to choose between Option A and Option B. The two options define how
many tokens the decider gets and how many the recipient gets.
In this example the decider gets 12 tokens and the recipient 5 tokens if the decider chooses Option A. If the
decider chooses Option B, the decider gets 10 tokens and the recipient 7 tokens.
In every decision task the computer will randomly match every decider with a different recipient. Thus the
decider-recipient pairs change in every decision task.
The decider will never know the identity of the recipient.
The recipient will never know the identity of the decider.
15
At the end of every table please press the “OK” button on the lower right hand side of your screen. Only after
pressing “OK” your decisions are saved and become effective. You will then be shown the next table.
Payoffs
At the end of experiment the computer will randomly pick one decision task out of every table. The computer
thus picks in total 4 decision tasks, one from every table. The corresponding token amounts from those 4
decision tasks will be added and changed into Euros.
If you are decider, your payoffs only depend on your own choices and on the random draw at the end of the
experiment.
If you are recipient, your payoffs only depend on the choices of the corresponding decider and the random draw
at the end of the experiment.
16
Appendix 2: Additional screen in REFLECT treatment
Before the computer randomly determines who will be Decider and who will be Recipient, we would like to
know your opinion.
We would like to know from you:
Which of the two Options (A or B) do you find morally right?
The answers to these questions will be kept anonymous. No other participant will get to know them at any time.
Your answers to these questions are not relevant for your payoffs.
Appendix 3: Additional screen in INFO treatment
This Experiment has been run before with more than 100 Deciders.
In the column on the right hand side of your screen you can see how the Deciders in those previous Experiments
decided. Specifically, you will be shown which percentage of Deciders chose Option A or Option B in the
corresponding Choice Task.
Appendix 4: Information about choices of previous players (by panel)
Ahead
Ahead
Behind
Behind
Create
Destroy
Create
Destroy
1
89% chose A
95% chose A
91% chose A
95% chose A
2
89% chose A
95% chose A
92% chose A
91% chose A
3
87% chose A
97% chose A
89% chose A
95% chose A
4
83% chose A
98% chose A
88% chose A
90% chose A
5
76% chose A
99% chose A
84% chose A
90% chose A
6
68% chose A
97% chose A
76% chose A
89% chose A
7
64% chose A
97% chose A
74% chose A
91% chose A
8
58% chose A
97% chose A
68% chose A
90% chose A
9
53% chose A
97% chose A
67% chose A
89% chose A
17
Appendix 5: Logic of the Modified Dictator Game (MDG)
Player i prefers allocation B to allocation A iff U iB ≥ U iA . Assuming Charness and Rabin (2002)
preferences and π i ≥ π j this implies:
(
(
π iB − π iA ≥ ρ (π iB − π iA ) − π Bj − π Aj
))
(A1)
which, for convenience, I rewrite as:
(
Δi ≥ ρ Δi − Δ j
)
(A2)
The same argument applies to π i ≤ π j by simply replacing ρ with σ . Assuming Δ i < 0 (i.e.
allocation B is less profitable to player i than allocation B) a person’s choice reveals her ρ and σ
parameters as depicted in Table A1.
Table A1: Parameter Space of the MDG
πi ≥ π j
πi ≤ π j
Δi < Δ j
ρ≥
Δi
>0
Δi − Δ j
σ≥
Δi
>0
Δi − Δ j
Δi > Δ j
ρ≤
Δi
<0
Δi − Δ j
σ≤
Δi
<0
Δi − Δ j
Table A2 illustrates how the experimental MDG devotes one decision panel to each of
these four situations. In the two panels on the left, the decider’s payoff is always higher than the
recipient’s ( π i ≥ π j ) whereas in the two right-hand panels it is the other way around ( π i ≤ π j ). In
the two upper panels the decider can create income for the recipient ( Δ i < 0 < Δ j ) whereas in the
two lower panels she can reduce ( 0 > Δ i > Δ j ).
In each panel, there are nine decision tasks. In each task the decider has to choose
between Option A and Option B, specifying two different payoff allocations for the decider and
the corresponding recipient. Option A is the same for every task within a given panel. Option B
creates or destroys income of the recipient at a cost of 10 tokens. Take for example task 1 of the
Ahead-Create panel. If the decider chooses Option A she receives 170 tokens and the recipient
70 tokens and if she chooses Option B she gets 160 and the recipient 82.
18
Ahead
Behind
Destroy
Create
π A = 170 π A = 70
i
j
π A = 110 π A = 120
i
j
Task
πB
i
πB
j
Δi
Δj
ρ≥
πB
i
πB
j
Δi
Δj
σ≥
1
2
3
4
5
6
7
8
9
160
160
160
160
160
160
160
160
160
82
84
88
94
102
112
124
138
154
-10
-10
-10
-10
-10
-10
-10
-10
-10
12
14
18
24
32
42
54
68
84
0.45
0.42
0.36
0.29
0.24
0.19
0.16
0.13
0.11
100
100
100
100
100
100
100
100
100
132
134
138
144
152
162
174
188
204
-10
-10
-10
-10
-10
-10
-10
-10
-10
12
14
18
24
32
42
54
68
84
0.45
0.42
0.36
0.29
0.24
0.19
0.16
0.13
0.11
Task
π A = 140 π A = 130
i
j
B
B
π
Δi
π
j
i
Δj
σ≤
1
2
3
4
5
6
7
8
9
130
130
130
130
130
130
130
130
130
-12
-14
-18
-24
-32
-42
-54
-68
-84
-5.00
-2.50
-1.25
-0.71
-0.45
-0.31
-0.23
-0.17
-0.14
118
116
112
106
98
88
76
62
46
-10
-10
-10
-10
-10
-10
-10
-10
-10
Δj
ρ≤
-12
-14
-18
-24
-32
-42
-54
-68
-84
-5.00
-2.50
-1.25
-0.71
-0.45
-0.31
-0.23
-0.17
-0.14
π A = 90 π A = 180
i
j
B
B
π
Δi
π
j
i
80
80
80
80
80
80
80
80
80
168
166
162
156
148
138
126
112
96
-10
-10
-10
-10
-10
-10
-10
-10
-10
Note: To ensure that stakes are comparable across panels, every panel has approximately (i.e. constrained on only
1 9
using integers) the same average pie size P = ∑ π i,tA + π Aj,t + π i,tB + π Bj,t . Ahead-Create has 254 tokens, Ahead18 t=1
Destroy 246, Behind-Create 244, and Behind-Destroy 246.
In each panel the relative price of creating/destroying decreases with every task. In task 1,
the decider has to give up 10 tokens to create/destroy 12 tokens whereas in task 9 for the same
cost the decider creates/destroys 84 tokens. Consequently, choosing Option B in task 1 and
Option A in task 2 of the same panel would violate the General Axiom of Revealed Preferences
(GARP). In the MDG, a GARP-consistent decider should have at most one switch from Option
A to Option B per panel, and no switch from B to A. In addition, consistency requires players
not to both create and destroy when they are ahead (or behind). If these consistency
requirements are met, the ρ and σ of a given decider are defined by the point at which she
switches from Option A to Option B.
For example, a player who chooses Option A in the first 3 tasks of the Ahead-Create
panel and Option B in the remaining 6 tasks, would have 0.36 > ρ ≥ 0.29 . The same player might
then for instance choose always Option A in the Ahead-Destroy panel and in the Behind-Create
19
panel but then switch to Option B in task 7 of the Behind-Destroy panel. This would yield
−0.31 > σ ≤ −0.23 . The type classification is straightforward: Selfish players will never choose
Option B since this is costly. Efficiency oriented players will create income for the recipient both
when ahead and when behind as long as the relative price of creating is low enough. Equality
oriented players will also create when ahead but destroy when behind. Competitive types will
destroy recipients’ income no matter whether they are ahead or behind.
20