Inequality and Happiness
Andrew E. Clark (Paris School of Economics and CNRS)
http://www.parisschoolofeconomics.com/clark-andrew/
[Partly based on Clark and D’Ambrosio (2015), Handbook of
Income Distribution, Vol. 2A, Chapter 13]
We all like invitations to publish that are based on
the MPU.
For the Handbook, the original e-mail from Tony
and François invited a chapter on “Attitudes to
Inequality”:
Isn’t it obvious that individuals – including us here
now – don’t like inequality?
So surely inequality reduces happiness?
And indeed, a very simple model of inequality and
happiness produces the same prediction:
i) Utility is concave in income
ii) Utility is a monotonic input into the social
welfare function
So greater inequality reduces social welfare.
Game over!
But can it really be that easy?
We suspect not…
Inequality is a social phenomenon: it refers
to disparities in incomes between
individuals (i.e. there is income inequality
when some individuals receive different
incomes than do others).
The key axiom in the measurement of
inequality is the Pigou-Dalton principle of
transfers: inequality increases whenever
there is a transfer of income from a
poorer to a richer individual.
2) We can have a dispassionate normative
opinion about any distribution of income,
which is independent of our own position
in that distribution.
In the sense that we’d have the same
normative view, wherever we appear in
the distribution, or even if we don’t.
Can think of this as a “veil of ignorance”
evaluation, or a normative opinion about
the distribution of income in our own
countries in the 19th century, even though
we do not appear in that distribution.
But our gut feeling may well be that there is
too much inequality too…
Over and above our correct fixation on the
diminishing marginal utility of income, we
may still conclude that inequality reduces
subjective well-being, right?
Not finished yet though…
3) Most of the time, we also appear in the
income distributions that are changing.
So Any change in incomes will affect not
only my own income, but also the gaps
between my income and the income of
others in my society (to whom I compare)
– my reference group.
This brings about a passionate response, as
it were.
Changing income inequality affects not only
how much income I receive (my absolute
income), but also how much richer and
poorer I am compared to others.
In this sense, we can think of the utility from
income (the indirect utility function), as
depending on not only my income but
also the income of my reference group:
V = V(Yi, Yi, ref)
The reference group
The term RG was first used by Hyman
(1942) in work on the evaluation of the
rankings that individuals assign to
themselves: it is the group or individuals
to whom they compare for the purpose of
self-appraisal.
The term RG has subsequently been refined
numerous contributions across the social
sciences, and various definitions of the
term are now proposed.
The comparative reference group
The comparative RG is in the spirit of the
original interpretation by Hyman.
The reference group acts as the standard of
comparison for self-appraisal: it is the
comparison of the individual’s own
income to that of others in this reference
group that yields status.
Income inequality in the comparative RG
Self-appraisal depends on comparisons to
those richer and poorer to the individual,
and thus to income inequality.
We can further usefully distinguish the
comparative reference group according to
whether the individual in question is or is
not a member of the reference group.
When the individual is in the reference
group, we typically think that higher
income for others reduces my subjective
well-being (status).
V = V(Yi, Yi, ref)
+ -
When the individual is not (yet) a part of the
reference group, but aspires to be, the
reference group acts as a relative
aspiration, that is as the group of which
the individual desires to be a member
(signal).
V = V(Yi, Yi, ref)
+ +
(cf. Hirschman’s tunnel effect).
Let’s take all of this at face value.
The correlation between inequality and
happiness will be
1) Negative via own income (concavity)
2) Negative (?) via normative evaluation (but
this depends on your views of fairness)
3) Ambiguous via comparisons when you are
a RG member, as it depends how Yi
changes relative to Yi, ref
4) Ambiguous via comparisons when you are
not a RG member, as it depends how Yi
changes relative to Yi, ref
The jury really is out.
The main culprit so far for this ambiguity is
income comparisons: inequality implies
changes in both absolute income, and
relative income.
Life would be so much easier without the latter
So how do we know that income comparisons
matter for individual happiness?
Happiness approach: Luttmer (2005)
US National Survey of Families and
Households
Clark (1996). BHPS (UK): very local comparisons
Log hourly pay: ln(HPi)
Log Hours
Log spouse's
hourly pay: (ln(HPs))
Dummy: HPi > HPs
Log spouse's hourly pay
(when HPs > HPi)
0.111
(0.060)
-0.251
(0.061)
-0.121
(0.044)
-----
0.039
(0.068)
-0.246
(0.061)
-0.056
(0.052)
0.171
(0.074)
---
0.060
(0.066)
-0.250
(0.061)
-0.047
(0.059)
---0.069
(0.037)
Estimated only on couples where both partners are in work.
Includes other standard control variables.
Some papers on non-membership.
E.g. Senik (2004). RLMS (Russia)
A smaller number of recent papers have uncovered
empirical results where some measure of individual wellbeing is positively correlated with reference group
income or earnings: the more others earn, the more
satisfied I am.
This finding has been interpreted as demonstrating
Hirschman’s tunnel effect (Hirschman and Rothschild,
1973): while others’ good fortune might make me jealous,
it may also provide information about my own future
prospects.
Some papers on non-membership.
E.g. Senik (2004). RLMS (Russia)
The distinction between status (-) and signal (+)
depends on how likely you are to end up in the
future with your reference group’s current
income, and thus on mobility between groups.
Subjective Well-Being Measures are not the only possible way of
showing comparison effects.
1) The Leyden approach. Invert the question: Who needs more
money to have a good life?
Initiated by Bernard Van Praag.
Ask individuals to assign income levels (per period) to six or nine
different verbal labels (such as "excellent“,"sufficient" and
"bad").
Van Praag can then estimate for each individual a
lognormal “Welfare Function of Income” (the jump in
income required to go from one label to another
becomes increasingly large).
The resulting individual estimated means () and
variances () – there is one curve for each person –
were then used as dependent variables in regressions
to show which types of individuals require a higher
level of income to be satisfied, and which individuals
have valuations that are more sensitive to changes in
income.
Those with higher reference group income, and who had
earned more in the past, had higher values of .
Those with higher values of μ use lower verbal well-being labels at any given
level of income
2) Ask people. Preference for rising income
profiles, and preferences for lower absolute
incomes:
• A: Your current yearly income is $50,000;
others earn $25,000.
• B: Your current yearly income is $100,000;
others earn $200,000.
Individuals have a marked preference for A over
B.
Positionality differs according to the domain. In
Alpizar et al. (2005) this is stronger for cars and
housing, and weaker for vacations and
insurance.
3) Experimental.
a) Ultimatum game: responders frequently
rejecting offers that are under 25% of the total
sum; as such the the vast majority of offers
are between 40% and 50% of the sum.
b) Dictator game: non-strategic. The survey of
616 such experiments in Engel (2011)
concludes that dictators give on average
28.35% of the sum of money to be divided.
This may inform us of normative views of
income distributions rather.
c) Zizzo and Oswald (2001) report the results of
an experiment whereby subjects can pay to
burn each other’s money. A majority of
subjects chose to do so, even though it costs
them real earnings. The average subject had
half of her earnings burnt, and richer subjects
were burnt more often.
d) McBride (2010) shows how comparisons to
income expectations matter, in a matching
pennies game, where individuals play against
computers.
The computer chooses heads or tails according to
(known) probability distributions (for example
80% heads, 20% tails).
After each round of playing, individuals report
Introduces social comparisons in some of the treatments
(by telling the individual the outcomes of the other
players).
Income expectation effect identified by varying the heads
and tails probabilities played by the computer. Each
subject has five pennies to play. When paired with a
80% heads, 20% tails computer, the best strategy is to
always play heads, which gives an expected payoff of
four pennies. When paired with a 65% heads, 35%
tails computer, the best strategy is still to always play
heads, but now the expected payoff is only 3.25
pennies.
Results: satisfaction is
i) higher the more one wins
ii) lower the more others win
iii) lower the higher was the aspiration level.
4) Natural Experiments
Card et al. (2012): the revelation of information on
others' earnings.
The natural experiment here is a court decision that
made the salary of any California state employee public
knowledge.
A local newspaper set up a website making it easy to
find this information.
Following this website launch, Card et al. informed a
random subset of employees at three UC campuses
about the site.
Some days later, all employees on the three
campuses were surveyed.
Compare the treatment group (informed
about the website) to others to reveal the
impact of information on others' salaries.
The reference group was defined here as
co-workers in the same occupation group
(faculty vs. staff) and administrative unit in
the university.
The survey found lower job satisfaction for
those with pay below the reference group
median and a greater intention to look for a
new job.
The effect on both for those who were
relatively well-paid was insignificant.
There is some evidence of an actual
quitting effect on those who were found to
be in the bottom earnings quartile in the
reference group.
This is not a banal effect of “low pay leads
to lower satisfaction and greater quits”.
Pay in the treated and untreated groups is
the same.
The treated group are instead more likely to
find out that they are relatively badly-paid
5) Neuro. Fließbach, K., Weber, B., Trautner, P., Dohmen, T., Sunde, U.,
Elger, C., & Falk, A. (2007). "Social comparison affects reward-related brain
activity in the human ventral striatum". Science, 318, 1305-1308.
Payoffs vary according to whether the individual gets the task right, and also randomly
when the task is correct
Brain activity measured via BOLD blood flow in
various voxels.
Particular attention paid to the ventral striatum:
the “neural circuitry of reward”
This kind of striatal activity has been shown to
predict both hedonic outcomes (subjective
well-being) and physiological outcomes
(cortisol output: the body’s response to stress)
Brain activation depends on relative income: compare C6, C8 and C11 (where the
individual receives 60 Euros), and C7 to C9.
The normative approach
Evaluation of the overall degree of income
inequality in the reference group, without
making any comparisons to others.
Experimental attitudes to inequality
The experimental economics contributions to
inequality aversion from the more aggregate
perspective have appealed to two different
approaches:
1) inequality and risk aversion with a parametric
social welfare function;
2) general social welfare functions.
Experimental attitudes to inequality
1) In inequality and risk aversion with a parametric
social welfare function two types of experiments
have been run.
The first is similar to that adopted in the
experiments on status or relative income that is
the choice between alternative societies with
different income distributions behind the veil of
ignorance (the “hypothetical grandchild”).
The second is
experiment.
based
on
the
leaky-bucket
Experiment type 1: The well-being of imaginary
grandchilden in alternative societies which are
characterized by different uniform income
distributions (e.g., Society A ranges from 10,000 to
50,000 Swedish kroner, but Society B from 19,400
to 38,800 Swedish kroner).
Expected income higher in Society A
Choose the society that is best for your grandchild.
Respondents were also told that they did not know
their grandchild’s position in the income
distribution, and that they should place equal
probability on all outcomes.
The more inequality-averse the individual is,
the more they are willing to trade-off
expected income in order to achieve a
more equal income distribution.
Individuals do exhibit a considerable amount
of
inequality
aversion
in
these
experiments
Amiel, Creedy, Hurn (1999) belongs to the second type of
experiment type 1, in which social inequality aversion is
estimated via the leaky-bucket experiment.
A sample of students were asked to indicate the amount of
‘lost money’ that they were willing to accept for a
transfer of money from a richer to a poorer individual,
loss due to administrative costs.
The median value of inequality aversion was much lower
than the existing estimates from hypothetical
grandchildren.
However, the circumstances of the two experiments are very
different, making a clear comparison of the results rather
difficult.
2) general social welfare functions.
In the income-distribution literature the indices that are
deemed appropriate to measure inequality are those
which conform to the Lorenz dominance criterion.
These indices fulfill four basic axioms: scale invariance,
symmetry, the population principle and the Pigou-Dalton
transfer principle.
The first three properties are commonly assumed in the
majority of indices of well-being; only the principle of
transfers is at the heart of inequality measurement.
Attitudes towards inequality have been interpreted by some
authors as the reaction of the general public to these
four basic properties.
The seminal book is this area is Amiel and Cowell (1999).
Given that the defining concept for inequality
measurement is the Pigou-Dalton transfer principle, we
discuss only those experimental results which cover this
aspect of inequality.
Verbal experiment:
“Suppose we transfer income from a person who has more
income to a person who has less, without changing
anyone else’s income. After the transfer the person who
formerly has more still has more.”
60% agree that this reduces inequality.
Numerical experiment:
Consider two income distributions:
Society A = (l, 4, 7, 10, 13)
Society B = (l, 5, 6, 10, 13).
Only 1/3 agree that Society B is more equal than Society A
(even though the “transfer” between the two
corresponds to the Pigou-Dalton principle, and to the
previous verbal experiment)
Individuals think of falling income inequality in Robin Hood
terms (and perhaps also of rising inequality in Sheriff of
Nottingham terms)
What is then the sum total of own income,
income comparisons, tunnel effects and
the normative evidence?
Inequality and well-being
There are many equations estimated such
as:
Ineq here is almost always Gini.
Table 1 in our chapter provides a representative
sample of estimation results for γ above.
There are 27 rows:
• In 14 γ is < 0
• In 5 it is > 0
• In 6 it is = 0
• In one we don’t know
• And in the last, it is both positive and
negative.
Probably fair to say that this is inconclusive (and
beware of the Moulton correction!).
Note 1
Is the Gini the “best” measure of the distribution for the
normative evaluation? Gini moves relatively little over
time, making multicollinearity a distinct possibility in
cross-country work.
Others are possible, such as the income share of the top
quintile, D9/D1, p95/p50, the percentage in poverty, or
even rank in the income distribution.
Ebert and Welsch (2009) is relatively unusual in comparing
the fit of different distribution measures, in explaining life
satisfaction, and concluding that “the overall degree of
inequality aversion is larger than that implied by the
standard measures applied in empirical analysis”
Most applied work doesn’t compare distribution measures
Note 2
Fairness and perceptions.
Above measures of income are objective: they
measure what others in the society actually earn.
This is of course not necessarily what individuals
believe that others earn.
ISSP question:
"We would like to know what you think people in these jobs
actually earn. Please write in how much you think they
usually earn each year, before taxes. (Many people are
not exactly sure about this, but your best guess will be
close enough. This may be difficult, but it is important, so
please try.). First, about how much do you think a
bricklayer earns?"
And then: what these individuals should earn each
year before taxes, regardless of what they do
actually receive.
Asked for a bricklayer, Doctor in general practice, bank
clerk, an owner of a small shop, the chairman of a large
national company, a skilled worker in a factory, a farm
worker, a secretary, a city bus driver, an unskilled factory
worker, and a cabinet minister in a national government
.
Can use these to create a subjective (individuallevel) measure of the fairness of the income
distribution. Take top and bottom occupations:
Legitimate Inequality =
ln[(PIMD/PIunskilled)/(JIMD/JIunskilled)]
Legitimate inequality > 0 for those who believe
income gaps should be smaller.
For example, Schneider finds an average value of the first
term in the square brackets of around 644, and the
average value of JIMD/JIunskilled just over 300. This yields a
value of legitimate inequality of around 0.75. LI is
negatively correlated with life satisfaction.
Osberg and Smeeding (2006) use the entire set of
individual responses to calculate the perceived and just
Gini. Most people are in favour of some levelling of
incomes, while almost no-one believes that all incomes
should be the same. On average, the ratio of the Gini
coefficients is 0.75. In some countries, such as the US
and Japan, this figure is around 0.8, in others such as
Spain and Sweden it is under 0.7.
Perceptions of the fairness of the income
distribution affect the relationship between
income inequality and life satisfaction.
They also affect behaviour in the lab (decisions
depend on whether income is earned or
allocated).
The “manna from heaven” aspect of much of
income in laboratory experiments may make
them less good predictors of behaviour in the
real world, where income is earned.
How good is your perception of your home
country’s income distribution?
The OECD's new web-tool Compare your income
allows you to see whether your perception is in
line with reality. In only a few clicks, you can see
where you fit in your country's income
distribution.
http://www.oecd.org/statistics/compare-yourincome.htm
[Not telling you how well I did]
Note 3.
Preferences for redistribution
If we want to know about individuals’ attitudes
regarding income inequality, why don’t we just ask
them if they want less of it?
Key predictions here regarding individuals’ positions in
the income distribution, both now and what they
expect in the future (POUM).
Empirically, both current income and predicted or
experienced mobility do predict the individual’s
desire to redistribute income.
Alesina et al. (2004) show that the effect of inequality
on happiness is larger in value in Europe than in the
USA: because of greater perceived social mobility in
the USA than in Europe
Fairness plays a role here too. Those who believe
that income is due to hard work (or that the poor
are lazy) are less willing to redistribute.
And the actual level of inequality does increase the
desire to redistribute.
Note 4.
To whom do we compare?
Almost all of the survey literature assumes that
everyone compares to everyone.
In
the experimental literature, which can
manipulate such things, comparisons to people
richer than you matter more than comparisons to
those poorer than you.
And it is unlikely that we all compare to the whole
income distribution, making the overall wellbeing effects of any change in the income
distribution even harder to predict.
Altruism
In addition, we may be compassionate about some
people in our reference group.
So we are happy when they become better off.
See the work of Elizabeth Dunn on altruism, and
the literature on charitable giving.
Note 5.
Other outcome measures. We have looked at SWB
and the desire to redistribute.
Other intriguing work has highlighted significant
empirical correlations between (almost always)
the Gini coefficient and:
• Agreeableness (Big Five):
-
Note 5.
Other outcome measures. We have looked at SWB
and the desire to redistribute.
Other intriguing work has highlighted significant
empirical correlations between (almost always)
the Gini coefficient and:
• Agreeableness (Big Five):
• Trust:
-
Note 5.
Other outcome measures. We have looked at SWB
and the desire to redistribute.
Other intriguing work has highlighted significant
empirical correlations between (almost always)
the Gini coefficient and:
• Agreeableness (Big Five):
• Trust:
• Political Participation:
-
Note 5.
Other outcome measures. We have looked at SWB
and the desire to redistribute.
Other intriguing work has highlighted significant
empirical correlations between (almost always)
the Gini coefficient and:
•
•
•
•
Agreeableness (Big Five):
Trust:
Political Participation:
Support for globalisation:
-
Note 5.
Other outcome measures. We have looked at SWB
and the desire to redistribute.
Other intriguing work has highlighted significant
empirical correlations between (almost always)
the Gini coefficient and:
•
•
•
•
•
Agreeableness (Big Five):
Trust:
Political Participation:
Support for globalisation:
Violent behaviour:
+
Note 5.
Other outcome measures. We have looked at SWB
and the desire to redistribute.
Other intriguing work has highlighted significant
empirical correlations between (almost always)
the Gini coefficient and:
•
•
•
•
•
•
Agreeableness (Big Five):
Trust:
Political Participation:
Support for globalisation:
Violent behaviour:
Self-enhancement:
+
+
Note 5.
Other outcome measures. We have looked at SWB
and the desire to redistribute.
Other intriguing work has highlighted significant
empirical correlations between (almost always)
the Gini coefficient and:
•
•
•
•
•
•
•
Agreeableness (Big Five):
Trust:
Political Participation:
Support for globalisation:
Violent behaviour:
Self-enhancement:
Female Preferences for facial masculinity:
+
+
+
Note 6.
Causality?
Let’s just say that this has been treated in a pretty
cavalier fashion in this literature.
Changes in income happen for a reason: could be
that it is this reason that affects well-being, not
income inequality as such.
Or that happiness causes inequality, rather than
inequality causing happiness.
So all we need is an exogenous movement in the
income distribution…
If only…
There is some interesting work that has appealed
to movements in the minimum wage.
Support for minimum-wage rises highest amongst
minimum wage workers.
And lowest amongst those who earn just above the
minimum wage (last-place aversion).
In unpublished work on the SOEP we find negative
correlation between life satisfaction and
minimum wage rises in Germany that is the
largest for those earning just over the minimum
wage.
Note 7.
And finally…
We don’t really care about income
Income is only instrumental: we like income because
it makes us happy/brings us utility.
Rather than looking about income inequality, what
about happiness inequality?
Economic growth reduces the inequality of subjective
well-being
Between countries
And within countries
This is (we think) a nice complement to the
Easterlin Paradox, where income growth at
the country level doesn’t seem to be
correlated with average happiness in that
country.
Here are some examples:
United Kingdom (BHPS): Real GDP per capita and life satisfaction
1996 - 2008
West Germany (SOEP): Real GDP per capita and life satisfaction
1984 - 2010
Australia (HILDA): Real GDP per capita and life satisfaction
2001 - 2009
Life Satisfaction in Five European Countries, (World Database of Happiness)
1973-2004
Average Life Satisfaction
4
3.5
3
2.5
2
1.5
UK
1
France
Germany
Italy
Netherlands
0.5
0
1973
1977
1980
1983
1986
Year
1988
1990
1993
1996
2000
2004
40000
3
2.5
Average Happiness
30000
2
20000
1.5
1
Happiness
Real Income Per Capita
10000
0.5
0
0
1973
1977
1981
1985
1989
Year
1993
1998
2003
Real Income Per Capita (2000 US$)
Happiness in 30 years of GSS data
We think that higher income reduces the dispersion
of well-being through the provision of public goods
(paid for by the rich, enjoyed by all)
Happiness inequality is determined by income
inequality as well.
Modern growth has come with greater income
inequality.
In most countries the level of income effect has
trumped the income inequality effect.
One exception: the US post-around 1990:
GSS Data