You are on page 1of 69

Why are we inequality averse?


Marcelo Bergolo1 , Gabriel Burdin2 , Mauricio De Rosa1 ,
Santiago Burone1 , Matı́as Giacobasso1 , and Martin Leites1
1 Universidad de la República (Uruguay)
2 Leeds University

February 1, 2019

Abstract
An overview of the different economic models available and recent
empirical findings on this topic suggest an overlapping between differ-
ent strands of the literature about the notion of inequality aversion.
As a result, there are different approaches and alternative methods
to measure inequality aversion, which makes it more difficult to reach
an agreement about the relevance of inequality aversion and to un-
derstand why individuals care about income distribution. In order to
explain the reasons why people might be affected by income inequal-
ity, we use an experimental survey design, carried out on a sample
of undergraduate students. This paper estimates income inequality
aversion and explores its relationship with individual notions of fair-
ness, differentiating between the role attributed to luck and effort on
income distribution outcomes. Furthermore, it explores whether a
higher chance of income mobility reduces individual inequality aver-
sion or increases it due to risk aversion. Finally, we address whether
individual’s inequality aversion is elastic to their position in the in-
come distribution.

This is an extremely preliminary draft. Please, do not cite or distribute

1
1 Introduction
Experimental economists have recently provided evidence that systematically
refutes the self-interest hypothesis and supports that individual behaviour is
strongly motivated by social preferences, concerns for fairness and reciprocity
(Charness and Rabin, 2002; Fehr and Schmidt, 2003; Clark and D’Ambrosio,
2014). Based on these foundations, there is agreement in the empirical litera-
ture that individual well-being depends on others’ individual income. People
dislike income inequality, defined as any disparities in income between in-
dividuals (Clark and D’Ambrosio,2014; Hopkins, 2008; Fehr and Schmidt,
2003; Heffetz and Frank, 2011). Nevertheless, there is a wide variety of the-
oretical interpretations of these results, which seek to explain why people
are concerned about the income of others or the income distribution. For
example, the original model of Fehr and Schmidt (1999) assumes that agents
dislike others having more (envy effect), but also dislike others having less
(compassion effect). However this latter effect may be opposite when the
pride effect dominates and agents like others having less than them. On the
other hand, some models suggest inequality aversion due to altruism, fair-
ness, risk aversion or social welfare (Alessina and Giuliano, 2009; Clark and
D’Ambrosio; 2014, Harsanyi, 1955; Amiel et al,1999 ).
Social preferences offer a framework to understand inequality concerns or
relative income concerns. However, there are different micro foundations to
explain these social preferences: altruism, envy, pride, fairness, competitive-
ness, etc. These micro foundations can have opposite signs on individual’s
utility and may compensate each other. Economic models have considered
these effects with different emphasis to explain empirical evidence. Using
alternative behavioural assumptions, these models can explain some results
found in empirical literature, but might be inconsistent with others. As a
consequence, it seems hard to reach an agreement about how relative or
inequality concerns should be modelled. Finally, which of these effects dom-
inates is key to determine the micro foundations of inequality aversion.
In order to explain the reasons why people might be affected by income
inequality, in this paper we focus on a particular notion of social preferences:
inequality aversion. An overview of the different economic models available,
and recent empirical findings on this topic, suggest an overlapping of differ-
ent strands of the literature about the notion of inequality aversion (Clark
and DAmbrosio; 2014). As a result, there are different approaches and alter-
native methods to measure inequality aversion, which makes it more difficult

2
to reach an agreement about the relevance of inequality aversion and to un-
derstand why individuals care about the distribution of income in a society.
Clark and D’Ambrosio (2014) distinguish between two broad groups of no-
tions to explain individual inequality attitudes: the Comparative approach
and the Normative approach. While the first establishes that individual
attitudes to income distribution in a society are self-centered, the second
evaluates the overall degree of income inequality in a society or in social
groups. Under the comparative notion, individuals care about their absolute
income, their position in the distribution and how much richer or poorer they
are compared to others. As a result, individual attitudes towards inequal-
ity depend on their income relative to others and their own position in the
income distribution, meaning that any changes in her position affects her
degree of inequality aversion, even if the overall inequality keeps unchanged.
On the other hand, in the normative approach individual attitudes towards
inequality are irrespective of where they appear on the distribution. In the
second case, an individual might care about income inequality even though
they do not participate in this society.
Alesina and Guliano (2010) suggest a useful model to discuss the foun-
dations of the normative approach. The authors distinguish two ways in
which inequality can be included in the utility function: directly and indi-
rectly. The direct channel assumes that individuals have views about social
justice, and they compare the income distribution with ”the desirable level
of income inequality” according with their notion of social justice. In this
case, the distance between the ”desirable” and the real income distribution is
directly included in the utility function. Two main implication suggests this
model. First, individuals dislike income inequality per-se, even beyond how
it affects their own income or well-being. In this sense, this model is deeply
related with the normative inequality aversion approach. Second, in a similar
context individuals will have different degree of tolerance towards inequality
as long as they have a different sense of fairness. Alesina and Guliano (2010)
advance in this sense, and suggest a model in which individuals consider two
components to explain income inequality: effort and luck. If individuals have
preferences for meritocracy, they will be reluctant to accept higher levels of
inequality that arise as a result of luck and not because of differential effort.
On the contrary, they will be less inequality averse when inequality rises as
a result of a relatively larger effort of one group compared to other. On
the other hand, the indirect channel assumes that income inequality can be
instrumentally valuable. This is the case when income inequality determines

3
negative externalities with potential consequences on the aggregate level of
social well-being. Alesina and Giuliano (2010) mention some mechanisms
to explain that negative externalities, as the investment in human capital,
the quality of institutions or levels of crime and violence in society. On the
other hand, inequality may improve social welfare if it improves individual
incentives. Alesina and Giuliano (2010) discuss the “incentive effects” and
suggest that inequality generates incentives for hard work mainly for people
below the top of income distribution.
An additional normative mechanism to explain the tolerance to income
inequality is related with perceptions about income mobility. The level of
income inequality in a society may be perceived as a less serious problem if
society offers higher chances of income mobility. A simple explanation is that
in this situation, the income dispersion across individuals in the long term
will be less than in the short term (Shorrocks, 1978; Jäntti and Jenkings,
2015 ). However, higher income inequality and mobility also increase the
chance for an individual to reach a lower position in the income distribution.
In this case inequality may be perceived as a higher individual risk, because
income movement over time represents unpredictability (Jäntti and Jenkings,
2015). As a result, individuals dislike income inequality because they are risk
averse (Johansson-Stenman et. al., 2003).
Note that perceptions about social mobility as well as the direct or in-
direct inclusion of inequality in the utility function as presented in previous
paragraphs all belong to the normative approach to explain individual at-
titudes towards inequality. They all represent peoples’ tastes for aggregate
income inequality and not from a relative perspective of the individual’s po-
sition compared to others. Namely, these foundations of inequality aversion
are not framed within the comparative approach.
The idea that these channels play a crucial role in explaining individuals
inequality aversion has a long story in social sciences and particularly in
economic models. Findings from behavioural economics and preferences for
redistribution studies indirectly support this idea. However, the extent to
which these channels affect the magnitude of inequality aversion has received
limited attention in empirical economic literature. The main contribution
of this paper is to fill this gap by providing new evidence on the micro-
foundation of inequality aversion.
To this end, first, we estimate inequality aversion parameter from the in-
dividual’s willingness to pay for lower income inequality in society. Second,
we analyse whether inequality aversion is related to the individual’s notion

4
of fairness using a strategy which allow us to distinguish between inequality
caused by luck or inequality caused by effort. Third, we also explore the role
of income mobility in terms of its interaction with income inequality aversion.
We test whether a higher chance of income mobility reduces individual in-
equality aversion or increases it due to risk aversion. Fourth, we also explore
if inequality aversion is related to the comparative notion and we analyse
whether the individual’s degree of inequality aversion depends on their po-
sition in the income distribution. Finally, how normative and comparative
components interact with each other and affect the magnitude of inequality
aversion is evaluated.
To this end, we use an experimental survey design, which is an adapta-
tion of Amiel and Cowell (1992). A sample of students from Universidad de
la Republica (UDELAR) in Uruguay, took part in the main experiment, in
which participation was voluntary and there was no show-up fee paid. Par-
ticipants were asked to choose, among several pairs of comparable societies
(without uncertainly), in which they would prefer their grandchild to live.
In order to explore the causal link between inequality aversion and its al-
ternative explanatory mechanisms, we implement a randomized information
treatment through which we aim to provide alternative hypothetical situa-
tions to the respondents. We elicited the respondents’ inequality aversion,
telling them that income inequality is only determined by effort or by luck.
An additional treatment provide to the respondents information that inequal-
ity is in part compensated by the possibility of income mobility. Importantly,
the treatments make no mention of social policies. Finally, in all treatments
the same individual is asked to choose between societies when her grandchild
is alternatively at the mean, at the maximum and the minimum of income
distribution. The questionnaires also contained several questions regarding
participant’s opinions, attitudes and preferences. This information, together
with the socioeconomic background of students provided by the University,
will be used to explore which mechanisms explain inequality aversion.
One of the main differences of this experiment compared to previous ones
is that we introduce a specific message to inform participants the reasons that
drive income inequality in the hypothetical society: effort or luck”. Another
difference with previous literature is that we also introduce different treat-
ment arms to vary the position in the income distribution as a mechanism
to identify the role of the comparative notion of inequality aversion.
Our results provide several findings that we believe are relevant to un-
derstand the foundations of inequality aversion relative to existing literature.

5
First, we find that inequality aversion of respondents is very elastic to the
notion of fairness. Second, income mobility is positively correlated with the
inequality aversion magnitude, which is interpreted as risk aversion. Finally,
the individual’s willingness to pay for lower income inequality in the whole
society is very elastic to the position of the individual in the income distri-
bution. We include a set of tests to understand why people care about these
two notions of income inequality.
This paper contributes evidence about importance of distinguishing be-
tween alternative foundations of inequality aversion. This evidence is a key
issue to design better tools in order to obtain a precise measure of inequality
aversion and to understand the individual norms and ethical preferences. The
distinction between the foundations of inequality aversion help us to better
understand people behaviour and social preferences. In this sense, a better
understanding of the micro-foundation of individual’s inequality aversion is a
matter of great importance to explain the support for re-distributive policies
as well as individuals responses to alternative policy designs. For example,
Aronsson and Johansson-Stenman (2016) prove that the structure of efficient
marginal income taxation is very sensitive to the exact notion of inequality
aversion that is assumed. Furthermore, the alternative notions of inequality
aversion generates very different incentives. For example relative wages is an
important determinant of job satisfaction and workers effort. More gener-
ally, the magnitude of inequality aversion parameter is key input in order to
address well-being evaluation and to measure the cost of income inequality
in social welfare function. However, as Amiel et al (2015) highlight, there is
not a simple connection between students’ abstract preferences for inequality
and their implications for public policy, which depends not only on ethical
preferences, but also on the causal factor driving the income distribution and
income mobility.
The relationship between inequality aversion and microfundations of in-
come distribution is also important for public policy. In this sense, evidence
about the magnitude of inequality aversion parameters and its sensibility to
alternative income distribution foundations, are key information to define the
optimal tax rates. More in general, better understanding of this issues could
help to answer if public policy should pursue policies that promote equality
rather than those which pursue equality of opportunity and higher mobility.
In the policy design to reduce income inequality, what is the relevance that
should be given to the meritocracy.
The rest of the paper is organised as follows, first we present the theo-

6
retical model, wue postulate the main hypotheses and review the previous
evidence about them ( section 2). The experimental design is presented in
section (3) and the data in section (4). Finally, we present the main results
(section 5) and some final comments in section (8).

2 Theoretical model
In order to explore the determinants of inequality aversion parameter, we
begin with a general model. For the individual i, who lives in the society j
we assumed that the individuals’ utility function can be written as:

uij (xij , Φj , γ̂j ) = h xij Φ−γ̂ij


 
(1)

with
γ̂ij = g(Fij , Pij , Mij , Tij ) (2)
where h is a function which defines any monotonically increasing transforma-
tion, xi is the individuals’ i income and Φ is a measure of income inequality
in society j. γ is a parameter of individual inequality aversion, which could
be interpreted as a constant inequality elasticity, given F , M , P and T . Note
that, when γ = 0 individuals do not care about income inequality and the
utility function corresponds to the conventional case. However, when γ < 0
individuals are inequality lovers. When γ > 0 individuals dislike inequality,
and they are willingness to pay for living in a more equal society and, in fact,
when γ > 1 it implies that a 1 % decrease in income inequality produces
more utility than a 1% increase one’s income.
Furthermore, the magnitude of γ varies between individuals, in which g
is a scalar function whose arguments are the notion of Fairness of the indi-
vidual i for the society j (Fj ), the individual beliefs about the opportunities
of income mobility in the society j (Mij ) and the individual i’s position in
the income distribution (Pij ). Finally, Tij reflects the heterogeneity in tastes
and we assume that it captures the indirect effect of income inequality ex-
ternalities on utility via consumption. To simply, we assume that Tij = 0
and we will focus on the first three arguments of γ̂ij , which will be explained
below.
F airness In order to consider the direct effect of income inequality
on individual utility, Alesina and Giuliano (2010) assume that individuals’
preferences about an acceptable level of inequality are mediated by a sense

7
of what is ”fair” and ”unfair”. The overall income inequality is decomposed
into the contribution of inequality derived from individual effort (Φe ) and luck
(Φl ). Individuals have heterogeneous preferences over both components:
ud ij = γie [Φej − Φe∗ 2 l l l∗ 2
ij ] − γi [Φj − Φij ] (3)
where ud captures how inequality directly affect the utility function, Φe,l
represents the income inequality induced by the source e or l, and Φe,l∗ rep-
resents the ideal level of inequality induced by that sources.1
For reasons of justice people might think that an individual is more en-
titled to retain the sources of its effort than income acquired by chance.
Namely, Φe∗ > Φl∗ . Furthermore, individuals might weight more deviations
from ideal for one or the other type of inequality. Alesina and Giuliano sug-
gests that γ e > γ l . These inequalities are anchored on two fundamental prin-
ciples of equality of opportunity: the compensation and reward principles.
The model rests on the idea that inequality due to circumstances beyond
individual control is ethically unjustified (Principle of compensation), while
inequalities due to differences in efforts are ethically legitimate (Principle of
reward). As a result, this model opens the possibility that the tolerance for
income inequality varies across individuals and in part is determined by the
origin of inequality. To simply, we rewrite the equation (3), in order to dis-
tinguish the role of inequality caused by luck and inequality caused by effort.
We define eij as

eij = [Φej − Φe∗ l l∗


ij ]/[Φj − Φij ] (4)
where eij is the ratio that represents the relationship between the unfair
income inequality that is due to effort and the unfair income inequality that
is due to luck. We could rewrite the equation (2) as follows

γ̂ij = g(eij , Mj , Pij ) (5)


∂γ̂
We expect that < 0. That sign would be capturing two effects: the
∂ej
difference between Φe and Φl and the difference between Φe∗ and Φl∗ .
M obility The second argument of eq (2) captures the individual beliefs
about income mobility (higher M reflects higher income mobility. On the
1
Note that we do not consider here the indirect effect of inequality on utility, since the
authors assume that externalities (as crime and education) depend on global inequality
rather than its components.

8
one hand, previous literature supposes that individual inequality aversion is
determined by the degree to which they are risk averse. This idea is related
with the degree to which marginal utility of income decreases with income,
namely with the concavity of utility function (Harsanyi, 1955). In other
words, behind a veil of ignorance, individuals would be willing to sacrifice
their own expected income in order to reduce inequality. Under uncertainty,
a more equal income distribution reduces the risk of being in a low position
in the distribution. On the other hand, income mobility could be socially
desirable because it is related with equality of opportunity (Jantti and Jenk-
ings; 2015, Amiel et. al.; 2015). Moreover, higher income mobility reduces
long term inequality. “Changes in relative incomes still tend over time to
equalise the distribution of total income receipts, and to this extent welfare
is improved” (Shorrocks; 1978a, pp. 392–393). If that is true, then higher
income inequality would be a greater problem when society presents greater
income rigidity and each individual stays in the same position at income
distribution period after period. If people consider that income mobility is
desirable because it partially compensates income inequality, a higher chance
of income mobility would be related with a lower inequality aversion param-
eter. These ideas suggest an ethical trade-off in preferences for equality and
mobility. However, higher income mobility means greater level of uncertainty
as individual income may vary between periods. As a result, the expected
∂γ
sign of the partial derivative is ambiguous and > or < 0.
∂Mj
P osition
Finally, we consider the relevance of the comparative notion of inequality
aversion on the attitudes towards income inequality.
We assume that the individual marginal sensitivity of the utility func-
tion to income inequality depends on the individuals position in the income
distribution. This mean that any changes in individual’s position affects
its degree of inequality aversion. Individual’s willingness-to-pay in order to
avoid income inequality is lower for those who face relative deprivation than
for those who have a positive relative income. This result suggest that in-
come inequality could be perceived as a luxury good. As a result we expect
that
∂γ
>0
P̃i,j
Following Carlsson et al (2005) a utility-maximizing respondent would be
indifferent between A and Bi, , when

9

ln(xi,A /xi,B )
γ̂i,j (ej , Mj , Pi,j ) =
ln(ΦA /ΦB )

0
e‘j ,Pi,j
‘ ,M
j

(6)

where 0 represents the individual perceptions about the magnitude of these


parameters.
An alternative approach to consider the comparative role of income in-
equality is parameterize the self-centered inequality aversion (Aronsson and
Johansson-Stenman, 2016) . In this case, we rewrite (1) as

Uij (xij , Φj , γ̂j , β̂ij , α̂ij ) = (xij )[RD]−α̂ [RA]−β̂ (Φj )−γ̂ (7)
where R x 
jmax
 xij (x−xji )f (xj ))dx if xij < xjmax
xij
RD=
1 if xij = xjmax

 R xij 
 xjmin (xij −x)f (xj )dx if xij > xjmin
xij
RA =
1 if xij = xjmin

and f (xj ) is the density function of the income distribution in the so-
ciety j. That parametrization is inspired in the inequity aversion model of
Fehr and Schmidt (1999). It assumes that utility depends on the difference
between individual’s own income and that of others and it distinguishes be-
tween relative affluence (RA) and relative deprivation effects (RD). Observe
that α is the weights on the average of other people’s incomes that are above
individual i and β is a weight on the average of incomes below i.
∂γ̂
This approach assume that = 0 . And the equation (2) is rewrit-
P̃i,j
ten as

  h i h i
XmaxB − xi,B XmaxA − xi,A
ln((xi,A /xi,B )) − α̂ + β̂ ln xi,B
/ xi,A

ˆ˜γi,j (ej , Mj , β̂, α̂) =
ln (ΦB /ΦA ))

0
e‘j ,x‘i,j ,Mj ,β̂ 0 ,α̂0

(8)

10
where 0 identify the magnitudes of these parameters for the individual i
and XmaxJ represents the maximum income in the society j.

2.1 Hypotheses and previous evidence


This section formulates the hypotheses to be tested and summarised the
previous evidence about them. We test three basic hypotheses about what
determines the inequality aversion parameter.
Hypothesis I: Fairness hypothesis. An individual’s valuation about an ac-
ceptable level of inequality depends on her views about what is fair or unfair.
The first hypothesis suggests that inequality aversion parameter is higher
when income distribution in a society is determined by luck rather than effort.
H I implies that, γ̂i,j (e∗j , Mj , Pi,j ) < γ̂i,j (e ∗ ∗j , Mj , Pi,j ), when(e∗j > (e ∗ ∗j ).
This hypothesis is related with Alesina and Giuliano (2010), who suggest
that individuals have defined preferences over the two components for a sense
of fairness: effort and luck. There is a lot of empirical evidence that people
have concerns regarding equality and fairness (see e.g. Fehr and Fischbacher;
2002, Fehr and Schmidt; 2003; Alesina and Giuliano, 2010). However there is
little evidence about what channel is more relevant to explain the magnitude
of inequality aversion: effort or luck.
Hypothesis II: Risk aversion vs preferences for mobility. The second hy-
pothesis establishes two competitive explanations for inequality aversion: in-
dividuals are risk averse or they prefer income mobility. On the one hand, em-
pirical evidence in support of risk aversion effect was provided by Johansson-
Stenman et al. (2002) and Carlsson et al (2005). Their results confirm
individuals’ preferences towards both risk and inequality in the same direc-
tion in society. On the other hand, Amiel et. al. (2015) provides novel
evidence about the existence of a trade-off in preferences for equality and
mobility based on a questionnaire experiment specifically designed to this
end. They confirm that people value mobility, but this is not enough to com-
pensate preferences for income inequality. Namely both income mobility and
equality are good socially desirable. They conclude that an equality-mobility
trade-off arise when some inequality is necessary for greater mobility, which
is interpreted by the author as consistent with an ethic of meritocracy.
Given M ∗j > (M ∗ ∗j ), the effect risk aversion dominates over the ef-
fects preferences for income mobility when , γ̂i,j (ej , M ∗j , Pi,j ) > γ̂i,j (ej , M ∗
∗j , Pi,j ).

11
Hypothesis III:Increasing sensitivity of income inequality. The individ-
ual marginal sensitivity of the utility function to income inequality depends
on the individuals position in the income distribution. This hypothesis
postulates that the negative effect of income inequality for richer individ-
s
uals on utility is larger than for poorer ones. Namely, γ̂i,j (ej , Mj , Pi,j ) >
s0 s s0
γ̂i,j (ej , Mj , P < i, j)whenPi,j > P i,j .

3 Experimental Design
Testing whether people value society’s income equality and exploring why
they are willing to pay for less inequality in an abstract way is not sim-
ple. Amiel et. al. (2015) emphasise that empirical strategies based on field
data face mainly two problems: they can not directly considering individ-
ual preferences; or they have not all necessary controls in order to test the
hypothesis about social preferences. In order to address these problems al-
ternative strategies, increasingly used in empirical economics, are based on
laboratory experiments or experimental surveys.
This paper address to the second strategy but combining two approaches.
First, it uses an experimental survey design in which individuals choose be-
tween hypothetical societies. These elections allow us to measure individual’s
inequality aversion. Second, this paper uses an experimental approach, which
allows us to test a causal relationship by using different variants of the exper-
imental survey in randomly selected samples of participants. Basically, we
manipulate relevant information about the hypothetical societies and their
income distribution foundations.
The experiment consisted of six modules answered by all respondents: (i)
the inequality aversion experiment at the mean of income distribution (ii)
the inequality aversion experiment at the bottom of income distribution (iii)
the inequality aversion experiment at the top of income distribution (iv) an
attention check question (v) background questions on socioeconomic status
(vi) individual attitudes, demographics, political beliefs and preferences for
redistribution.
The randomized information treatment is described in detail in section
3.2, and it focuses on the modules related with the inequality aversion ex-
periment. The respondents were given information and instructions in the
digital questionnaire.

12
3.1 Elicitation of inequality aversion parameter
For the inequality aversion experiment we adapted the strategy used in Amiel
and Cowell (1992) and Carlsson et. al. ( 2005). We follow Carlsson et. al.
( 2005) and Aronsson and Johansson-Stenman (2016) assuming individual’s
utility function can be written as (1)
First we assume that Φ is the coefficient of variation, which is defined
σ
as: Φ = |x| . There are alternatives measures of inequality but the coefficient
of variation allow us comparability with previous papers. Furthermore, it
presents some advantages: is symmetric, scale-invariant and satisfying the
principle of transfers (Lambert, 1993). When γ > 1 it implies that a 1
% decrease in the coefficient of variation produces more utility than a 1%
increase one’s income.
Respondents know the level of income and the position of their grandchild
in each society as well as their income distribution. In these choices there
is no uncertainty, which allows us to identify parameters of individual in-
equality aversion in a risk-free setting. Respondents choose between pairs of
hypothetical societies with different income distributions, where their grand-
child’s income is known. The idea behind asking for grandchilds instead of
oneself is to abstract participants from their current situation when choos-
ing. Nevertheless, as they do not know their grandchild preference, necessary
assumption is that when they choose, they revel their own preferences.
Participants made repeated choices between Society A and alternative
societies Bi with (i=1....9). Participants are told that in these societies there
is no welfare state and that there are no dynamic effects. Both societies are
equal in everything (prices, quality of goods and services), but differ in the
level of income inequality and the grandchild’s income. Participants are also
told that there are no people whose income is not enough to cover basic needs
in order to avoid possible effects of aversion to poverty.
For all choices, society A remained unchanged. Income distribution is
uniform and varies between 10000 and 50000 per month. This means that
in society A the mean income is 30000 per month while the coefficient of
variation, ΦA , is 0.385.
On the other hand, in all the alternative societies Bi (with 1 < i < 9) the
coefficient of variation ΦB is 0.1925. The income distribution is also uniform,
but the range of the distribution changes among societies Bi . However, the
grandchild’s position in the distribution is fixed and their income is higher in
Bi than in Bi+1 ∀i. Inequality aversion is estimated via individual’s choices

13
between A and Bi . When an individual revels that she prefers society Bi over
society A, it means she is willingness to sacrifice certain amount of income
of their grandchild in order to reduce the income inequality in the society in
which her grandchild would live. For example, if a participant choose B in
the first three choices (i.e: B1 , B2 , B3 > A), but in choice number four society
A is chosen (A > B4 ), then we can say that in order to reduce inequality
in 50%, this participant is willing to sacrifice up to 1050 but less than 1800
(Table 7 presents information about income distribution in all societies of
the experiment). As a result, this choice implies a trade-off between an
individual’s income and equality in society.
Since we know the grandchild’s income, and the inequality income in
each society (ΦA < ΦB <), based on equation (5) we can elicit the inequality
aversion parameter for an individual who revels to be indifferent between
Society A and Society B. Table 7 in the annex presents the values for each
society. The elicitation of parameter γ are based on intervals. Furthermore,
the magnitude of the parameter is conditioned by the individual value of eij ,
Mj and Pij .
For example, when the individual is at mean of the income distribution,
this parameter is estimated based on the following equation:

log(xmean
B /xmean
A )
γ̂ij (eij , Mj , Pij = M eanj ) =

log(ΦB /ΦA ) ‘ ‘ 0

ej ,Pi,j ,Mj (9)

Respondents were asked to make repeated choices between societies A


and Bi when their grandchild is in different positions of income distribution.
When the respondent’s grandchild is at the bottom of the distribution, and
when is at the top of the distribution, which allow us to elicit γ̂ij (eij , Mj , Pij =
M inj ) and γ̂ij (eij , Mj , Pij = M axj )
Alternatively, we use the alternative approach presented in and we pa-
rameterize the role of self-centered inequality aversion we use (12), which
0 0 0
has three unknowns: γ̂ , β̂ and α̂ . However, because each individual make
repeated choices in three position of income distribution, we have a systems
of three equations over three unknowns. With some additional assumptions,
we can recover the parameter γ̂i,j (ej , Mj , β̂, α̂). Burone (2018) discuss this
strategy in detail and demonstrates that it allows a very accurate estimate
of this parameter. This strategy estimates values in a continuous range (very
small intervals), which is an advantage compared with the process based on
the equation (6), Furthermore, note that estimating aversion to inequality

14
by this way provides an unbiased estimation of not self centered inequality.
For this reason we expect to obtain a a more accurate measure of normative
inequality aversion. This is appealing because our randomized information
experiment focuses on normative variables.

3.2 Randomized Information Experiment


We use an experimental survey design, which is an adaptation of Amiel &
Cowell (1992). Participants were asked to choose, among several pairs of
societies (without uncertainly), in which they would prefer their grandchild
to live. The main difference with previous papers is that in our experiment
participants are provided with information about what determines the in-
come distribution and the position of their grand child. Besides, the same
individual is asked to choose when her grandchild is at different points of the
income distribution. Through choices, participants must choose if they are
willingness to sacrifice (or not) their grandchilds income to reduce inequality,
but each respondent carried out their decision at three alternative points of
income distribution: The Mean, the Top and the Bottom.
Figure 1 in the annex provides a screen-shot when participants are told
that their grandchild is at the mean of income distribution. For commu-
nicative purposes We used pictures of buildings to represent the alternative
societies in each choice. These pictures represent differences on income distri-
bution between both societies thought the amount of coins displayed together
with the main statistics of each income distribution. Furthermore, each pic-
ture presents the grandchild’s position in each building and shows the gap
with top and bottom of distribution. As a result, each picture combines intu-
itively information about grandchild’s absolute income, her relative income
and societys inequality.
In the control group (no information about reasons behind position reached
by grandchild is provided) participants always choose their preferred society
knowing position of their grandchild and income distribution at each soci-
ety. However, they do not have any information about what determines
income distribution in these hypothetical societies. Three treatment groups,
for which allocation was randomized, were provided alternative explana-
tions about the causes of differences in income distribution between societies:
(Treatment I) Effort; ( Treatment II) Luck; (Treatment III) Mobility.
a) Effort treatment: Participants are said that their grandchild position
at income distribution is the result of the effort made

15
b) Circumstances/luck treatment: Participants are said that their
grandchild position at income distribution is the result of luck.
c) Mobility treatment: Participants are said that their grandchild
grandchild has certain chances of achieving upward (or downward mobility)
at distribution.
D) Independently of treatment or control group, all participants were
asked to answer three sets of choices changing position of grandchild at in-
come distribution. In the first set grandchild is at the mean, in the second at
the bottom and in the third at the top of income distribution. This variation
in the position will be considered as fourth treatment, in this case, Whithin
individuals.
Screen-shots of instruction for each treatment and as well as each set of
choices are presented in the Annex.

3.3 Estimating the causal effect


In this section, we present our identification strategy to measure the impact
of the informational treatment in order to test our hypotheses. The aim
is to identify what influences the degree of individual inequality aversion
using a randomised information treatment which describes what determines
the income distribution in society or the individual’s position in the income
distribution. Based on equation (5), for the individual i, who lives in the
society j, the magnitudes of γ̂ij depends on its a prior beliefs about the
0
parameters: e‘j ,P i, j ‘ and Mj . . In our data set, estimates of γ̂ based on
the control groups provide evidence for the conditioned prior values of these
parameters. Additionally, this prior beliefs represents the baseline. The
informational treatment changes one of those parameters, keeping the rest
constant. This treatment will allow us to assess the impact of a change in
one argument of the function γ̂ on the magnitude of the inequality aversion
parameter. It is possible to identified the treatment effect analytically from
the partial derivative:

γ γ̂ 0 − γ̂
= 0 lim
∂z zj −zj →0 zj0 − zj
where z represents one of the arguments of the function γ̂ij . Namely this
derivatives measure how respond the individual inequality aversion parameter
to change in e or Mj or Pij . If we assume monotony of that function, we

16
can identify the sign of that derivative when zj0 − zj 6→ 0. The coefficient of
interest is the estimated parameter
τ
, which represents the causal effect of receiving the informational treatment
on inequality aversion parameter.

T T C C
ln(x i,A /x )
i,j ln(x i,A /x )
i,j
τz = (γ̃ T −γ̃)−(γ̃ C −γ̃) = γ̃ T −γ̃ C = T T
− C
ln(ΦA /Φj ) T T T ln(ΦA /Φj )C C C

ej ,Pi,j ,bj ej ,Pi,j ,MjC

(10)

where τ is the effect for the treatment Z, T y C represents the status of the
individuals regarding the informational treatment ( Treatment Z vs control)
and xTi,j is the income of a treated individual i and ΦTj the inequality in the
society J. Analogously, the superscript vC identifies the same parameters
for the individuals that belong to the control group. Observe that ΦTj = ΦC j ,
because is not a treatment variable. This allow us to elicit the parameter
γΦj following the following the proceeding that we were described in section
(3.1).
Let γef f ort denotes inequality aversion parameter in equation (1) esti-
mated using answers provided by participants who were assigned to effort
treatment. Similarly, γluck is the parameter for those under luck treatment,
γmobility is the one for those assigned to mobility information treatment and
finally γcontrol denotes the same parameter estimated for those assigned to
control group.
Following this notation, hypotheses presented in previous section imply
the following inequalities for parameters and τ :

Hypothesis I (Fairness hypothesis): implies that, γef f ort < γcontrol <
γluck . Namely we expect that
τe f f ort > 0
and τl uck < 0.

Hypothesis II ( Risk aversion vs. preferences for mobility): risk aver-


sion dominates preferences for income mobility when γmobility > γcontrol . In

17
this case, τm obility > 0

We will address if the magnitude and sign of taue f f ort > 0, τl uck and
τm obilityvarywiththepositionof theindividualintheincomedistribution.
Finally, we can adapt the equation (11) to identify the impact of a change
in the position on the magnitude of the inequality aversion parameter. In
this case, he information treatment was defined as change in the position
of the individual i in the income distribution, keeping the rest constant.
As Table 7 in the annex shows, the values in societies A and Bj were de-
fined to meet the following requirements: (xPi,A s /xPi,j s ) = (xPi,A s0 /xPi,j s0 ) and
ΦPA s /ΦPj s = ΦPA s0 /ΦPj s0 , where P s and P s0 represent two position in the in-
come distribution and P s > P s0 .

ln(xTi,A /xi,j ) ln(xi,A /xi,j )
τp = −
ln(ΦA /Φj ) s T ln(ΦA /Φj ) C s0 C C

ej ,Pi,j ,bj ej ,P i,j ,Mj

(11)

First, we kTable 7 in the annex presents the values for each society.
In this case, τp indicates the effect of the position on γ. By Hypothesis
III we expect that τp ¿. Namely we expect to confirm an increasing sensitiv-
ity of income inequality, which implies that, γmax > γmean > γmin .

4 Data
4.1 Data collection procedure and Questionnaire
We recruited a sample of undergraduate students from Universidad de la
Republica, sending an e-mail to invite them to participate in a survey. Par-
ticipation was voluntary and there was no remuneration. E-mail included
information about the research team and about the institutional affiliation
(Research Economic Institute). This mail redirected students to the survey
which we built using Qualtric’s online survey software. We collected data
from five distinct sessions. Students have the option to reject participation
in the first screen.

18
The invitation was received by 2956 students from first year of Facultad
de Ciencias Económicas y de Adminitración (cohort 2018).A reminder was
sent 18 days later, and a second reminder for those who had never entered
the survey was sent three months later.
Total time to carry out the experiment (including the instructions and
socioeconomic questions) varied between 25 to 30 minutes.
A total of 1049 students started the survey of whom 191 rejected to par-
ticipate. The proportion of rejection is balanced in each treatment. Final
number of completed questionnaires is 854, of which 662 were valid (con-
sistent) responses for the inequality-aversion experiments in the three set of
elections. Details are presented in tables 12 and 13.
Following Carlsson et al (2005) respondents are considered inconsistent
if they switch from choosing alternative A to alternative B in later choices,
keeping fixed their grandchild position in the distribution (the mean, mini-
mum or maximum). As elections are made one to one this case was possible
but inconsistent. The consistence of a response demands consistence in the
three sets of choices, namely when her grandchild is at mean, at minimum,
at maximum of the income distribution, respectively. This definition is much
more demanding than which is used in Carlsson et. al. (2005). A series
of tests were carried to explain being inconsistent, finding no evidence of
any kind of bias (i.e: groups of inconsistent and consistent participants are
balanced in observable outcomes, across treatments and do not show differ-
ent behaviour in others sets of choices where there were not inconsistent).
Based on this evidence inconsistent responses might be explained by learning,
attention or fatigue effects.
As it was already explained, we defined 4 groups (3 treatments and 1
control). Each student has the same probability of being assigned to each
group, namely this probability is 0,25.
We collected data from five different modules. (1,2 and 3) the main
inequality aversion experiment (9 elections in the mean; 9 elections in the
minimum and 9 elections in the maximum); (4) attention check question (5)
Socioeconomic and demographic characteristics (6) several questions regard-
ing participant’s opinions, attitudes and preferences.

4.2 Additional questions


In modules 5 and 6 the questionnaire contained questions regarding partici-
pant’s socioeconomic background, households’ demographics characteristics,

19
their individual opinions, attitudes, beliefs and political preferences. That
information was used to explore the anatomy of individual responses to in-
formational treatment in the main inequality aversion experiment. This in-
formation was used later in the robustness analysis.
A commonly used question in inequality of opportunity approach, is in-
cluded to measure the perceptions about role of luck and effort on people
achievement’s. Participant’s were asked the following question:

Which of these statements do you feel more identified with?


• Level of income and position reached by people in society are largely the
result of personal effort.
• Level of income and position reached by people in society are mostly
the result of factors that can not be controlled (for example family,
luck, etc).

Two additional questions were included in the questionnaire to explore


the effect of considering inequality as a problem on its aversion. More related
to the comparative notion, participants were asked if in their opinion income
inequality is a problem in Uruguay. Five options were given to participants:
i) It is not a problem, ii) it is a minor problem, iii) it is a problem, iv) it is
a serious problem, v) it is a very serious problem.
On the other hand, opinions about inequality for society in general was
asked to explore the normative notion of inequality aversion. In this case
participants were asked if they consider inequality is a problem for: i) not a
problem, ii) a few people iii) the poorest in society iv) almost all people in
society iv) all society.
In Figures 3 to 15 at the Annex screen-shots of questions used in the
questionnaire are presented.

4.3 Potential problems


Sample
Our sample is clearly not representative of Uruguayan population. Advan-
tages of applying the experimental survey to a significant sample of university
students are that, (i) we can assume that they understand the abstract ques-
tion included in the survey and they make reasoned choices. (ii) students are

20
easily recruited and that sample provides heterogeneity of individual’s char-
acteristics; (iii) we can merge the data with Faculty administrative records
(background and their academic performance). As a result, this sample is
useful to test our hypotheses and to provide evidence on individual social
preferences, in order to improve economic-theoretical models.

Experimental approach
Designing an experimental survey to elicit individual preferences is a great
challenge and faces many potential problems. The survey used in this paper
is based on Amiel and Cowell (1992) and Carlsson et al. (2005). Strat-
egy adopted faces characteristic problem when individual’s preferences are
elicited using a set of discrete choices. Samuelson (1938, 1948) has al-
ready postulated the basic foundations of the revealed preferences approach.
Beshears et al. (2008) argue that an individual’s behavior might deviate from
their preferences due to diverse factors, such as analytic errors, myopic im-
pulses, inattention, passivity, and misinformation. They suggest alternative
methods, none of which are ideal, that might mitigate the potential dispar-
ity between revealed preferences and normative preferences. Authors argue
that an active decision mechanism, one in which there is no default option,
forces individuals to explicitly state their preferences. On the other hand,
they suggest that aggregated revealed preferences provide complementary
information about the central tendencies of preferences, when respondents’
choices include an idiosyncratic error with a mean that is close to zero.
According to instructions, when a respondent makes a choice, they have
to consider the best situation for their imaginary grandchild. This strategy
seeks to avoid the influence of respondent’s personal circumstances or the
environment in which they complete the survey (Johansson-Stenman et al,
2002). As they do not have a grandchild (and have no information about their
preferences), the necessary assumption is that they use their own preferences
when making a choice about their grandchild’s future.
A methodological issue concern is that the respondents might provide
strategic responses, for example, in order to buy ‘moral satisfaction’ (Kah-
neman and Knetsch 1992), the desire to make a good impression on the
experimenter (Haertner and Schokkaert (2008), or to reinforce certain char-
acteristic of their identity as ‘self-image motive’ (Akerlof and Kranton, 2000)
or because they perceive themselves as extreme and subsequently adopt less
extreme choices to avoid harming their grandchild. Beshears et al (2008)

21
argue that real behaviour can also be driven by signalling motives, so this
result is not an unexpected consequence.
Finally, another potential problem, which to some extent is related with
the previous comments, is that questionnaire studies use hypothetical ques-
tions and do not provide financial incentives for individuals to respond truth-
fully. Haertner and Schokkaert (2012) noticed that this problem is more rel-
evant if the purpose of the empirical research is to predict current behaviour
which reflects a mixture of self-interest, norms, and signalling motives.
Beshears et. al. (2008) argue that self-reports provide useful information
about an agent’s norms, values and goals. Amiel et al (2012) and Haertner
and Schokkaert (2012) argue that experimental surveys focus on an individ-
ual’s opinion and their ethical preferences being ambiguous how and which
financial incentives might be relevant to address this issue. Moreover, real
world incentives are very different from the incentives in a questionnaire en-
vironment, so they would not be enough to predict an individual’s behaviour.
Finally, as a long tradition in the theory of justice argues, in a study about
norms and ethical preferences the impartiality of the respondents could be
an essential component. When they respond to the questionnaire we do not
need subjects be guided exclusively by self-interested considerations.
If the questionnaire is very complex and extensive, financial incentives
might improve the quality of the respondents’ answers. However, as Amiel
et al (2012) argue, why should the respondents not try to pay attention
and answer carefully and truthfully if the instructions are clear and students
volunteer to participate?
An alternative problem could be that respondents assume lexicographic
strategies (they always respond the minimum or maximum). To mitigate
this problem, instructions say that there are no extremely poor people in the
society and we could check and control this issue in the results.
Another problem could arise if respondents don’t pay attention or don’t
understand the picture they are looking at and the income inequality implied.
We address this issue by incorporating a specific question to check attention.
Also, at the end of the survey, an image of two societies similar to the ones
showed during the grandchild questions was presented and participants were
asked which society was more unequal. In this case it was stated that there
was only one correct answer and this question was included to find out if
participants have understood the exercise. Finally, a question was included
in order to ascertain if participants considered that they have paid enough
attention to the questionnaire and that therefore their answers should be

22
used in the analysis.
A complementary strategy is to consider the valid answer, so we identify
those respondents who provide consistent responses (consistency is defined
in the previous section). We then explore whether the consistency of indi-
viduals’ responses is correlated with the two check questions: self-reported
attention check and the income inequality understanding exercise. We carried
out a simple OLS model, and both control check variables are significant in
explaining consistent responses. Moreover, time spent completing the ques-
tionnaire has no significant effect2 . Considering only consistent responses,
528 students finished entire the survey3 .
Respondents were randomly assigned to each treatment. Nevertheless,
being consistent was not known at time of assignment. For this reason, final
number of consistent answers is not exactly equal in each treatment. To
know if there is any kind of bias in the final amount of consistent answers
carried out in each treatment, balance in socioeconomic characteristics was
checked and the results are shown in Table 21. As can be seen, does not
seem to be any unbalanced characteristic4 .
As an additional control, for each position (minimum, maximum and
mean) we compare aversion to inequality using the responses of participants
who were consistent in every set of choices and results using consistent an-
swers in each position. Results are presented in Tables 9, 10 and 11 in the
Annex. As can be seen, distribution with or without participants who were
inconsistent in other positions are almost the same. This is important as it
means that participants who were inconsistent in some position do not have
a different behaviour in those set of choices where they did not choose in an
inconsistent way.
A final point is that our inequality aversion estimate could be overesti-
mated, because our sample is based on volunteer students. As a result, their
behavior could be guided more by altruism than by selfish considerations.
2
As the questionnaire was online, it is possible that participants who spent very little
time completing the questionnaire did not pay enough attention. On the other hand,
spending too much time could also indicate not paying enough attention as it could be
the case that participants completed a part of the questionnaire, left it uncompleted and
finished later. For this reason time in not a good proxy of attention paid during completion
of the survey
3
528 students were never inconsistent. 245 were inconsistent only in one set of questions
(mean, minimum or maximum), 63 were inconsistent in two sets and 18 were inconsistent
in every set of questions
4
Further tests to check for this potential bias will be carried out

23
We could check this issue by comparing our result with those of previous
papers. A specific bias might arise due to environmental effects, because our
survey is on-line. We tested this issue by comparing with data from Burone
(2018, forthcoming), in which students complete the same questionnaire in a
class room. We did not find a significant difference between the results based
on both strategies

5 Main results
In this section we present the main findings of the empirical analysis and
we interpret them. Table 1 shows the inequality aversion parameter (γ)
for respondents using responses from the set of choices where grandchild is
hypothetically at mean of income distribution.
First we focus on the Control Group. The mean5 of γ estimated useing
responses when grandchild is at mean and participants are from control group
mean mean
is denoted as γcontrol . The mean value for γcontrol is 0.18946 . This value is
in line with results found in Burone (forthcoming) for a similar sample of
Uruguayan students, who made an in-class version of this experiment with
no treatment (equal to our control group). He found a mean value for γ mean ,
which is, in mean, 0.1965. Carlsson et. al. (2005) found a mean value of
inequality aversion of 0.2989 for Sweden, also carrying out the experiment in
an in-class context. In our case the magnitude of the parameter is slightly
lower, which could be an effect of carrying out the experiment online and
5
Strictly, participants who choose society A in the first election have a parameter
−∞ < γ < −0.09. In the same way, it is known that participants who always prefer
society B have an aversion to inequality parameter γ > 0.78. In order to calculate the
mean value of gamma we always use the upper bound. In the case of γ > 0.78 we suppose
a hypothetical upper bound given by the sum of the lower bound plus the distance of the
widest interval
6
From our experiment we can say the minimum and the maximum value of aversion to
inequality for each individual consistent with their decision, but we do not know the exact
value. In order to calculate the mean values, the midpoint of the individual’s interval
was imputed. We assume that the distribution of this parameter is uniform within each
interval, as we do not know the real distribution. Under this assumption we can use the
mean of each interval. Alternatively, our results were also calculated using the upper
bound and conclusions are basically the same but with a change of level (higher values).
In the rest of the document we follow this strategy (imputing the midpoint of the interval
mean
in
P calculations). Midpoint of the individual’s interval was imputed as follows: γ =
1≤j≤9 (F req(j − 1 ≤ γ ≤ j) ∗ (γint−min + γint−max )/2)

24
the anonymity of the respondents. However, there is no reason to suppose
that this deferentially affects the mechanisms that explain inequality (and the
informational treatment), which is the focus of this study. Moreover, it is not
obvious whether taking the experiment online could lead to underestimating
the real parameter or whether on the other hand, taking it face to face could
lead to overestimating it. Finally, this potential bias is an aspect to take into
account when interpreting the magnitude of the results.
As shown in Table 1, 26% of participants are inequality lovers as they
showed willingness to sacrifice personal income to live in a more unequal
society by choosing society A in the first election. Note that the maximum
income in society A is much bigger than in B (50000 vs 42600). As a result,
this decision may indicate that these respondents are willing to renounce
some of their grandchild’s money in order to increase the maximum income
in society, even if this implies that their grandchild would live in a more un-
equal society. Namely, they value efficiency (or growth) more than equality.
Furthermore, this higher potential income might motivate a ”tunnel effect”,
i.e: it would generate expectations that their grandchild would obtain a
higher income in the future.
Close to 19% of respondents from the control group have a relatively low
inequality aversion parameter, allocated between -0.09 and 0.05. Namely,
their willingness to sacrifice their grandchild’s income in order to live in a
more equal society is close to 0. On the other hand, 55% of the respondents
are clearly willing to pay in order to reduce income inequality. Some of them,
17%, are extremely averse to inequality as they never choose society A, a
share which is slightly lower than that of inequality lovers. It is interesting
to note that in our sample there are more extreme respondents, as the shares
of inequality and equality lovers are two times more than the shares found
by Carlsson et. al. (2005).
In Table 1 estimates of aversion to inequality are presented also for every
treatment. As can be seen, the participant’s response to 2 of the 3 infor-
mation treatments is very high. In particular, Effort treatment is the one
which seems to affect less behaviour of respondents. Figure 1 presents this
information graphically.

Figure 1: Inequality aversion by treatment

25
Table 1: Aversion to inequality for every treatment. Responses when grand-
childs income is equal to the mean society income

Control Luck Effort Mobility

γ <0,09 n 39 24 37 29
% 26% 22% 27% 18%
n 12 11 18 8
-0,09< γ <0
% 8% 10% 13% 5%
n 16 12 25 18
0< γ <0,5
% 11% 11% 18% 11%
n 18 4 10 11
0,05< γ <0,09
% 12% 4% 7% 7%
n 9 4 8 17
0,09< γ <0,15
% 6% 4% 6% 11%
n 11 14 9 16
0,15< γ <0,21
% 7% 13% 6% 10%
n 18 18 13 16
0,21< γ <0,34
% 12% 17% 9% 10%
n 17 12 10 27
0,34< γ <0,51
% 11% 11% 7% 17%
n 11 10 9 19
0,51< γ <0,78
% 7% 9% 6% 12%
n 26 35 23 18
>0,78
% 17% 32% 17% 11%
Total N 177 144 162 179
Mean - 0,1894 0,3012 0,1524 0,2252
Median - 0,15 0,34 0,09 0,2

26
When participants are told that the position reached by their grandchild
is mainly due to luck, willingness to sacrifice income to live in a more equal
society is greater. As a result, the whole distribution of γluck moved to the
right with respect to γcontrol , increasing the magnitude of the inequality aver-
sion parameter. The mean of inequality aversion parameter is 59% greater
in contrast with control group. When people are told that effort is mainly
behind their grandchild’s position, the aversion to inequality is reduced by
19.5% with respect to control group. However, the shape of the distribution
of γcontrol and γef f ort are very similar, which suggests that effort treatment
seems to be the treatment that least affects the aversion to inequality pa-
rameter distribution.
The proportion of participants who are extremely averse to inequality
(γ > 0.78) is also the greatest for Luck Treatment. On the other hand, Effort
Treatment leads to the highest proportion of inequality lovers, although the
difference with Control is really small.
Two-tail mean tests were carried out to test differences in mean aversion
to inequality by treatment. In addition, the one-tailed test is used to provide
evidence with respect to the inequalities in the magnitude of the inequality
aversion parameter suggested by the hypotheses. The results are presented
in Table 17 in Annex 2.
The most interesting and robust result from Table 1 is the difference

27
between the Luck and Effort treatments. When participants are told that
luck mainly explains income distribution, aversion to inequality is almost
two times greater than if they are told that effort explains income distri-
bution (Hypothesis I). With a one-tailed test (and two-tailed test), the null
hypothesis γluck < γcontrol is not rejected (P-value=0.01).
The one-tailed test does reject the null hypotheses γef f ort = γcontrol and
γef f ort < γcontrol . However, the hypothesis of no difference in γluck and γef f ort
is rejected (Pvalue 0.003 and 0.001). These results confirm our first Hypote-
hesis , which implies: γef f ort ≤ γcontrol < γluck ).
On other hand, mobility treatment seems to be the treatment that most
affects the tails of the inequality aversion parameter distribution. Mobility
treatment is the one which shows the lowest proportion of extreme values in
inequality aversion distribution. Namely, only 18% of participants show an
extreme aversion to inequality, which suggests the trade-off between inequal-
ity aversion and preferences for income mobility. However, it also shows the
lowest proportion of participants with extreme aversion to inequality which
suggests the domain of risk aversion effect. The domain of this effect is
in accordance with the movements in the mean of γmobility , which is 18.9
% greater than γcontrol , even though the one-tailed test rejects the null hy-
potheses γmobility = γcontrol and γmobility ¿ γcontrol . However, note that the
magnitudes of γmobility , γef f ort and γluck are different and the p-values show
that the differences are statistically significant at 1% (see Table 1).
In sum the Hypotehesis II implies γmobility > γcontrol . Our estimates of
the inequality aversion parameter confirm this, but no evidence of statistical
significance was found. This result suggests that preferences for mobility
and risk aversion affect the inequality aversion parameter in two opposite
directions, which potentially compensate the effects of one another. However,
there is weak evidence on the domain of the risk aversion effect. This result
is obtained if we observe the choices made when a grandchild is at the mean
in the income distribution, but, as we will describe in the next section, it is
not true when the grandchild is in other positions. We will come back to
these results when we later discuss Table 2.

5.1 When position matters


As it was explained, individuals’ aversion to inequality may be affected by
their position in the income distribution (Hypothesis III). A variant in our
experiment consists of changing the grandchild’s position in the income distri-

28
bution. This strategy allows us to test if the effect of informational treatments
on an individual’s inequality aversion parameter depends on the knowledge
about their grandchild’s position in the income distribution. In order to ex-
plore this mechanism, Table 2 shows aversion to inequality by Treatment in
different positions.

Table 2: Inequality aversion by informational treatments when the position


in the income distribution changes

Mean Minimum Maximum


Mean S.D Median Mean S.D Median Mean S.D Median
Control 0.19 0.43 0.12 0.02 0.43 -0.05 0.34 0.40 0.28
Luck 0.30 0.46 0.28 0.18 0.53 -0.05 0.43 0.44 0.43
Effort 0.14 0.43 0.03 -0.03 0.47 -0.36 0.25 0.40 0.18
Mobility 0.24 0.37 0.18 -0.22 0.33 -0.36 0.49 0.33 0.43

First, the magnitude and sign of the individuals’ inequality aversion pa-
rameter is very sensitive to their position in the income distribution. Observe
min mean max
that γcontrol < γcontrol < γcontrol and the magnitude of these differences are
statistically significant at 99% (mean test for significance are presented in Ta-
ble 16 at the Annex). This result is confirmed for each treatment, providing
evidence which confirms Hypotesis III, which implies γmin < γmean < γmax .
Burone (forthcoming) interprets this result as a cost of being concerned to
inequality inversely related to position. On one hand, for the poorest in so-
ciety concern for redistribution is more costly in relative terms. This result
also could be explained by a fairness effect, because poorer individuals might
perceive that those richer individuals should assume the cost of the reduction
of income inequality. On the other hand for the richest individuals, the cost
of redistribution is lower in relative terms so their own position enable them
to concern for equity and be willingness to pay in order to live in a more
equal society. This results suggests that income inequality concern behaves
as a luxury good.
This exercise also provides robust evidence with respect to Hypothesis
I. Regardless of the grandchild’s position in the income distribution, the
evidence provided in Table 2 confirms γef f ort ≤ γcontrol < γluck ). The null
min min max max
hypotheses γluck ≤ γef f ort and γluck ≤ γef f ort are rejected (P-values are both
0.001). In addition, when grandchild is at the top of income distribution
max max
in her society, the one-tail test reject the null hypotheses γcontrol ≤ γluck

29
max max
and γcontrol ≤ γef f ort (P-value 0.002 and 0.07 respectively) (mean test are
presented in Tables 17, 18 and 19 in the Annex).
In sum, luck treatments is related to the highest degrees of inequality
aversion, while effort treatment is associated with a lower magnitude of in-
dividuals’ inequality aversion.
This exercise also provides additional evidence with respect to Hypothesis
II. The highlight result is that the conclusion about what effect dominates
in mobility treatment, preferences for mobility or risk aversion, depends on
the position: when participant’s grandchild is on the top of the income dis-
tribution, they show the highest degrees of inequality aversion. Note that
max
the mean of γmobility is the highest estimated value (of every treatments) and
max
it presents a significant gap with respect to γcontrol ( is 43% greater and the
mean
P-value is 0) and with respect to γmobility ( is 103% greater and the P-value
is 0). However, when participants assume that their grandchild is at the bot-
tom of income distribution, the inequality aversion parameter is the lowest.
min mean
γmobility is lower than γmobility . This indicates that preferences for mobility
effect dominates the risk aversion effect.
This evidence suggest that aversion to inequality is sensible to position.
Table 3 shows proportion of answers sensible to position by treatment7 . As
can be seen, Mobility treatment is the one where individuals are more sensible
to position. On the other hand, luck is the Treatment that is less affected by
position.
Mobility treatment seems to be the one in which people are more sensi-
ble to position. Note that in this case γ could be interpreted as inequality
aversion but also as risk aversion as participants are said that there is some
chances for their grandchild to change her position in the future. Which
of these effects dominates depend on the grandchild position in the income
distribution. Despite different degrees of sensibility to position are found for
each treatment, in general aversion to inequality seems to increase with posi-
tion, which suggest the idea that inequality concern behaves as a luxury good.
Finally, effort treatment is the case in which inequality aversion parameter
is less sensible to the grandchild position in the income distribution.
7
An individual answer is considered to be sensible to position if there is a difference
greater than two between the point in which preference for Society A over Society B is
revealed by position. For example, if Society A is preferred in the forth choice at mean,
individual is considered sensible to position if at minimum or maximum Society A is
preferred in choices 1, 7, 8 or 9

30
Table 3: Sensitivity of aversion to inequality by treatment

Treatment
Sensible Control Luck Effort Mobility Total
0 46 61 61 52 193
33% 58% 50% 33% 37%
1 95 45 62 133 335
67% 42% 50% 84% 63%
Total 141 106 123 158 528

5.2 Consistency analysis: Aversion to inequality, atti-


tudes and beliefs
In order to advance in the direction of better understanding the effects of
informational treatment and different degrees of sensibility to position we
explore how there results are associated with the individuals notion of fairness
and their concern about income inequality. A set of questions about that
issues were included in the questionnaire.
First, we consider individual’s perceptions about the role of luck and
effort. We explore if the response to treatment depends on that perceptions.
Aversion to inequality depending on answers to this question are presented
in Table 4
As can be seen, inequality aversion is strongly affected by participants
opinion about the role of luck and effort. In general, those who attributes
a greater role to luck are more averse to inequality in every treatments and
every positions. This is a potential channel to understand why people are
more or less concerned about inequality and what is more, more or less
wiliness to pay in order to reduce inequality.
Another interesting issue to investigate is if considering inequality as a
problem affects aversion to inequality and its relation with both normative
and comparative notions. Two questions were included in the survey to
consider this issue. One is clearly more related to the normative notion of
inequality aversion and other is more related to the comparative one. From
answers to this questions two variables were built (Var 1 and Var 2). Var 1
identifies those participants with 1 when they consider that income inequality

31
Table 4: Inequality aversion for question about role of luck and effort

Treatment Role* Mean S.D


At mean

Control Luck 0.33 0.44


Effort 0.15 0.41
Luck 0.47 0.45
Luck
Effort 0.28 0.45
Luck 0.34 0.45
Effort
Effort 0.08 0.37
Mobility Luck 0.34 0.28
Effort 0.21 0.40
At minimum

Control Luck 0.03 0.46


Effort 0.03 0.43
Luck 0.31 0.54
Luck
Effort 0.15 0.53
Luck 0.15 0.53
Effort
Effort 0.06 0.54
Mobility Luck -0.18 0.37
Effort -0.25 0.28
At Maximum

Control Luck 0.45 0.43


Effort 0.29 0.37
Luck 0.52 0.44
Luck
Effort 0.44 0.45
Luck 0.50 0.35
Effort
Effort 0.17 0.36
Luck 0.57 0.29
Mobility
Effort 0.46 0.32
*Role: refers to role assigned to luck and effort in question presented in
section 4.2. In Table 4 Effort indicates first alternative was chosen while
Luck indicates second alternative was chosen to question mentioned.

32
is a serious problem in Uruguay. Var 2 takes the value of 1 when participants
respond that inequality is a problem for society.
Aversion to inequality depending on answers to questions described in the
two paragraphs above are presented in Table 5. To present results answers
were grouped into two categories8 : options i, ii and iii vs options iv and v.
As can be seen, systematically those who think inequality is a problem in
Uruguay are more averse to inequality. This result is robust to treatment and
position. Another interesting finding is the fact that Effort is the treatment in
which considering inequality a problem in Uruguay implies a greater increase
in degree of aversion to inequality9 .
If we focus in aversion to inequality depending on question about con-
sidering inequality a problem for the society all (normative notion) at mean
and maximum the effect is clear: those who consider inequality as a problem
are more averse to inequality. At minimum it is true only for luck and effort
treatments. Nevertheless, difference in Control and Mobility treatments is
not significant (in tables 17, 18 and 19 in the Annex 2 test to check signifi-
cance are presented).
Considering inequality a problem, both from a normative as well as a
comparative notion, is related to being more averse to inequality.

6 Aversion to inequality and position


The economic literature has provided evidence on the degree to which in-
equality affects the well-being of individuals. However, the correct form to
model the effect is a subject not closed. Some authors model aversion to
inequality not considering position while other model this effect exclusively
through the distance between individuals income and the income of the rest of
individuals in society. Aronsson and Johansson-Stenman (2016) discuss this
difference and denominate self centered inequality aversion when inequality
aversion is position dependent (note it is related to the comparative notion)
8
Var 1: option i (it is not a problem), option ii (it is a minor problem), option iii (it
is a problem), option iv (it is a serious problem), option v (it is a very serious problem).
Var 2: option i (not a problem), option ii (it is a problem for a few persons, option iii (it
is a problem for the poorest in society) option iv (it is a problem for almost all society)
option v (it is a problem for all society)
9
Maximum is the only position in which the increase is greater for the Control group
but even in this case Effort is the treatment with the second greater effect

33
Table 5: Aversion to inequality by treatment and position depending on
considering inequality a problem

Treatment Var 1 Mean Var 2 Mean


Grandchilds position = mean

Control 1 0.27 1 0.27


0 0.17 0 0.08
1 0.41 1 0.38
Luck
0 0.28 0 0.18
1 0.18 1 0.29
Effort
0 0.06 0 0.08
Mobility 1 0.28 1 0.38
0 0.23 0 0.18
Grandchilds position = minimum

Control 1 0.06 1 0.02


0 0.15 0 0.04
1 0.27 1 0.23
Luck
0 0.15 0 0.16
1 0.09 1 0.01
Effort
0 -0.13 0 -0.17
Mobility 1 -0.19 1 -0.23
0 -0.27 0 -0.22
Grandchilds position = maximum

Control 1 0.46 1 0.39


0 0.26 0 0.23
1 0.50 1 0.50
Luck
0 0.45 0 0.42
1 0.35 1 0.28
Effort
0 0.23 0 0.20
Mobility 1 0.55 1 0.51
0 0.44 0 0.45
Var1: Dummy = 1 if consider income inequality is a problem in Uruguay.
Var2: Dummy = 1 if consider income inequality is a problem for society

34
and not self centred inequality aversion when this effect is independent of po-
sition of the individual (note this notion is more related with the normative
one).
Burone (forthcoming) provides evidence that this distinction is relevant.
Moreover, the author estimates aversion to inequality for the same indi-
vidual varying position and finds evidence suggesting that position affects
significantly the degree of aversion to inequality, same result we confirm by
Hypotesis III. As a consequence, estimations of inequality aversion that do
not consider the self centered component would be biased.
Based on this results, in this paper, we use a biased measure of normative
inequality aversion as it does not depend on the position of the grandchild.
We also find suggestive evidence that position affects this measure of inequal-
ity aversion. Using the same model and strategy adopted by Burone in this
section we calculate inequality aversion corrected for bias due to position
(modeling position and using choices made for participants in each position
jointly). In section A.1 we briefly explain the model used, for a deeper dis-
cussion see Burone (forthcoming). Results are presented in Table 6.

Table 6: Unbiased inequality aversion

Mean S.D Min Median Max


Control 0.35 0.44 -0.81 0.32 1,31
Luck 0.44 0.44 -0.62 0.38 1,69
Effort 0.28 0.38 -0.54 0.18 1,69
Mobility 0.45 0.38 -0.44 0.42 1,24

Results when aversion to inequality is estimated unbiased show that in-


dividuals have a concern for inequality that is highly affected by Treatment.
When individuals are told that exist high chances of changing position in fu-
ture, aversion to inequality is the greatest (this also could be interpreted as
risk aversion). Very similar results are found for Luck treatment. Individuals
dislike inequality more if Luck explains position reached. On the other hand,
if Effort is the main component that explains position reached, aversion to
inequality is the lowest.
Results presented in Table 6 are an unbiased estimation that consider
position. A simple comparison between this results and those presented in

35
Table 2 show that Mobility Treatment is extremely affected by position,
but when bias due to position is controlled, then aversion to inequality is
the greatest for mobility treatment. These results show the importance of
position in individuals willingness to sacrifice income for equity.
Finally, note that estimating aversion to inequality by this way, which
is an unbiased estimation of not self centered inequality aversion as it takes
into account the self centered effect, we confirm our Hypothesis I and II.

7 Exploring the anatomy of inequality aver-


sion
Last, but not least, we used MCO models to explain aversion to inequality.
Results are presented in Table 14 of the annex.
Results presented in Table 14 are indicative that effectively we are cap-
turing aversion to inequality as the effects found are consistent. Nevertheless,
in some cases we do not find significance, result we attribute to number of
observations in most cases.
First we used a model considering only Treatments (specification (1)). As
can be seen Treatments of Luck and Mobility have a positive effect on aversion
to inequality while Effort has a negative effect. This result is consistent with
previous results.
Second, we carried out models with questions included in the survey (spec-
ifications (2 and 3). We find that considering inequality a problem as well as
identifying as left voter have positive effect on aversion to inequality, while
attributing effort as the main factor to explain position reached (instead of
factors out of control like family or luck) has a negative effect.
Third, we explored the relation of inequality aversion with socioeconomic
characteristics (specification (4)). We find that age, sex (being a woman)
and work, have positive effects.
Fourth, we explored the relation with affirmations included in the survey
in order to capture possible mechanisms that could affect inequality aver-
sion (specification (5)). Results are extremely consistent and confirm we are
effectively capturing aversion to inequality.
Finally, specification (6) is our favorite model which combines previous
variables. We find a significant and positive effect of treatments Luck and
Mobility on aversion to inequality and a negative effect of treatment Effort.

36
Age and identifying as left voter also show a positive and significant effect on
aversion to inequality, as well as considering inequality a problem because it
generates crime and violence. Other consistent relations are observed in this
specification but significance is not found at usual confidence levels.

8 Final comments
This paper provides several findings that we believe are relevant in order to
understand the foundations of inequality aversion relative to existing litera-
ture. First, we find that inequality aversion of respondents is very elastic to
the notion of fairness. Specifically, for effort treatment (ie: individuals are
said that effort is the main factor that explains position reached) aversion to
inequality is the lowest for all the treatments carried out. This result seems
to be related with prior beliefs about the role of effort on income inequality.
On the other hand, luck treatment (ie: individuals are said that luck is the
main factor that explains position reached) is related with greater aversion to
inequality. These two results together suggest that individuals in our sample
have preferences for meritocracy. They are reluctant to accept higher levels
of inequality that arise as a result of luck and not because of differential
effort. In this sense, inequality aversion is higher when inequality rises as a
result of a relatively larger effort of one group compared to another.
Second, income mobility is positively correlated with inequality aversion
magnitude, which is interpreted as risk aversion. Our results suggest that
individual risk aversion dominates preferences for income mobility. However,
the trade-off between both effect depends on the position of the individual
in the income distribution.
Finally, individual’s willingness to pay for lower income inequality is very
elastic to their position in the income distribution. This result suggest that
income inequality concern behaves as a luxury good.
An alternative unbiased measure of income inequality was proposed. Again
we find suggestive evidence that supports our hypothesis both comparing es-
timations of aversion to inequality as well as using a model to explain the
effect of treatments on it.
We carried out alternative robust checks which confirms our results. By
using alternative subjective questions to check if we are effectively capturing
aversion to inequality, results proved to be robust and extremely consistent.
We explored the anatomy of inequality aversion, confirming the role of the

37
normative and the comparative notion of inequality aversion. Additionally,
the inequality aversion parameter is higher when the respondents consider in-
equality a problem because it is unfair or because it generates crime, violence
or other negative externalities explored.

9 References
Alesina, A, Guliano, P. (2009). Preferences for Redistribution. National
Bureau of Economic Research. Working Paper No. 14825
Amiell, J, Cowell, F. (1992). Measurement of income inequality: Experi-
mental test by questionnaire. Journal of Public Economics. Vol. 47, issue 1,
3-26
Aronsson, T., Olof Johansson-Stenman, Wender, R. Redistribution through
Charity and Optimal Taxation when People are Concerned with Social Sta-
tus. No 919, Umeå Economic Studies from Umeå University, Department of
Economics.
Beshears, John Choi, James J. Laibson, David Madrian, Brigitte C.,
2008. ”How are preferences revealed?,” Journal of Public Economics, Else-
vier, vol. 92(8-9), pages 1787-1794, August.
Burone, Santiago. (2018) ”Aversión a la Desigualdad: Aportes para
su medición y descomposición del efecto de la posición. Evidencia para
Uruguay”. Master’s thesis in economics. Facultad de Ciencias Económicas
y de Administrción. Universidad de la República. Manuscirt in preparation.
Carlsson, F. Daruvala, D., Johansson-Stenman, O. (2005) Are People
Inequality-Averse, or Just Risk-Averse? Economica Vol 72, pp. 375-396.
Clark, A., D’Ambrosio, C.. (2014). Attitudes to Income Inequality: Ex-
perimental and Survey Evidence. IZA Discussion Paper No. 8136.
Charness, G; Rabin, M. (2002) Understanding Social Preferences with
Simple Tests The Quarterly Journal of Economics Vol. 117, No. 3, pp.
817-869
Fehr, E., Schmidt K. M. (1999). A Theory of Fairness, Competition, and
Cooperation. The Quarterly Journal of Economics, Vol. 114, No. 3 (Aug.,
1999), pp. 817-868.
Fehr, E., Schmidt K. M. (2003). Theories of Fairness and Reciprocity-
Evidence and Economic Applications. Advances in Economics and Econo-
metrics, Econometric Society Monographs, Eighth World Congress, Vol. 1,
pp. 208-257. Harsanyi, J. C. (1955). Cardinal welfare, individualistic ethics,

38
and interpersonal comparisons of utility. Journal of Political Economy, 63,
309–21.
Heffetz, O. Frank, R. (2011). Preferences for Status: Evidence and Eco-
nomic Implications. Handbook of Social Economics Vol. 1, pp. 69-91.
Hopkins, E. (2008). Inequality, happiness and relative concerns: What
actually is their relationship?. Journal of Economic Inequality. Vol. 6, pp.
351-372.
Jantti, M., Jenkins, S. (2015). Income Mobility en Atkinson y Bour-
guignon. Handbook of income distribution.
Johansson-Stenman, O., Carlsson, F., & Daruvala, D.. (2003). Are Peo-
ple Inequality Averse, or Just Risk-Averse? Economica, 2005(72), pp. 375-
396. Samuelson PA. A Note on the Pure Theory of Consumer’s Behaviour.
Economica. 1938;5:61–71.
Samuelson PA. Consumption Theory in Terms of Revealed Preference.
Economica. 1948;15:243–253.
Shorrocks, F. (1978). The Measurement of Mobility. Econometrica. Vol.
46, No. 5, pp. 1013-1024

A Annex 1
A.1 Unbiased estimation of inequality aversion
We use the model and strategy proposed in Burone (forthcoming) in order
to estimate unbiased inequality aversion. In this section we briefly present
the model and strategy.
The model is presented in equation 12.

Uij (xij , Φj , γ̂j , β̂ij , α̂ij ) = (xij )[RD]−α̂ [RA]−β̂ (Φj )−γ̂ (12)
where R x 
jmax
 xij (x−xji )f (xj ))dx if xij < xjmax
xij
RD=
1 if xij = xjmax

 R xij 
 xjmin (xij −x)f (xj )dx if xij > xjmin
xij
RA =
1 if xij = xjmin

Where Ui is the utility of individual i, which depends on income of individuals
1 (x1 ) to N (xN ). N is the number of individuals in society.

39
vxi is a function of individuals i income. A and B in the model capture
the effect of distance between individuals income and rest of society income,
while the last component of equation 12 is the unbiased inequality aversion
component as Φ is a measure of inequality in society which does not depend
of position (we use the variation coefficient). f (x) is the income distribution
function.
Note that RA is the sum of income of all individuals in society that have
more than individual i while RD is the sum of income of all individuals in
society who have less than individual i.
In the model α is the parameter which captures the effect on utility of
the distance between individuals income and the income of all individuals
in society who have more. β captures the effect on utility of the distance
between individuals income and the income of all who have less. So α and
β capture the effect on aversion to inequality due to position while γ is the
unbiased measure of inequality aversion.
If we know the choices made for individuals between societies A and B in
three different positions, and if societys values (i.e: f (x), xmax and xmin ) are
known (as we do in the context of the questionnaire made), solving for each
position where individuals have to choose:

UiA = UiB

 
(xmaxB −xB )/xB
log (xB /xA ) − α log (xmaxA −xA )/xA
−β log (xmaxB /xmaxA )
ˆ˜γi,j (ej , Mj , β̂, α̂) ==
log (ΦB /ΦA )

When xi = xmin :

 
(xmaxB −xminB )/xminB
log (xminB /xminA ) − α log (xmaxA −xminA )/xminA
−β log (xmaxB /xmaxA )
ˆ˜γi,j (ej , Mj , β̂, α̂) ==
log (ΦB /ΦA )

When xi = xmax :

ˆ˜γi,j (ej , Mj , β̂, α̂) == log (xmaxB /xmaxA ) − β log (xmaxB /xmaxA )
log (ΦB /ΦA )

40
We obtain three equation which represent the indiference preferences be-
tween society A and Bj for the three position. We get a system of no linear
equations that can be solved. Doing this, we get γi,j (ej , Mj , β̂, α̂) for each in-
dividual, a parameter that captures unbiased inequality aversion taking into
account position and combining all choices made for participants. The sys-
tem allow us to identify the values ofˆ˜γ, β̂, and α̂ that are compatible with the
preferences of individuals. The parameters are estimated based on a situation
of indiferences among the societies A and Bj , which implies an assumption.
However Burone (2018) discusses the implication of this assumption and he
used simulations to demonstrated that this strategy provides an accurate
measure of ˆ˜γ, β̂, and α̂.

B Annex 2

41
Table 7: Experiment Hypothetical Societies Values
Society Min Mean Max Rel. deprivation Inequality Strd. Dev.
When grandchild is in the mean of income dist
A 10000 30000 50000 0,3333 0,385 11547
B1 21300 31950 42600 0,6667 0,1925 6149
B2 20000 30000 40000 0,6667 0,1925 5774
B3 19300 28950 38600 0,6667 0,1925 5571
B4 18800 28200 37600 0,6667 0,1925 5427
B5 18000 27000 36000 0,6667 0,1925 5196
B6 17200 25800 34400 0,6667 0,1925 4965
B7 15800 23700 31600 0,6667 0,1925 4561
B8 14000 21000 28000 0,6667 0,1925 4041
B9 11600 17400 23200 0,6667 0,1925 3349
When grandchild is in the bottom of income dist
A 10000 30000 50000 0,3333 0,385 11547
B1 10650 15975 21300 0,6667 0,1925 3074
B2 10000 15000 20000 0,6667 0,1925 2887
B3 9650 14475 19300 0,6667 0,1925 2786
B4 9400 14100 18800 0,6667 0,1925 2714
B5 9000 13500 18000 0,6667 0,1925 2598
B6 8600 12900 17200 0,6667 0,1925 2483
B7 7900 11850 15800 0,6667 0,1925 2281
B8 7000 10500 14000 0,6667 0,1925 2021
B9 5800 8700 11600 0,6667 0,1925 1674
When grandchild is in the top of income dist
A 10000 30000 50000 0,3333 0,385 11547
B1 26625 39938 53250 0,6667 0,1925 7686
B2 25000 37500 50000 0,6667 0,1925 7217
B3 24125 36188 48250 0,6667 0,1925 6964
B4 23500 35250 47000 0,6667 0,1925 6784
B5 22500 33750 45000 0,6667 0,1925 6495
B6 21500 32250 43000 0,6667 0,1925 6207
B7 19750 29625 39500 0,6667 0,1925 5701
B8 17500 26250 35000 0,6667 0,1925 5052
B9 14500 21750 29000 0,6667 0,1925 4186
Table 8: Societies in Inequality aversion experiment

Table 21: Respondent’s characteristics by treatment group

42
Mean Control Luck Effort Mobility
Total - 100% 100% 100% 100% 100%
Toma enserio - 88% 92% 85% 83% 94%
0cm 0 46% 45% 47% 42% 51%
Weakly 0-10 5% 7% 8% 2% 2%
work 11-20 5% 6% 3% 5% 6%
hours 21-30 8% 7% 10% 7% 9%
31-40 13% 13% 15% 13% 12%
> 40 22% 22% 16% 30% 20%
0cm 1 6% 6% 5% 5% 7%
How 2 23% 22% 16% 28% 24%
Many 3 28% 26% 34% 27% 25%
Peo- 4 24% 27% 27% 23% 21%
ple 5 12% 17% 8% 12% 13%
live 6 3% 3% 3% 1% 4%
with 7 1% 0% 3% 0% 2%
you 8 0% 0% 0% 1% 0%
9 0% 0% 0% 1% 0%
10 3% 1% 3% 1% 5%
0cm 1 9,87% 10,09% 9,59% 13,04% 6,77%
Fa- 2 12,11% 8,26% 16,44% 8,70% 15,04%
ther 3 5,13% 6,42% 5,48% 1,09% 7,52%
ed- 4 13,26% 12,84% 10,96% 11,96% 17,29%
u- 5 10,56% 11,01% 9,59% 14,13% 7,52%
ca- 6 1,00% 1,83% 0,00% 2,17% 0,00%
tion 7 0,00% 0,00% 0,00% 0,00% 0,00%
8 0,50% 0,92% 0,00% 1,09% 0,00%
9 4,00% 4,59% 5,48% 2,17% 3,76%
10 0,84% 0,00% 0,00% 1,09% 2,26%
11 11,27% 7,34% 12,33% 14,13% 11,28%
12 6,63% 6,42% 8,22% 4,35% 7,52%
13 2,18% 3,67% 1,37% 2,17% 1,50%
14 2,11% 0,92% 2,74% 3,26% 1,50%
15 14,06% 18,35% 15,07% 13,04% 9,77%
16 6,49% 7,34% 2,74% 7,61% 8,27%
0cm 1 9,84% 10,09% 6,85% 14,13% 8,27%
Mother
ed-
u- 43
ca-
tion
2 17,43% 11,01% 23,29% 17,39% 18,05%
3 7,81% 7,34% 10,96% 5,43% 7,52%
4 10,49% 7,34% 10,96% 10,87% 12,78%
5 4,30% 8,26% 4,11% 1,09% 3,76%
6 3,22% 6,42% 1,37% 4,35% 0,75%
7 4,42% 4,59% 1,37% 8,70% 3,01%
8 0,46% 0,00% 0,00% 1,09% 0,75%
9 0,00% 0,00% 0,00% 0,00% 0,00%
10 11,15% 8,26% 13,70% 7,61% 15,04%
11 2,49% 2,75% 1,37% 4,35% 1,50%
12 7,04% 10,09% 4,11% 8,70% 5,26%
13 1,83% 2,75% 2,74% 1,09% 0,75%
14 12,07% 11,01% 10,96% 9,78% 16,54%
15 7,44% 10,09% 8,22% 5,43% 6,02%
0cm 0-16.000 3,95% 2,75% 4,11% 2,17% 6,77%
House- 16.001-23.900 7,74% 7,34% 9,59% 6,52% 7,52%
hold 23.901-30.000 8,93% 11,01% 5,48% 8,70% 10,53%
In- 30.001-36.000 8,35% 6,42% 9,59% 7,61% 9,77%
come 36.001-42.500 4,85% 6,42% 4,11% 4,35% 4,51%
42.501-50.000 10,61% 6,42% 10,96% 13,04% 12,03%
50.001-58.000 7,91% 11,01% 9,59% 6,52% 4,51%
58.001-70.000 9,13% 8,26% 10,96% 9,78% 7,52%
70.001-89.000 10,35% 7,34% 12,33% 11,96% 9,77%
Mas de 89.000 10,74% 16,51% 5,48% 11,96% 9,02%
No sabe/No contesta 17,44% 16,51% 17,81% 17,39% 18,05%
0cm Men 43,40% 43,12% 53,42% 40,22% 36,84%
Sex Women 56,60% 56,88% 46,58% 59,78% 63,16%
0cm Min 17,5 18 17 18 17
Age 1Q 18 18 18 18 18
Mean 20,625 20 21 22,5 19
Mean 23,6 23,5 23,4 24,9 22,6
3Q 26,5 26 26 29 25
Max 54,25 55 53 56 53
0cm Foreign 7,40% 6,03% 9,64% 7,27% 7,30%
Sec- Public school 65,25% 63,79% 72,29% 60,91% 65,69%
ondary Public tech. school (UTU) 7,17% 6,03% 4,82% 10,91% 6,57%
school Other university 0,00% 0,00% 0,00% 0,00% 0,00%
in-
sti-
tu- 44
tion
Private-Lay 17,71% 19,83% 12,05% 18,18% 18,98%
Private-religious 2,47% 4,31% 1,20% 2,73% 1,46%

References Education levels: 1 Bachillerato completo 2 Bachillerato in-


completo 3 Ciclo ba¡sico completo 4 Ciclo basico incompleto 5 Enseñanza tec-
nica (UTU) 6 Maestria/Doctorado 7 Magisterio/IPA 8 Militar 9 No sabe/No
contesta 10 Policial 11 Primaria completa 12 Primaria incompleta 13 Tercia-
ria no universitaria completa 14 Terciaria no universitaria 15 Incompleta 16
Universidad completa 17 Universidad incompleta

45
Table 9: Aversion to inequality for every treatments. Position of the child
equal to mean income in society. Values in (%).

C1 C2 L1 L2 E1 E2 M1 M2
γ <0,09 24 22 18 17 26 23 15 16
-0,09< γ <0 7 7 8 8 13 11 4 4
0< γ <0,5 10 9 9 8 14 15 10 10
0,05< γ <0,9 6 10 3 3 4 6 6 6
0,09< γ <0,15 5 5 2 3 2 5 10 9
0,15< γ <0,21 7 6 7 10 5 6 9 9
0,21< γ <0,34 9 10 12 13 10 8 9 9
0,34< γ <0,51 9 10 7 8 7 6 16 15
0,51< γ <0,78 7 6 8 7 5 6 10 11
>0,78 16 15 25 24 15 14 11 10
N 141 177 106 144 123 162 158 179
Mean 0,19 0,19 0,30 0,30 0,14 0,15 0,24 0,23
Treatment: C=Control, L=Luck, E=Effort, M=Mobility. For each
treatment (1) is the distribution excluding observations that were
inconsistent in other sets of decisions, (2) is the distribution using all the
observation consistent at mean choices, no matter if then participant was
inconsistent in other positions.

46
Table 10: Aversion to inequality for every treatments. Position of the child
equal to minimum income in society. Values in (%).

C1 C2 L1 L2 E1 E2 M1 M2
γ <0,09 38 39 35 36 54 51 76 77
-0,09< γ <0 27 27 18 18 17 19 16 15
0< γ <0,5 9 7 5 4 4 4 0 0
0,05< γ <0,9 3 3 1 1 1 1 0 0
0,09< γ <0,15 2 2 2 4 1 1 0 0
0,15< γ <0,21 4 3 3 2 2 2 1 0,005
0,21< γ <0,34 1 1 3 3 2 2 1 0,005
0,34< γ <0,51 1 1 1 1 3 3 1 0,005
0,51< γ <0,78 2 4 5 4 2 2 0 0
>0,78 14 15 28 27 15 15 6 13
N 141 186 106 165 123 170 158 207
Mean 0,02 0,02 0,18 0,15 -0,03 -0,0 -0,22 -0,22
Treatment: C=Control, L=Luck, E=Effort, M=Mobility. For each
treatment (1) is the distribution excluding observations that were
inconsistent in other sets of decisions, (2) is the distribution using all the
observation consistent at minimum choices, no matter if then participant
was inconsistent in other positions.

47
Table 11: Aversion to inequality for every treatments. Position of the child
equal to maximum income in society. Values in (%).

C1 C2 L1 L2 E1 E2 M1 M2
γ <0,09 7 6 8 7 12 12 3 3
-0,09< γ <0 7 6 6 7 7 6 2 2
0< γ <0,5 16 15 15 15 24 24 4 3
0,05< γ <0,9 4 6 2 4 2 4 3 2
0,09< γ <0,15 4 6 4 5 4 5 5 6
0,15< γ <0,21 9 8 6 7 6 8 4 6
0,21< γ <0,34 13 13 7 8 14 13 21 19
0,34< γ <0,51 6 7 7 5 8 8 15 14
0,51< γ <0,78 11 12 9 9 7 5 18 19
>0,78 23 21 37 33 17 15 26 25
N 141 185 106 169 123 186 158 207
Mean 0,34 0,34 0,43 0,40 0,25 0,22 0,49 0,48
Treatment: C=Control, L=Luck, E=Effort, M=Mobility. For each
treatment (1) is the distribution excluding observations that were
inconsistent in other sets of decisions, (2) is the distribution using all the
observation consistent at maximum choices, no matter if then participant
was inconsistent in other positions.

Table 12: Survey details about invitations sent

Date Sent Bounced N Answers Rejected Completed Observations


28/05/18 3095 139 2956 618 29 589 First invitation
15/06/18 2335 2 2333 292 110 182 Reminder I
20/08/18 1994 0 1994 139 52 87 Reminder II

Table 13: Survey response rate

N No response rate Rejection rate Realization rate


2954 0,65 0,06 0,29

48
Table 14: Regresios to explain aversion to inequality

Estimation to inequality aversion


Estimation (1) (2) (3) (4) (5) (6)
Constant .3543339* .3904194* .3663004** .1288186 .3484811** .2543607**
(.0566831) (.0549335) (.0499359) (.1227001) (.0377911) (.0608407)
T1 .0811744* .1179923* .127675* .1587159**
(.0504059) (.0607522) (.0607496) (.061714)
T2 -.0787776** -.0551063* -.0416478* -.0259115*
(.047616) (.0512218) (.0511745) (.0519062)
T3 .1005668** .116016** .1222268** .1325689**
(.0370873) (.0485126) (.0482074) (.0476845)
Ineq 1 .0097241**
(.0419099)
Ineq2 .107556** .0737018**
(.0463777) (.0470921)
Effort -.0844929** -.0710128* -.0542485**
(.038579) (.0392819) (.0409067)
Left .3918812** .3610504* .2344971*
( .0316973) (.0556878) (.0639357)
Wage .0023451** -.0124846**
(.0373526) (.0385683)
Age .0483738** .0524966**
(.028924) (.0221436)
Sex .0416694* .0297698*
(.0393653) (.0381918)
Work .0223886**
(.0500632)
Educ 1 .2604696**
(.1267257)
Educ 2 .1546602
(.1230995)
Educ 3 .176974
(.1250742)
Educ 4 .1617082
(.1260689)
Educ 5 .2686249
(.1226334)
Educ 6 .1664703
(.1202775)
Afir 1 .0379793**
(.041536)
Afir 2 -.0210287**
(.0399059)
Afir 3 .1148386** .0975289**
(.0395971) (.0387912)
Afir 4 -.0195778**
(.0406253)
Afir 5 -.0575516** -.0531923**
(.0394849) (.0408238)
Afir 6 -.0878748** -.0653959**
(.0453835) (.0457771)
Obs 528 478 478 454 528 455
R2 0.03 0.04 0.06 0.04 0.04 0.1
Significance levels: *p < .5 **p < .025 ***p < .0125
Note:Table 15 contains a description of the variables used in regression

49
Table 15: Description of variables used in regression models

Variable Description
T1 Dummy =1 if treatment is luck
T2 Dummy =1 if treatment is effort
T3 Dummy =1 if treatment is mobility
Ineq 1 =1 if agree inequality is a problem (first question presented in Figure 9)
Ineq 2 =1 if agree inequality is a problem second question presented in Figure 9)
Effort =1 if attributes effort more than luck to results (question presented in Figure 8)
Left Dummy = 1 if answer 4 or less in last question presented in Figure 14
Wage =1 if consider minimum wage should be increased (question presented in Figure 8)
Age Age in years
Sex Dummy =1 if female
Work Dummy = 1 if individual works
Educ 1-6 Dummies for maximum educational level reached by parents
Afir 1-6 Dummies for affirmations included in the questionnaire (1 is agreement)

50
Table 16: Mean test for aversion to inequality for each treatment when grand-
children income position varies

Ho P r(T < t) P r(|T | > |t|) P r(T > t)


Control
mean max
γcontrol = γcontrol 0.0000 0.0000 1.0000
mean min
γcontrol = γcontrol 1.0000 0.0000 0.0000
min max
γcontrol = γcontrol 0.0000 0.0000 1.0000
Luck
mean max
γluck = γluck 0.0000 0.0000 1.0000
mean min
γluck = γluck 0.9959 0.0081 0.0041
min max
γluck = γluck 0.0000 0.0000 1.0000
Effort
mean max
γef f ort = γef f ort 0.0006 0.0013 0.9994
mean min
γef f ort = γef f ort 1.0000 0.0000 0.0000
min max
γef f ort = γef f ort 0.0000 0.0000 1.0000
Movility
mean max
γmobility = γmobility 0.0000 0.0000 1.0000
mean min
γmobility = γmobility 1.0000 0.0000 0.0000
min max
γmobility = γmobility 0.0000 0.0000 1.0000

51
Table 17: Mean test for aversion to inequality when grandchildren income is
equal to mean income in Society

n Mean S.E Ha: diff < 0 Ha: diff 6= 0 Ha: diff > 0
Control 177 0.1893907 0.0313574
Luck 144 0.3011844 0.0370148 0.011 0.021 0.990

Control 177 0.1893907 0.0313574


Effort 162 0.1524805 0.03264 0.792 0.416 0.208

Control 177 0.1893907 0.0313574


Mobility 179 0.2252442 0.0280627 0.197 0.395 0.803

Luck 144 0.3011844 0.0370148


Effort 162 0.1524805 0.03264 0.999 0.003 0.001

Luck 144 0.3011844 0.0370148


Mobility 179 0.2252442 0.0280627 0.952 0.097 0.049

Effort 162 0.1524805 0.03264


Mobility 179 0.2252442 0.0280627 0.045 0.090 0.955

52
Table 18: Mean test for aversion to inequality when grandchildren income is
equal to minimum income in Society

n Mean S.E Ha: diff < 0 Ha: diff 6= 0 Ha: diff > 0
Control 186 0.0235278 0.0323705 0.006 0.013 0.994
Luck 165 0.1527512 0.0407531
Control 186 0.0235278 0.0323705 0.827 0.347 0.173
Effort 170 -0.0211263 0.0347093
Control 186 0.0235278 0.0323705
1.000 0.000 0.000
Mobility 207 -0.224522 0.0226858
Luck 165 0.1527512 0.0407531
0.999 0.001 0.001
Effort 170 -0.0211263 0.0347093
Luck 165 0.1527512 0.0407531
1.000 0.000 0.000
Mobility 207 -0.224522 0.0226858
Effort 170 -0.0211263 0.0347093
1.000 0.000 0.000
Mobility 207 -0.224522 0.0226858

53
Table 19: Mean test for aversion to inequality when grandchildren income is
equal to maximum income in Society

n Mean S.E Ha: diff < 0 Ha: diff 6= 0 Ha: diff > 0
Control 185 0.336463 0.0282734 0.071 0.142 0.929
Luck 169 0.4001 0.0329788
Control 185 0.336463 0.0282734
0.998 0.004 0.002
Effort 186 0.2206651 0.0274658
Control 185 0.336463 0.0282734
0.000 0.000 1.000
Mobility 207 0.4775045 0.0234787
Luck 169 0.4001 0.0329788
1.000 0.000 0.000
Effort 186 0.2206651 0.0274658
Luck 169 0.4001 0.0329788 0.026 0.051 0.975
Mobility 207 0.4775045 0.0234787
Effort 186 0.2206651 0.0274658
0.000 0.000 1.000
Mobility 207 0.4775045 0.0234787

54
Table 20: Mean test for aversion to inequality useing unbiased estimation of
aversion to inequality

n Mean S.E Ha: diff < 0 Ha: diff 6= 0 Ha: diff > 0
Control 141 .3543339 .0370783 0.0767 0.1534 0.9233
Luck 106 .4355083 .0429064
Control 141 .3543339 .0370783
0.9385 0.1231 0.0615
Effort 123 .2755564 .0341459
Control 141 .3543339 .0370783
0.0169 0.0338 0.9831
Mobility 158 .4549008 .0298445
Luck 106 .4355083 .0429064
0.9982 0.0035 0.0018
Effort 123 .2755564 .0341459
Luck 106 .4355083 .0429064 0.3510 0.7019 0.6490
Mobility 158 .4549008 .0298445
Effort 123 .2755564 .0341459
0.0000 0.0001 1,000
Mobility 158 .4549008 .0298445

55
C Annex 3
C.1 Screen-shots of survey questions

Figure 2: Screen shoot

¿Usted prefiere que su nieto/a viva en la sociedad A o en la sociedad


B?

Figure 3: Screen shoot

56
Figure 4: Screen shoot

57
Figure 5: Screen shoot

58
Figure 6: Screen shoot

59
Figure 7: Screen shoot

60
Figure 8: Screen shoot

61
Figure 9: Screen shoot

62
Figure 10: Screen shoot

63
Figure 11: Screen shoot

64
Figure 12: Screen shoot

65
Figure 13: Screen shoot

66
Figure 14: Screen shoot

67
Figure 15: Screen shoot

68
69

You might also like