Income Inequality and Voter Turnout
Daniel Horn
GINI Discussion Paper 16
October 2011
Growing Inequalities’ Impacts
Acknowledgement
Earlier version of this paper was presented at the GINI Year 1 conference in Milan, February 4-5. 2011. Comments
from the participants and comments from Márton Medgyesi and Tamás Keller are warmly acknowledged. Remaining errors are solely mine.
October 2011
© Daniel Horn, Amsterdam
General contact: [email protected]
Bibliographic Information
Horn, D. (2011). Income inequality and voter turnout - evidence from European national elections.
Amsterdam, AIAS, GINI Discussion Paper 16.
Information may be quoted provided the source is stated accurately and clearly.
Reproduction for own/internal use is permitted.
This paper can be downloaded from our website www.gini-research.org.
Income inequality and voter turnout
Evidence from European national elections
Daniel Horn
TÁRKI Social Research Institute (TÁRKI)
Institute of Economics, Hungarian Academy of Sciences
17 October 2011
DP 16
Daniel Horn
Page • 4
Income inequality and voter turnout
Table of contents
Abstract.......................................................................................................................................................................7
1. Voter turnout and inequality..................................................................................................................................11
1.1.Hypotheses................................................................................................................................................................16
2. Data and method. ..................................................................................................................................................17
3.Results. ..............................................................................................................................................................21
3.1.Declining turnout – hypothesis 1............................................................................................................................21
3.2.Income bias – hypothesis 2a....................................................................................................................................22
3.3.Inequality on the top and at the bottom – hypothesis 2b......................................................................................24
3.4.Universal welfare states – hypothesis 3.................................................................................................................25
4.Conclusion and further comments. ...........................................................................................................................29
References..................................................................................................................................................................31
Appendix.....................................................................................................................................................................33
GINI Discussion Papers.................................................................................................................................................47
Information on the GINI project.....................................................................................................................................49
Page • 5
Daniel Horn
Page • 6
Income inequality and voter turnout
Abstract
The paper looks at the link between inequality and voter turnout, and derives three hypothesis from previous
literature. It is shown that inequality associates negatively with turnout at the national elections (hypothesis 1).
Although this is not a very strong effect, but it is net of several factors affecting voter turnout that are empirically
well proven – such as individual characteristics or different features of the political system. The literature suggests
that this negative association is either due to the lower turnout of the poor relative to the rich in high inequality
countries (hypothesis 2) or due to the effects of the universal welfare state, which increases turnout through altered
social norms as well as decreases inequality through government intervention (hypothesis 3). Although none of
the hypotheses were refuted, neither was really supported by the data. I also tested whether inequalities at the top
or at the bottom have a different affect on turnout. Although the results, again, are not very robust, it seems that
larger differences in income between the very rich and the middle decreases overall turnout, while higher difference between the middle and the very poor increases turnout. This is just the opposite of what is expected from the
Downsian rational voter model.
JEL codes: D72, D63
Page • 7
Daniel Horn
Page • 8
Income inequality and voter turnout
While effects of individual resources and institutional characteristics on political participation have long been
established, there is no knowledge of the effect of inequality on political participation. In particular, does an increase in inequality mobilize or de-mobilize citizens to participate in politics? We know only a little how societal
environment affects voter political participation. Inequality, for instance, can have an impact through changing
social norms (Lister 2007), through altered political agenda (Solt 2010; Mueller and Stratmann 2003) or through
other chanels (Paczynska 2005). It is thus likely that in societies with greater income inequalities, we should observe polarization of participation modes, where higher inequality will be associated with a larger divergence of
participation. This paper looks only at one of these modes of political participation: voter turnout. Although voting
can be seen as the least unequal type of participation, it is still far from being unbiased (Lijphart 1997). We know
that voting is strongly conditioned on socio-economic position (Geys 2006b; Lijphart 1997; Blais 2006; Gallego
2007), on civic resources (Verba et al 1995), and also on country level factors (Geys 2006b, 2006a; Blais 2006).
But we know less about the link between inequality and voter turnout. This paper will look at this link, and speculate about the possible reasons why inequality might influence voter turnout.
In the following, I first review the current literature on the link between inequality and voter turnout, and
derive some hypotheses from these. The next section introduces the European Election Survey (EES) data which
is used to test the hypotheses, and executes the tests. The third section speculates about the results, while the last
section concludes.
Page • 9
Daniel Horn
Page • 10
Income inequality and voter turnout
1. Voter turnout and inequality
Voter turnout has been low and steadily declining, especially in the developed countries, throughout the last
decades (e.g. Lijphart 1997). The average European voter turnout was around 85% up until the mid-80’s, and
has dropped a massive 10-15 percentage points ever since. This drop is partly due to the introduction of the ten
Central-Eastern European (CEE) countries into the European community, but can also be observed within the
Western-European states, although to a smaller extent, as well as within the CEE part of Europe (see 1. figure
below and 10. figure in the appendix).
CEE
50
60
70
80
90
non-CEE
1940
1960
1980
2000
20201940
1960
1980
2000
2020
Year
Source:IDEA
Voter turnout, parlamentary elections
60
70
80
90
Voter turnout, parlamentary elections
1. 1.figure–AveragevoterturnoutinEuropeancountries
figure – Average voter turnout in European countries
all Europe
1940
1960
1980
Year
2000
2020
Source:IDEA
Similarly, income inequalities have grown during the last couple of decades. This trend is
Similarly, income inequalities have grown during the last couple of decades. This trend is observable in more
observable in more than two-thirds of the OECD countries, independent of the utilized
than two-thirds of the OECD countries, independent of the utilized measure (OECD 2008). Income inequali-
measure (OECD 2008). Income inequalities tend to fluctuate much less than voter turnout
ties tend to fluctuate much less than voter turnout (see 2. figure below and 11. figure in the appendix). Also, it is
(see 2. figure below and 11. figure in the appendix). Also, it is questioned whether the observed
questioned
whether
observed variance
of the indicators
of inequality
are due to
the actualof
variance
social
variance
of thetheindicators
of inequality
are due
to the actual
variance
socialofinequalities
or
inequalities
or rather
due to measurement
bias.
rather due
to measurement
bias.
Nevertheless, both voter turnout and measures of income inequality vary considerably
between countries. The argument that we should only observe income measures to change
very little or very slowly would question the adequacy of time series models (unless data for a
long period were available). But variance between countries offers the possibility to identify
the relation between inequality and voter turnout using cross-country models.
Page • 11
Daniel Horn
Nevertheless, both voter turnout and measures of income inequality vary considerably between countries. The
argument that we should only observe income measures to change very little or very slowly would question the
adequacy of time series models (unless data for a long period were available). But variance between countries offers the possibility to identify the relation between inequality and voter turnout using cross-country models.
2.figure‐AverageGinicoefficientinEuropeancountries
2. figure - Average Gini coefficient in European countries
CEE
26
28
Gini
30
32
non-CEE
1995
2000
2005
2010
1995
2000
2005
2010
year
Source:Eurostat
Gini
28 28.5 29 29.5 30 30.5
all Europe
1995
2000
year
2005
2010
Source:Eurostat
In order to minimize measurement bias, I use several indicators of inequality (see 2. table in the appendix).
In order to minimize measurement bias, I use several indicators of inequality (see 2. table in
Beside the Eurostat’s income Gini coefficient I use a Gini of earnings (SSO 2009), an s80/s20 ratio (SSO 2009), the
the
appendix). Beside the Eurostat’s income Gini coefficient I use a Gini of earnings (SSO
mean distance
from the median
Lancee and
vanmean
de Werfhorst
(2011),from
a poverty
ratemedian
from the Statistics
2009),
an s80/s20
ratioindicator
(SSOof2009),
the
distance
the
indicator of Lancee
on Income
Conditions(2011),
(SILC) database
and two p95/p5
one from
the Luxembourg
Income and Living
and
van and
de Living
Werfhorst
a poverty
rate measures,
from the
Statistics
on Income
Study (LIS) (Tóth
and Keller
2011) and and
one from
thep95/p5
SILC database.
These all indicate
overall income
inequaliConditions
(SILC)
database
two
measures,
one from
the Luxembourg
Income
Study
(LIS)
and above
Keller
and
oneI from
database.
all indicate overall
ties. I also
look (Tóth
at inequalities
and2011)
below the
median.
use the the
aboveSILC
and below
the medianThese
MDMI indices
income
inequalities.
alsoand
look
at inequalities
above
below
the
median.
I use the above
(Lancee and
van de WerfhorstI 2011)
the p95/p50
and p50/p5 figures
fromand
the LIS
and from
the SILC
databases
and
median MDMI indices (Lancee and van de Werfhorst 2011) and the p95/p50
(see below
3. table inthe
the appendix).
and p50/p5 figures from the LIS and from the SILC databases (see 3. table in the appendix).
3. figure below shows the association between voter turnout and different measures of income
inequalities. Apparently, the association is not very strong, and negative – if any.
This mild association is unsurprising if we consider that voter turnout is directly influenced
byPagemany
factors, mostly unrelated to inequalities. Below I summarize the main driving forces
• 12
of voter turnout and also present some hypotheses about the relation of inequality and
Income inequality and voter turnout
3. figure below shows the association between voter turnout and different measures of income inequalities.
Apparently, the association is not very strong, and negative – if any.
This mild association is unsurprising if we consider that voter turnout is directly influenced by many factors,
mostly unrelated to inequalities. Below I summarize the main driving forces of voter turnout and also present some
hypotheses about the relation of inequality and turnout.
3.figure–theassociationbetweenmeasuresofincomeinequalityandvoterturnout
Voter turnout
40
60
80
HU
CZFI
SI
IE
SK
MT
GR
IT
100
BE CYLU
DK
SE
AT NL DE
ES
PT
FR EEUK
BG
PL
LT
LU MTBE
CY GR
DK
SENL
AT
IT
DE
ES
HU
FISICZ IEPT
EE
FR UK
SK
Voter turnout
40
60
80
100
3. figure – the association between measures of income inequality and voter turnout
LV
LT
BG
RO
20
20
RO
LV
PL
Voter turnout
40
60
80
DK
SE NL
AT
BE LU
CY
DE
GR
IT
ES
HU
IE
EE
UK LV
PT
PL
10
20
BE
DK
30
40
Poverty, SILC
50
LU
CY
GR
SE
ITNLDEAT
ES
IEHU
FI CZ
SI
UK EE LV
FR
SK
PL
LT
60
PT
20
LT
20
SI CZ FI
FR
SK
40
100
30
35
Gini, Eurostat 2009
Voter turnout
40
60
80
25
100
20
.3
.4
.5
MDMI
.6
.7
.8
.2
.25
.3
.35
earning Gini, SSO
.4
(source: European Election Study/Piredeu, Eurostat, SILC, Lancee-v.d.Werfhorst, SSO)
(source:EuropeanElectionStudy/Piredeu,Eurostat,SILC,Lancee‐v.d.Werfhorst,SSO)
The most
often
model
to predict
turnout
is the
Downsian
rational vote
The most
oftenused
used model
to predict
individualindividual
voter turnout voter
is the Downsian
rational
voter
model (Downs
model
1957).
statesthey
that
vote or not base
1957). (Downs
The model states
that The
peoplemodel
decide whether
willpeople
vote or notdecide
based onwhether
an expectedthey
utility.will
The expected
onutility
an expected
is versus
the benefit
from
theirparty
choice
(party) being th
is the benefit utility.
from their The
choiceexpected
(party) beingutility
the winner
the disutility
of another
being elected,
winner
versus the disutility of another party being elected, multiplied by the probability o
multiplied by the probability of their vote being decisive and the costs of voting subtracted from this. The paradox
their vote being decisive and the costs of voting subtracted from this. The paradox of (no
of (not) voting is thus the fact that this “equation” is likely to be negative if many people vote (since the prob-
voting is thus the fact that this “equation” is likely to be negative if many people vote (sinc
ability of a vote being decisive is almost nil, while the costs of voting is likely to be greater than zero), but if few
the probability of a vote being decisive is almost nil, while the costs of voting is likely to b
people vote the expected utility is certainly positive (since the probability of a vote being decisive is great). Many
greater than zero), but if few people vote the expected utility is certainly positive (since th
resolutions for this paradox have been developed (see Geys 2006b for a comprehensive review). The addition of
probability of a vote being decisive is great). Many resolutions for this paradox have bee
consumption benefit (Riker and Ordeshook 1968), taking ethical or altruistic preferences into account (Goodin
developed
(see Geys 2006b for a comprehensive review). The addition of consumptio
benefit (Riker and Ordeshook 1968), taking ethical or altruistic preferences
into accoun
Page • 13
(Goodin and Roberts 1975), the minimax regret strategy (Ferejohn and Fiorina 1974) or gam
Daniel Horn
and Roberts 1975), the minimax regret strategy (Ferejohn and Fiorina 1974) or game theoretical approaches (Palfrey and Rosenthal 1983, 1985; Ledyard 1984) have all tried to address the paradox of voting. Indeed, the “pure”
Downsian model of voting addresses the questions of marginal changes (why a middling person might vote) much
better than the aggregate level of turnout (how many people vote) (Geys 2006b, p18).
Although using the Downsian framework it is hard to explain aggregate levels of turnout, there are several,
empirically well documented factors that increase or decrease one’s probability to cast a vote. Individual characteristics certainly matter: richer, more affluent people are much more likely to vote, just as higher education leads to
a higher probability of voting (Lijphart 1997). The literature, understandably, is more occupied with country level
factors that affect turnout. For instance Blais (2006) and Geys (2006a) both provide comprehensive reviews about
these factors. Geys (2006a) clusters country level factors into three groups: socio-economic, political and institutional variables. Examples of socio-economic factors are population size, population concentration, population
stability, population homogeneity or previous turnout level. Political variables can be the closeness (or marginality) of an election (i.e. how close the outcome of the election is), campaign expenditures, or political fragmentation. Institutional variables are the electoral system (majority, proportional representation or plurality voting),
compulsory voting, concurrent elections, registration requirements… etc. Geys (2006a) in his review concludes
that little agreement has been reached with many of the above factors. Institutional factors are the most consensual:
compulsory voting, easier registration procedures, concurrent elections and proportional representation all foster
higher turnout. Population size and electoral closeness also seem to be affecting turnout in general, although several of the analyzed papers had not found any link between them.
The review also notes that population heterogeneity (homogeneous groups within the society) seem to have no
effect on turnout, although theoretically “as cohesion increases group solidarity (and ‘social pressure’), political
participation in communities with high degree of socio-economic, racial or ethnic homogeneity should be higher
than in areas where this is not the case” (Geys 2006a p.644-645, emphasis in original). The question similar to population heterogeneity is in the focus of this paper as well, so this no-relationship finding is discouraging. However,
the reviewed papers are mostly using a Herfindahl-Hirschmann concentration index to proxy heterogeneity, which
is quite distant from the measures of inequality (used by this paper). Moreover, there are convincing new studies
(e.g. Kaniovski and Mueller 2006; Yamamura 2009; Funk 2008) that argue that more heterogeneous communities
Page • 14
Income inequality and voter turnout
are less likely to vote, in line with the expectations of the group-based model (see Uhlaner 1989; Grossman and
Helpman 2002; Filer, Kenny, and Morton 1993).
There are some studies that directly test the association of inequality and voter turnout. The most comprehensive study is Solt’s (2010) testing the Schattschneider hypothesis (Schattschneider 1960). In his book, Schattschneider wrote that large economic inequalities lead to low participation rates as well as a high income bias in
participation. “As the rich grow richer relative to their fellow citizens […] they consequently grow better able
define the alternatives that are considered within the political system and exclude matters of importance to poor
citizens” (Solt 2010 p.285) Hence poor will less likely to cast a vote, as inequality goes up, since their expected
benefit from voting declines. Solt (2010) uses American gubernatorial elections data to test the association between turnout and inequality. He uses state level Gini coefficient calculated for three years (1980, 1990, 2000) to
proxy income inequality, while voter turnout is also for these years. Solt shows that income inequality associates
negatively with electoral participation, while higher income people tend to vote relatively more as inequality rises.
A similar conclusion is presented by Mueller ad Stratmann (2003), but with a different theoretical approach.
They argue that if upper classes have higher participation rates than lower classes, and upper classes favor right
of center parties, lower classes left of center parties, and right of center parties adopt policies that benefit the upper classes, while left of center parties adopt policies that favor the lower classes, then lower participation rates
will lead to higher income inequalities. Hence their conclusion: voter turnout associates negatively with income
inequality, but it is the decreasing participation rate that drives inequalities and not vice-versa. That is, their result
is the same, but the line of argument is different from that of Schattschneider (1960). The Muller and Stratmann
(2003) argument fits the Meltzer and Richard (1981) logic, namely that “when the mean income rises relative to
the income of the decisive voter, taxes rise, and vice versa.” If fewer people vote, then relatively more rich people
vote, so median voter income will be larger (with mean income unchanged), which decreases taxes (preferences
for redistribution is smaller).
One might also argue oppositely. Based on the Meltzer and Richard (1981) logic, if government decides only
about the size of redistribution, then voter turnout should relatively be low if inequality is low, and turnout be high
if inequality is high. When inequality is low, then poor people have little to gain, and rich little to lose if government redistributes, so why would they vote? Similarly if inequality is high, then poor have a lot to gain, and rich
Page • 15
Daniel Horn
have a lot to lose from redistribution, hence they will vote. This, of course, is an overly simplified argument not
taking into account several other incentives driving one to vote.
While both Solt (2010) and Muller and Stratmann (2003) base the negative association of inequality and voter
turnout on differences in participation rates between people with different incomes, Lister (2007) uses differences
in social norms between countries to explain the negative association. He argues that the missing link (omitted
variable) between inequality and turnout is institutions. Institutions affect social norms, which affect individual
behavior. Universalist welfare states encourage solidarity and participation, and thus foster higher voter turnout
than other types of welfare states. Nevertheless, his argument also leads to a negative relation between inequality
and turnout: universal welfare states tend to have lower income inequalities and higher turnout.
The Downsian median voter logic, on the other hand, might lead one to argue in a different way. Growing
inequalities might increase the probability of the lower income/ lower class people to influence politics more, if
they can coalesce with the middle. In other words, if rising inequalities are due to rising income on the top, then
the redistributive preferences of middle will be closer to the bottom than to the top; thus the middle might unite
with the lower income/lower classes to “conquer” the upper classes. This would mean that higher inequality on the
top would lead to a relatively higher turnout for the lower class/lower income. On the other hand, if rising income
is due to a relatively decreasing income for the poor (as compared to the middle), then the coalition of the middle
with the upper classes seem theoretically more likely (Lupu and Pontusson 2011). Hence, when looking at the relation between inequality and turnout one has to look not only at measures of general income inequality but also at
differences between the bottom and the middle and the top.
1.1.
Hypotheses
From the above literature I derive three separate hypotheses for testing:
1. Inequality associates negatively with voter turnout, ceteris paribus the other factors that are shown to
influence turnout.
2. The reason for this negative association is that
a. turnout for lower income people tend to be relatively smaller, when overall inequality is high
(i.e. if inequality is high poor people tend to vote less, while rich tend to vote more, but this latter
does not counterbalance the drop of “poor-votes”), or alternatively
b. turnout for lower income people tend to be relatively smaller if inequality at the bottom is high,
but turnout for lower income tend to be relatively higher if inequality on the top is high.
or
3. Universal welfare states have higher turnout as well as lower income inequality.
Page • 16
Income inequality and voter turnout
2. Data and method
I use the 2009 PIREDEU European Election Study (EES 2010; van Egmond et al. 2010) to test the association between inequality and turnout. The study was conducted right after the 2009 European parliamentary elections with the aim to research the EU elections. The main advantage of these surveys is that they contain all 27
European countries, with approximately 1000 responses from each. Besides the turnout measure it also contains
a modest background questionnaire about individual characteristics, including education, gender and a subjective
income measure (see below). The EES also provides substantial amount of data about the institutional system. The
EES data was collected at one point in time in each country, thus the time since the last national election varies
across countries, as a consequence the responses about actual turnout will also be differently overstated (people
remember harder to an earlier election). 1. table below shows the participating countries and their aggregate voter
turnout. The right column shows the actual turnout at the 2009 national elections. Unfortunately, the questionnaire
did not have a question about actual turnout, but rather asked about the party vote. Nevertheless this question had
the option “did not cast a vote” but many have refused to answer (e.g. in Italy, where voting is compulsory more
than 26% of the voters did not answer), and also many had not remembered the action (e.g. in Latvia almost 20%
did not know the answer). Nevertheless, I relied on those, who had definite answer; hence the turnout measure is
those who voted over those who voted plus those who did not vote (reported figure column). Since surveys tend to
overestimate the actual voter turnout – and as can be seen, differences between the reported and the official figures
are sometimes substantial (e.g. Slovakia or Romania) – I used the ratio of the 2009 official/reported voter turnout
ratio as weights in the estimations below.
Page • 17
Daniel Horn
1. table – Voter turnout, actual and observed
2009
refused to answer
not voted
voted
not eligible
don't know
reported
figure*
official figure
AUSTRIA
BELGIUM
BULGARIA
CYPRUS
CZECH REPUBLIC
DENMARK
ESTONIA
FINLAND
FRANCE
GERMANY
GREECE
HUNGARY
IRELAND
ITALY
LATVIA
LITHUANIA
LUXEMBOURG
MALTA
NETHERLANDS
POLAND
PORTUGAL
ROMANIA
SLOVAKIA
SLOVENIA
SPAIN
SWEDEN
UK
13,60
17,47
10,60
9,00
5,59
1,10
3,57
4,80
16,40
12,75
5,60
11,24
5,39
26,30
3,20
6,90
8,29
30,90
3,18
4,29
14,60
7,78
6,50
8,30
10,90
3,29
6,80
2,90
4,79
19,80
4,50
21,67
3,40
19,86
8,10
7,70
5,48
6,10
14,53
6,79
5,80
15,08
23,40
9,69
2,50
5,57
22,65
8,80
18,64
14,86
7,60
8,30
2,00
11,60
77,60
68,86
56,80
78,70
65,49
92,50
64,15
76,40
61,70
71,41
84,50
69,75
74,53
57,60
58,14
58,10
62,64
60,80
86,17
62,97
65,00
64,61
69,78
76,60
76,80
88,22
73,40
1,30
1,90
2,20
2,80
2,55
1,00
2,48
1,80
3,50
3,39
2,20
1,19
3,50
1,00
4,40
1,90
8,19
2,00
1,49
2,20
4,40
0,50
2,46
0,80
1,60
2,50
3,50
4,60
6,99
10,60
5,00
4,71
2,00
9,93
8,90
10,70
6,97
1,60
3,28
9,79
9,30
19,18
9,70
11,19
3,80
3,58
7,88
7,20
8,47
6,40
6,70
2,40
3,99
4,70
0,96
0,93
0,74
0,95
0,75
0,96
0,76
0,90
0,89
0,93
0,93
0,83
0,92
0,91
0,79
0,71
0,87
0,96
0,94
0,74
0,88
0,78
0,82
0,91
0,90
0,98
0,86
0,79
0,91
0,56
0,89
0,64
0,87
0,62
0,65
0,60
0,78
0,87
0,68
0,67
0,78
0,62
0,49
0,92
0,93
0,80
0,54
0,64
0,39
0,55
0,63
0,74
0,82
0,62
mean
9,56
10,46
70,49
2,47
7,02
0,87
0,74
*voted/(voted+ not voted)
source: European Election Study/Piredeu 2009.
The EES data allows for numerous individual controls. The base model (see 6. table in the appendix) includes
individual level controls as well as country level controls. The individual controls are age, age squared, gender,
age when the respondent finished education and a within country standardized “subjective standard of living”.1
Country level controls are compulsory voting, multiple election at the same time, size of population, existence of a
threshold for a party to get in the parliament, electoral system (from proportional (0) to plurality (5)), presidential
system, federalism, time since last national election (years), percentage of other nationalities, GDP as percentage
1
The question for the subjective standard of living was: „Taking everything into account, at about what level is your family’s standard of
living? If you think of a scale from 1 to 7, where 1 means a poor family, 7 a rich family, and the other numbers are for the positions in
between, about where would you place your family?” Since the country mean for this question tend to correlate with income inequalities
I standardized the answers within country (0 mean, 1 sd).
Page • 18
controls
are for
age, aage
squared,
gender,
ageparliament,
when the respondent
finished
education
and a
threshold
to get
in when
the
system
(from
are squared,
age,party
agegender,
squared,
gender,
whenelectoral
the
respondent
finished
education (0)
and to
a
controlscontrols
are age, age
age
the age
respondent
finished
education
and aproportional
1
within
country voting,
standardized
“subjective
standard
ofsame
living”.
Country
level
controls are
compulsory
multiple
election
at
the
time,
size
of
population,
existence
of
a
1 Country
1 Country
pluralitystandardized
(5)), presidential
system,
federalism,
time
national
election
(years),
within country
“subjective
standard
of living”.
level last
controls
are level
within
country
standardized
“subjective
standard
of since
living”.
controls
are
compulsory
voting,
multiple
election
at
the
same
time,
size
of
population,
existence
ofandavoter turnout (0) to
Income inequality
threshold
for
a
party
to
get
in
the
parliament,
electoral
system
(from
proportional
compulsory
voting, multiple
at the
same
size
of population,
of a all variables
percentage
ofvoting,
otherelection
nationalities,
GDPtime,
as the
percentage
of mean
EU
27. See
compulsory
multiple
election
at
same
time,
sizeexistence
of population,
existenceinofthe
a
threshold for a party to get in the parliament, electoral system (from proportional (0) to
plurality
(5)),
presidential
system, electoral
federalism,
time
since
last national
election (years),
threshold
for
a
party
to
get
in
the
parliament,
system
(from
proportional
(0)
to
appendix
4.
table
andtosystem,
5.gettable
. I parliament,
use all
of since
these
variables
controls
in each of
threshold
forpresidential
a party
infederalism,
the
electoral
systemas
(from
proportional
(0)the
to
plurality
(5)),
time
last
national
election
(years),
of mean
EU
27.
See
all
variables
in
the
appendix
4.
table
and
5.
table.
I
use
all
of
these
variables
as
controls
in
each
plurality
(5)),
presidential
system,
federalism,
time
since
last
national
election
(years),
percentage of other nationalities, GDP as percentage of mean EU 27. See all variables in the
estimations
below.
plurality
(5)),
presidential
federalism,
time
nationalinelection
(years),
percentage
of
other
nationalities, system,
GDP as percentage
of mean
EUsince
27. Seelast
all variables
the
percentage
of
other
nationalities,
GDP
as
percentage
of
mean
EU
27.
See
all
variables
in
the
of the estimations
appendix below.
4. table
and 5. table. I use all of these variables as controls in each of the
appendix
4. table
and nationalities,
5. table. I useGDP
all ofasthese
variablesofas
controls
in each
of the
percentage
of other
percentage
mean
EU 27.
See all
variables in the
appendix 4. table and 5. table. I use all of these variables as controls in each of the
estimations
below.
below.
I estimations
will
useuse
three
types
of models
to test the
association
between inequality
andinequality
voter turnout.
A simple
logit
I will
three
types
of models
to test
the association
between
and
voter turnout.
A
appendix
estimations
below. 4. table and 5. table. I use all of these variables as controls in each of the
simple
logitcountry
regression
(1)standard
with country
standard
errors will
base, a 2 step
estimations
below.
regression
(1) with
clustered
errors willclustered
be the base,
a 2 step estimation
(2) be
and the
a hierarchical
I Iwill
use
three
types
of models
to testto
the
association
between inequality
and voter turnout.
A turnout. A
will
use
three
types
of
models
test
the
association
between
inequality
and
voter
(2)
and to
a test
hierarchical
model
(3) inequality
will provide
robustness
checks for the logit
I will useestimation
three types of
models
the association
between
and voter
turnout. A
modelsimple
(3) willlogit
provide
robustness
checks
for the logit
model.standard errors will be the base, a 2 step
regression
(1)
with
country
clustered
simple
logit regression
(1) with
country
clustered
be the base, a 2 step
simple logit
(1) with of
country
clustered
standard
errorsstandard
will
be theerrors
base,
awill
2 step
model.
I willregression
use (2)
three
models
to test(3)
thewill
association
between
inequality
andlogit
voter turnout. A
estimation
andtypes
a hierarchical
model
provide robustness
checks for the
The
logit
model
will
be
the
following:
estimation
(2)
and
a
hierarchical
model
(3)
will
provide
robustness
checks
for the logit
estimation
(2)
and
a
hierarchical
model
(3)
will
provide
robustness
checks
for
the
logit
The logitlogit
modelregression
will be the following:
simple
(1) with country clustered standard errors will be the base, a 2 step
model.
model. model.
(1)
The
logit model will
the following:
estimation
(2)beand
a hierarchical model (3) will provide robustness checks for the logit
The logit model
will model
be the following:
The(1)logit
will be the
where
V is the dummy
for following:
voting, Y is a factor of individual characteristics, while Z represents country level
model.
V is the dummy for voting, Y is a factor of individual characteristics, while Z represents country level
(1)where
(1)
where
V is including
the dummyafor
voting, Yofisincome
a factor inequality;
of individuali characteristics,
while
Z represents
features
measure
is an index for
individual
andcountry
c for level
country.
are
The
model
will beYthe
where V is
thelogit
dummy
for voting,
is afollowing:
factor of individual characteristics, while Z represents country level
features
including
athe
measure
of for
income
inequality;
i isi is
anan
index
forforindividual
and
are
features
including
measure
of while
income
inequality;
index
individual
andccfor
forcountry.
country.Z represents
are paramwhere
V is to
voting,
is idiosyncratic
a factor
of
individual
characteristics,
while
country level
parameters
beadummy
estimated,
isY an
error
term.
features including
are
(1) a measure of income inequality; i is an index for individual and c for country.
parameters
to
be
estimated,
while
is
an
idiosyncratic
error
term.
features
including
a measure
of binary
income
inequality;
i is an index
for with
individual
and cspecifications
for country. that differ
are
estimation
with
dependent
variables
struggle
stochastic
parameters
to be
estimated,
while
an
error
term.
eters toPooled-sample
be
estimated,
while
isfor
anidiosyncratic
idiosyncratic
term.
where
V is the
dummyis
voting,dependent
Y is
aerror
factor
of struggle
individual
while Z that
represents
Pooled-sample
estimation
with
binary
variables
withcharacteristics,
stochastic specifications
differ country level
parameters
to In
bewith
estimated,
while
isvariables
an idiosyncratic
error
term. specifications
Pooled-sample
estimation
binary
struggle with
stochastic
that use
differ
across
levels.
these
casesdependent
Franzese
(2005)
suggests
the
researcher
to consider the
of a 2 step estimation. In
features
including
awith
measure
income
inequality;
i is an
index
for
andestimation.
c that
for differ
country.
are
across
levels.
In these cases
Franzese
(2005)
suggests
the researcher
towith
consider
the individual
usespecifications
of a 2 step
In across
Pooled-sample
estimation
binaryof
dependent
variables
struggle
stochastic
across levels.
In
Franzeseprocess
(2005)
thedependent
researcher
to
consider
the
use offor
awith
2 each
stepstochastic
estimation.
In within each
Pooled-sample
estimation
withsuggests
variables
struggle
specifications
thatcountry
differ
the
2 these
step cases
estimation
Ibinary
estimate
a predicted
probability
respondent
the
2 step estimation
process Iwhile
estimateisaanpredicted
probability
each respondent within each country
parameters
to be estimated,
idiosyncratic
errorfor
term.
thelevels.
2 stepacross
estimation
process
a suggests
predicted
probability
forthe
each
respondent
within
each
country
stI estimate
nd Inestimation.
levels.
In
cases
Franzese
(2005)
suggests
researcher
consider
a 2is,
In
In
these
cases
Franzese
(2005)
the
researcher
to
consider
thetouse
of avariable
22the
step
estimation.
the 2
stthese
nd use
step,
then
I
use
the
country
means
as
the
dependent
inof
the
2step
separately
in
the
1
step, then Iwith
use the
country
means asvariables
the dependent
variable
in the
step.
That
I step. That is, I
separately
in the 1 estimation
st
nd stochastic specifications that differ
Pooled-sample
binary
dependent
struggle
with
separately in the 1 step, then I use the country means as the dependent variable in the 2 step. That is, I
the
2 the
step
estimation process I estimate a predicted probability for each respondent within each country
estimate
the
estimate
step estimation
process
I estimate
predicted (2005)
probability
for each
within
eachthe
country
in the
across levels.
In these
casesa Franzese
suggests
the respondent
researcher to
consider
use ofseparately
a 2 step
estimation.
In
estimate
the
st
nd
separately
(2a)
(2a) in the 1 step, then I use the country means as the dependent variable in the 2 step. That is, I
(2a) the 2 step estimation process I estimate a predicted probability
for
each
each country
1st step,
then I and
use
the country
means asofthe
dependent
variable male
in thevoter,
2nd step.
That
is,school
I respondent
estimate
the within
equation,
predict
the probability
voting
for
a 40-year-old
who went
to
untilto
age
of 18, until age of 18,
estimate
the
equation,
and
predict
the
probability
of
voting
for
a
40-year-old
male
voter,
who
went
school
equation, separately
and predict the
of voting
a 40-year-old
malemeans
voter, who
wentdependent
to school until
age of in
18,the 2nd step. That is, I
then for
I use
the country
as the
variable
in probability
the 1st step,
with mean
standard of living. In order to correct for the different efficiency of the first step estimates I use the
(2a)
with
mean
standard
of
living.
In
order
to
correct
for
the
different
efficiency
of
the
first
with mean standard of living. In order to correct for the different efficiency of the first step estimates I use thestep estimates I use the
estimate
the deviation of the predicted probabilities as weight in the second step.
inverse
standard
equation,
and predict
the probability
of voting
forin athe
40-year-old
who
went to school until age of 18,
inverse standard
of deviation
the predicted
as weight
second
step.male
inversedeviation
standard
of probabilities
the predicted
probabilities
as weight
in thevoter,
second
step.
equation, (2b)
and
(2a)predict the probability of voting for a 40-year-old male voter, who went to school until age of 18,
(2b) with(2b)
mean standard of living. In order to correct for the different efficiency of the first step estimates I use the
where
PrVoteand
is the
predicted
meanof
ofvoting
voting. for
The atheoretical
difference
this went
approach
and the until age of 18,
equation,
predict
the country
probability
40-year-old
male between
voter, who
to school
where
PrVote
is
the
predicted
countryInmean
of to
voting.
Theprobabilities
theoretical
difference
between
this
approach
andestimates
the
with mean
standard
of
living.
order
correct
for
the
different
efficiency
of
the
first
step
I use the and the
inverse
standard
deviation
of
the
predicted
as
weight
in
the
second
step.
where
PrVote
is the
the 2predicted
country
meanforofdifferent
voting. effects
The theoretical
difference
between this
approach
logit
model
is that
estimation
allows
of the individual
characteristics
within
with
meanthe
standard
ofstep
living.
In order
to correct
for the
different
efficiency
of
the within
first step
estimates I use the
logit model
is that
2 step estimation
allows
for different
effects
of the
individual
characteristics
(2b)
countries,
i.e.deviation
theis2 that
step is
athe
more
flexible
model
compared
thedifferent
logit.
logit
model
the
2 predicted
step
estimation
allowsastofor
of the individual characteristics within
inverse
standard
of
probabilities
weight
in the effects
second step.
deviation
the predicted
countries,inverse
i.e. the 2standard
step is a more
flexibleofmodel
compared probabilities
to the logit. as weight in the second step.
Finally,
estimate
where
respondents
are nested
in
countries:
whereIPrVote
isa hierarchical
the
predicted
country
mean
of voting.
Theto
theoretical
countries,
i.e. the
2 step
is amodel,
more
flexible
model
compared
the
logit. difference between this approach and the
Finally, I estimate
(2b)a hierarchical model, where respondents are nested in countries:
logit
model
is thata hierarchical
the 2 step estimation
allows
for different
effects
the individual characteristics within
Finally,
I estimate
model, where
respondents
are nested
in of
countries:
where
PrVotefor
is the
the subjective
predicted standard
country mean
of was:
voting.
The theoretical
difference
approach and the
The question
of living
„Taking
everything into
account,between
at aboutthis
what
1 The
where
PrVote
the
predicted
country
mean
voting.
theoretical
difference
approach
countries,
i.e.
the 2standard
stepstandard
is aofmore
model
compared
to 1the
logit.
question
foristhe
subjective
of flexible
living
was:
„Taking
everything
into
account,
at about
what
level
is your
family’s
living?
If of
you
think
ofThe
a scale
from
to
7, where
1between
means
athis
poor
family, 7 and the
logit
modelstandard
is thatother
the
2numbers
step
estimation
forin1different
effects
of athe
individual
within
level is your
family’s
of
living?
If you
think
a allows
scale
from
to 7, where
1 means
poor
7 characteristics
aFinally,
family,
and the
are
for of
the
positions
between,
about
where
wouldfamily,
you place
your at about what
1rich
The
question
for
the subjective
standard
of
living
was:
everything
into
account,
I estimate
a hierarchical
model,
where
respondents
are„Taking
nested
in
countries:
a rich
family,
and
the
other
numbers
are
for
the
positions
in
between,
about
where
would
you
place
your
logit model
is that
the
2 2step
estimation
allows
for
different
effectstoof
theincome
individual characteristics
within counfamily?”
country
mean
for this
question
tend compared
correlate
with
I standardized
countries,
i.e.the
the
step
isthis
a more
model
the
logit.
level
isSince
your
family’s
standard
offlexible
living?
you to
think
ofincome
a scale
from 1inequalities
to I7,standardized
where
1 means a poor family, 7
family?”the
Since
the
country
for
question
tend toIfcorrelate
with
inequalities
answers
withinmean
country
(0 mean,
1 sd).
aFinally,
rich
family,
and(0athe
other
numbers
are
forrespondents
the positions
between,
about where would you place your
the answers
within
mean,
1 sd).
I country
estimate
hierarchical
model,
where
arein
nested
in countries:
tries, i.e.
the 2question
step
is a for
more
flexible
model
compared
to the logit.
1 The
the
subjective
standard
of
was:
„Taking everything
account, Iatstandardized
about what
family?”
Since the
country
mean for
this question
tend
to correlate
with incomeinto
inequalities
10living
10
level
is
your
family’s
standard
of
living?
If
you
think
of
a
scale
from
1
to
7,
where
1
means
a
poor
family, 7
the answers within country (0 mean, 1 sd).
Finally,a1 Irich
estimate
a hierarchical
model,
where respondents
are nested in
in between,
countries: about where would you place your
family,
and
the
other
numbers
are
for
the
positions
The question for the subjective standard of living was: „Taking everything into account, at about what
family?”
Sincefamily’s
the country
mean
this question
tend
correlate
inequalities
standardized
10
level is your
standard
offor
living?
If you think
of to
a scale
fromwith
1 to income
7, where
1 means a Ipoor
family, 7
(3)
the
answers
within
country
(0
mean,
1
sd).
a rich family, and the other numbers are for the positions in between, about where would you place your
where u is
a country
level error
term.
family?”
Since
the country
mean
for this question tend to correlate with income inequalities I standardized
10
where the
u is answers
a countrywithin
level error
term.(0 mean, 1 sd).
country
A big handicap of the 2 step estimation procedure is that it cannot handle interaction effects between individual
A big handicap
of thelevel
2 step
estimation
is that
handle
interactioncountry
effects between
anda group of
and country
variables
(byprocedure
definition),
andit cannot
also 10
that
by predicting
means, individual
I pre-define
1
whose turnout
will be the
variable
in the
secondmeans,
step. When
testinga hypotheses
2 I have to use
countrypeople,
level variables
(by definition),
anddependent
also that by
predicting
country
I pre-define
group of people,
interaction terms: how the country level inequality affects the association between income and voter turnout.
whose turnout will be the dependent variable in the second step. When testing hypotheses 2 I have to use interWithin hierarchical models, as well as within simple logit models, the interaction can easily be done. 2 step
nd voter turnout. Within
action estimation,
terms: how on
the the
country
affects theasassociation
betweenfrom
income
step can easily be depicted (see
otherlevel
hand,inequality
has its advantage
well. The results
the 2and
below),
unlikeasthe
estimates
the logit
ormodels,
the hierarchical
estimates.
hierarchical
models,
well
as withinofsimple
logit
the interaction
can easily be done. 2 step estimation, on
Results
Declining turnout – hypothesis 1
Page • 19
The point estimates of the individual controls in the base model (6. table) are all as expected.
Daniel Horn
the other hand, has its advantage as well. The results from the 2nd step can easily be depicted (see below), unlike
the estimates of the logit or the hierarchical estimates.
Page • 20
Income inequality and voter turnout
3. Results
3.1.
Declining turnout – hypothesis 1
The point estimates of the individual controls in the base model (6. table) are all as expected. Age associates
with higher turnout but at a declining rate, years spent in school education also increases turnout, richer tend to
vote more, and women are just as likely to vote as men, if all above socio-economic characteristics are controlled
for. From the country level features fewer factors are significant. The existence of a threshold decreases turnout,
and presidential systems also have fewer people to cast a vote, while federalist states have a higher turnout. Nevertheless, since these country level characteristics are not in the focus of the paper, I leave them in the models as
controls, even if they do not significantly associate with turnout.
Tests for hypothesis 1 are in 7. table (logit), 8. table (2 step), and 9. table (hierarchical) in the appendix. All
estimation procedures seem to provide the same results. Except the Eurostat Gini variable, and the s80/s20 ratio all
income measures associate negatively with voter turnout, but very few are significant. Only the poverty rate shows
significant association with the turnout across estimations, while the MDMI and the earnings Gini coefficients are
also weakly (10% or less) significant in all models. The other four proxies (income Gini, s80/s20 and the two
p95/p5 ratios) are all insignificantly related to voter turnout.
However, if I consider that the models have taken into account several known influences of turnout, and that
the number of countries are not very high I should conclude that these results are in line with the 1st hypothesis.
It seems that inequality associates negatively with voter turnout, if we control for other factors which are claimed
to influence turnout.
4. figure below depicts this association using the predicted probabilities from the 1st step of the 2 step procedure. It is apparent that the association between inequality and predicted voter turnout is not very strong, but
negative. Especially the poverty rate associates closely with turnout, but all other indicators show a negative rather
than a positive relation.
Nonetheless, the reason for this negative association is not straightforward. The hypotheses above could allow
for three different reasons: a) turnout for the poor declines as inequality goes up, b) turnout for the rich declines
as inequality goes up or c) there is no change in the relative turnout of the different income people but universal
welfare states drive the results.
Page • 21
goes up, b) turnout for the rich declines as inequality goes up or c) there is no change in the
relative turnout of the different income people but universal welfare states drive the results.
Daniel Horn
4.figure–Associationofinequalitywithturnout–predictedprobabilitiesfromthe1ststepestimatesofthe2
4. figure – Association of inequality with turnout – predicted probabilities from the 1st step estimates of the 2 step
stepprocedure
procedure
EE
.6
PL
voter turnout, pred. pr.
.7
.8
.9
1
20
25
LT
30
35
Gini, Eurostat 2009
SEAT MT CY
DK BE
GR
IT
IE
NLDE
SI
FR ESPT
LU
FI
UK
HU
SK
CZ
BG
EE
40
RO
LV
LT
BG
.6
PL
10
20
voter turnout, pred. pr.
.7
.8
.9
1
CZ
LV
30
40
50
Poverty rate, SILC
SE
DK
AT
NL
SI
BE CY
IT
DE
FR
LU ES
FI
60
GR
IE
PT
HU UK
SK
LV
CZ
EE
LT
PL
.6
SI
MT
.3
voter turnout, pred. pr.
.7
.8
.9
1
SEAT
BE
DK CY
GR
IT
NL IE
DE
FR
PT
LU
ES
FI
RO
UK
HU
SK
.4
.5
.6
.7
Mean distance from the median
SE
BE CY
DK AT
IE DE
NL
SI FRLU
FI
HU
SK
CZ
IT
4
.8
GR
PT
ES
RO
UK
LV
EE
LT
PL
.6
voter turnout, pred. pr.
.7
.8
.9
1
lowess smoother
6
8
p95/p5, SILC
BG
10
note: predicted probabilities are for a 40 year old man with average income, who finished education at age 18
note: predicted probabilities are for a 40 year old man with average income, who finished education at age 18
Incomebias–hypothesis2a
3.2. Income bias – hypothesis 2a
Unfortunately the EES dataset does not contain an absolute measure of income or class. The
income
measure in the dataset is a subjective standard of living. The respondents place
Unfortunately the EES dataset does not contain an absolute measure of income or class. The income measure
themselves within seven categories as compared to others in the society, and thus are
in the dataset is a subjective standard of living. The respondents place themselves within seven categories as com-
endogenous with the inequality measures (e.g. the greater the inequality, the more people are
pared to others in the society, and thus are endogenous with the inequality measures (e.g. the greater the inequality,
likely to consider themselves poor). For this reason the interaction between the subjective
the more people are likely to consider themselves poor). For this reason the interaction between the subjective
standard of living and the inequality measure might be biased.
12The higher the inequality the more people tend to
be poor (because it is a subjective / self evaluated measure). So a person, with similar probability of voting might
consider herself poor in one country with high inequality and not poor in a country with low inequality. And viceversa a person might consider herself rich in a low inequality country while not rich in a high inequality country,
assuming identical turnout probabilities. Hence the effect of the interaction of income with inequality on turnout
could be biased. Unfortunately the direction of the bias is also not clear. It depends on our assumption of inequality on the subjective evaluation of income. If average income people tend to “de-valuate” their income more in an
unequal country than rich people, then the effect of income on voter turnout will be downwardly biased. But if rich
Page • 22
Income inequality and voter turnout
tend to look at themselves as lesser rich in an unequal country as compared to the subjective income “decline” of
an average income person, then income effect on voting will be upwardly biased. So the direction of bias of subjective income on turnout will depend on the relative evaluation of income across income groups.
I could not find any suitable instrument that could solve this problem. I would need a variable that explains
why one might consider herself poorer meanwhile being uncorrelated with voter turnout and only unconditionally
correlated with inequality. Hence, the estimates of the interaction effect of lower income and inequality on turnout
might be biased. In order to minimize bias I standardized the income proxy (subjective standard of living) within
countries. By this, the cross-country correlation of the standardized income and the inequality measure will be
zero, by definition. However, we do not rule out the fact that there will be more relatively poor people in higher
income countries. Nevertheless, I believe that the size of this bias will be small (see argument about the direction
of the bias), and thus only marginally affecting the substantial results.
The lack of absolute income could also be a problem if we assume that absolute income matters as well as
relative income (see Solt 2010). Within the Downsian framework, the lack of absolute income might not be a
problem, if we disregard the “hard” costs of voting: people vote more likely when the probability of their vote being decisive goes up; hence their relative position within the society matters. However, if we assume that absolute
income matters as well – poorer people have troubles paying the costs of voting, e.g. traveling to the voting booth
is costly – the point estimates will be biased, due to an omitted variable bias. Although I must make the assumption that absolute costs does not matter, I believe that in developed countries casting a vote is not very expensive.
10. table and 11. table in the appendix shows that the subjective standard of living does not associate significantly with voter turnout through increased inequality: richer people are not more likely to vote as inequality
goes up. None of the interaction effects are significant, moreover their signs are also not consistently negative or
positive.
Another way of looking at this income bias is to use the marginal effect if the subjective income on turnout
from the 1st step estimation. 5. figure depicts the association of this marginal effect (estimated for the same 40
year old male schooled until age 18) with the different inequality measures. Apparently the same conclusion can
be drawn: we cannot straightforwardly conclude that higher income people tend to vote more in more unequal
countries.
Page • 23
countries.
Daniel Horn
5.figure–Incomebias–associationofinequalitywiththemarginaleffectofincome(standardofliving)from
individuallevelregressions
5. figure– Income bias – association of inequality with the marginal effect of income (standard of living) from individual
level regressions
3.3.
NL
DK
CZ
HU
SKATBE
SE
SI
20
25
MT
PL
LT
ESBG
FR EE
DE
LU
CY
IE
UK
IT
LV
PT
RO
GR
30
35
Gini, Eurostat 2009
40
FI
NL
DK MT
ESEE
FR
CZ UK
HU
DE
AT
SK
SI BE PT
LU
CY
IT
IE GR
SE
10
20
LT
PL
BG
LV
RO
30
40
50
Poverty rate, SILC
60
marginal effect, subj. st. of living
0 .02 .04 .06 .08
FI
marginal effect, subj. st. of living
0 .02 .04 .06 .08
marginal effect, subj. st. of living
0 .02 .04 .06 .08
marginal effect, subj. st. of living
0 .02 .04 .06 .08
lowess smoother
FI
NL
DK
PL
SE
SI
.3
CZ
AT
LT
ESEE
FR
LV
HU UK
DE
SK
BE LU
CY
IT
IE
GR
PT
.4
.5
.6
.7
Mean distance from the median
.8
FI
NL
DK
PL
EE ES
FR
CZHU
DE
AT
SK
SI BE LU
CY
IE
SE
4
LT
UK
IT GR
BG
PT
6
8
p95/p5, SILC
LV
RO
10
Inequality on the top and at the bottom – hypothesis 2b
Inequalityonthetopandatthebottom–hypothesis2b
between income inequality on the top and at the bottom with voter
12. table below shows the association
turnout. Results are not very robust, mainly due to the fact that the general indicators of inequality (MDMI and
p95/p5), which can be separated along income distribution, are themselves poor explanators of turnout. Thus we
see no strong association between inequality on the top and inequality at the bottom with voter turnout. However,
the point estimates, as well as mild significance of the p95/p50 indicators and the p50/p5 LIS measure shows that
higher inequality at the top associates with lower turnout, while higher inequality at the bottom goes together with
higher turnout (or rather no association at all at the bottom). That is, the higher the difference between the very rich
15between the middle and the very poor does not
and the middle decreases overall turnout, while higher difference
change (or mildly increases) turnout. 13. table shows no indication of income effects at all. So the relative turnout
of the rich and the poor, as shown by the interaction terms, does not change when inequality changes on the top or
at the bottom. Although the point estimates are insignificant, they are against hypothesis 2b. i.e. richer people tend
to vote more if inequality on the top is higher, while poorer tend to vote more if inequality at the bottom is higher.
Page • 24
Income inequality and voter turnout
This is just the opposite of what hypothesis 2b has assumed. These effects, however, are all insignificant,
which might be due to the small number of countries, as well as the relatively unimportant effect of inequality.
This is also what we see in 6. figure below: the different measures of income inequality on the top and at the
bottom associate mildly with the predicted probability of voter turnout. Inequality on the top tend to go weakly and
negatively together with turnout, while inequality at the bottom has no relation with turnout.
6. figure - Association of inequality on the top and at the bottom with turnout – predicted probabilities from the 1st
6.figure‐Associationofinequalityonthetopandatthebottomwithturnout–predictedprobabilitiesfrom
step estimates of the 2 step procedure
the1ststepestimatesofthe2stepprocedure
PT
LU UK
SK HU
LV
EE
CZ
1
CZ
RO
EE
PL
LT
LV
BG
HU
NL
SI
SK
LV
EE
CZ
-.35
-.3
-.25
-.2
MDMI, below the median
GR
ES IT
UK
LU
HU
EE
PL
3
SE
DK
AT
CY
NL BE
IE DE GR
FI
SI
ITES
FR
PT
UK
LU
HU
SK
LV
CZ
EE
PL LT
1.5
1
2
2.5
p95/p50, SILC
RO
BG
.7
.7
PL LT
AT
BE
IE
DE
FISI
NL
2
2.5
p95/p50, LIS
SE
DK
voter turnout, pred. pr.
.8
.9
FR
PT
UKLU
DK
DKSE
NL
AT
FI
BE
DE
SI
3
IE GR
IT
ES
UK
LU
HU
EE
PL
.7
ES IT
SE
AT
voter turnout, pred. pr.
.8
.9
GR
BE CY
IE DE
FI
1.5
1
.6
.8
1
1.2
MDMI, above the median
1
.4
voter turnout, pred. pr.
.8
.9
PT
UK
LU
SKHU
.7
.7
PL LT
AT
CY
BENL
GR
IE
DE
SI FI
IT
FR ES
.7
GRIE
SE
DK
voter turnout, pred. pr.
.8
.9
1
AT
NL BE CY
DE
SI FI IT
FR ES
voter turnout, pred. pr.
.8
.9
SE
DK
voter turnout, pred. pr.
.8
.9
1
lowess smoother
2
2.5
3
3.5
p50/p5, SILC
4
2
2.2
2.4 2.6 2.8
p50/p5, LIS
3
Universalwelfarestates–hypothesis3
3.4. Universal welfare states – hypothesis 3
Although Lister (2007) uses the Gini coefficient to proxy universal welfare states – arguing
that the lower the inequality the higher the state intervention is – for the purposes of this
Although Lister (2007) uses the Gini coefficient to proxy universal welfare states – arguing that the lower the
paper this proxy would obviously not be useful. Therefore, in order to test hypothesis 3, I
inequality the higher the state intervention is – for the purposes of this paper this proxy would obviously not be
have to use alternative measures of the welfare state. I will utilize government spending as
useful. Therefore,
in orderand
to test
hypothesis 3, Ispending
have to use alternative
of the welfare
state. I will utilize
percentage
of GDP
government
on socialmeasures
protection
as percentage
of GDP.
Both
are similar,
widely used
measures
for the
size ofonstate
and thus
for the
government
spendingbut
as percentage
of GDP
and government
spending
socialinterventions,
protection as percentage
of GDP.
welfare
Esping-Andersen
1990).
that the and
larger
spending,
the(e.g.
more
Both are state
similar,(e.g.
but widely
used measures for
the sizeI ofassume
state interventions,
thus the
for the
welfare state
universal the welfare state is. 7.figure and 8.figure below shows that government spending
Esping-Andersen 1990). I assume that the larger the spending, the more universal the welfare state is. 7. figure and
associates negatively with inequalities (the higher the spending the lower the inequalities), as
8. figure below shows that government spending associates negatively with inequalities (the higher the spending
expected, and it also associates positively with voter turnout (8.figure and 9.figure). Thus
Lister’s (2007) argument could hold: we have observed that inequality associates negatively
Page • 25
with turnout, and also that inequality associates negatively with the welfare state, which
associates positively with turnout. Hence the link between inequality and turnout could
Daniel Horn
the lower the inequalities), as expected, and it also associates positively with voter turnout (8. figure and 9. figure).
Thus Lister’s (2007) argument could hold: we have observed that inequality associates negatively with turnout,
and also that inequality associates negatively with the welfare state, which associates positively with turnout.
Hence the link between inequality and turnout could indeed be driven by the welfare state. If universal welfare
states are indeed an omitted variable, we should see the unbiased effect of income inequality on voter turnout after
controlling for government spending.
7.figure‐governmentexpenditurevs.incomeinequality
7. figure - government expenditure vs. income inequality
IELU
BG
GR
UK
PL
CY
IT
FR
DE
NL
CZ
SK
FI BE
AT
DK
HU
SE
RO
LV
LT
PL
SK
EEIE
ES
LU
GR
CY UK PT IT
BE
MT
CZ DE
SI
NL FI AT
HU
10
SI
40
45
50
55
35
40
10
35
RO
LV
LT
IE
LV
EE
LU
ES
PL
CY
SK
GR
UK
DE
CZ
SI
35
40
NL
45
IT HU
BE
FI
p95/p5, SILC
6
8
PT
FR
50
55
FR
SE
50
55
BG
EE
ES
CY
SI
CZ
SK
35
GR
UK
PL
IT
DE
IE
LU
DK SE
45
DK
PT
LT
AT
4
Mean distance from the median
.3
.4
.5
.6
.7
.8
BG
60
PT
ES
EE
Poverty rate, SILC
20 30 40 50
MT
LV
LT RO
20
Gini, Eurostat 2009
25
30
35
40
Government expenditure, Total, % of GDP, 2007
40
FR
BE
NL FI AT HU
DK SE
45
50
55
14.14.table
and15.15.
shows
the logit
where
the government
spending
table and
tabletable
shows the
logit models,
wheremodels,
the government
spending
and the government
spending and the
government spending on social protection is included. Since there were no substantial
on social protection is included. Since there were no substantial differences between the results of the logit, the 2
differences between the results of the logit, the 2 step and the hierarchical estimations I show
step and the hierarchical estimations I show only the results from the logit regression for the tests of hypothesis 3.2
only the results from the logit regression for the tests of hypothesis 3.2
It is apparent that government spending has a not very strong but positive effect on turnout, ceteris paribus
It is apparent that government spending has a not very strong but positive effect on turnout,
individual and other country level factors. If we accept that the size of government spending proxies welfare state
ceteris paribus individual and other country level factors. If we accept that the size of
entrenchment well, we can conclude that universal welfare states tend to foster voter turnout. However, the point
government spending proxies welfare state entrenchment well, we can conclude that
estimates of the different inequality measures did not change significantly after including government spending.3
universal welfare states tend to foster voter turnout. However, the point estimates of the
different inequality measures did not change significantly after including government
3 All
2
Results from
the 2 indicators,
step and hierarchical
estimations
were, again,
almost
– and could
be requested from
the author. but lost
spending.
but
the Gini
from
theidentical
Eurostat,
remained
negative
3
Only the p95/p5, LIS inequality indicator became significant but stayed negative after controlling for government spending on social
protection (but controlling for the total spending did not have his effect), this is probably due to some outlier or high leverage case.
a bit of
significance, due probably to increased multicollinearity between the variables.
Page • 26
From this I conclude that although welfare states tend to have higher voter turnout, it does
not seem to be the omitted variable that would explain the effect of inequality on turnout.
Income inequality and voter turnout
All indicators, but the Gini from the Eurostat, remained negative but lost a bit of significance, due probably to
increased multicollinearity between the variables.
From this I conclude that although welfare states tend to have higher voter turnout, it does not seem to be the
8.figure–governmentexpenditureonsocialprotectionvs.incomeinequality
omitted variable that would explain the effect of inequality on turnout.
8.figure–governmentexpenditureonsocialprotectionvs.incomeinequality
expenditure,
Social
prot,
8. figureGovernment
– government expenditure
on social protection vs.
income inequality
% of GDP, 2007
60
LT
RO
IE
CY
EE
SK
IE
CY
SK
5
10
5
10
PovertyPoverty
rate, SILC
rate, SILC
2010 3020 4030 5040 6050
LV EE
PT
BG
ESMT UK
PL
GR
IT
PT
FR
LU
DE
GR
NL
DK
UK
PL BEIT AT
FI
SE
HU
FR
LU
DE
SI
NL
DK
BE
FI
AT
SE
HU
BG
ES
CZ
CZ
15 SI
20
25
15
20
25
10 10
ROLT
LV RO
LT
SK
LV RO
LT
EE
CY
IE
SK
EE
CY
IE
5
10
5
p95/p5,p95/p5,
SILC SILC
46
68
810
PT
LV
IE LT
EE
CY
IE
LV SK
EE
CY
GR
UK PT
PL HUIT
LU
BE GR DE
FI
FR
UK
ES
CZ
ES
SK
5
10
5
10
CZ
AT
PL
SI
NL HUIT
LU
BE
DK
DE SE
FI
FR
15 SINL
20AT
15
20
SE
DK
25
25
BG
LV
PL
PL HU
GR
PTIT
BE
FR
SI HU
LU
FIDE DK
NL
GRAT SE
PT
ES UK BEIT
CZMT SI
FR
DE DK
AT
15 LUNL
20FI
SE
ES
CZMT
10RO
UK
15
LV
LT
20
25
25
BG
RO
LT
EE
LT
IE
CY
EE
SK
IE
CY
5
10
5
10
4
Mean distance
Mean distance
from thefrom
median
the median
.3
.4 .3 .5 .4 .6 .5 .7 .6 .8 .7
.8 20
Gini, Eurostat
Gini, Eurostat
2009 2009
25
20
30
25
35
30
40
35
40
Government
expenditure, Social prot, % ofBGGDP, 2007
MT
LV
SK
PT
BG
GR
IT
PT
ES
UK
PL
ES
GR
UK
LU
PL
IT
FR
BE
AT
HU
FI
NL
SI
DE SE
DK
CZ
CZ
15
LU
DE
20AT
FI
BE
HU
SINL
15
FR
SE
DK
20
25
25
9.figure–Governmentexpenditurevs.predictedturnout
9. figure – Government expenditure vs. predicted turnout
55
HU
15
BG
PL
CZ
LU
UK
10
LT
BG
10
LT
PT
SK
CZ
EERO
BEMT
ES
SI
NL
MT
ES IE CY
LV
SK
EERO
IE CY
5
LV
.8
.9
voter turnout, pred. pr.
1
.7
.8
.9
voter turnout, pred. pr.
1
Conclusionandfurthercomments
55
50
50
IT
FI
HU
PT
UK
PL
BG
PL
BG
PT
CZ
CZ
.7
LT
.7
SI
LU
CY
MT
ES IE
SK
.8
.9
RO
LU
voterLV
turnout, pred.
pr. IE
EE
CY
NLMT
DE
GR
SI
ES
EE
BE AT
DK
NL
BE AT
IT
FIDE
GR
UK
RO
LV
LT
35
.7
SE
FR
45
LU
UK
DE BE
FI
AT
NL
SI GR
IT
HU
45
15
PL
PT
DK
SE
40
20
HU
IT
AT
GR
DK
40
20
DE
FI
FR
SE
FR
35
DK
SE
Governmen
G
to
eve
xprn
en
m
de
itu
ntre
e,xp
To
eta
nd
l,itu
%
re
o,fT
G
oD
taP
l,,%
20o
0f7GDP, 2007
25
FR
5
Governmen
G
to
eve
xprn
en
m
de
itu
ntre
e,xp
So
ecia
nditu
l pre
ro,t,S%
ocia
oflG
pD
roP
t,,%
20o
0f7GDP, 2007
25
9.figure–Governmentexpenditurevs.predictedturnout
1
SK
.8
.9
voter turnout, pred. pr.
1
The paper addressed the issue of the effect of inequality on voter turnout. Using the 2009
Conclusionandfurthercomments
PIREDEU European Election Study dataset I tested three different hypotheses. These
Page • 27
The paper addressed the issue of the effect of inequality on voter turnout. Using the 2009
hypotheses were derived from previous literature. The analyses could show that inequality
PIREDEU European Election Study dataset I tested three different hypotheses. These
Daniel Horn
Page • 28
Income inequality and voter turnout
4. Conclusion and further comments
The paper addressed the issue of the effect of inequality on voter turnout. Using the 2009 PIREDEU European
Election Study dataset I tested three different hypotheses. These hypotheses were derived from previous literature.
The analyses could show that inequality associates negatively with turnout at the national elections (hypothesis
1). This is not a very strong effect, but it is net of several factors affecting voter turnout that are empirically well
proven – such as individual characteristics or different features of the political system. The literature suggests
that this negative association is either due to the lower turnout of the poor relative to the rich in high inequality
countries (hypothesis 2) or due to the effects of the universal welfare state, which increases turnout through altered
social norms as well as decreases inequality through government intervention (hypothesis 3). None of these were
really supported by the data. Although none of the hypotheses were refuted, I did not find significant association
of the interaction effect of the individual income with inequality – i.e. income associates similarly with turnout in
different inequality countries. Similarly, it seems that universal welfare states have a higher turnout, but this does
not influence the association of inequality with turnout. I also tested whether inequalities at the top or at the bottom
have a different affect on turnout. Although the results, again, are not very robust, it seems that larger differences
in income between the very rich and the middle decreases overall turnout, while higher difference between the
middle and the very poor increases turnout. This is just the opposite of what I have expected from the Downsian
rational voter model.
Page • 29
Daniel Horn
Page • 30
Income inequality and voter turnout
References
Blais, André. 2006. “What Affects Voter Turnout?” SSRN eLibrary. Retrieved July 8, 2010.
Downs, Anthony. 1957. An economic theory of voting. New York: Harper and Row.
EES. 2010. “European Parlament Election Study 2009, Voter Study, Advance release.”
van Egmond, Marcel H., Eliyahu V. Sapir, Wouter vad der Brug, Sara B. Hobolt, and Mark N. Franklin. 2010.
“EES 2009 Voter Study Advance Release Notes.”
Esping-Andersen, Gosta. 1990. The Three Worlds of Welfare Capitalism. Princeton, EUA : Princeton University
Retrieved May 5, 2011.
Ferejohn, John A., and Morris P. Fiorina. 1974. “The Paradox of Not Voting: A Decision Theoretic Analysis.” The
American Political Science Review 68(2):525-536. Retrieved July 7, 2010.
Filer, John E., Lawrence W. Kenny, and Rebecca B. Morton. 1993. “Redistribution, Income, and Voting.” American Journal of Political Science 37(1):63-87. Retrieved July 7, 2010.
Franzese, Robert J. 2005. “Empirical Strategies for Various Manifestations of Multilevel Data.” Political Analysis
13(4):430 -446. Retrieved May 5, 2011.
Funk, Patricia. 2008. “Social Incentives and Voter Turnout: Evidence from the Swiss Mail Ballot System.” SSRN
eLibrary. Retrieved July 13, 2010.
Gallego, Aina. 2007. “Unequal Political Participation in Europe.” International Journal of Sociology 37(4):10-25.
Retrieved July 13, 2010.
Geys, Benny. 2006a. “Explaining voter turnout: A review of aggregate-level research.” Electoral Studies 25(4):637663. Retrieved July 7, 2010.
Geys, Benny. 2006b. “‘Rational’ Theories of Voter Turnout: A Review.” Political Studies Review 4(1):16-35. Retrieved July 7, 2010.
Goodin, R. E., and K. W. S. Roberts. 1975. “The Ethical Voter.” The American Political Science Review 69(3):926928. Retrieved July 7, 2010.
Grossman, Gene M., and Elhanan Helpman. 2002. Special interest politics. MIT Press.
Kaniovski, Serguei, and Dennis C. Mueller. 2006. “Community Size, Heterogeneity and Voter Turnouts.” Public
Choice 129(3/4):399-415. Retrieved July 8, 2010.
Lancee, Bram, and Herman van de Werfhorst. 2011. “Income Inequality and Participation: A Comparison of 24
European Countries.” GINI Discussion paper 6.
Ledyard, John O. 1984. “The Pure Theory of Large Two-Candidate Elections.” Public Choice 44(1):7-41. Retrieved July 7, 2010.
Lijphart, Arend. 1997. “Unequal Participation: Democracy’s Unresolved Dilemma.” The American Political Science Review 91(1):1-14. Retrieved July 8, 2010.
Lister, Michael. 2007. “Institutions, Inequality and Social Norms: Explaining Variations in Participation.” The
British Journal of Politics & International Relations 9(1):20-35. Retrieved April 22, 2011.
Lupu, Noam, and Jonas Pontusson. 2011. “The Structure of Inequality and the Politics of Redisribution.” American Political Science Review Vol. 105(No. 2):316-336.
Page • 31
Daniel Horn
Meltzer, Allan H., and Scott F. Richard. 1981. “A Rational Theory of the Size of Government.” The Journal of
Political Economy 89(5):914-927. Retrieved July 8, 2010.
Mueller, Dennis C., and Thomas Stratmann. 2003. “The economic effects of democratic participation.” Journal of
Public Economics (87):2129–2155.
OECD. 2008. “Growing Unequal? Income Distribution and Poverty in OECD Countries.”
Paczynska, Agnieszka. 2005. “Inequality, Political Participation, and Democratic Deepening in Poland.” East European Politics & Societies 19(4):573 -613. Retrieved April 23, 2011.
Palfrey, Thomas R., and Howard Rosenthal. 1983. “A Strategic Calculus of Voting.” Public Choice 41(1):7-53.
Retrieved July 7, 2010.
Palfrey, Thomas R., and Howard Rosenthal. 1985. “Voter Participation and Strategic Uncertainty.” The American
Political Science Review 79(1):62-78. Retrieved July 7, 2010.
Riker, William H., and Peter C. Ordeshook. 1968. “A Theory of the Calculus of Voting.” The American Political
Science Review 62(1):25-42. Retrieved July 7, 2010.
Schattschneider, Elmer Eric. 1960. The Semisovereign People: A Realist’s View of Democracy in America.
Solt, Frederick. 2010. “Does Economic Inequality Depress Electoral Participation? Testing the Schattschneider
Hypothesis.” Political Behavior 32(2):285-301. Retrieved January 12, 2011.
SSO. 2009. “Annual Monitoring Report 2009, Social Situation Observatory.”
Tóth, István György, and Tamás Keller. 2011. “Income Distributions, Inequality Perceptions and Redistributive
Claims in European Societies.” GINI Discussion paper 7.
Uhlaner, Carole J. 1989. “Rational Turnout: The Neglected Role of Groups.” American Journal of Political Science 33(2):390-422. Retrieved July 8, 2010.
Yamamura, Eiji. 2009. “Effects of social norms and fractionalization on voting behaviour in Japan.” Applied Economics. Retrieved July 9, 2010.
Page • 32
Income inequality and voter turnout
Appendix
2. table – indicators of overall iunequality
Gini
Gini Earning
cnt
Gini coeffi- Gini of gross
cient, Euroearnings in
stat 2009
cash for full(source:SILC) time workers,
SSO (Source:
SILC)
AUSTRIA
BELGIUM
BULGARIA
CYPRUS
CZECH REPUBLIC
DENMARK
ESTONIA
FINLAND
FRANCE
GERMANY
GREECE
HUNGARY
IRELAND
ITALY
LATVIA
LITHUANIA
LUXEMBOURG
MALTA
NETHERLANDS
POLAND
PORTUGAL
ROMANIA
SLOVAKIA
SLOVENIA
SPAIN
SWEDEN
UK
25,7
26,4
33,4
28,4
25,1
27,0
31,4
25,9
29,8
29,1
33,1
24,7
28,8
31,5
37,4
35,5
29,2
37,8
27,2
31,4
35,4
34,9
24,8
22,7
32,3
24,8
32,4
0,321
0,248
0,331
0,315
0,264
0,256
0,319
0,275
s80/s20
MDMI
Poverty rate p95/p5, LIS p95/p5, SILC
s80/s20,
Mean Distance Population p95/p5, LIS p95/p5, SILC
SSO 2009 from the meat risk of
(Tóth-Keller 2007-2008
(Source: dian (Lancee- poverty or
2011)
(Medgyesi)
SILC)
v.d.Werfhorts social exclu2011)
sion, Eurostat
2005 (Romania 2007,
Bulgarian
2006)
3,658
3,893
6,459
4,072
3,395
3,425
4,869
3,709
0,330
0,318
0,322
0,334
0,284
0,384
0,347
0,342
4,540
5,370
3,557
4,395
4,887
7,058
5,658
3,904
0,309
0,348
0,377
0,295
0,250
0,301
0,293
0,305
0,371
3,739
5,014
6,076
6,966
3,309
3,262
4,926
3,321
5,346
16,8
22,6
61,3
25,3
19,6
17,2
25,9
17,2
18,9
18,4
29,4
32,1
25,0
25,0
45,8
41,0
17,3
20,6
16,7
45,3
26,1
45,9
32,0
18,5
23,4
14,4
24,8
16,8
22,6
61,3
25,3
19,6
17,2
25,9
17,2
18,9
18,4
29,4
32,1
25,0
25,0
45,8
41,0
17,3
20,6
16,7
45,3
26,1
45,9
32,0
18,5
23,4
14,4
24,8
4,8
4,9
3,6
7,9
4,2
5,2
7,1
5,8
5,9
7,4
5,0
3,8
6,6
4,8
6,9
3,9
6,7
4,4
4,5
9,1
4,9
4,1
3,8
6,1
4,3
4,5
5,6
6,7
4,3
5,1
6,3
9,5
7,0
4,9
4,2
6,4
7,7
10,0
4,0
4,1
6,5
3,9
6,6
Page • 33
Daniel Horn
3. table – indicators of inequality above and below the median
MDMI, above p95/p50, LIS p95/p50, SILC MDMI, below p50/p5, LIS
cnt
MDMI, above the p95/p50, LIS p95/p50, SILC MDMI, below the p50/p5, LIS
median (Lancee(Tóth-Keller
(Medgyesi) median (Lancee- (Tóth-Keller
v.d.Werfhorts
2011)
v.d.Werfhorts
2011)
2011)
2011)
0,47
2,19
2,122
-0,27
2,2
AUSTRIA
0,55
2,11
2,027
-0,31
2,3
BELGIUM
2,855
BULGARIA
0,65
2,174
-0,29
CYPRUS
0,51
2,070
-0,22
CZECH REPUBLIC
0,38
1,78
1,839
-0,24
2,04
DENMARK
0,71
2,94
2,394
-0,31
2,69
ESTONIA
0,55
1,97
2,035
-0,28
2,12
FINLAND
0,55
2,101
-0,29
FRANCE
0,58
2,24
2,296
-0,29
2,34
GERMANY
0,79
2,56
2,469
-0,36
2,77
GREECE
0,67
2,44
2,012
-0,31
2,37
HUNGARY
0,84
2,21
2,269
-0,3
2,67
IRELAND
0,63
2,53
2,328
-0,34
2,91
ITALY
2011,01,06
2,945
-0,36
LATVIA
0,78
2,574
-0,33
LITHUANIA
0,64
2,24
2,255
-0,33
2,24
LUXEMBOURG
MALTA
0,46
1,89
2,153
-0,24
1,99
NETHERLANDS
0,67
2,49
2,554
-0,35
2,66
POLAND
0,95
2,971
-0,34
PORTUGAL
2,719
ROMANIA
0,57
1,926
-0,26
SLOVAKIA
0,43
2,01
1,919
-0,25
2,38
SLOVENIA
0,65
2,38
2,340
-0,36
2,9
SPAIN
0,38
1,89
1,848
-0,27
2,05
SWEDEN
0,72
2,7
2,546
-0,34
2,49
UK
Page • 34
p50/p5, SILC
p50/p5, SILC
(Medgyesi)
2,054
2,215
3,174
2,254
1,962
2,068
2,557
2,115
2,156
2,436
2,697
2,137
2,255
2,721
3,215
2,736
2,163
1,956
2,512
2,602
3,661
2,086
2,129
2,784
2,130
2,595
Income inequality and voter turnout
4. table - – country level indicators 1
cnt
Multiple
(concurrent)
Compulsory
Threshold
voting
elections
AUSTRIA
BELGIUM
BULGARIA
CYPRUS
CZECH REPUBLIC
DENMARK
ESTONIA
FINLAND
FRANCE
GERMANY
GREECE
HUNGARY
IRELAND
ITALY
LATVIA
LITHUANIA
LUXEMBOURG
MALTA
NETHERLANDS
POLAND
PORTUGAL
ROMANIA
SLOVAKIA
SLOVENIA
SPAIN
SWEDEN
UK
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
1
0
1
0
0
0
0
0
0
1
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
1
0
1
0
0
0
1
1
0
1
0
0
1
1
0
0
0
1
0
0
1
0
0
1
0
Elect. System: Presidential
Proportional
system
(0) vs. plurality (5)
3
3
3
3
3
3
3
3
1
4
3
4
5
3
3
2
3
5
3
3
3
4
3
3
3
3
0
3
0
3
1
0
0
0
0
2
0
0
0
3
0
0
3
0
0
0
3
3
3
0
3
0
0
0
Federalism
1
1
0
0
0
0
0
0
0
1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Source: European Election Study/Piredeu 2009
Page • 35
Daniel Horn
5. table – country level indicators 2
cnt
Years from national
election
(own research)
AUSTRIA
BELGIUM
BULGARIA
CYPRUS
CZECH REPUBLIC
DENMARK
ESTONIA
FINLAND
FRANCE
GERMANY
GREECE
HUNGARY
IRELAND
ITALY
LATVIA
LITHUANIA
LUXEMBOURG
MALTA
NETHERLANDS
POLAND
PORTUGAL
ROMANIA
SLOVAKIA
SLOVENIA
SPAIN
SWEDEN
UK
Total population,
2008
GDP, per capita, % of
EU 27 (Eurostat ) Government expenditure,
Total, % of GDP, 2007
3,25
8318592
124
1,00
10666866
116
0,08
7640238
44
1,92
789269
98
0,92
10381130
82
2,42
5475791
121
1,75
1340935
64
1,83
5300484
113
0,30
63982881
108
0,42
82217837
116
0,75
11213785
93
0,83
10045401
65
1,92
4401335
127
2,84
59619290
104
1,33
2270894
52
3,00
3366357
55
0,00
483799
271
2,75
410290
81
1,00
16405399
131
2,42
38115641
61
0,42
10617575
80
3,42
21528627
46
1,00
5400998
73
3,34
2010269
88
2,75
45283259
103
1,25
9182927
118
0,92
61179256
112
Source: European Election Study/Piredeu 2009, unless otherwise noted
Page • 36
48,4
48,4
41,5
42,9
42,6
51
34,7
47,3
52,3
44,1
44,1
49,9
35,6
47,9
35,9
35
36,2
42,2
45,5
41,9
45,8
36,3
34,6
42,3
38,7
52,5
44,4
Government
expenditure, Social
protection, % of
GDP, 2007
19,9
17,1
13,1
9,9
12,9
21,7
9,6
19,9
22,2
20,3
18,7
17,4
10,1
18,2
8,4
11
15,3
13,8
16
15,6
17,5
9,8
10,6
15,5
13
21,6
15,3
Income inequality and voter turnout
10.figure–VoterturnoutinEuropeancountries
10.figure–VoterturnoutinEuropeancountries
40 40
60 60
80 100
80 100 40 40
60 60
80 100
80 100 40 40
60 60
80 100
80 100 40 40
60 60
80 100
80 100
40 40
60 60
80 100
80 100
Voter
Voter
turnout,
turnout,
parlamentary
parlamentary
elections
elections
10. figure – Voter turnout in European countries
Austria
Bulgaria
Czech Republic
Denmark
Estonia
Finland
Austria
Bulgaria
Czech Republic
Denmark
Estonia
Finland
France
Germany
Greece
Hungary
Ireland
Italy
France
Germany
Greece
Hungary
Ireland
Italy
Latvia
Liechtenstein
Lithuania
Malta
Netherlands
Norway
Latvia
Liechtenstein
Lithuania
Malta
Netherlands
Norway
Poland
Portugal
Romania
Slovakia
Slovenia
Spain
Poland
Portugal
Romania
Slovakia
Slovenia
Spain
Sweden
Switzerland
United Kingdom
Sweden
Switzerland
United Kingdom
1940 1960 1980 2000 2020 1940 1960 1980 2000 2020 1940 1960 1980 2000 2020
1940 1960 1980 2000 2020 1940 1960 1980 2000 2020 1940 1960 1980 2000 2020
1940 1960 1980 2000 2020 1940 1960 1980 2000 2020 1940 1960 1980 2000 2020
Year
Year
1940 1960 1980 2000 2020 1940 1960 1980 2000 2020 1940 1960 1980 2000 2020
Source:IDEA
Source:IDEA
Gini
Gini
20 25
20 30
25 35
30 40
35 40
20 25
20 30
25 35
30 40
35 40
20 25
20 30
25 35
30 40
35 40
20 25
20 30
25 35
30 40
35 40
20 25
20 30
25 35
30 40
35 40
11.figure–incomeGinicoefficientofEuropeancountries
11. figure – income Gini coefficient of European countries
11.figure–incomeGinicoefficientofEuropeancountries
AUSTRIA
BELGIUM
BULGARIA
CYPRUS
CZECH REPUBLIC
DENMARK
AUSTRIA
BELGIUM
BULGARIA
CYPRUS
CZECH REPUBLIC
DENMARK
ESTONIA
FINLAND
FRANCE
GERMANY
GREECE
HUNGARY
ESTONIA
FINLAND
FRANCE
GERMANY
GREECE
HUNGARY
IRELAND
ITALY
LATVIA
LITHUANIA
LUXEMBOURG
MALTA
IRELAND
ITALY
LATVIA
LITHUANIA
LUXEMBOURG
MALTA
NETHERLANDS
POLAND
PORTUGAL
ROMANIA
SLOVAKIA
SLOVENIA
NETHERLANDS
POLAND
PORTUGAL
ROMANIA
SLOVAKIA
SLOVENIA
SPAIN
SWEDEN
UK
SPAIN
SWEDEN
UK
1995
2000
2005
2010
1995
2000
2005
2010
1995
2000
2005
1995
2000
2005
2010
1995
2000
2005
2010
1995
2000
2005
Source:Eurostat
Source:Eurostat
1995
2000
2005
2010
1995
2000
2005
2010
1995
2000
2005
2010
1995
2000
2005
2010
1995
2000
2005
2010
1995
2000
2005
2010
2010
2010
year
year
27
27
Page • 37
Daniel Horn
6. table – Logit Base model (ORs)
VARIABLES
Age
Age squared
Female
Age when finished education
Subjective standard of living
(within country Z-score)
Compulsory voting
Multiple (concurrent) elections
Population
Threshold
Elect. System: Proportional (0) vs. plurality (5)
Presidential
Federalism
Years from national election
% of other nationalities
GDP per capita, % of EU27
Constant
Observations
Odds ratios, robust clustered se in parentheses,
** p<0.01, * p<0.05, + p<0.1
Page • 38
(1)
vote=1
1.089**
(0.0124)
1.000**
(0.000120)
1.040
(0.0629)
1.056**
(0.0149)
1.196**
(0.0320)
1.308
(0.284)
0.731
(0.583)
1.000
(4.21e-09)
0.500**
(0.128)
1.053
(0.0913)
0.847*
(0.0680)
1.772*
(0.427)
1.076
(0.0761)
0.970+
(0.0173)
1.004
(0.00522)
0.142*
(0.112)
20,202
Income inequality and voter turnout
7. table – Different measures of income inequality on turnout, logit (ORs)
(1)
(2)
(3)
(4)
VARIABLES
vote
vote
vote
vote
Subjective standard of living
1.194**
1.186**
1.203**
1.201**
(within country Z-score)
(0.0315)
(0.0316)
(0.0347)
(0.0338)
Gini, Eurostat 2009
1.017
s80/s20, SSO 2009
(5)
(6)
(7)
vote
vote
vote
1.198**
(0.0318)
1.204**
(0.0430)
1.190**
(0.0318)
(0.0268)
1.041
(0.170)
Gini, earning, SSO 2009
0.982*
(0.00702)
Mean distance from the median
0.218+
(0.180)
Poverty rate, SILC
0.976+
(0.0129)
p95/p5, LIS
0.751
(0.338)
p95/p5, SILC
Constant
Observations
0.0852*
(0.105)
20,202
0.138+
(0.147)
18,964
0.249
(0.215)
20,202
0.181+
(0.177)
18,112
0.251
(0.218)
20,202
0.861
(1.581)
12,979
0.942
(0.0816)
0.157+
(0.157)
19,603
Robust seeform in parentheses
** p<0.01, * p<0.05, + p<0.1
8. table – Different measures of income inequality on turnout, 2 step
(1)
(2)
(3)
(4)
(5)
VARIABLES
Predicted probabiliya, 2nd step
Gini, Eurostat 2009
0.000854
s80/s20, SSO 2009
0.00224
(0.0332)
-0.00174
(0.00168)
Mean distance from the median
-0.338
(0.217)
Poverty rate, SILC
-0.00608*
(0.00265)
p95/p5, LIS
-0.0400
(0.0711)
p95/p5, SILC
Observations
R-squared
(7)
(0.00541)
Gini, earning, SSO 2009
Constant
(6)
0.792**
(0.192)
0.831**
(0.170)
0.864**
(0.137)
0.875**
(0.136)
0.943**
(0.127)
1.024*
(0.376)
-0.00779
(0.0167)
0.849**
(0.153)
27
0.575
25
0.602
27
0.603
24
0.648
27
0.685
17
0.627
26
0.587
Standard errors in parentheses, a – for a 40 year old average income men, who finished education at age 18
** p<0.01, * p<0.05, + p<0.1
Page • 39
Daniel Horn
9. table – Different measures of income inequality on turnout, hierarchical logit (ORs)
(1)
(2)
(3)
(4)
(5)
VARIABLES
Vote
Vote
Vote
Vote
Vote
Gini, Eurostat 2009
1.007
s80/s20, SSO 2009
0.944
(0.165)
0.979+
(0.0121)
Mean distance from the median
0.0943+
(0.114)
Poverty rate, SILC
0.964*
(0.0141)
p95/p5, LIS
0.713
(0.291)
p95/p5, SILC
Random effects parameters
sd(Constant)
Observations
Number of groups
seEform in parentheses
** p<0.01, * p<0.05, + p<0.1
Page • 40
(7)
Vote
(0.0346)
Gini, earning, SSO 2009
Constant
(6)
Vote
0.166
(0.207)
0.272
(0.262)
0.380
(0.327)
0.306
(0.273)
0.501
(0.415)
0.986
(2.091)
0.895
(0.0803)
0.300
(0.265)
0.529**
(0.0795)
20,202
27
0.507**
(0.0799)
18,964
25
0.499**
(0.0758)
20,202
27
0.474**
(0.0770)
18,112
24
0.473**
(0.0718)
20,202
27
0.458**
(0.0911)
19,603
26
0.499**
(0.0774)
19,603
26
Income inequality and voter turnout
10. table – Income bias, logit (ORs)
(1)
VARIABLES
vote
Subjective standard of living
1.029
(within country Z-score)
(0.130)
Gini, Eurostat 2009
1.018
* standard of living
s80/s20, SSO 2009
(0.0269)
1.005
(0.00428)
* standard of living
Gini, earning, SSO 2009
(2)
(3)
(4)
(5)
(6)
(7)
vote
vote
vote
vote
vote
vote
0.773
(0.122)
1.042
(0.170)
1.006
(0.0188)
* standard of living
Mean distance from the median
1.011
(0.157)
0.982*
(0.00726)
0.998
(0.00236)
* standard of living
Poverty rate, SILC
1.118
(0.0987)
0.222+
(0.184)
1.155
(0.202)
* standard of living
p95/p5, LIS
1.223*
(0.0986)
0.976+
(0.0129)
0.999
(0.00256)
* standard of living
p95/p5, SILC
1.239
(0.189)
0.750
(0.338)
0.995
(0.0252)
1.182
(0.186)
Constant
0.0833*
(0.103)
0.137+
(0.146)
0.251
(0.218)
0.179+
(0.175)
0.252
(0.219)
0.866
(1.593)
0.942
(0.0819)
1.001
(0.0125)
0.156+
(0.158)
Observations
20,202
18,964
20,202
18,112
20,202
12,979
19,603
* standard of living
Robust seeform in parentheses
** p<0.01, * p<0.05, + p<0.1
Page • 41
Daniel Horn
11. table – income bias, hierarchical logit (ORs)
(1)
(3)
VARIABLES
vote
vote
Subjective standard of living
(within country Z-score)
1.057
0.775
Gini, Eurostat 2009
* standard of living
s80/s20, SSO 2009
(0.176)
1.008
(0.0346)
1.005
(0.00545)
(0.194)
(5)
(7)
(9)
(11)
(13)
vote
vote
vote
vote
vote
1.046
(0.207)
1.147
(0.136)
1.274**
(0.0772)
1.231
(0.190)
1.228**
(0.0945)
0.944
(0.166)
1.003
(0.0189)
* standard of living
Gini, earning, SSO 2009
0.979+
(0.0121)
0.998
(0.00316)
* standard of living
Mean distance from the median
0.0959+
(0.116)
1.139
(0.265)
* standard of living
Poverty rate, SILC
0.964*
(0.0141)
0.999
(0.00179)
* standard of living
p95/p5, LIS
0.713
(0.291)
0.999
(0.0251)
* standard of living
p95/p5, SILC
0.162
(0.202)
0.271
(0.261)
0.385
(0.331)
0.303
(0.270)
0.505
(0.418)
0.988
(2.095)
0.895
(0.0803)
0.998
(0.0120)
0.300
(0.265)
0.529**
(0.0795)
20,202
27
0.507**
(0.0799)
18,964
25
0.499**
(0.0758)
20,202
27
0.474**
(0.0769)
18,112
24
0.474**
(0.0719)
20,202
27
0.458**
(0.0911)
12,979
17
0.499**
(0.0774)
19,603
26
* standard of living
Constant
Random effects parameters
Observations
Number of groups
seEform in parentheses
** p<0.01, * p<0.05, + p<0.1
Page • 42
Income inequality and voter turnout
12. table - Different measures of income inequality below and above the median on turnout, logit (ORs)
(1)
(2)
(3)
(4)
(5)
vote
vote
vote
vote
vote
VARIABLES
MDMI, below the mean
2.357
p50/p5, LIS
(6)
vote
(7.290)
17.76**
(17.43)
p50/p5, SILC
0.880
(0.342)
MDMI, above the median
0.466
(0.409)
p95/p50, LIS
0.102+
(0.134)
p95/p50, SILC
Constant
0.108+
(0.142)
0.000592**
(0.00119)
0.160
(0.229)
0.108*
(0.0977)
27.45
(74.54)
0.550+
(0.191)
0.382
(0.467)
Observations
Robust seeform in parentheses
** p<0.01, * p<0.05, + p<0.1
18,112
12,979
19,603
18,112
12,979
19,603
13. table – Income bias, below and above the median, logit (ORs)
(1)
(2)
VARIABLES
vote
vote
Subjective standard of living
(within country Z-score)
1.006
1.271
MDMI, below median
* standard of living
p50/p5, LIS
(0.162)
2.174
(6.760)
0.566
(0.315)
* standard of living
p50/p5, SILC
(0.432)
17.66**
(17.83)
0.980
(0.138)
* standard of living
MDMI, above the median
(3)
(4)
(5)
(6)
vote
vote
vote
vote
1.195
(0.156)
1.160*
(0.0846)
1.198
(0.267)
1.102
(0.174)
0.880
(0.344)
0.998
(0.0520)
* standard of living
p95/p50, LIS
0.469
(0.412)
1.051
(0.108)
* standard of living
p95/p50, SILC
0.102+
(0.134)
1.004
(0.0900)
Constant
0.105+
(0.140)
0.000600**
(0.00124)
0.160
(0.230)
0.108*
(0.0971)
27.40
(74.71)
0.552+
(0.193)
1.035
(0.0715)
0.378
(0.465)
Observations
18,112
12,979
19,603
18,112
12,979
19,603
* standard of living
Robust seeform in parentheses
** p<0.01, * p<0.05, + p<0.1
Page • 43
Daniel Horn
14. table – welfare state test 1 – logit (ORs)
(1)
VARIABLES
vote
Government expenditure,
Total, % of GDP, 2007
1.075*
Gini, Eurostat 2009
s80/s20, SSO 2009
(0.0345)
1.023
(0.0258)
(2)
(3)
(4)
(5)
(6)
(7)
vote
vote
vote
vote
vote
vote
1.067*
(0.0343)
1.070*
(0.0319)
1.059+
(0.0360)
1.065+
(0.0353)
1.098+
(0.0592)
1.068*
(0.0330)
0.981
(0.133)
Gini, earning, SSO 2009
0.986+
(0.00832)
Mean distance from the median
0.551
(0.576)
Poverty rate, SILC
0.986
(0.0131)
p95/p5, LIS
0.990
(0.326)
p95/p5, SILC
Constant
Observations
Robust seeform in parentheses
** p<0.01, * p<0.05, + p<0.1
Page • 44
0.00259** 0.00695**
(0.00507) (0.0126)
20,202
18,964
0.0101**
(0.0164)
0.00850*
(0.0174)
0.0113*
(0.0208)
20,202
18,112
20,202
0.957
(0.0744)
0.000777* 0.00709**
(0.00276) (0.0120)
12,979
19,603
Income inequality and voter turnout
15. table – welfare state test 2 - logit
(1)
VARIABLES
vote
Government expenditure,
Social prot, % of GDP, 2007
1.104*
Gini, Eurostat 2009
s80/s20, SSO 2009
(0.0515)
1.009
(0.0247)
(2)
(3)
(4)
(5)
(6)
(7)
vote
vote
vote
vote
vote
vote
1.108*
(0.0541)
1.098+
(0.0539)
1.076
(0.0543)
1.091+
(0.0567)
3.773**
(0.603)
1.101*
(0.0511)
0.897
(0.128)
Gini, earning, SSO 2009
0.986+
(0.00733)
Mean distance from the median
0.300
(0.252)
Poverty rate, SILC
0.983
(0.0130)
p95/p5, LIS
0.106**
(0.0387)
p95/p5, SILC
Constant
Observations
Robust seeform in parentheses
** p<0.01, * p<0.05, + p<0.1
0.0241**
(0.0314)
0.0351**
(0.0425)
0.0538**
(0.0599)
0.0528*
(0.0698)
0.0572*
(0.0726)
20,202
18,964
20,202
18,112
20,202
0.931
(0.0754)
1.51e-10** 0.0385**
(4.13e-10) (0.0444)
12,979
19,603
Page • 45
Daniel Horn
Page • 46
Income inequality and voter turnout
GINI Discussion Papers
Recent publications of GINI. They can be downloaded from the website www.gini-research.org under the
subject Papers.
DP 15
Can higher employment levels bring down poverty in the EU?
Ive Marx, Pieter Vandenbroucke and Gerlinde Verbist
October 2011
DP 14
Inequality and Anti-globalization Backlash by Political Parties
Brian Burgoon
October 2011
DP 13
The Social Stratification of Social Risks. Class and Responsibility in the ‘New’ Welfare State
Olivier Pintelon, Bea Cantillon, Karel Van den Bosch and Christopher T. Whelan
September 2011
DP 12
Factor Components of Inequality. A Cross-Country Study
Cecilia García-Peñalosa and Elsa Orgiazzi
July 2011
DP 11
An Analysis of Generational Equity over Recent Decades in the OECD and UK
Jonathan Bradshaw and John Holmes
July 2011
DP 10Whe Reaps the Benefits? The Social Distribution of Public Childcare in Sweden and Flanders
Wim van Lancker and Joris Ghysels
June 2011
DP 9
Comparable Indicators of Inequality Across Countries (Position Paper)
Brian Nolan, Ive Marx and Wiemer Salverda
March 2011
DP 8The Ideological and Political Roots of American Inequality
John E. Roemer
March 2011
DP 7
Income distributions, inequality perceptions and redistributive claims in European societies
István György Tóth and Tamás Keller
February 2011
DP 6
Income Inequality and Participation: A Comparison of 24 European Countries + Appendix
Bram Lancee and Herman van de Werfhorst
January 2011
DP 5
Household Joblessness and Its Impact on Poverty and Deprivation in Europe
Marloes de Graaf-Zijl
January 2011
DP 4
Inequality Decompositions - A Reconciliation
Frank A. Cowell and Carlo V. Fiorio
December 2010
DP 3
A New Dataset of Educational Inequality
Elena Meschi and Francesco Scervini
December 2010
Page • 47
Daniel Horn
DP 2
Are European Social Safety Nets Tight Enough? Coverage and Adequacy of Minimum Income Schemes in 14 EU Countries
Francesco Figari, Manos Matsaganis and Holly Sutherland
June 2011
DP 1
Distributional Consequences of Labor Demand Adjustments to a Downturn. A Model-based Approach with Application to
Germany 2008-09
Olivier Bargain, Herwig Immervoll, Andreas Peichl and Sebastian Siegloch
September 2010
Page • 48
Income inequality and voter turnout
Information on the GINI project
Aims
The core objective of GINI is to deliver important new answers to questions of great interest to European
societies: What are the social, cultural and political impacts that increasing inequalities in income, wealth and
education may have? For the answers, GINI combines an interdisciplinary analysis that draws on economics,
sociology, political science and health studies, with improved methodologies, uniform measurement, wide
country coverage, a clear policy dimension and broad dissemination.
Methodologically, GINI aims to:
●● exploit differences between and within 29 countries in inequality levels and trends for understanding the
impacts and teasing out implications for policy and institutions,
●● elaborate on the effects of both individual distributional positions and aggregate inequalities, and
●● allow for feedback from impacts to inequality in a two-way causality approach.
The project operates in a framework of policy-oriented debate and international comparisons across all EU
countries (except Cyprus and Malta), the USA, Japan, Canada and Australia.
Inequality Impacts and Analysis
Social impacts of inequality include educational access and achievement, individual employment opportunities and labour market behaviour, household joblessness, living standards and deprivation, family and
household formation/breakdown, housing and intergenerational social mobility, individual health and life
expectancy, and social cohesion versus polarisation. Underlying long-term trends, the economic cycle and
the current financial and economic crisis will be incorporated. Politico-cultural impacts investigated are: Do
increasing income/educational inequalities widen cultural and political ‘distances’, alienating people from
politics, globalisation and European integration? Do they affect individuals’ participation and general social
trust? Is acceptance of inequality and policies of redistribution affected by inequality itself ? What effects
do political systems (coalitions/winner-takes-all) have? Finally, it focuses on costs and benefi ts of policies
limiting income inequality and its effi ciency for mitigating other inequalities (health, housing, education
and opportunity), and addresses the question what contributions policy making itself may have made to the
growth of inequalities.
Support and Activities
The project receives EU research support to the amount of Euro 2.7 million. The work will result in four
main reports and a final report, some 70 discussion papers and 29 country reports. The start of the project
is 1 February 2010 for a three-year period. Detailed information can be found on the website.
www.gini-research.org
Page • 49
Amsterdam Institute for Advanced labour Studies
University of Amsterdam
Plantage Muidergracht 12 ● 1018 TV Amsterdam ● The Netherlands
Tel +31 20 525 4199 ● Fax +31 20 525 4301
[email protected] ● www.gini-research.org
Project funded under the
Socio-Economic sciences
and Humanities theme.
Income inequality and voter turnout
Page • 51
Amsterdam Institute for Advanced labour Studies
University of Amsterdam
Plantage Muidergracht 12 ● 1018 TV Amsterdam ● The Netherlands
Tel +31 20 525 4199 ● Fax +31 20 525 4301
[email protected] ● www.gini-research.org
Project funded under the
Socio-Economic sciences
and Humanities theme.
Income inequality and voter turnout
Page • 53