Rethinking voter turnout as a collective action problem : the effects of social capital on voter turnout

0

Rethinking Voter Turnout as a Collective

Action Problem

The Effects of Social Capital on Voter Turnout

Submitted by: Kristine Marie Saksenvik Næss

Submitted to: Dr. Wouter van der Brug

Second reader: Dr. Armèn Hakhverdian

Date: July 2019

i

Abstract

This paper aims to reconcile instrumental voting theory with theories which explain voting behaviour through the lens of solving a collection action problem. It does so by studying the relationship between social capital and voter turnout in two ways: (1) voters whose utility function includes the utility function of more individuals gain a higher utility from electoral success than voters caring for fewer individuals, and (2) voters exposed to more community pressure to vote are more likely to gain utility from voting than individuals exposed to less community pressure. Therefore, an individual’s social capital should positively correlate to the probability that the individual participates in an election. Two separate sets of statistical models are constructed to gain a better understanding of the relationship between social capital and voter turnout in developed democracies on both individual and aggregate levels. The individual models are based on two World Value Surveys. The results show a positive relationship between organisational participation and the individual’s perceived belonging to a community on one side and voter turnout on the other side, in almost all the countries studied. Municipal data in the Netherlands between 2012 and 2017 is used for the aggregate model. The model finds evidence partially contradicting to the paper hypotheses, but also indications that social capital affects different types of elections in different ways. More research is needed to further verify the results.

ii

Content

Abstract ... i

Acknowledgements ... iii

1. Introduction ... 1

2. Defining Social Capital ... 4

3. Theory ... 6

4. Previous Scholarship ... 8

4.1 Altruistic Voting ... 8

4.2 Pressure to Vote ... 9

4.3 Levels of Electoral Participation ... 10

4.4 Other Theories ... 10

5. Methodology ... 12

5.1 Methodology and Justification ... 12

5.2 Model 1: Survey Data from Developed Democracies ... 12

5.3 Model 2: Intermunicipal Migration and Homeownership in the Netherlands ... 14

6. Results ... 19

6.1 Results Model-Set 1: World Value Survey ... 19

6.2 Results Model-Set 2: Homeownership and Internal Migration ... 21

7. Limitations... 24

7.1 General Assumptions and Limitations ... 24

7.2 Limitations for First Model-Set: World Value Survey ... 24

7.3 Limitations for Second Model-Set: Homeownership and Internal Migration ... 25

8. Discussion ... 27

9. Conclusion ... 31

References ... 34

Appendix ... 40

A.1 Model-Set 1: Plot means ... 40

A.2 Model-Set 1: Independent Country Models ... 42

A.3 Model-Set 1: Logistic Regression Models ... 46

B.1 Model-Set 2: Scatterplots for Main Independent Variables ... 48

iii

Acknowledgements

This work would not have been possible without the help and guidance of my current

supervisor Dr. van der Brug, and my former supervisor and instructor Dr. Hakhverdian. I am very grateful for the help and supervision these professors have given me. Their doors were always open for me when I needed direction.

I must also thank my professors at University of Amsterdam Dr. Berkhout and Dr. Medeiros for guiding me in the direction of the research topic through seminars on collective action and intragroup conflict, as well as Dr. Zicha from Leiden University College and Dr. Underhill from University of Amsterdam for developing my understanding of and passion for political economy.

I also wish would express my profound gratitude to my family for continuous support during my years of study and their ability to inspire me whenever I have been in the need of

guidance. I would also like to thank my partner Joes de Natris, MSc., for his support, help and encouragement during the writing process.

1

1. Introduction

The paradox of voting is one of the grand puzzles in rational choice theory: why do rational individuals vote when there is a cost and no assured benefit? The question has been studied from many perspectives, yet the puzzle remains to a large extent unsolved. This issue is becoming increasingly important, as political participation is decreasing in developed states (Mair 2013). From a traditional rational choice perspective, the cost of voting outweighs the benefits one may acquire from voting if the group of voters is large (Riker and Ordeshook 1968). This is because the ability of any given voter to influence a collective decision is very low (Downs 1957), and the cost of gaining information about an election and the act of voting itself tend to be high. This makes voting a collective action problem and we would expect citizens not to vote unless selective incentives such as third-party enforcement or benefits are present (see Olson 1965). This paper proposes a new theory of voter turnout as a collective action problem, where social capital is the main driver of selective incentives for electoral participation.

This leads to the research question: To what extent does social capital influence voter turnout

in developed electoral democracies? Here, developed electoral democracies refer to states

with non-fraudulent voting systems, where the political system is stable and reliable. Social capital is defined here as the positive (Putnam 1995) and negative externalities (Portes 1998) from social interactions in groups. An individual with more and stronger social connections has more social capital, and vice versa. This paper further makes a distinction between more social connections (quantitative social capital) and stronger social connections (qualitative social capital). Quantitative social capital is caused by larger networks of connections, whilst qualitative social capital is caused by smaller inter-group interactions. Higher levels of social capital have been shown to affect democracy and society positively through engaging civil society in politics (Putnam et al. 1994).

From this the paper hypothesise that individuals with stronger networks have more individuals to care about and more information about these individual’s preferences and therefore have a stronger incentive to vote, in addition to more individuals that can function as third-party enforcers incentivising voters to vote. This is because the utility gained from electoral success will be higher if the individual has more social relationships that would

2

benefit from such electoral success, than if the individual has less strong relationships. Moreover, the costs associated with voting may be lower amongst voters with strong communities because these voters have a higher access to community information, and the cost of not voting might be higher if voters are exposed to community pressure (Gerber, Green and Larimer 2008). The paper further questions whether the effects of social capital influence voter turnout differently in different types of elections, as national elections and local elections have differing policy domains that impact the individual and the community in different ways. The paper will answer the research question and the sub-question through testing two hypotheses in two cross-sectional model-sets. The first model-set operationalises social capital and voter turnout through World Value Survey answers. The second model-set measures social capital through intermunicipal migration and homeownership in Dutch municipalities.

Further understanding the paradox of voting is important because voter turnout is diminishing, which can affect the quality of democracy and undermine electoral outcomes (Cavanagh 1981). If the preference of the voter is different to that of the non-voter (Lutz and Marsh 2007), then changes in voter turnout matters for the electoral outcome. The preferences of individuals systematically not partaking in elections may not be revealed to parties, thereby limiting the implementation of non-voter’s policy preferences. This is may also be

problematic for the society in general because it may create segments that receives less political attention, that can increase other forms of inequality (Cavanagh 1981, p.63), and reduce overall economic growth and prosperity (Stiglitz 2015; Wilkinson and Pickett 2009) putting the quality of democratic institutions in peril (North, Wallis and Weingast 2009; Lijphart 1997).

The paper contributes to the scholarship on electoral participation through providing further understanding of why individuals vote in developed electoral democracies. Previous

scholarship has either taken an atomistic approach, where individuals were considered to operate in a social vacuum with little regard to enforcement of social norms (informal institutions) or attempted to understand voter turnout purely through normative social

dimensions, assuming little self-interest. This paper aims at further bridging the gap between such an instrumental voting approach and informal institutions, by expanding the notion of selective incentives to include social interactions. This is important as social connections (Putnam 1993a, Tocqueville 1835) and normative institutions (North 1990) can impact society overall and the formal institutions that maintain democracy. The paper also aims at

3

merging previously distinct explanations on voter turnout, such as that of community pressure to vote and that of voting as a form of altruism. The paper also brings value to academia through furthering the knowledge on how social capital can be measured.

The paper first explores previous definitions of social capital and defines the concept.

Thereafter, the paper puts forward the theory and reviews previous literature on the topic. The paper then presents the methodology, results and the analysis and discussion of the results. The paper ends with a conclusion answering the research question and finds evidence that partially supports the theory.

4

2. Defining Social Capital

The meaning of social capital must be clarified before its relation to voter turnout can be put forward. The word social capital initially referred to the potential economic profitability of social interactions (Field 2017). The concept was further developed and gained recognition through influential authors such as Bourdieu, Coleman and Putnam. Coleman (1988, p.98) approaches the concept broadly and defines social capital as the “structure of relations between actors and among actors”. Bourdieu (1986) linked social capital to the reproduction of inequality, as social capital can maintain and limit access to social networks. Further scholarship on the negative consequences of social capital includes conformity to group ideals (Portes 1998, p.17), violent sub-groups (Villalonga-Olives and Kawachi 2017), rent seeking (Fukuyama 2001) and other forms of patronage or clientelism (North, Wallis and Weingast 2009; Medina and Stokes 2007; Stokes et al). Nevertheless, the voting theories on patronage and clientelism assumes that social interaction is limited to within groups relationships, and as such limits the social capital of both included and excluded participants. In contrast, Putnam defines social capital primarily as positive externalities of aggregate group interactions (Putnam 1995); focusing primarily on social norm as one such externality. In my paper externalities of social capital is extended to include care for others, group pressure and access to information that may be considered unintended yet beneficial consequences of social interactions. Direct benefits of social capital include: lowered transaction costs (North 1990), knowledge of how a population of agents acts (Platteau 1994, p.802), administrative

efficiency (Putnam 1993b), as well as health, wellbeing (Helliwell and Putnam 2004) and child development (Putnam 2015). Based on this, the paper defines social capital as the positive and negative externalities of social interactions, where more social capital indicates more and stronger social connections with other individuals.

We should further distinguish between two forms of social capital. The first being formed in larger networks and the latter in groups or networks with more frequent internal interactions (Durkheim 1893). The effects of these two networks on voter turnout will likely differ, as one relies on interacting with a range of different individuals, whilst the other requires knowledge of a smaller group and network, and the ability to trust its members. This should not be confused with bonding and bridging social capital (see Putnam 2000). Bonding social capital is the act of socialising with one’s own group, whilst bridging social capital is when

5

Durkheim (1893) and distinguishes between frequent and non-frequent interactions. The paper defines frequent interactions with groups as qualitative social capital, and large-scale social networks as quantitative social capital. The prior relies to a greater extent on social norms, and the latter on the enforcement of formal rules and property rights.

One may also distinguish between material and immaterial social networks. Material

networks consist of exchanges of goods and services, whilst immaterial networks are caused by shared identities and informal relations (Botterman, Hooghe and Reeskens 2012). Material networks may be considered part of quantitative social networks, as they primarily rely on larger groups of people and depend on property rights. The distinctness of material social capital is important because this type of social capital can impact economic and democratic institutions directly through increasing and improving economic transactions that improves overall wellbeing (North, Wallis and Weingast 2009, Acemuglu and Robinson 2012). As such, the paper defines social capital as positive and negative externalities of the social interaction, and further divides social capital into a qualitative and quantitative type; where the former is the consequence of smaller group interactions and the latter is caused by larger networks, that may be economic in nature.

6

3. Theory

Previous studies have found that the probability of an individual’s vote affecting the electoral outcome is insignificant in larger groups (Gelman, Silver and Edlin 2012; Downs 1957). This means that an individual must instead perceive the vote to matter for its preferred policy outcome (or not lose utility from the act of voting). An individual is more likely to vote if it believes that the vote will bring the status quo closer to its preferences. This means that for the individual to vote, the individual must believe that the vote is counted, and the vote can affect the policy outcome in towards the individual’s preferred policy position. This includes both having trust in the political system serving the individual’s preferences (Inglehart 1997), and a belief that the individual’s vote will affect the electoral outcome. This is more common in developed democracies; however, voter turnout varies also within these states (Mair 2013). This paper’s first theory builds upon remarks made by Fowler and Kam (2007): individuals vote altruistically to benefit society overall, vote to benefit their social group, or vote to

benefit themselves. The development of the theory that belonging to a social group matters for electoral participation is limited. Assuming that the conditions for participating in an election are (perceived to be) met, my paper theorises that individuals with a stronger social network have more individuals to care about and therefore have a stronger incentive to vote, than if the individual did not care for others. This is because the possible utility gained from electoral success will be higher than if the individual had a less strong network. As such, the voting individual would be more likely to vote if it has close connections that it cares about, because the utility gained from helping these individuals through voting is higher than the utility gained from helping less close connections.

In a closer-knit community, individuals might have a greater ability to gain the necessary information to vote. An individual may have more knowledge about other people’s policy preferences and can utilise this knowledge to satisfy its own and/or its groups preferences. In contrast, individuals that do not live in close-knit communities do not have the same access to information, and the transaction costs to obtaining this information reduces the ability to satisfy other people’s or once own preferences. This brings the paper to theorise that voters with higher access to information as a result of community strength are more likely to partake in voting than individuals with less strong communities, because these individuals have better understanding of other’s policy preferences. In addition to this, individuals with more

7

community relations, are more likely to know whether group members vote or not. This contributes to solving the collective action problem of voting, as knowing that others will participate (and vote aligned with your preferences) increases the probability that the individual’s vote matters for the electoral outcome.

These theories do not exclude the effects of community pressure on turnout. The existence of community pressure, whether through community enforcement, or indirectly through

democratic norms may further impact the cost of voting. Building on Miller (1952) and Levitsky and Ziblatt (2018), this paper also theorises that: individuals exposed to community pressure to vote are more likely to vote than individuals not exposed to community pressure (Gerber, Green and Larimer 2008), because exposure to community pressure increases the cost of non-participation. Moreover, individuals exposed to community pressure might gain more utility from the act of voting, as other individuals in the community can provide positive feedback to the complying voter. This theory nonetheless assumes that individuals can

monitor one another. In larger networks where individuals have lower social capital, individuals are likely less able to monitor one another, and therefore have a lower ability to apply community pressure. Whilst the theory cannot explain why all individuals vote, it might hold merit when combined with the theory on altruism and information accessibility.

Moreover, the effect of participation in national elections may differ from participation in local elections. An individual with more social capital in the community is more likely to vote in local elections because partaking in community elections more directly affects other

individuals in the community, and such voters may share more information about local issues with one another. This does not exclude the possibility that the individual perceives national elections to greatly affect its social group and therefore votes in such elections as well. The decision to vote will also be impacted by other costs of voting. These costs include for example: the time spent gaining information about candidates and parties and their impact on the voter’s policy preferences, and the time and money spent getting to the voting station and casting the vote (Haspel et al. 2005).

Based on this, the paper hypothesises that:

H1: Social capital affects voter turnout positively.

H2: The effect of social capital on voter turnout is stronger in local elections than in national elections.

8

4. Previous Scholarship

Whilst much research is conducted on party choice in relation to social capital in developed democracies, little research is conducted on the connection between why citizens vote and social capital. The theory presented in the previous section is grounded in such previous literature.

4.1 Altruistic Voting

The theory is supported by previous scholarship on the subject. According to Edlin et al. (2007) Fowler (2006) and Fowler and Kam (2007), voters may vote altruistically. Edlin et al. (2007) finds that when caring for the general good, it becomes rational to vote in larger elections where the potential to affect the greater good is larger, thereby laying the theoretical foundation for this paper’s theory section. Edlin et al. assumes that voters only care about all people’s wellbeing, rather than caring about only one’s own or one’s group’s wellbeing (see Fowler and Kam 2007, p.813). Whilst it is possible that an individual can act altruistically, within the individual’s own understanding of what altruism entails, this is not necessarily the norm (Olson 1965). This paper therefore seeks to further the understanding of individual voters as acting partially altruistically and partially self-interestedly.

Additionally, Fowler and Kam (2007) do not consider that the electorate in most elections is too large for an individual voter to plausibly influence the election outcome and that the cost of participating in an election is high (Downs 1957; Riker and Ordeshook 1968; Edlin et al. 2007). This means that whether an individual participates in an election (or not) does not matter for the electoral success (i.e. utility does not increase with participation), and therefore participation in large elections cannot be attributed purely to individual gains. As such, it is probable that other factors, such as having knowledge about other group members voting behaviour and whether the group members participate affects whether an individual participates.

Nevertheless, the findings of such papers come with restrictions, as the studies conducted are limited to experimental tests played out as games in enclosed environments, where individuals do not already know each other (See Fowler and Kam 2007; Fowler 2006), thereby the

experimental setting bears little resemblance to the actual setting in which people vote. Some may argue that field experiments are considered very authoritative in terms of causal

9

(outside the experimental setting) if variables that could interact with the variable of interest are excluded or additional treatments that alter the individual’s mindsets through feeding them expectations are included. Such theories also assume that that the voter has the (perceived) information required to make judgements about the general good, and that parties are distinct enough that the voter believes that one provides more general good (see Katz and Mair 1995). This means that the theory on altruistic voting cannot be fully verified from previous

empirical testing. 4.2 Pressure to Vote

Community pressure to vote has been studied (Miller 1952; Gerber, Green and Larimer 2008). The theory on community pressure has been tested and corroborated in experimental settings. Examples include to publish individuals’ voting behaviour to family and neighbours (Gerber, Green and Larimer 2008). Whilst this finds that community pressure can be created, it does not find that community pressure develops naturally. Other scholars found that some

individuals gain utility from imposing social pressure on other group members (Schram and van Winden 1991), that voters gain utility from conforming to social pressure (Rosenstone and Hansen 1993), and that enforcement in the form of social isolation increases participation (Grossman and Helpman, 2001, p. 85). Nevertheless, as discussed in section 5, pressure to vote is not the only causal factor of participation, as it assumes that citizens have the capacity to monitor each other’s voting behaviour.

In contrast to the previous arguments, Blais (2000, p.104) finds that individuals expressing a duty to vote, rather than only feeling a group pressure, increases participation. Similarly, scholars have argued that democratic norms may affect voter turnout (Riker and Ordeshook 1968). Levitsky and Ziblatt (2018) argue that informal institutions can enforce democratic norms. This idea may be furthered to political participation through feeling a civic duty to vote (Campbell, Gurin, and Miller 1954, p.199; Riker and Ordeshook 1968), as democratic norms may foster intrinsic value to voting. However, such hypotheses attribute the causal mechanism to normative behaviour, rather than explaining what causes these democratic norms. This means that civic norms are likely an externality of other factors causing citizens to vote. Social capital can be one such causal factor, as more social capital can cause

community pressure that can increase the civil duty to vote. As such, one should not include the civic duty to vote as a control variable. The theory may, nonetheless, have some merit in underpinning that individuals may receive utility from the act of voting, or that individual may feel pressured into voting regardless of enforcement.

10 4.3 Levels of Electoral Participation

The effects on social capital on local electoral participation may differ from that of national participation. Differences between local and national electoral voter turnout has already been conducted and shows that elections that are perceived to be more important tends to be higher and that national election turnout tend to be higher than local election turnout (Hajnal and Lewis 2003). Nevertheless, the effects of social capital on different election types are less studied.

Caring about more individuals in local community may cause citizens to vote more in local elections. For example, citizens that recently moved are less likely to vote than citizens that have lived in the community longer (Rosenstone & Hansen, 1993; Squire et al. 1987 pp.157-159; Cassell and Hill 1981), due to costs of registration (Highton 2000) and the marginal loss of social capital over time with increased mobility (Glaeser, Laibson and Sacerdote 2002, p.F450). If a citizen moves, their social ties weakens, and the social ties of the citizens at the previous lactation weakens. Moreover, it takes time to establish new connections. If one follows the theory on voting to benefit one’s connections, then one would expect that newly moved individuals care less for participation in elections that does not affect their social group: thereby voting less in local elections.

Moreover, individuals with more social capital may have more information about local elections. Voters with more information about the candidates have a higher predisposition to vote because these voters are more likely to vote aligned with own preferences (Matsusaka 1995); whilst such a theory does not explain why people vote, it indicates which individuals are more likely to vote (Geys 2006, pp.24-25). This builds the grounds for the theory

presented previously, namely that voters with more access to information about local elections should be more likely to vote in local elections.

4.4 Other Theories

A range of other theories have also been presented on why citizens vote. For example, people tend to vote in groups. Organisational membership (Verba, Schlozman and Brady 1995), and interpersonal relationships, especially that of parents to children, matter for voting (Campbell et al. 1960), and people tend to vote with regards to who is in their social network

(Zuckerman 2005). This is even more common in ethnically homogenous groups (Förster 2018). Wilkinson and Pickett (2009) argued that poorer and excluded groups have less trust in the political system and causing decreased voter turnout. Additionally, youth tends to

11

It is thereby likely that voters in certain neighbourhoods or social groups have higher levels of voter turnout because they assimilate group behaviour. Whilst not explaining why voting matters, this is important to consider when studying groups of individuals, as an individual voter may not be independent from other group members. This also contributes to the theory that individuals with more knowledge of group member’s voting patterns matters for voter turnout, as such knowledge increases the probability of an individual’s vote mattering for electoral success.

Many other factors impact whether people vote or not vote. Acevedo and Krueger (2004) argued that people vote because they perceive their vote to matter. This theory is further backed by social research where when the perceived benefit from voting is increased, the rationality of voting is enhanced (Kanazawa 2000). Whilst the theory explains why voting may be regarded as a rational by the individual, it does not explain why some individuals choose not to vote. Individuals may also vote because parties or individuals buy their votes. However, such practices are less common in developed electoral democracies, as the cost of buying votes increases when the size of the electorate increases (Bueno de Mesquita et al. 2003).

Other variable shown to impact voter turnout is education level, income, age and gender. More educated citizens, citizens with higher income level, older citizens or citizens taking care of children and females are more likely to participate in elections (Leighley and Nagler 2014; Lahtinen, Mattila and Martikainen 2017, Larcinese 2007). These variables should as such be controlled for when estimating the causes of voter turnout.

12

5. Methodology

5.1 Methodology and Justification

This paper will conduct analysis of two different data sets. The first is based on individual survey responses from the World Value Survey. The latter uses municipal level data on internal migration and homeownership in the Netherlands. This approach utilises two of the most common methods for measuring social capital; individual surveys and aggregate proxies (Botterman, Hooghe and Reeskens 2012). Municipal aggregates allow us to understand the aggregates of many individuals and their collective behaviour; whilst losing individual data and intra-municipal differences. On the other hand, survey responses give a more detailed view of the individual and its immediate surroundings; whilst also being limited to the subjective and often flawed responses of individuals. As such, this study finds it beneficial to complement the survey responses with municipal level data based on actual human

interaction. If all results of the triangulation find similar results, the results will be further validified.

5.2 Model 1: Survey Data from Developed Democracies

Conceptualisation and operationalisation: Survey data from the World Value Survey

(2018) Wave 5 (2005-2009) and 6 (2010-2014) is used to measure social capital and voter turnout. Social capital is measured as whether individuals consider themselves to be a part of a community and whether they participate in organisations. The latter measurement is

commonly used as a measurement for social capital (Putnam 1995; Shiell, Hawe & Kavanagh 2018). This allow us to measure whether individuals that feel like they are a part of a

community and partake more in community building activities are more likely to vote.

Case/data selection: Data is limited to developed liberal democracies, because in such states

there is a significant probability that electoral participation causes the electoral outcome. This may not be the case if the voting system is fraudulent or the political system is unreliable or unstable. The paper limits these states to V-DEM (2018) states with a code higher than 0.7 on the liberal democracy index in the year the survey was taken. This dataset was chosen because it measures the quality of restraints placed upon governments; including civil liberties, rule of law, independent judiciary and checks and balances (World Value Survey Methodology, 2018); that allows for fair elections (North, Wallis and Weingast 2009). The models utilise the

13

World Value Survey. This data was chosen because it is a large-scale dataset with individuals from a range of different countries, asking questions closely related to social capital.

Variables: Data on electoral participation can be found in the 5th and 6th wave. There are three dependent variables utilised in three separate models: whether the individual votes in local elections and in national elections (yes, usually and never) from the 6th wave and whether the individual voted in recent parliamentary elections (yes or no) from the 5th wave. The main independent variables are: ‘I see myself as a member of my local community’ and membership of a group, membership of labour union and membership of political party. Local community feeling is coded as: 1=strongly disagree, 2=disagree, 3=agree, 4=strongly agree, and organisational participation is coded as the number of organisations an individual is member of (active or inactive). Labour union and party membership is dummied in separate variables. Organisational membership includes: Religious, recreational, arts and education, environmental, professional, charitable, consumer, self-help or other organisations.

Membership of self-help groups is not included in the parliamentary model. This variable is log transformed, however, is still right skewed as most individuals do not have organisational memberships.

The community feeling variable is included to measure whether people with larger levels of qualitative social capital vote more. Social capital is measured through the number of organisations an individual is a member of, and thereby measures primarily the quantity of social relationships, and whether these organisations are political in nature. Union

membership is important to distinguish from the others as union membership and union density is shown to increase voter turnout through increasing mobilisation (Leighley and Nagler 2007), and political participation is included for similar reasons.

Control variables include education level (1 to 10, where 10 is the highest educated), scale of income (1 to 10, where 10 is the richest), age and gender (0=male and 1=female).

Method: Three multivariate regression models are created to include the three types of

participation: national, local and parliamentary, where the two first variables are coded as 1 for never voting, 2 for usually voting and 3 for always voting, and the third is coded as 0 for non-participation and 1 for participation. Each of the three models have one election type as the dependent variable. All available datasets with electoral data were employed to further determine the generalisability of the results: ideally all models show similar results after

14

interpretation. This particularly applies to the national and parliamentary models, as these models both measure national level data. Missing answers were removed from the regression. The survey data is based on scales. When the categories on the scales are equidistant; treating the scale as interval data can be justified (Hair et al. 2017, p.26), however, the results may not be completely accurate. The paper applies ordered logistic regressions and a logistics

regression (appendix A.3), to further verify the results as these regressions do not rely on linear regression assumptions. An ordered logistics model is applied to answers from survey wave 6 as the model allows for more than two response variables. There are also large

differences between state turnouts (appendix A.1). Some of this variation can be explained by compulsory voting in certain states (Australia, Brazil, Chile (until 2011), Uruguay and

Cyprus). These variations are controlled for using country-fixed-effects (for all states names see appendix A.1). To determine whether the results have cross-country validity the

regression analysis was estimated also on individual states (appendix A.2).

The variables will be used to test hypothesis one through whether: (1) individuals that feel like they are a part of their community votes more than individuals that does not feel this. This tests whether individuals with more qualitative social capital are more likely to vote. The paper further tests whether (2) individuals that participate in groups vote more. Here, group participation refers primarily to non-political participation, but the paper expects a similar trend with these memberships as well. The hypotheses tests whether individuals with more group connections are more likely to vote.

The paper further tests hypothesis two through whether people with more social capital vote more in local elections than in national/parliamentary elections. The paper tests whether (3) higher levels of community feeling impacts local elections more than national elections, and (4) whether more group participation impacts local elections more than national elections. These tests do not consider whether these individuals were pressured into voting or not, nor can we determine whether people vote because they care more about the community.

5.3 Model 2: Intermunicipal Migration and Homeownership in the Netherlands

Conceptualisation and operationalisation: The paper applies the measure of

homeownership and furthers the theoretical implications of this measure. The paper also proposes a new measurement for social capital, namely that of intermunicipal migration, to test the relationship between social capital and voter turnout in the Netherlands. From this data one cannot determine whether social capital at an individual level affects voter turnout,

15

but whether municipalities with differing levels of social capital have higher levels of voter turnout. Previous aggregate measures of social capital includes wealth inequality measured through income, birth numbers, wealth benefits, unemployment, house price, higher education (Botterman, Hooghe and Reeskens 2012), crime figures (Sampson et al. 1997), within

community voluntary organisations (Putnam 2000; Shiell, Hawe & Kavanagh 2018), and homeownership (Manturuk, Lindblad and Quercia 2009; Dipasquale and Glaeser 1999), as well as trust and values. The effectiveness of the trust and values is disputed as these are generally based on psychological traits rather than social ties (Carpiano and Fitterer 2014). As previously discussed, internal migration; migration within a state; can be considered a measurement for social capital when individuals moving across electoral borders, because such individuals have reduced social capital. The moving individual loses previous

connections and the individuals left at departure lose the connections to this individual. Furthermore, the establishment of new connections takes time (Glaeser Laibson and Sacerdote 2002, p.F450). However, this measurement has certain drawbacks. Firstly, the measure considers some quantitative social capital, as membership of more groups can include less strong community connections. Secondly, factors not related to social capital might affect the propensity to vote for internal migrants. Individuals intending to move across political borders have a lower proclivity to vote, because people intending to move will not be affected by the election results pertaining to the previous political unit (Dowding, Hohn and Rubenson 2012). Additionally, there is an increased cost of voting caused by voter

registration and gaining information about politics (Rosenstone & Hansen 1993 p.157; Dowding, John and Rubenson 2012).

Little research is conducted on the link between homeownership as a proxy for social capital in a model with voter turnout. Homeownership contributes to more social capital as

homeowners invest more in social integration into the community. This is partially because homeowners have reduced mobility (Dipasquale and Glaeser 1999). Other studies have found that homeownership increases voter turnout (Manturuk, Lindblad and Quercia 2009), and that people having invested in their communities are less likely to not vote (Dowding, Hohn and Rubenson 2012). Nevertheless, also this measurement has drawbacks, as the home is in most cases homeowner’s largest asset (Fischel 2001) and the vote can influence its value through local service levels and taxation (Allers and Geertsema 2016; Fischel 2001). Despite this, some forms of social capital developed through community organisations or neighbourhood watches can increase the value of the home (Glaeser, Laibson and Sacerdote 2002, p. F452);

16

making the distinction between whether social capital or the home as an asset causes voter turnout less clear.

Moreover, the difference between local and national electoral participation in relation to social capital is considered: in the Netherlands the definition of local extends to a great extent beyond municipal elections, because provincial and water board elections also concern local issues, and are therefore included to increase the accuracy of the results. As previously mentioned, increased intermunicipal migration is expected to lower voter turnout at the local level, as moving citizens care less for, and have less information about other individuals in the municipality. Similarly, citizens with houses have an increased ability to develop care and information about individuals in the municipality. Additionally, distinguishing between the effects of social capital on different election levels can differentiate types of social capital. If homeowners vote more in local elections only, one cannot determine whether this is caused by the home as an asset or social capital. However, if homeowners vote more in both local and national elections or only national elections, then homeowners vote not only because of the home value. This is because mainly local policies affect the home value, and this is

determined by local elections. If the homeowner only cares for the home as an asset, voting in national elections should not correlate with homeownership. We may further hypothesise that citizens living in more densely populated communities, rely more on weaker social ties, and therefore are less likely to vote in all types of elections, especially that of local elections, as these individuals likely have less strong community ties.

Case/data selection: The Netherlands is chosen as the state of analysis because of its large

range of data availability at a municipal level. Data from Het Centraal Bureau voor de Statistiek (Statline 2019) and Verkiezingsuitslagen (2019) will be used. Other case studies could be added to the study. Voter turnout is the citizens in each given Dutch municipality that voted in a given election between 2012 and 2017 as a percentage of the total population eligible to vote in the given election. All available electoral data that also overlaps with the independent variables was included to test the hypothesis that social capital influences voter turnout. The hypothesis on the differing effects of social capital on national and local levels, is primarily concerned with second chamber and municipal elections, in addition to water board and provincial elections.

Variables: The main dependent variable includes voter turnout as a percentage of the

electorate in municipalities in the 2015 water board election, municipal elections between 2012 and 2017, the provincial election in 2015, the European Parliament election in 2014

17

including the Ukraine Referendum in 2016. These variables are also used as dummy variables to control for differences in electoral outcomes between election types. Note that all

municipal elections are coded as municipality elections and that EU and the referendum election are coded as European elections.

Two models are included with two different main independent variables. The first variable is the number of people moving to a municipality from another Dutch municipality, as a

percentage of the total population in the given municipality. Ideally the number of people moving away from a municipality should be included, however, it was left out due to high collinearity with the previous variable. Due to a right skew, this variable is log transformed. The second variable is the number of homeowners in a given Dutch municipality, as a percentage of the total population in the given municipality. Homeownership is calculated as the percentage of lived in homes out of the total number of rented and owned homes in a municipality.

Control variables includes: The percentage of each age group that lives in a given municipality. The age group 0-15 are combined into one variable to control for whether municipalities with more young people have different voter turnout. The age group 20-25 is log transformed due to a significant right skew. Ideally the adult age groups would start at 18, as this is the age citizens are eligible to vote. Income is measured as the average household income excluding student households in the given municipality per 1000, the variable is log transformed due to a right skew. Population density per km2 is included as a control measure of the quantity of social capital, this variable is also log transformed. This variable does not measure whether people in more densely populated areas form more closely-knit bounds, but it allows us to measure whether the potential to have more social interaction affect voter turnout. Note that an education variable is not included due to lacking data.

Method: Two models were created, one for homeownership and one for internal migration.

The models were not merged due to collinearity between the variables, as homeownership likely affects the ability of individuals to migrate (Dipasquale and Glaeser 1999). The number of municipalities per year varies. Therefore, each set of elections has been transformed to fit 2017 municipal borders, using the COELO (2019) transformation calculator. All other variables are also set to 2017 municipal borders. The dependent variable consists of all election outcomes, and data on election types is dummied to control for differences in voter turnout between elections. An alternative model could be to change what constitutes each election type to for example national and local elections or to election years. This method was

18

not selected as the prior model would leave out significant differences between different elections, and the latter model would not allow us to see trends as easily. The model selected overlaps to a great extent with actual election trends (see appendix B.1). Another possibility was to run separate models on each election type (appendix B.2). Moreover, water board is used as the base category and is therefore not included in the regression models. One age category is also removed to reduce multicollinearity between the age variables. The age group 25-45 was chosen. Because the number of years represented in the dataset are limited, a municipal and year fixed effects could not be applied; however, controlling for election years could have been a possibility. Voter turnout, homeownership and internal migration outliers were not removed. This is because these variables include important information that should not be excluded.

Interaction effects between election types and homeownership and election type and migration is included. If correlations between elections type and homeownership or immigration differs from one another, it means that certain election types have different effects on homeownership or migration. This is included to further understand whether homeownership affect voter turnout in local (and provincial) elections more than in other elections. All interaction effects were included in the main models, except for the interaction with water board as the water board dummy is the base category.

The paper tests whether: (1) municipalities with higher levels of homeownership have higher voter turnout, and further tests whether (2) homeownership has a stronger effect on local elections than national and European elections. The paper also tests whether: (3)

Municipalities with more citizens settling in the municipality have higher levels of voter turnout. The paper tests whether (4) the effect settling in the municipality will have a stronger effect on local voter turnout such as water board, municipal and possibly provincial elections, than on national (and European elections).

19

6. Results

6.1 Results Model-Set 1: World Value Survey

The VIF score results show very low levels of multicollinearity when dummy variables are excluded, similarly, the number of unanswered questions for each included country is low. However, the organisational variable is significantly right skewed, and the model show some heteroskedasticity that impacts the validity of the results negatively. Nevertheless, two ordered logistics models and one logistics model (see appendix A.3) further verify the results discussed below.

The results in table 1 show correlations that to a great extent align with the predictions. Note that national and local elections are measured on a scale from 1 to 3, and parliamentary elections on a scale from 0 to 1. However, model limitations, including the low model fit reduces the reliability of the results. The model shows that for every point increase in an individual’s community feeling, we can expect a 0.10 point increase in local electoral

participation a 0.09 point increase in national electoral participation, on a three point scale (or (0.10/2) 0.05 for local elections and (0.09/2) 0.05 for national elections on a two point scale), and 0.03 point increase in parliamentary elections at a two point scale at a 99.9% confidence level. The parliamentary results differ from that of local and national elections as the prior only measures the past parliamentary elections whilst the latter measures participation more generally.

Organisational participation shows similar results, for one percent increase in the number of organisational memberships, local electoral participation by (0.02/100) 0.0007 points, national participation by (0.01/100) 0.0006 points (or (0.0006/2) 0.00 for local elections and

(0.0007/2) 0.00 for national elections), and parliamentary elections by (0.02/100) 0.0002 on a two point scale. Furthermore, having a political party membership increases local

participation by 0.21 points and national participation by 0.24, and labour union membership increases participation by 0.05 ((0.05/2) 0.03 on a two-point scale) on a local level, and 0.03 on a parliamentary level. Increased age and education have a somewhat positive effect on voter turnout. All the results show no large difference when the model is run on individual countries, indicating that the results have cross-country validity within the sample of countries (see appendix A.2).

20

Table 1: Table Presenting the relationship between Social Capital and Voter Turnout

Significance code: ‘***’ significant on a 0.001 confidence level, ‘**’ significant on a 0.01 confidence level, ‘*’ significant on a 0.05 confidence level, ‘ ’ ‘ significant on a 0.1 confidence level.

(1) Local Elections 2010-2014 (2) National Elections 2010-2014 (3) Parliamentary Elections 2005-2009 Intercept 1.31*** 1.38*** 0.18*** Community Feeling 0.10*** 0.09*** 0.03*** Log(Organisational Membership) 0.07*** 0.06*** 0.02***

Labour Union Membership 0.05** 0.02 0.03***

Political Party Membership 0.21*** 0.24*** 0.02’

Income Scale 0.02*** 0.03*** 0.01***

Gender 0.01 0.02 -0.01

Age 0.01*** 0.01*** 0.01***

Highest Education Level 0.03*** 0.04*** 0.02***

Residual standard errors 0.64 0.62 0.39

Degrees of freedom 13534 13525 13339

Adjusted R-Squared 0.18 0.19 0.10

21

6.2 Results Model-Set 2: Homeownership and Internal Migration

For model 1 and model 2, the VIF score shows that there is significant multicollinearity between the electoral dummies and the interaction variables. Model 3 and 4, have much lower levels of multicollinearity. The high levels of multicollinearity in the first models can be partially explained by the correlation between the election type dummies and the interaction variables. There is also some multicollinearity caused by the correlation between

homeownership and income level, and some multicollinearity still persist between the age variables. The Durbin Watson test indicate some degree of autocorrelation (1.36) for the homeownership models and (1.36-1.37) for the internal migration models. The models do not have high levels of heteroscedasticity. Homeownership is somewhat left skewed, and internal migration is somewhat right skewed, thereby reducing the validity of the results.

The models including interaction effects in table 2 shows that a one percent increase in homeownership decreases voter turnout by 0.21 percentage points, but the internal migration coefficient is not significant. The models without interaction effects show similar results for homeownership. A one percent increase in homeownership decreases voter turnout by 2.24 percentage points, and that a one percent increase in internal migration increases voter turnout by (3.21/100) 0.03 percentage points. The results also show that voter turnout in national elections are significantly higher than voter turnout in provincial, and European elections, and that municipal electoral turnout is in between. Figures in Appendix B.1 furthers this belief, as well as indicates that water board elections have higher turnout than European elections. Together, the interaction effects do not have a significant effect on voter turnout, with the exception of second chamber elections on the internal migration model. However, individual level electoral models show that homeownership has a negative relationship on voter turnout in Second Chamber and European elections, and internal migration has a positive effect on all election types except municipal elections (See appendix B.2).

The variable on population density shows that a one percent increase in population density causes a (2.62/100) 0.03 percentage point increase in voter turnout. The age variables show a significant negative correlation between municipalities with more individuals in the age groups 0-15, 20-25 and 45-65, and a positive correlation between individuals in the age-group 15-20. This furthers previous research indicating that the older population tends to vote more. However, the findings also show that municipalities with more young individuals and

individuals with newly acquired voting rights vote more. As the latter often tend to vote less (Lahtinen, Mattila and Martikainen 2017), it could be that these influence others in the

22

community to vote. A one percent increase in labour participation also has a 1.47 percentage point increase in voter turnout in the first homeownership model, 1.48 in the second

homeownership model, and a 1.31 percentage point increase in the internal migration model. Furthermore, females are more likely to vote than men in both models. Moreover, the model fit is relatively high at 0.49 or 0.50 r-squared for all models.

23

Table 2: Table Presenting the relation between Social Capital and Voter Turnout in Dutch elections Homeownership (1) Homeownership (2) Internal Migration (3) Internal Migration (4) Intercept -28.71 -29.44 -30.23 -36.16 Homeownership -0.21** -0.24*** Log(Internal Migration) 1.47 3.21** Municipality dummy 4.56 4.41** -2.93 4.41** Provincial dummy -5.32 -4.03*** -22.51* -4.12***

Second Chamber dummy 29.25*** 23.76*** 1.60 23.80***

EU dummy -8.19 -18.69*** -21.49* -18.68*** Age: 0-15 -1.00* -1.00* -1.14** -1.15** Age: 15-20 4.20*** 4.19**** 3.57*** 3.58*** Log(Age:20-25) -18.77*** -18.6*** -16.85*** -16.91*** Age: 45-65 -1.23*** -1.22*** -1.43*** -1.44*** Age: 65-80 0.08 0.08 -0.18 -0.16 Age:80+ 0.17 0.16 0.23 0.21 Labour Participation 1.47*** 1.48*** 1.31*** 1.31*** Percent Female 1.58** 1.63** 2.07*** 2.05*** Log(Population Density) -2.71 -2.73*** -2.64*** -2.62*** Log(Income) -1.81*** -1.93 -6.96’ -6.47’ Homeown.:Municipal -0.00 2.02 Homeown.:Provincial 0.02 5.02’ Homeown.:Second Chamber -0.09 6.05* Homeown.:EU -0.17’ 0.77

Residual standard error 13.01 13.01 13.02 13.03

Degrees of freedom 2453 2457 2543 2467

Adjusted R-squared 0.49 0.50 0.49 0.49

Number of observations 2473 2473 2473 2473

Significance code: ‘***’ significant on a 0.0 confidence level, ‘**’ significant on a 0.01 confidence level, ‘*’ significant on a 0.05 confidence level, ‘’‘ significant on a 0.1 confidence level.

24

7. Limitations

Except for the variable on community feeling, the paper assumes that humans will develop social capital primarily because of their geographical location. Whilst this can be true, humans interact with one another differently, and their geographic location is as such not adequate to determine whether they will get along (Fafchamps 2004).

Moreover, the paper does not consider the effect of voter mobilisation. The decision to vote may be determined by a party’s ability to mobilise individuals (Verba, Schlozman and Brady 1995) through political advertising (Gosnell 1927; Gerber and Green 2000). This is

troublesome, as voter turnout may be higher in areas where people invest more in voter turnout. Voters located in the same geographical area or voters with similar preference may therefore be mobilised similarly, independent of social capital. Moreover, it would be expected that parties with more funding can mobilise more voters. As such it could be that poorer municipalities spend less money on voter mobilisation. Whilst this affects the survey model to a lesser degree, as this model accounts for individual level differences, neither model fully addresses this.

Reverse causation is another issue concerning all models. From the results one cannot discern whether social capital causes voter turnout, or whether voter turnout also affects social capital. For example, increased voter turnout could improve societal functions and democratic values that in turn increase social capital. Moreover, it is possible that the relationship is only two components of a casual chain. For example, increasing social capital could increase

democratic norms, which again could increase voter turnout.

All results are limited to a short time span. The survey models are limited to the given survey years, and municipal results are limited to seven years. Extending the longitude of the models would improve our understanding of how social capital changes over time. This could be important because social interaction and therefore social capital changes over time. 7.2 Limitations for First Model-Set: World Value Survey

The survey models have several limitations that affect the results. These models are based on survey results, and therefore the answers are error prone. Firstly, this is particularly the case with the questions on community feelings and income scale. The answers to both questions are highly subjective. The first question is limited to the individuals understanding of

25

community and feelings towards the community, and the latter question is limited to the individuals understanding of its own social standing within the state. Secondly, not everyone is willing to participate in a survey, if these individuals have something in common that could affect social capital or voter turnout this could affect the results.

In addition to this, there are issues with the regression. There is some multicollinearity, the adjusted r-square is very low, indicating a poor model fit, and possibly omitted variable bias. The results should therefore be used with caution. One missing variable could be that of population density.

7.3 Limitations for Second Model-Set: Homeownership and Internal Migration The data on homeownership and intermunicipal migration have limitations. If migrating to a new municipality affects voter turnout, then we may expect that the effects have a diminishing effect over time often lasting more than one year. Similarly, if homeownership affects voter turnout due to social capital, then homeowners that have lived in the home longer have likely built a stronger social network and become more likely to vote. This means that the effects of immigration and emigration from previous years affect current results, and that current homeownership is affected by previous homeownership. Moreover, it is not unlikely that renting a home for an extended period has similar effects on voter turnout if one could control for the home as an asset. The study cannot determine whether pre-data collection affects the results, as the data on homeownership and migration is limited to between 2012 and 2017, and the longitudinal evidence is limited. This means data availability is limited and tests for

autocorrelation on the given data could not be conducted. This also means that, intermunicipal constant differences could not be controlled for through fixed effects. Additionally, data is measured at an aggregate level, and differences at the neighbourhood level could affect the results.

Using internal migration has some drawbacks. Firstly, the variable on internal migration from municipalities could not be included in the data due to multicollinearity issues. This means that some variance in the data is probably not accounted for in the model. Secondly, migration to and from the Netherlands may affect election results. However, the effects of this is smaller than for migration to the Netherlands, as (non-EU) individuals moving to the Netherlands may not be eligible to vote. Lastly, the percentage of internal migrants in the Netherlands is relatively low. This means that the effects of migration on social relations could be quite low, and therefore might not affect social capital to the extent that it impacts voting decisions for

26

non-migrants. Other complicating factors include some multicollinearity. This is a lesser issue in model 2 and 4 (see table 2), as well as in the individual election models (see appendix A.2).

27

8. Discussion

The two model-sets give somewhat different results, that both align with and contradict the paper’s hypotheses. The survey model showed that both community feeling and community participation on a national, parliamentary and local level had a positive effect on voter turnout. In contrast, the municipal models showed a negative correlation between homeownership and overall voter turnout and a positive correlation between internal migration and overall voter turnout. If we consider all independent variables to be good predictors of social capital, then the models provide contradictory results.

Some of this difference can be explained by the distinctness between participation in different election types, as well as model limitations. The paper operationalised social capital through homeownership to measure whether community pressure or care for one’s group matters for voter turnout. The negative relationship between homeownership and voter turnout was unexpected and contradict the paper’s hypothesis that social capital matters for voter turnout and previous scholarship on the topic. Unlike Manturuk, Lindblad and Quercia (2009), the results show a negative relationship between homeownership and electoral participation. This can to some extent be explained by Manturuk, Lindblad and Quercia’s control for

neighbourhood disadvantage, where they find that citizens in more disadvantageous neighbourhoods vote more in local elections than renters, as the exit option of these individuals are limited due to lower asset value (see Hirschman 1970). Nevertheless, the individual election models (appendix B.2) indicates that homeownership is significant at the national and European level, despite also showing a negative correlation with voter turnout. This shows that social capital, rather than the home as an asset (as discussed in the

methodology) is the factor impacting voter turnout. This means that despite homeownership having a negative effect on voter turnout, we can deduce that it is also social capital and not only home as an asset that causes this relationship. Nevertheless, the validity of the results is reduced by model limitations, including lack of normal distribution of homeownership and multicollinearity. This means that the validity of the results could be compromised, and therefore the paper cannot determine whether social capital has a positive, negative or no effect on voter turnout.

In all but municipal elections, internal migration had a positive effect on voter turnout. This contradicts the prediction that internal migration should affect voter turnout, especially at the

28

local level, and contradicts previous literature on the subject. The results are surprising, as the paper predicted a negative relationship, as internal migrations should weaken social ties (Glaeser, Laibson and Sacerdote 2002) and decrease voter turnout through increasing the costs of voting and lowering the utility derived from voting. This also contradicts previous literature on the topic expecting geographical mobility to decrease voter turnout (Rosenstone and Hansen, 1993; Squire et al. 1987; Cassell and Hill 1981). However, a partial reason for why the effects of internal migration on voter turnout is not negative could be that some costs associated with voting, such as re-registration (see Squire et al. 1987 and Highton 2000) are not present in the Netherlands. Moreover, there might be economic differences between individuals that move across state lines, as geographical mobility often indicates a certain level of wealth (Gimpel and Schukneck 2001), a factor increasing propensity to vote. In addition to this, the models have significant limitations, as outlined in both the method and limitations section. Therefore, the paper cannot determine whether these results are caused by limitations of the internal migration model, or whether municipalities receiving more internal migrants vote more.

Nevertheless, if one considers the results valid, then the causes of increased voter turnout could be attributed to a larger flow of information as a result of migration. More internal migrants could also be indicative of migration causing more quantitative social capital, as more mobility increases the number of connections, or that more relocations to a municipality indicates that this municipality already has more social capital. Another possibility could be that migrants moving to municipalities with higher voter turnout might want to conform to social norms and thereby vote more. Whilst the paper theory does not cover this aspect of social capital, it is nonetheless interesting, as it can indicate that quantitative social capital can matter for voter turnout, as internal migrants may contribute to increasing the number of social connections. Nevertheless, the results also show that increased population density lowers voter turnout. This indicates that people living closer to one another affects social capital negatively, and thereby that the potential for gaining more qualitative social capital has a negative effect on voter turnout. Moreover, the findings further the theoretical argument that community pressure and care for community members could matter for voting, as more social density indicates larger groups that are more difficult to monitor.

The results from the survey models indicate that individuals with more community feeling or who are members of more organisations have higher participation levels. This means that qualitative social capital and to some extent quantitative social capital matters for voter

29

turnout. The latter is derived from the organisational variable, as the number of organisations an individual is a part of influences voter turnout positively. This aligns with the hypothesis that social capital matters for voter turnout. Nonetheless, the effects of organisational participation on voter turnout is relatively small, in contrast to previous studies that have found a significantly strong relationship between membership and political participation (Putnam 1993). This is partially due to organisational membership being measured as the number of organisations an individual is a member of rather than the type of membership, as well as measuring voter turnout rather than proxies for the quality of democracy. Furthermore, as community pressure to vote and care for the group is not directly measured, my paper cannot conclude that these are the direct causes of the positive relationship between the main variables. However, following the theoretical arguments based on Fowler and Kam (2007) and Miller (195), Gerber, Green and Larimer (2008) it is likely that community pressure and care for some group members matters for voter turnout. In addition, the results show that the effects of community feeling have a stronger correlation with voter turnout than

organisational participation. Previous scholars (see Putnam 1995) have operationalised social capital through organisational participation, but not studied whether organisational

participation causes social capital. Here, the results also show that community feeling, which can be considered a more concise measurement of (qualitative) social capital, also matters for social capital. Thereby, adding to the theoretical assumption that community produces social capital.

Despite this, the results are impacted by theoretical implications and model limitations mentioned in previous sections. One of the most significant limitations being the low model fit; likely caused by missing variables; furthering the belief that qualitative social capital is only a partial explanation of voter turnout. Nevertheless, the reliability of the results is further strengthened by that all states except for Chile and Uruguay showed similar results when statistically significant. Nevertheless, the correlation between organisational participation and voter turnout was weaker than between community feeling and voter turnout. Despite this, the organisational variable is significantly right skewed after log transformation, and the results may be compromised. To further verify these results, a model measuring whether individuals participating in organisations or not could be added. This would also allow for measuring whether individuals with more community involvement, regardless of quantity, vote more. The relationship between voter turnout and labour union participation is strong. This furthers the prediction that labour union participation matters for voter and supports previous

30

scholarship (Leighley and Nagler 2007). Moreover, party membership at a local and national level has a strong positive effect on participation, implying that party membership affects participation positively. Whilst it is difficult to determine whether this is purely due to members of these organisations having a stronger interest in politics, it is not unlikely that relationships established in these organisations could also matter for voter turnout.

The paper also hypothesised that social capital affects participation in local elections more than national elections. However, the differences between local, national and parliamentary elections were not as strong as expected. The results show some difference between the effects of organisational participation on the local and national level; however, the difference may not be significant. The effects on parliamentary elections were less strong than both local and national elections (after converting the scale). However, these models are measured slightly differently, and as such, the comparison between them is flawed. The municipal model-sets showed no statistical significance for the impact of social capital on local electoral participation but showed that the effects of social capital in national and EU elections often has a negative effect on voter turnout. Despite not being significant, the models showed similar trends for other election types. As such, the paper could not reject the hypothesis that the effect of social capital on different election types matters. However, the paper

recommends further analysis of the issue as little research has been conducted on the matter and the results cannot be compared to other studies.

From the results, one can derive that social capital measured as individual community feelings affects voter turnout positively at a local, national and parliamentary level, and that

organisational participation has similar effects. Similarly, the results show that the effect of community feeling, and organisational participation are slightly stronger, but not significantly stronger, at the local level than at the national level when comparing results within survey wave 6. This means that the paper should not reject hypothesis one or two. However, due to the low model fit, more research should be conducted. The second model-set indicates that homeownership decreases overall voter turnout at a municipal level in the Netherlands. The model cannot determine whether the results are caused only by social capital or also by the home as an asset. This means that prediction hypothesis one and two should be rejected if a decision was based purely on these results. However, due to the model limitations especially that of missing variable bias and multicollinearity, more research needs to be conducted on this area as well.