Economic Inequality and Voting Participation

Economic Inequality & Voter Turnout 1 Economic Inequality and Voting Participation "Well, because poor people don t vote. I mean, that s just a fact," - Bernie Sanders. 2016 Authors: Nils Brandsma & Olle Krönby Supervisor: Stig Blomskog Södertörns högskola Institution of Political Economics Bachelor Thesis: 15 hp Economics Fall 2016

Economic Inequality & Voter Turnout 2 Abstract The following paper assesses a statistical relationship between Economic Inequality and Voting Participation among a sizable amount of nations across the world representing all continents. With an deductive approach, three theoretical standpoints of interest are presented: one that describes a negative, another inconclusive, and one with a positive relationship between the variables of interest. Through panel data analysis the study finds support in favour of a negative relationship in that as economic inequality rises, voting participation in parliamentary elections decreases.

Economic Inequality & Voter Turnout 3 Table of Contents ABSTRACT... 2 1. INTRODUCTION... 4 1.1 RESEARCH QUESTION... 4 1.2 PREVIOUS RESEARCH... 5 2. THEORETICAL FRAMEWORK... 7 2.1 CONFLICT THEORY (MELTZER & RICHARD 1981):... 7 2.2 RELATIVE POWER THEORY (GOODIN & DRYZEK 1980):... 8 2.3 RESOURCE THEORY (VERBA, SCHLOZMAN & BRADY 1995)... 9 2.4 SUMMARY OF THEORIES... 10 4.1 VARIABLES... 15 5. ANALYSIS... 17 5.1 SUMMARY OF VARIABLES... 17 7. CONCLUSIONS... 25 8. TOPICS FOR FUTURE RESEARCH... 26 REFERENCES... 28 APPENDIX... 30

Economic Inequality & Voter Turnout 4 1. Introduction A negative aspect of the modern democracy is the often low rate in participation by its citizens (Pew Research Center 2016). The 2016 election in the United States drew attention to a number of facts, one being that the president-elect only received approximately 26% of the popular vote, and if the candidate did not vote would have been a physical person, he or she would ve won by a landslide. Why is it the case that people chose to abstain from voting, even though it is an event only occurring every few years and has massive impact on peoples lives? Our paper looks into one suggested factor, income inequality, since the increase in average real income has been largest for the richest quintile and lowest for the poorest (Bartels 2009 2). As expressed by Robert Dahl in 1961: In a political system where nearly every adult may vote but where knowledge, wealth, social position, access to officials, and other resources are unequally distributed, who actually governs? (Dahl 1961 1) Although the subject has been studied before, most of it has been focused on a select few countries, often based in Western Europe and North America. Even then, the results has ranged from finding a positive correlation between voter turnout and income inequality (Brady 2003), inconclusive results (Geys 2006) and negative correlations (Solt 2008). In order to avoid drawing conclusions from outdated information, this study will examine data from the millennia and forward in the sole purpose of generating results at the more present time of the 21st century. In this paper, we intend to include countries that have previously been excluded from similar studies, many of them developing countries, in which some are regarded as partly free by the Freedom House index. Previous research has mostly been focused on western industrialized democracies, but has suggested a future study with a broader sample, which is what we aim to do. 1.1 Research Question This study is of deductive nature and will therefore go about introducing three theories relating to voting participation and economic inequality, to later test these using panel data analysis in order to answer our research question. The research question of this paper is as follows:

Economic Inequality & Voter Turnout 5 To what extent, and in which direction does economic inequality affect the civil participation of voters in national elections? The corresponding equation will be set up as follows: Y!"#$%!"#$%"& = α + βx1!"#" + ε 1.2 Previous research So far, previous research has been inconclusive regarding the effect of income inequality on political participation. Frederick Solt (2008) examines three dependent variables, political interest, participation in political discussion and electoral participation. He finds that income inequality affects all three negatively. Most notable for us is the finding that higher income inequality lowers political participation across the board, and that the results show on average, likelihood for participation is lowered by 12,9 percentage points for the lowest quintile and slightly less drastic number for each subsequent higher quintile. In other words, economic inequality depresses voter turnout for all income groups in this study. (Solt 2008) Benny Geys (2006) provides us with an aggregate-level survey study which uses the results of 83 studies on turnout, and provides valuable information on variables relating to population and political institutions. He also briefly mentions income inequality, and finds 13 significant negative results, 6 significant positive and 13 insignificant results (Geys 2006 645). Geys updated this research in 2016 with João Cancela. It builds on Geys earlier work previously mentioned, and points us towards a potential risk we face going into this study, that our results may be inconclusive. Obviously, this is not a unique predicament, but the study made by Cancela and Geys points to a quite harsh apparent truth. When examining 18 previous studies regarding income inequality over turnout, only 11% were successful (Cancela & Geys 2016 267). With this in mind they establish that income inequality is a poor variable for explaining turnout. This means that we should expect a low R 2 value, if this pattern should hold. It also means that we have a low chance of success for the study, and any result producing statistically significant relationships should be considered a pleasurable outcome. The tiny light in this darkness is that the interest for studies on income inequality

Economic Inequality & Voter Turnout 6 has increased, perhaps thanks to thinkers such as Thomas Piketty, and we are likely to see an increase in research trying to figure out the link between these variables. Henry E. Brady (2003) performs an analytical study of income inequality and turnout, but focuses solely on the US. Like others, his study does not find any definitive results. In line with Solts methodology, he considers other factors than voting turnout, looking at participation in terms of writing letters or joining protests (Brady 2003 3). The problem with considering non-electoral factors, both lawful and unlawful, is that they are in general not a sign of a well performing democracy. The lack of demonstrations can either mean that the state is very repressive, or that there simply isn t anything to complain about. Demonstrations can be targeted towards foreign countries actions (like demonstrations in Europe against the Vietnam war) or towards their own government, or even as a hail to historical political event, such as the first of May parade in Sweden. Bluntly put, it often measures the political activity related to being unsatisfied with political circumstances, and increasing this type of participation is generally not preferable. Having low voter turnouts are however more universal as a measure of lack of democratic elements or political apathy. It seems abundant providing research to a topic already so thoroughly researched, especially since we are going into a study knowing that the results are likely to be inconclusive. But from what we have observed from earlier work is that many studies has had a broad scope. Either it is lumping positive and negative effects before conducting the research, such as Brady 2003, or it is a study conducted on positive effects but with far too many variables: the Solt 2008 study uses 28 control variables which inevitably ends up telling us nothing. The use of a large number of control variables is a debated issue in political economics, the critics has given it the quite negative sounding name trashcan regression or garbage can regression. The method of previous studies is not the only issue we consider troubling, but the selection as well. First of all, none of the aforementioned studies have represented democracies from all continents. Solt, who provides the most inclusive research, studies 22 countries, all of them except five are European states. The five non-european states are Canada, United States, Australia, Israel and Taiwan. Thus, there is a substantial under representation of states outside of what is considered the "western civilization". However, this is recognized by Solt and he has suggested further research with a more inclusive dataset. This is not the only problem of selection Solt has, he has also included states with compulsory voting laws, which damages explanatory power of his study which he himself admits to (Solt 2008 57). Our

Economic Inequality & Voter Turnout 7 study attempts to capture the result of a larger set of states, including countries from previously missing continents such as South America and Africa. However, it is most likely the case that Solts selection is a matter of which data he used. Solt relies on the Luxembourg Income Study to provide him with trustworthy data on income inequality, which in concern to validity is a good approach since it is a more comprehensive database than the data we are using, from the world bank. The purpose of our study is to expand the range of countries in this type of research, as encouraged by Solt, since he suggests it as a topic for future research in his article (Solt 2008 50). Sadly, due to the available data we can not perform a study showing the difference among income groups, which we will discuss further later. 2. Theoretical framework 2.1 Conflict Theory (Meltzer & Richard 1981): Conflict theory presents a classical economical rational choice-type theory to explain voter turnout over income inequality. They assume that individuals are rational and that the income dictates political preferences. The way that Meltzer & Richards measure political preferences is in quite simple terms, basically arguing that the essence is being against or for redistribution of income and resources within society. With a rational-choice type perspective, this is simplistic but not necessarily a far-fetched assumption. They divide voters into three groups, the middle one is called the decisive voter and essentially is the median of the combined voting population, and the other two groups are above and below the decisive voter respectively. The allocation and size of these groups (the decisive voter being in theory just one person, but practically probably a large group of people) decide the outcome of the election. The driving force for turning up to vote is the protecting of income for the richer group, and for the below-median group the reason to go to vote is the possibility of redistributing in their favor. In a more unequal society, these forces will be amplified as there are larger potential gains or losses in play. In a more unequal society, voters with lower incomes will gain more utility by voting for the candidate arguing for more redistribution. Conversely, in the same society the richer voters would potentially lose more in the event of an electoral win by the candidate favoring redistribution. Turnout should then be lower in very equal societies, where more voters are indifferent about redistribution. (Meltzer & Richard 1981)

Economic Inequality & Voter Turnout 8 2.2 Relative Power Theory (Goodin & Dryzek 1980): An alternative to conflict theory is Relative power theory by Goodin and Dryzek (1980) who argues that with wealth comes relative power. Goodin and Dryzek draws on earlier work in social psychology where social efficacy is regarded as subjective and irrational behaviour by individuals on deciding to participate in politics or not, meaning that depending on their feeling of capacity in getting the job done. Goodin and Dryzek agrees with the basic thought of subjective efficacy but continues to argue that worse off socio-economic individuals relinquish participation with rationality. An example of this is that a more wealthy individual has been rewarded by the system, therefore this particular individual attains trust to the system and will continue to participate. This leads to a concentration of power to a certain class of people, who can further decide the agenda of politics. To this class, it becomes more rational to attain information which they use to form an opinion which they present at the voting booths. On the flip side, lower socio-economic groups will have a lower probability of winning due to the fact that their interests may not be represented. Therefore, they are less likely to participate in the elections. The relative wealth is a deciding factor for turnout, according to Goodin & Dryzek, as the quote below shows: Thus it seems that people are not indifferent to the standing of others, and do not look exclusively to their own economic position, when deciding whether to vote (Goodin & Dryzek 1980 284) Thereby, worse off citizens will lose their incentive to vote due to the relative position of power they have compared to their neighbor. Furthermore they argue that rational choice creates incentives to vote depending on every individuals possible utility from participating. The higher probability of being successful, the more rational it is for individuals to acquire information and participate in politics. Another important aspect of this theory is that due to political power being concentrated, the issues on the political agenda are fewer and more specific, which should lower voter turnout for high income individuals as well, since politics will appear to be less meaningful. In conclusion, Goodin and Dryzek argues that with greater socio-economic inequality, political participation declines, mostly for low income groups, but also to a lesser extent for high income groups. (Goodin & Dryzek 1980)

Economic Inequality & Voter Turnout 9 2.3 Resource Theory (Verba, Schlozman & Brady 1995) Resource theory measures political activity as a result of three resources; civic skills, money and free time. Civic skills include language proficiency, educational attainment, participation in higher school governments and activity in adult institutions such as the job, church or another organization. Resource theory suggests three answers to the question of why people do not vote: because nobody asked them to, because they don't care, and finally because they can't. The theory proceeds to research the latter variable, and identifies three factors of resources to be an essential part of why citizens are unable to vote: time, money and civic skills. They distinguish between three types of political activity, voting, making campaign contributions and engaging in time consuming political activity, such as campaigning or attending meetings. This paper has made the conscious decision to only focus on electoral participation in the voting booths, so we will mainly focus on the findings related to that. However, a brief comment on the other two will be made. Verba et al. find that the main reason to donate money to a campaign is having money, whereas time, civic skills and political interest only has a marginal effect. For time consuming activities, political interest, having free time and civic skills are both important factors, but income is not. For voting however, their findings suggest that political interest is the most important indicator for turnout, followed by the significant variable of having free time. Education, they find, has been overestimated in previous work as an indicator for voting, but they still conclude that education is an important explanatory variable in determining political interest, along with past participation in high school government and having a good vocabulary (Verba, Schlozman & Brady 1995 p.283). Income has little impact on voting according to this theory. Worth noting about this study is that it is a research entirely conducted in the United States of America, and the correlation between attaining education and having money is high relative to other states, most likely due to the high cost of private education (OECD 2014). In states with a lower entry cost of education might have a different correlation. Another point stated in the Verba et al paper is that family income and civic skills are important determinants to political interest across a multitude of tested models, that included a multitude of variables and testing. In no case could the importance of these two variables be evaporated from having a significant effect on political interest (Verba, Schlozman & Brady 1995 282).

Economic Inequality & Voter Turnout 10 2.4 Summary of theories Our three theories expect different outcomes, conflict theory expects inequality to be positively correlated with turnout, as it creates more incentives for the poor population to vote for redistributive policies and for the richer population to vote against such policies (Melzter & Richards 1981). Relative power theory expects the opposite, that with concentration of wealth, power is also concentrated to a certain socio-economic class. This leads to political apathy among the poorer citizens and decreases their incentives to vote. Consequently, the concentration of power will lead to less impactful politics, and therefore even people with higher incomes will vote less (Goodin & Dryzek 1980). Resource theory is less certain of which direction voting turnout will go in as a result of increased inequality. They argue that people vote as a result of three resources, money, free time and civic skills. When identifying where these resources are concentrated they conclude that high socioeconomic individuals has more money and civic skills, but free time might go either way. There is a possibility that their high paying job demands a lot of time, and long hours leaves less free time. However, the opposite can also be true for low paying jobs, such as nursing or having several jobs. The variables of money and civic skills should however, according to Verba et al, effect turnout in different ways for different income groups, where high socioeconomic individuals are more likely to vote, and low income individuals are less likely to vote (Verba, Schlozman & Brady 1995). Our expectation is that economic inequality will have a negative relationship with voting participation. However, in our final models we won't be able to tell if this difference is mainly among lower socio-economic groups or for the entire population. This will provide a difficulty in determining whether the resource theory or the relative power theory is the more correct theory, if inequality proves to be negatively correlated with turnout. 4. Data & Method Instead of hand picking countries as subjects of study, we have used the freedom house database to exclude non-democratic, excessively corrupt or non-functioning states. Furthermore we are excluding states who employ any kind of compulsory voting laws. Even though compulsory voting systems have been proven successful in increasing voter turnout (Blais 2006), compulsory voting is not of interest since this study is aimed to find a

Economic Inequality & Voter Turnout 11 correlation between factors without legislation forcing the participation to increase. The data analyzed is collected from three reputable sources: The World Bank International Institute for Democracy and Electoral Assistance (IDEA) United Nations Development Programme, Human Development Reports (UNDP, HDR) The data for freedom house scores is included in the IDEA dataset and provides the score at the time of the relevant election, which enables us to use it in a more effective manner than using the scores from the latest edition of the freedom house datasets. Before describing our method more in depth, a few clarifications and explanations should be done. Firstly, we have performed separate regressions depending on electoral system, presidential or parliamentary. Since we are running panel data regressions, it is important not to have repeated time variables within the data. From our perspective, we could see two options. Either run two separate regressions for each system, or give them two panel data IDs for each system and run them together. Either option has it s pros and cons. The latter approach, dividing the countries examined with unique panel IDs for each type of election has the risk of having repeated data for some years. This can occur when countries which has both parliamentary and presidential elections (which is most presidential systems) have their elections for president and parliament the same year. This essentially means that the value for GINI, population and education is repeated for that year but with two potentially different values for voting turnout, which would give us biased results. Separating the regression models effectively solves this problem, as no country in our data has had either a parliamentary or a presidential election twice in one year. Using this method would also result in more complexly presented data, as the grouping system would be both country and electoral system to create one group variable, rendering the number of entities to be 177, which is a confusing number given the fact that we only examine 115 countries. The drawback of performing regressions on the separated data is that the chapter regarding results will be quite model heavy. This last point is only amplified by another methodological choice, that we are examining two dependent variables. The first one is self explanatory, voting turnout. It is based on officially reported statistics and is the total amount of votes over the amount of people registered to vote.

Economic Inequality & Voter Turnout 12 Total Votes Registered Voter = Y1!"#$%!"#$%"& At first glance, this seems like a perfectly decent measure of electoral participation, but the problem is that it does not consider the people not registered to vote, which intuitively is a choice made by the most disenfranchised individuals among the voting age population. International IDEA provides us with a remedy to this; they have made an estimate of how many voted among the total voting age population, i.e. anyone over the minimum voting age. Total Votes Voting Age Population = Y2!"#$%&!"#!"#$%&'(")!"#$%"& This leaves us with three options: either we look at only the officially reported number, which would result in a potentially less accurate depiction of reality, we look at only the voting age population or we look at both. In the best of worlds, the voting age population estimates would be accurate enough to be the only measurement we needed, but sadly we do not live in such a world. Later, in the chapter where the data is described you may notice why and it will be explained further there. The short version is that the data has it s flaws, leaving us with results that go beyond 100% turnout in a few cases, in other cases the amount of registered voters is larger than there are people in the voting age population. The third option is to analyze them both separately, which is what we have chosen to do given these explained issues. Random Effect Models: Equation 1 Y1!"#$%!"!"#$% = α + βx1 GINI + βx2 ln( GDP Capita,PPP) + βx3 ln(population) + βx4 Freedom House + βx5 Education + (υ it + T i + ε) Equation 2 Y2!"#!"#$%"& = α + βx1 GINI + βx2 ln( GDP Capita,PPP) + βx3 ln(population) + βx4 Freedom House + βx5 Education + (υ it + T i + ε)

Economic Inequality & Voter Turnout 13 Fixed Effect Models: Equation 3 Y1!"#$%!"#$%"& = α + βx1 GINI + βx2 ln( GDP Capita,PPP) + βx3 ln(population) + βx4 Freedom House + βx5 Education + υ it + T i + ε Equation 4 Y2!"#!"#$%"& = α + βx1 GINI + βx2 ln( GDP Capita,PPP) + βx3 ln(population) + βx4 Freedom House + βx5 Education + υ it + T i + ε VAP=Voting age population, α=intercept, β=coefficient, υit=fixed individuality term, Ti=Time, ε=error term The models that will be analyzed are the random effect models, which are deemed more appropriate through the Hausman test (appendix 1). In theory, we might suspect an omitted variable bias when looking at the relation between Freedom House scores and GINI, or perhaps education and GINI, but a few examples we know of might also offset that assumption. The most immediate example is the case of the United States, where freedom levels, GDP per capita and education levels are all quite high, and yet they suffer from a higher level of inequality than other similar states located in Scandinavia. Sources: Human Development Reports, The World Bank. The above graph shows similarities between the Scandinavian countries and the US in GDP/capita and Education levels, yet the difference in inequality is higher. The Republic of Korea has also succeeded in creating economic growth, enjoying similarly high levels of education, GDP per capita and freedom scores but may be considered lacking

Economic Inequality & Voter Turnout 14 in the equality department (Denney, Steven 2014). Rather than following the assumption that inequality is reduced by market powers as a result of economic growth, high levels of civil and political rights and having a highly educated population, we could attribute the differences to political and social culture. States in Scandinavia has achieved lower inequality through the means of active government participation, such as progressive taxes, and Japan is an example of how social culture has worked to prevent high inequality levels (The Economist 2015). Then again, these are just a handful of states across our dataset, and we should be careful before assuming that this is all true for the rest of the world. One option would be to try to capture these individualities within a fixed effect regression, but finding the necessary data across the world for political and social culture around economic inequality will prove to be most difficult. The variable that comes closest to capturing these differences may be our variable for freedom house scores, but since it s entirely possible to have political rights and civil liberties without the government pursuing active measures to redistribute income it is far from capturing all of these effects. The intuition in this case, is unable to provide us with clear answers, and because of this fact we have chosen to let the Hausman test decide for us which model should be used. For the data we are using in this study, the Hausman test does not observe significant bias among the independent and control variables and therefore the analysis will be performed with the random effects regression method. However, since the intuition is divided on the issue, we have chosen to include the results of the fixed effect models in the appendix. We have also chosen to analyze the population and GDP per capita variables log-transformed with the natural base e. This is because these variables in their normal state matter differently across the countries observed. An increase with a thousand people in India has different implications than an increase by the same amount in Finland. However, if they increase their population by 1% then that has more similar implications for both, because their institutions are relatively adapted to their current population. Growth in GDP also makes more sense to analyze in terms of percent increase or decrease, for a few reasons. The main reason is similar to the one stated about population, that we are analyzing economies that has different sizes, and one units increase in GDP matters differently for different economies, whereas a percentage increase accounts for size in a more adequate manner. The second reason is because when discussing GDP in general, we usually discuss it in terms of percentage change and not with absolute numbers.

Economic Inequality & Voter Turnout 15 4.1 Variables Voter Turnout This variable is gathered from the International Institute for democracy and electoral assistance (IDEA) and consists of officially published voting results, measured by percentage of votes from registered voters. Voting age population turnout (VAP) This second turnout variable measures the turnout of the entire population above the age of voting. The variable above measures the turnout based on the registered voters. Our research attempts to measure voter apathy and disenfranchisement based on income inequality, and not registering to vote should be regarded as non-participation as much as not turning up to the voting booths. There is a reason we chose to have both voting age population and regular voter participation, which is because this measure also has its downsides. One problem is that this measure does not take into account that people might have legal barriers to registering their vote, and that there are potential problems with the actual data since it is merely an estimate by IDEA. Measured in percentage points. GINI-index The variable GINI refers to the statistical measure of dispersion of income in a nation. This means that the variable measures income inequality. The GINI coefficient is usually a number between zero and one, where one is the maximum inequality. Our data uses the equivalent conversion of numbers between 0 and 100, so instead of a value of 0,47 the number would be 47. Using the GINI index as our measure for inequality puts us in the position of having to defend the measure system. The GINI has been criticized for a number of reasons, and alternative measures have been proposed. The Luxemburg income study is an alternative index for measuring inequality, which is used by cited professor Frederick Solt, who we criticized for using too few countries and mostly western ones at that. The problem however, is not his ambitions, but the data he was using. The Luxemburg income study is a great database for wealth and income of countries, done through extensive use of different surveys to produce trustworthy data. The thoroughness of the LIS data is a good thing for validity, but this data takes time to produce, and since we aim to provide data on more countries than the LIS can provide we have unfortunately not been able to use this database.

Economic Inequality & Voter Turnout 16 It should be noted that the LIS bases it s inequality measure on the GINI coefficient still. Another alternative is using the Palma index, which calculates inequality a different way than the GINI, and according to some it provides more telling results. There is a problem here in that it is a relatively new index, and is ill suited for pre-2015 studies. Using the Palma would exclude almost all of our available data on turnout results; hence we have chosen to perform this study with the GINI index. (Cobham 2013) Gross Domestic Product per capita with Purchasing Power Parity The variable for GDP per capita with respect to Purchasing Power Parity is measuring the general wealth of the countries examined. Using the addition of per capita and purchasing power parity is because this more accurately describes the available resources. We are interested in the wealth of individuals in the state, and their political behavior, not in the wealth of a country. Population Size of population, gathered from IDEA. There are a few different reasons to include a measure of population, but the main one is that we want to control for it because it might have an impact in the sense that a higher population may lead to a lower turnout since each individual vote matters less, as suggested by Geys (2006 642). There are a few other population type variables that relates to socio-economic factor we could use, such as the level of urbanization, rate of population growth or population density that could control for similar things. The argument for a simple population size is that it this study aims to understand what makes people vote or not, not what they intend to vote for. Urbanization and population density might be variables more affecting a left-right decision rather than a vote-abstain decision, whereas the individual votes carries equal weight in most states. Of course, there are differences across our countries as well, the most accessible example being the difference in voting power between American states (FairVote 2016). Population density would also be an interesting variable, as higher density should in theory lower the cost of gathering information about options, and therefore increasing turnout. (Geys 2006 642-644). Education The variable education refers to mean years of education for men and women aged 25 and above. The data is collected from the World Bank, and is more incomplete than the rest of the variables observed, except for GINI. We have chosen to run the variable last in each regression, to ensure our first four models have the maximum amount of observation.

Economic Inequality & Voter Turnout 17 Freedom house Data from international organization Freedom House and their reports on civil and political liberties. Score ranges between 1 and 5 where the former is the most free and the latter the least free. 5. Analysis In this chapter we will firstly present our regression analyses starting with the bivariate model and then advancing to several multivariate models. After the presentation of the regressions we will analyse and describe the result to distinguish whether or not we can answer our research question. 5.1 Summary of Variables N: observations counting each country+year as one n: Number of groups (countries) T: time variable, elections. Period of analysis is 2000-2013 Description 1: Panel data: presidential Panel variable N n T Min T Max T Median T Country + year 160 62 13 1 4 3 Variable N n Mean Min Max Std.deviation Voter Turnout (%) Voting age population turnout (%) 158 62 64.05 22.36 95.7 14.12 159 62 60.21 18.93 97.85 14.36 GINI 67 36 36.27 8.1 61 9.88 GDP/capita PPP 156 61 10450 530 51433 11644 Education 110 59 7.4 1.3 12.3 3.2 Population 160 62 24,7 (million) 19092 313 (million) 52.6 (million) Freedom House 160 62 2.6 1 5 1.2

Economic Inequality & Voter Turnout 18 Description 2: Panel data: parliamentary Panel variable N n T Min T Max T Median T Country + year 371 115 13 1 7 3 Variable N n Mean Min Max Std.deviation Voter Turnout (%) Voting age population turnout (%) 351 115 65.03 22.77 95.7 13.57 351 114 62.10 12.7 107.56 * 17.42 GINI 132 62 35.61 23.7 61 8.28 GDP/capita PPP 356 111 16148 530 88250 16156 Education 241 110 8.6 1.3 12.9 3 Population 371 115 28.8 (million) 10267 1.16 (billion) 93.9 (million) Freedom House 371 115 2.1 1 5 1.2 The 107% VAP turnout makes our methodological choices more clear, since the estimate is a more accurate description of reality for countries that require registration before voting. This might increase the cost of voting, depending on the process for registering to vote. The voter turnout variable is also only based on amount of voters among those who registered, which in some cases can be very different from what is actually the number of people eligible to vote. This is however, still an estimate, and it has more errors than the official statistics. Therefore, we have chosen both variables for analysis. The second thing that is important to point out is the lack of observations for GINI, which decreases the amount of observations we eventually end up with. The findings of Geys and Cancela (2006, 2016) report, that most inequality over turnout studies fail can most likely be * This result requires an explanation. The turnout was not 107.56% in the election, as that is impossible. IDEA explains this phenomenon with several explanations relating to actual estimation of figures but also the process of registration. IDEA firstly points out that the voting age population figures are based on estimates, which might differ from the true values (as with all estimates). Another issue springs from the fact that data used for the variable may be gathered from different sources, one for VAP, and another for registration. With the problem of estimates in mind, the different sources for data may have different estimates, which lead to discrepancies between the two measures. Secondly, the lists presented by governments or organizations may be flawed in the number of registered voters. Examples of this can be individuals listed two or more times, or that no longer eligible voters are not removed from the voting lists. (IDEA 2016)

Economic Inequality & Voter Turnout 19 traced back to this. The studies of economic inequality has been historically cursed with obstructions such as lack of data and questionable measuring systems, as Piketty mentions in the first chapter of The Capital in the 21st Century (Piketty, Thomas 2013). Further back in time, even less observations for GINI is available, which in the case of adding more length to the dataset would have caused even more missing observations and on these grounds we have chosen to not expand the time period of analysis. To some extent this is also true for the variable regarding education, but as the latter plays the role of a control variable is easier to work around. As you will notice, it is included after every other control variable in the regressions to keep the first four models observations at the highest possible level. Since education is mentioned in all our mentioned previous studies, as well as being a central part of the theories we have still chosen to include this variable in our regressions. The population data ranges between relatively small countries, such as S:t Kitts & Newis, S:t Vincent and the Grenadines and The Federated States of Micronesia and larger ones such as India. This might lead to an overestimation in how much population matters for turnout, since the difference in population can be several million people in some cases. This is also a reason why we have chosen to log-transform our population and GDP per capita variables.

Economic Inequality & Voter Turnout 20 Model 1: Presidential voter turnout Model 1 2 3 4 5 Dependent variable: Voter Turnout Constant (Standard error) 65.56*** (7.17) 93.56*** (20.53) 97.67*** (29.58) 91.42*** (35.51). 122.14*** (43.68) GINI -.03 (.18) -.14 (.20) -.16 (.16) -.16 (.21) -.21 (.21) ln(gdp/capita, PPP) -2.62 (1.83) -2.83 (1.88) -2.32 (2.48) -.79 3.43 ln(population) -.10 (1.34) -.09 (1.34) -2.38 (1.62) Freedom House.66 (2.25) Education 1.21 (2.48) -1.10 (1.09) Country + year (Countries) 66 35 66 35 66 35 66 35 43 28 R2 within between overall 0.22 0.0005 0.01 0.22 0.0005 0.01 0.23 0.0005 0.01 0.21 0.002 0.02 0.2 0.14 0.11 Significance level codes: * - 10% level, ** - 5% level, *** - 1% level of significance. For our first model, there is little to be said since almost nothing is statistically significant. Most likely, there are too few observations and the data is sporadic at best.

Economic Inequality & Voter Turnout 21 Model 2: Presidential voting age population turnout Model 1 2 3 4 5 Dependent variable: Voting Age Population Turnout Constant (Standard error) 68.26*** (6.21) 79.00*** (17.94) 120.73*** (24.77) 100.67*** (29.21). 93.09** (41.73) GINI -.15 (.16) -.18 (.17) -.15 (.21) -.17 (.16) -.17 (.19) ln(gdp/capita, PPP) -1.03 (1.63) -1.14 (1.57).49 (2.05) 2.62 (3.31) ln(population) -2.63** (1.11) -2.61** (1.09) -3.03** (1.53) Freedom House 2.14 (1.84) Education 3.03 (2.35) -.90 (1.05) Country + year (Countries) 67 36 67 36 67 36 67 36 44 29 R2 within between overall 0.04 0.001 0.03 0.09 0.0002 0.03 0.17 0.12 0.1 0.08 0.17 0.13 0.06 0.21 0.16 Significance level codes: * - 10% level, ** - 5% level, *** - 1% level of significance. In model two, the only significant relationship observed is from the Population variable. Since nothing else but the constant is significant, there is little to be said other than that having a large population tends to depress turnout according to our data.

Economic Inequality & Voter Turnout 22 Model 3: Parliamentary voter turnout Model 1 2 3 4 5 Dependent variable: Voter Turnout Constant (Standard error) 82.96*** (6.15) 103.47*** (15.39) 132.43*** (20.21) 163.34*** (22.74). 175.81*** (27.52) GINI -.54*** (.16) -.64*** (.17) -.61*** (.17) -.57*** (.16) -.51** (.20) ln(gdp/capita, PPP) -1.85 (1.28) -1.40 (1.24) -3.96** (1.55) -4.61** (2.34) ln(population) -2.15** (.94) -2.14** (.90) -2.56** (1.02) Freedom House -3.82** (1.40) Education -4.84** (1.67) -.02 (.75) Country + year (Countries) 130 60 130 60 130 60 130 60 98 52 R2 within between overall 0.06 0.09 0.14 0.14 0.06 0.11 0.13 0.15 0.16 0.15 0.23 0.19 0.15 0.29 0.22 Significance level codes: * - 10% level, ** - 5% level, *** - 1% level of significance. For our first model that analyzes the parliamentary election we see more of significant relationships. First of all, in this model our variable for inequality is significant and shows a negative correlation, and for every one unit increase on the GINI scale, voter turnout in the official statistics is depressed by roughly 0,5-0,6 percentage units. GDP per capita, Population and Freedom House scores all display a negative significant relation with voter turnout, which will be subject to further analysis in the results chapter. Our explanatory power measured in R-squared is 22%, which in the case of voter turnout is not necessarily bad. Turnout is dependent on a lot of variables, some very specific to the country or time of election. Some countries has election day as a public holiday to make sure voting does not mean losing income from work, other do not. There is simply no model big enough to account for all these specific laws and practices that affect turnout. In these circumstances, 22% explanatory power should not be considered a bad result.

Economic Inequality & Voter Turnout 23 Model 4 : Parliamentary voting age population turnout Model 1 2 3 4 5 Dependent variable: Voting Age Population Turnout Constant (Standard error) 81.84*** (6.08) 73.23*** (15.33) 117.36*** (19.73) 141.37*** (23.04). 153.26*** (26.75) GINI -.59*** (.16) -.56*** (.17) -.55*** (.16) -.53*** (.16) -.37** (.18) ln(gdp/capita, PPP).79 (1.29) 1.20 (1.13) -.81 (1.60) -2.65 (2.29) ln(population) -3.05** (.91) -3.02*** (.88) -3.31*** (.97) Freedom House -2.83** (1.44) Education -3.64** (1.62).76 (.73) Country + year (Countries) 130 61 130 61 130 61 130 61 98 53 R2 within between overall 0.02 0.15 0.17 0.01 0.17 0.18 0.02 0.32 0.25 0.04 0.36 0.25 Significance level codes: * - 10% level, ** - 5% level, *** - 1% level. 0.05 0.35 0.24 Similarly to the model for voting turnout in parliamentary systems, the above model for the voting age population turnout displays similar characteristics. Perhaps surprisingly, inequality matters less in all five models in the VAP model than for the model examining official statistics. This may indicate a few different things. First of all, it could mean that registration does not increase the the cost of voting. However, since this is in contrast to studies made on the subject, this is most likely not the case (Rosenstone & Wolfinger 1978). It could also be a result of turnout being lower overall, and the control variables account for less of a difference. When measuring voting age population turnout, GDP per capita is no longer statistically significant, in any of the models. Having a high population still has a negative effect on voter turnout, and having less political and civil freedoms measured by the freedom

Economic Inequality & Voter Turnout 24 house index also show a negative correlation. The R-squared values tell us that we can in this model approximately account for 24% of the effect on turnout. 6. Results The models 1 & 2 on presidential elections show insignificant results most likely because of lack in the data. Hence this thesis will continue to focus on models 3 & 4 for parliamentary elections where we find significant and effectful results. The following chapter will present the result of the analysis with regard to the earlier stated research question and theory. To recap, the research question that will be answered in this thesis is as follows: To what extent does economic inequality affect the civil participation of voters in national elections? In models 3 & 4 we find that a one-unit change in the GINI index results in an approximate decrease in voter turnout by 0.4 to 0.6 percentage points. As seen in the models, the explanatory power of the regressions increases with the adding of more control variables except for education. The insignificant result of education can be explained by the drop in observations when the variable is added; therefore the quality and R2 value of the whole model decreases. The variable for population size seems to be negatively related to turnout, which indicates that the idea presented by Geys (2006), that with a higher population the relative power of the vote decreases, may be true. It could however also be a result of the inclusion of micro states, as discussed earlier. In the theory chapter we presented three different views on economic inequality and the effect on turnout. Firstly, there was Conflict Theory, which argues that in more unequal societies individuals with lower income will have a higher utility of voting, and thus the theory predicts turnout to be higher in more unequal nations (Meltzer & Richards 1981). Secondly, we discussed Relative Power Theory by Goodin & Dryzek (1980), which predicts that in a more economically unequal society, people will lose their incentive to vote due to lower relative power and therefore a higher GINI score would decrease turnout. Thirdly, we had the Resource Theory by Verba, Schlozman & Brady (1995) who argue that political participation depends on three resources: civic skills, money and free time. The theory finds three forms of political participation but for the purpose of this study we focus on actual voting, which is argued not to be affected by income.

Economic Inequality & Voter Turnout 25 Out of these three theories, only Relative Power Theory by Goodin & Dryzek (1980) is able to relate to the findings in the models of this paper. Since the variable for GINI (a higher GINI indicates a more economically unequal society/nation) has a negative relationship with turnout and voting age population turnout, this paper supports this theory. But this is not to be mistaken for a causal relationship since turnout is affected by multiple more variables than the ones included in this paper. Conflict Theory is the only theory fully rejected, because the relationship seen in every model contradicts the theory of higher inequality leading to higher voting participation. As for Resource Theory, the variable included in our paper that would give an indication of this theory being supported is education as a part of its resources that affect political interest. But the variable for education is not statistically significant and therefore inconclusive. Unanswered is the question of who deters from voting, whether it is the richer or poorer population of a nation. But as GINI increases, the resources of a country are concentrated to a smaller portion of its citizens and thus the number of richer individuals is smaller. As a result, when the GINI-score is high, the population deterring from voting should be the ones with less income. There is a possible point of concern given that we only find significant results for the parliamentary models, namely that the presidential election may have a larger force of voter attraction in systems that employ voting for both parliament and president. This could lead to an overestimation of the effect on voter turnout. However, in theory this election should be as important to people as the elections for president, and a result showing lower turnout could still be a result of low access to information or political interest, possibly because of voter disenfranchisement and economic inequality. 7. Conclusions Our findings reflect mainly those of Solt (2008) in that inequality is negatively related to voter turnout. Since this study separates itself from the previous work in the sense that it includes countries previously excluded from inequality versus political participation, so the difference in results from earlier work is of interest. Since the Solt study is the most similar in terms of results to ours the difference here is interesting. We can conclude that the theory that seems to be most related to our findings is the relative power theory. We have in this study not made any regressions for different income

Economic Inequality & Voter Turnout 26 percentiles, where we measure the difference among different income groups for each country. Neither have we controlled for measures of free time and civic skills, other than education. Were such data available for our sample, we would have gladly ventured in that direction, to more clearly distinguish which of relative power and resource theory is more applicable. One point of criticism towards the paper, which to some extent can be applied to many other turnout-studies, is the aspect of a low explanatory capacity. This can be seen throughout all models of this paper where R2 is closer to zero rather than one. The support that we find in favour of Solts arguing and the relative power theory of lower turnout in more economically unequal societies is of importance and statistically significant, however the results are only partially explanatory to voter turnout and there are many more variables that affect turnout than those included in the models of this paper. But essentially this papers research question is not what affects voter turnout, it is rather how economic inequality pushes turnout in a certain direction of increasing or decreasing. Thus, the paper has the capacity of answering the research question even though its given low values for R2. The result of achieving a higher value for R in this study would consist of adding more variables to the models, by doing so the models would have a better explanatory power but would not benefit the capacity of answering our research question. For a final conclusion, this study has observed that for parliamentary elections, economic inequality has a negative effect of participation in democracies. 8. Topics for future research There is still much to be found in the future regarding the effect of inequality over turnout. First of all, the future looks promising in terms of providing trustworthy and accurate data on inequality, both with datasets such as the Luxemburg Income Study, and new measuring systems such as the Palma index. Future research is encouraged with the same research question and intuition as this but with a shifted focus of size and geography. Of interest would be similar studies with more detail but with smaller focus of specific continents or groups of nations, perhaps not with the focus on western societies to illuminate the difference. The wide focus of this paper allows for a more