Polarization of political elites : a cause or a consequence of mass polarization?

Polarization of political elites: a

cause or a consequence of mass

polarization?

Student: Daniël de Rijke

Student number: 10186859

Specialisation: Political economy

Professor: A. Hakhverdian

Second reader: E. Harteveld

Assignment: Master thesis political science

Word count: 9040

Date: June 2019

Abstract - Using European elections and data from the European Elec-tion Study since 1999, the process of ideological polarizaElec-tion is clearly visible in Europe on the level of political elites and the public. Polariza-tion of attitudes toward European integraPolariza-tion is also visible at the mass level, but not at the level of political parties. Political parties remain very homogenous in their attitudes toward European integration. The research shows that ideological elite polarization is mainly an effect from mass level polarization, although leadership effects have also been found. In bigger party systems, political parties respond stronger to changes in the electorate in comparison to smaller party systems. Political elites have proven non-responsive to polarizing attitudes on European integration, suggesting that the electoral connection between the public and the elites has been lost in this policy field.

Key words: elite polarization, mass polarization, congruence, causality, ideology, European integration

Table of contents

Part I: Introduction 3

Part II: Theoretical framework and hypotheses 5

Part III: Methods 12

Part IV: Data collection and measurement 16

Part V: Results 21

Part VI: Conclusion and discussion 32

Part I: Introduction

Tom-Jan Meeus (2019), an established political journalist in the Netherlands, wrote in NRC that we are living in the times of the polarization scam. Nowadays, politicians desire polarization above empathy or compromise to maximize their voteshare. A si-milar observation is made by Andy Becket in The Guardian (2018) about British poli-tics: "Politics is raw in Britain. Remainers against Brexiteers and vice versa. Pensio-ners are set against millennials; nationalists against immigrants; populists against eli-tes; rural traditionalists against city liberals. Party politics is characterised by con-tempt and dogma." On the other side of the Atlantic Ocean the Financial Times (2018) wrote that moderate politicians have lost the latest elections in the United Sta-tes (US) and that the present Congress is "the most ideologically polarized in US his-tory".

It makes sense that journalists are using the term 'polarization' to describe po-litics, considering that politicians themselves are also using the term. The Dutch pri-me minister, Mark Rutte, said that the Dutch people should focus more on the simila-rities than their differences to put a check on polarization (NOS, 2019). Merkel (2018) went even further when she warned that the current world order is under threat from, among other factors, polarization.

Of course, when politicians and journalists start talking about polarization, political scientists will follow. In political science the number of publications featuring the term

'polarization' in the title has been rising steadily, as can be seen from figure 1. Note that the number of publications featuring polarization in the title is low for this year, but at the moment of writing it is only June 2019.

The academic debate concerning polarization is far from settled. Are the elites polarized? Are the people? And what causes polarization? Most of these questions are still heavily debated, or as Lelkes (2016: 392) so elegantly puts it: "The debate on mass polarization is itself polarized".

The aim of this research is to contribute to these debates by examining whether pola-rization is a top-down or a bottom-up proces. If we observe polapola-rization at the elite level, is that a consequence of mass level polarization? Or when elites are polarized, do they cause polarization at the public level? Or simpler, as Steenbergen et al. (2007) put it in their research: "who is cueing whom?". This will be investigated for ideological polarization and for polarization of attitudes toward European integration.

Part II: Theoretical framework and hypotheses

When researching polarization, it is extremely important to define what type of polari-zation you're investigating. This research is about the link between mass and elite polarization. Before going any further, the different types of polarization have to be conceptualized:

1. Ideological mass polarization: the divergences of policy preferences among the

public. In the literature most of the research is focussed on the US, where one side observes a culture war (Abramowitz & Saunders, 2005; Abramowitz & Saun-ders, 2008; Abramowitz, 2010) and the other side claims it is not soo bad (Fiorina et al., 2008; Dimaggio et al., 1996). There is not a lot of literature focussing on mass polarization in Europe. From Iversen and Soskice (2015: 1803) some infor-mation about mass levels of polarization can be distilled. Within their research they find a lot of variation of polarization, meaning that there are more and less ideologically polarized countries. Norway, the Netherlands, Sweden, Denmark, France and Germany are among the more polarized countries in Europe accor-ding to their data. Counter intuitively, the United Kingdom is the least polarized country after the republic of Ireland and Greece (Iversen & Soskice, 2015: 1782).

2. Elite polarization: the process where political elites become ideologically divided

groups without any overlap. A process which is, once again, visible in the US (He-therington, 2001; Layman & Carsey, 2002; McCarthy et al., 2006), since it is stu-died a lot over there. Elite polarization in Europe is stustu-died in Dalton's (2008) arti-cle, where he developed an index to measure elite polarization. When developing this index, Dalton (2008: 907) calculated the elite levels of polarization for a num-ber of European countries at two moments in time. According to this research, for eight of the twelve European countries of which he collected sufficient data, the elite polarized. But since his goal was to develop an index for measuring elite po-larization, the results are not extensive enough to say something substantive about the levels of elite polarization in Europe. Nevertheless, his results suggest the process of elite polarization is ongoing.

3. Affective polarization: this form of polarization is about the feelings of one group in

society versus another. In versus out group identities are the main driver between this form of polarization. Affective polarization is probably best visible in the US, where democrats and republicans increasingly dislike one another (Ivengar et al., 2019: 4). For example, the number of democrats and republicans who would be disappointed when their child were to marry someone from the opposite party has grown significantly (Klar et al., 2018: 384), this can be interpreted as growing in-terparty hostility. This is even more visible from Mason (2015: 141), where people in the USA dislike people from the other party, even if they agree on the policy subject. Affective polarization is not going to be part of this research, but affective polarization is connected to elite and ideological mass polarization, where those forms of polarization could cause affective polarization (Rogowski & Sutherland, 2016: 488). Rising levels of mass and elite polarization can therefore result in ri-sing levels of affective polarization.

This research will focus on the connection between ideological mass polarization and elite polarization in Europe. The contribution to the academic literature will therefore be twofold, providing insight in the levels of polarization in Europe and a better un-derstanding of the connection between mass and elite polarization.

Congruence

Mass and elite levels of polarization should be linked. In a proper functioning demo-cracy, the preferences of the majority and the political elite should be aligned (Dahl, 1998; May, 1974). Congruence is the alignment of those preferences. If the ideologi-cal distance between the elite and the public is smaller, the system is more congruent (Golder & Stramski, 2010: 91). Interestingly, the literature on congruence shows a clear link between public and elite preferences (Golder & Stramski, 2010; Powell, 2013; Hakhverdian & Schele, 2017), but doesn't provide a clear answer to the ques-tion which came first.

There are roughly two models which can explain congruence of the public and the elite preferences. The responsiveness model, where the elite responds to

chan-ges in the public, and the democratic leadership model, where the public responds to changes in the elite.

Causal mechanism 1: the responsiveness mechanism hypothesis

The responsiveness mechanism, explaining why elite preferences follow mass prefe-rences, consists of two more mechanisms. The first is referred to as the electoral tur-nover mechanism. In a democracy, voters get to vote public officials into office. As-suming that voters are informed properly about the policy preferences of politicians, they will vote for officials who will serve their interests best. Therefore, the preferen-ces of the politicians in office will match the preferenpreferen-ces of the electorate (Stimson et al., 1995: 545). Hibbs (1977: 1468) proved this for macro economic policies in the US and Great-Britain. They are, for the most part, in line with the core constituents of the governing party. With that he showed that people choose representatives that will serve their interests best and therefore that government can be responsive to the pe-ople. This mechanism can also work for polarization. When people's preferences di-verge toward the extremes of the political spectrum, they will be more likely to vote for parties who articulate policy alternatives in line with extreme ideological positions.

The second democratic responsiveness mechanism is called rational anticipa-tion. The main assumption of this mechanism is that politicians are rational actors who try to maximize their share of the votes. While in office, politicians will try to adapt their policy according to changes in the electorate. When their policy is aligned with the policy positions of the electorate, that will enhance their chances of re-electi-on (Stimsre-electi-on et al., 1995: 545-546). This mechanism has proven to work when the changes in the electorate are great enough that it will decrease the changes of re-election. This is confirmed by Hobolt and Klemmensen (2008: 332) who found that this mechanism can explain political behavior of elites in contested elections.

In conclusion, the first hypothesis of this research is: elite level polarization is a response to mass level polarization.

Causal mechanism 2: democratic leadership

Opposed to responsiveness mechanism is the democratic leadership mechanism. Peoples' policy preferences are not completely independent of the opinions of

politi-cians. Evaluations are based on an interplay between the person and the outside world (Druckman & Lupia, 2000: 4). Preferences evolve over time when new informa-tion is encountered, but the formainforma-tion of new preferences is highly dependent on previously held beliefs (Druckman & Lupia, 2000: 6). Politicians are a source of which people can receive new information. Therefore, politicians are capable of influencing peoples' preferences. Since politicians are just one of the sources and preference formation is influenced by a lot of different factors, they can only alter attitudes to a certain degree.

A causal effect of elite opinions on voter opinions has been found on the issue of European integration (Gabel & Scheve, 2007: 1027). The same effect was found for ideological positioning, although this effect was conditional. Popular governments in Great-Britain were able to converge the ideological position of the masses toward their own (Hakhverdian, 2012: 1401). This mechanism could also work for polarizati-on. If political parties take increasingly polarizing positions on a range of policy prefe-rences, people who identify with that party will, generally speaking, follow (Layman & Carsey, 2002: 799-800). In sum, political elites are able to lead the public's preferen-ces in a certain direction. This can happen on specific issues, but also on ideological positioning.

In conclusion, the second hypothesis is: mass level polarization is a response to elite level polarization.

Different contexts: polarization on attitudes toward European integration

The different causal mechanisms may hold different explanatory strength in different situations. There is no clear reason to assume that for ideological positioning and po-licy preferences the causal relationship works exactly the same. European integration may be a policy issue where the public is influenced stronger by elites, since EU in-tegration is always viewed upon as an elite driven project (Gabel & Scheve, 2007: 1016). But while Gabel and Scheve (2007: 1027) do find that elite polarization on EU integration does have an effect on the preferences of individuals on EU integration, they don't test the model the other way around: what is the effect of individuals' prefe-rences on elite polarization? Steenbergen et al. (2007: 29) found that the nature of the relation between elite and public preferences toward European integration is

re-ciprocal: the public influences the elites, but at the same time the elites also influence the public. But, this doesn't settle the academic debate. According to Toshkov's (2011:185-186) research the connection between EU policy and the electorate has been lost. Until the middle of the nineties, EU policy seemed to respond to changes in the public's preferences, but that relationship has disappeared since then.

Because the relationship between elite and mass preferences toward Eu-ropean integration is still debated and the causal mechanism might work different for ideological polarization and for polarization of policy preferences, both causal me-chanisms will also be tested for polarization of attitudes toward European integration. While the evidence on ideological congruence from the academic literature (Huber & Powell, 1994; Golder & Stramski, 2010; Powell, 2013; Hakhverdian & Schele, 2017) is unambiguous, the evidence for congruence on preferences toward European inte-gration (Gabel & Scheve, 2007; Steenbergen et al., 2007; Toshkov's, 2011) is mixed. Therefore, the third hypothesis is: the causal mechanism behind congruence works better for ideologic polarization than for polarization of preferences toward European integration.

Different contexts: number of political parties

The number of parties can influence the relation between mass and elite preferen-ces. The number of parties can influence the rational behavior of the political elites. Downs (1957: 143) predicted that political parties in a two party system would con-verge, while parties in a multi party system would be polarized. The reasoning behind this prediction is that when two parties compete for votes in a one dimensional spa-ce, they have the highest potential vote share when they converge toward the centre. Both parties will receive all the votes on their respective sides of the spectrum, while the contested votes lie in between the two parties. The parties will converge ideologi-cally to gain the most of those centre votes (Downs, 1957: 142).

In a multiparty system, this mechanism works opposite to the two party logic. In a multiparty system, voters have more widespread options. Because of this wider range of options for voters, it is more likely that the vote will be linked stronger to the ideological position, making the political parties more polarized (Downs, 1957: 144). But that does not mean that there are no dualist tendencies in multiparty systems.

According to Sartori (1976: 131), the turning point is between five and six parties. Party systems with five or less relevant parties will show dualist tendencies and can therefore be classified as moderate pluralism, while systems with six or more parties are classified as extreme pluralism (Sartori, 1976: 127). According to this model, poli-tical elites will become more polarized in systems with six or more parties.

To test whether the number of parties influence the causal mechanism, the fourth hypothesis is: in democratic systems with more parties, the causal mechanism is better at predicting levels of polarization. This will only be tested for the causal me-chanism that proved to be the best working meme-chanism of the two.

Different contexts: proportionality

The second institutional setting that can influence the levels of polarization is propor-tionality. According to Ezrow (2007: 191), systems with more proportional electoral laws are less responsive to changes in the voter distribution. Meaning that dispropor-tionality leads to stronger responses of political parties. One of the potential reasons behind this behavior is that parties in more proportional systems are not just vote seeking parties, but also policy seeking. Responding just to changes in voter distribu-tion may make parties too different from each other. This would make coalidistribu-tion buil-ding harder and therefore the parties chances of entering a coalition would decrease. On the one hand, in disproportional systems, vote seeking behavior would be of stronger influence in party behavior, resulting in stronger responses to changes in the voter distribution (Ezrow, 2007: 184-185). On the other hand, Golder and Stramski (2010: 104) found something completely different. They showed that ideological con-gruence between the public and their representatives is higher in democratic systems with higher levels of proportional representation, suggesting that parties in larger par-ty systems are also mainly vote seeking.

Since there seems to be a relation between proportionality and congruence, proportionality has to be taken into account when doing research into the causal me-chanisms behind congruence. A measure for proportionality will therefore be added as a control variable.

Normative implications

There are different important normative implications for both causal mechanisms. When elite level polarization is a response to mass level polarization, political parties can be seen as channels of expression for the public. Parties react to changes in the public and therefore behave exactly how political parties ought to behave in a demo-cracy (Sartori, 1979: 24).

The normative implications for the democratic leadership model aren't as posi-tive in the context of polarization. These negaposi-tive normaposi-tive implications are based on the assumption that division in society is not desirable. If elite polarization is a cause of mass polarization, political parties actively divide the electorate. Political parties who divide for electoral gain are what Sartori (1979: 11) called factions. Alt-hough polarized political elites provide clearer cues for the electorate, making it ea-sier for voters to distinguish between parties (Levendusky, 2010: 124), the quality of public preferences decreases (Druckman et al., 2013: 74). More specifically, voters are more convinced of their own opinions and become more intolerant to alternative arguments (Druckman et al., 2013: 75). This leads to a less informed public, which again is harmful to the democratic proces. Voters should learn about relevant policy alternatives before casting their votes (Dahl, 1998: 37). Polarization as a strategy of political parties does not only lead to a more divided public, but also to a less tolerant and less informed public. In turn, this can lead to higher levels of affective polarizati-on, where different groups in society increasingly dislike, or even become hostile to, each other (Iyengar & Westwood, 2015: 705). So when elite polarization is not cau-sed by mass polarization, but a strategy from political parties, they really act as a di-visive force in society.

Part III: Methods

To test whether changes in the public precede changes in the elite or vice versa, pa-nel data is needed, which includes mass and elite polarization indicators on multiple points in time. The European Election Study (EES) provides data on respondents' left-right position, respondents' position on European integration, party positions on the left-right scale and party positions on European integration. The survey is carried out in all EU member states and the questions are identical. In 1999 questions on party and respondents' positions on EU integration were added. Therefore, since 1999 information about every EU member state on respondents' and parties' ideolo-gical positions and positions toward EU integration can be extracted.

The survey is carried with an interval of five years, every time when European elections are held. The 2019 survey is not ready yet, so the studies from 1999, 2004, 2009 and 2014 are used. At four different points in time, data can be extracted on the levels of mass and elite polarization for every EU member state. For the years, 1999, 2004, 2009 and 2014 data is collected for all participating countries, creating a data-set with an N of 87.

The challenge of this research is testing if elite polarization happens before or after mass polarization. Because there is data available at four different points in time, this is now possible.

The compiled dataset consists of panel data, containing polarization indicators for every country at four points in time. In panel data, the country specific data is not in-dependent over time, therefore the standard errors are clustered by country.

Testing for causality

With a time series regression it is possible to investigate the temporal ordening of va-riables. The goal is to prove a causal relation, but with a time series regression you can still only prove a correlation. A correlation, as is widely known, does not imply causation. But, by investigating variables over time one condition of causality can be proven.

The condition that is being investigated in this research is the following: when phenomenon A causes phenomenon B, it has to occur before phenomenon B in time. The time lag can be tiny, but there has to be some time between the two phenomena if there is a causal relationship (Granger, 1980: 330; Sims, 1972: 541). Therefore, a time series analysis "makes it possible to investigate causal connections by compa-ring the temporal ordecompa-ring of variables, whether a change in public opinion occurs after (leadership) or before (responsiveness) a change in policy (Jacobs & Shapiro, 1994: 13)".

For a regression model this means your dependent variable on t=0 is compa-red to your independent variables on t=-1. This way you can find out if A happened on t=-1, B also happens on t=0, suggesting a causal relationship.

When comparing present cases to past cases, the overall number of observa-tions does drop. This is because the first observaobserva-tions in time can't be compared to earlier observations. Therefore, with data on four elections, there are only three time comparisons possible. The first is between 1999 and 2004, the second between 2004 for and 2009 and the third between 2009 and 2014. Although I gathered data on 87 elections, the N consequently drops to 60 when doing a time series regression. When researching polarization on EU integration, the N drops even further, down to 36. This is because there was less data available on party positions toward European integration. The most important reason being, that, in the 2014 EES, only the res-pondents in France were asked to score the political parties on their positions toward European integration.

The responsiveness and democratic leadership model

To test the responsiveness and the democratic leadership hypotheses, regression models are built. Both for the European integration as the ideology context these models are created. For the responsiveness model, elite polarization is the dent variable. For the democratic leadership model, mass polarization is the depen-dent variable.

The independent variables in both models are exactly the same. There will be a time lag for all variables, because the focus of this research is on causality. There-fore, most of the independent variables are lagged, measured at t-1. This goes for

the mass and elite polarization variables and the economic control variables. The responsiveness model, with elite polarization at t=0 as a dependent variable, will also include elite polarization at t-1 as an independent variable, since preferences are highly dependent on previously held believes (Druckman & Lupia, 2000: 4). The same goes for the democratic leadership models, where mass polarization at t-1 is also an independent variable.

The proportionality variable is not lagged. When testing if the elite or the mass responds stronger or weaker in more or less proportional systems, the focus is on the immediate effect. Proportionality isn't investigated as a cause of polarization, but as a control variable to correct for omitted variable bias.

When four different models are created, both models for the EU context and the ideo-logical context, most of the hypotheses and the main research question can be ans-wered. When the beta coefficients in the responsiveness models are significant and above zero, the responsiveness hypothesis can be accepted. When the beta coeffi-cients of the democratic leadership models are higher than zero and significant, the democratic leadership hypothesis can also be accepted. With this information, the third hypothesis, if the causal mechanisms work different in the context of EU integra-tion and ideology can also be rejected or accepted.

To answer the research question, is polarization a top-down or a bottom-up proces, the beta coefficients and the R-squared values of both models are compared. The R-squared value will provide information about the model itself, to what extend the models are able to explain variation of levels of polarization. The beta coefficients provide information of the strength of the relationship, to what extend the elite or mass polarization values are influences by the predictors.

Number of parties

To test if the number of effective parties influence the causal mechanism, a variable for effective number of parties can't just be added to the regression model. The main reason is because one more party does not necessarily make the elite and the mass more congruent. There is supposed to be a turning point somewhere. According to Sartori (1976: 131) the turning point is between 5 and 6 parties. But this seems a bit

arbitrary. So, to test whether bigger or smaller party systems influence the causal mechanism, I take a slightly different approach. The line which would function as the turning point in this research is the median value of the effective number of parties. The best working regression model, responsiveness or democratic leadership, will now be tested again. Once for party systems with less than the median number of parties and once for party systems with more than the median number of parties. The benefit of this approach is that the number of observations for both models is the same. The disadvantage is that the turning point is still somewhat arbitrary. This ap-proach will therefore not enable me to find the turning point, but it will be clear if party size matters.


Part IV: Data collection and measurement

In the EES survey, respondents are asked four questions which are relevant for this research and necessary to measure levels of polarization. Respondents are asked to place themselves on the left-right scale, where 1 is extremely leftwing and 10 is ex-tremely rightwing. They are asked if they think, on a scale from one to ten, if Eu-ropean integration has gone too far or should go further. On this scale position 1 is European integration has gone too far, while position 10 is European integration should go further. Since the survey is repeated, this data is now available for 4 points in time. The changes of the ideological positions and attitudes toward European of respondents lie at the basis of measuring polarization on the public level.

In the same survey respondents are asked to place the political parties on tho-se scales too. The respondents are asked to do that for up to eight political parties. The aggregate data, the mean value of all answers, of these question provide 'objec-tive' party positions for the competing political parties (Dalton, 2008: 904). Therefore, data can be extracted from the EES on party positions on European integration and on the ideological positions of up to eight political parties. The shifts in ideological party positions and party positions toward European integration form the basis of operationalizing elite polarization.

Knowing ideological positions and policy preferences doesn't provide a mea-sure for polarization just yet. When operationalizing polarization, the first problem you come across is captured well by Dimaggio et al. (1996: 693): "Polarization is both a state and a proces. Polarization as a state refers to the extent to which opinions on an issue are opposed in relation to some theoretical maximum. Polarization as a pro-cess refers to the increase in such opposition over time". Polarization as a state would be the upper situation in figure 2, polarization as a process would be the chan-ge from the lower situation to the upper situation in figure 2 (Fiorina & Abrams, 2008: 566).

Another way to operationalize polarization is measuring ideological consisten-cy, the preferred method of Abramowitz and Saunders (2008), but this method for measuring polarization has been criticized widely for exaggerating the levels

polari-zation (Lelkes, 2016: 394). Ideological consistency means the degree to which ideo-logical preferences become internally consistent. When people with different ideolo-gical positions are ideoloideolo-gical consistent, there is less overlap between the groups with different ideological positions. This lack of overlap between people can be inter-preted as a sign of polarization. The consistency measurement for polarization shows polarization in the USA, but this is also highly dependable on American partisanship: democrats becoming more consistently liberal and republicans becoming more con-sistently conservative (Baldassarri & Gelman, 2008: 439-440).

Since I am not investigating whether party members in Europe have become more polarized, but investigating the public in a broader sense, I use the divergence method for measuring polarization.

Measuring divergence

Divergence, as an indicator for polarization, can be measured with the standard de-viation (Ezrow et al., 2013; Lelkes, 2016). Measuring the standard dede-viation within the public is relatively easy, by calculating the standard deviation for the left-right po-sition of the respondents and for the popo-sition toward European integration. Calcula-ting the standard deviation of the left-right positions and positions toward European integration for political parties is harder. Dalton (2008: 906) developed a polarization index to measure the level of polarization of political parties. This means, when tal-king about elite polarization in this research, this is operationalized as the polarization of political parties. The math behind Dalton's polarization index is similar to the math behind a standard deviation. Dalton (2008: 906) developed the following formula:

To calculate Dalton's (2008: 906) polarization index, data from different sources has to be combined. From the EES data is derived about the 'objective' party positions. To give relative strength to those party positions, the party positions need to be weig-hed according to their share of the votes. Data about election results are extracted form the Parliaments and governments database (website parlgov.org, 2019).

Measuring the number of political parties

When measuring the number of parties, it makes no sense to count all the competing parties. Small and irrelevant parties should not be part of the analysis, when investi-gation the number of parties. Therefore, parties should be weighed according to their relative size. To calculate the number of effective parties, Laakso & Taagepera (1979: 3) developed the following formula:

In the dataset provided bij the Parliaments and government database (website parlg-ov.com, 2019), the effective number of parties is already calculated according to the formula provided by Laakso and Taagepera (1979). Data on the number of effective parties is therefore extracted from this database.

Disproportionality

A variable for proportionality is added as a control variable. At first sight, differences in proportionality might not be expected in various European elections, but there are several reasons for the European parliamentary elections to be more or less propor-tional. First of all, the distribution of seats between countries is not proporpropor-tional. Seats in the European parliament are divided according to the principle of degressive proportionalism. Members of the Europeans parliament (MEPs) from bigger countries have bigger constituencies, but bigger countries will always have more MEPs than smaller countries (website europarl.eu, 2019). Secondly, the number of parties that compete in different countries varies. Since there are a limited amount of seats to be won in European elections, the percentage of votes a party needs for one seat is higher than in regular elections. The differences between the percentage of votes and the percentage of seats a party gets, can therefore increase, but also vary across countries and time.

Gallagher (1991) developed an index to measure disproportionality, which is one of many, but seems to be the index which is most preferred in comparison with other indices (Taagepera & Grofman, 2003: 673). To calculate Gallagher's least-squares index:

The Parliaments and government database (website parlgov.com, 2019) calculated the disproportionality according to Gallaghers least-squares index. Data on dispro-portionality will thus be extracted from the Parliaments and government database.

Other control variables

The dataset is completed by a set of control variables. The economic control varia-bles include economic growth, gross domestic product per capita (GDP) and the un-employment level. All the economic data is extracted from the World Bank's database (website worldbank.org 2019). Finally, a dummy variable will be added for former communist countries, where Western European countries are scored 0 and former communist countries are scored 1. In these younger democracies the polarization mechanism might work differently.

Final dataset and control variables

To sum it all up, a dataset is created which includes data on 28 different European countries over four European elections. For every country, per election, it contains a measurement for polarization on the elite and on the public level, the number of ef-fective political parties and an index for disproportionality. The economic control va-riables, all extracted from the World Bank (website worldbank.org, 2019), include GDP per capita, GDP growth and the percentage of the workforce which is unem-ployed. To complete the dataset, a dummy variable is added for former communist countries.

Part V: Results

Before diving into the regression models, I will start by describing the data. Because of all the data that had to be combined, like objective party positions, polarization measures for the public on the left-right scale and for attitudes toward European inte-gration and election outcomes, a unique dataset was created.

Ideological polarization since 1999

Per European election, from 1999 to 2014, and for every country there is data availa-ble for ideological polarization on the elite and the public level, as is visualized in figu-re 3. A couple of diffefigu-rent things catch the eye. First of all, the public polarization va-lues are higher than the elite polarization vava-lues. According to Dahl (2008), his pola-rization index, which is used as a measure for elite polapola-rization, is calculated in a si-milar manner to the calculation of standard deviations. Although this is sisi-milar, it is not the same. Therefore, the absolute differences might not be of importance, but they are noteworthy.

The increase of ideological polarization, in both the public and the political eli-tes, is the most important information provided by figure 3. This points to an ongoing process of polarization in Europe. For the public, figure 3 shows the highest average polarization values in 2009. Although the average polarization in 2014 was lower than in 2009, overall ideological polarization among the public has increased between 1999 and 2014. A possible explanation for this peak in ideological polarization might be the economic crisis of 2007-2009. It is possible that in times of economic crisis, left wing people become more left wing, while right wing people become more right wing. But this is only one possible explanation.

The peak in public polarization in 2009 is not visible at the level of elite polari-zation. At the same time, a clear increase of elite polarization is visible between 1999 and 2014. The elite polarization box plot also shows three outliers. In 2004, 2009 and 2014 the polarization value for Cyprus is significantly higher than for the other coun-tries. It seems that the political elites in Cyprus are more polarized than the political elites in other countries. In 1999 Cyprus was not part of the European Union yet,

the-refore there is no data on Cyprus before 2004. So it is not clear if Cyprus has always been more polarized or that the process of polarization happens faster in Cyprus.

Concluding, ideological polarization is happening in Europe on the elite and the pu-blic levels.

Polarization of attitudes toward European integration since 1999

Figure 4 shows the same box plots, only on the y-axis the polarization values for pre-ferences toward European integration are portrayed. On the left side, the elite polari-zation values are seen per year. The first thing that stands out is the lone observation for elite polarization in 2014. In 2014, in the EES only the respondents in France were asked to score the different parties on their attitudes toward European integrati-on. Therefore, there is only data available on elite polarization toward European inte-gration for one country in 2014. When evaluating if elite polarization increased over

1. 5 2 2. 5 3 3. 5 0 0. 5 1 4 4. 5 5 Mass polarizat ion Lef t-Right 1999 2004 2009 2014 Year 0 1 2 3 4 5 0. 5 1. 5 2. 5 3. 5 4. 5 E lit e polarizat ion Lef t-Right 1999 2004 2009 2014 Year

time, the 2014 box plot does not add a lot of information to the overall figure, but for completeness sake it was not removed.

A second important attribute of figure 4 is the overall low level of elite polarization on attitudes toward European integration. Although European integration has been characterized as an elite driven project, it is still surprising that the variation in attitu-des is so low. From 2004 to 2009 the elite polarization even decreased, while there was an economic crisis happening.

The overall low elite levels of polarization also point to a different problem. Ei-ther the positions of the different political parties toward European integration are very uniform or the public is not aware of the different positions political parties hold toward European integration. In the first case, this points to a democratic problem. Although, as is clearly visible on the right side of figure 4, the public is becoming more polarized on their attitudes toward European integration, they don't have wide range of options to vote accordingly. There could be two reasons for this lack of

pola-0 1 2 3 0. 5 1. 5 2. 5 3. 5 4 4. 5 E lit e P olarizat ion E U 1999 2004 2009 2014 Year 2 2. 5 3 3. 5 4 0 0. 5 1 1. 5 4. 5 Mass polarizait ion E U 1999 2004 2009 2014 Year

rization. One, the political elites are really homogenous. Two, the political elites are more polarized, but the public is unaware of these differences in attitudes toward Eu-ropean integration, which would also result in a homogenous distribution of elite pola-rization. This is because the party positions are determined by what the respondents of the EES think the party positions are. Either way, this is a problem for the demo-cratic proces. Or the information flow is flawed and people don't know what the par-ties stand for or the parpar-ties don't provide enough policy alternatives. Therefore, the political elites and the public are not connected well on this topic.

At the same time the public seems to polarize along the same path on the atti-tudes toward European integration as ideologically. Overall, the process of polarizati-on is visible and there is a peak at the year of 2009. The peak in both ideological po-larization and popo-larization of attitudes toward European integration suggest that there was something else happening which caused the public to be more polarized around that time. The peaks in ideological polarization and polarization of attitudes toward European integration also suggest that the polarization values are related. One logi-cal explanation for the polarization peaks could be the economic and the Eurocrisis, but this is just speculation.

Finally, the overall mass polarization values are higher for attitudes toward Eu-ropean integration than for ideological mass polarization. The reason could be that the public is more divided on their preferences toward the European integration. This could also be caused by the formulation of the questions in the EES. For the questi-ons in the EES concerning European integration, the scale on which you can answer is finite: score 1 if you think European integration has gone too far and 10 if it has not gone far enough. On the left-right scale the extremes aren't defined so clearly and are therefore up for interpretation. People will probably be less likely to score them-selves on the extremes of the left-right scale, than on the European integration scale. So comparing the mass polarization for ideology and European integration values is not very useful.

Congruence of polarization

Ideologically the mass and the elite are congruent, as is seen in figure 5. Whenever elite polarization is higher, the mass polarization is, generally speaking, also higher.

There are of course exceptions. Romania has one of the most polarized publics, but at the same time the elite is hardly polarized ideologically. The cohesion is confirmed by Pearson's R, who shows a correlation coefficient of 0,45 (p<0,001) and the β-coefficient of 0,22.

At the same time the polarization of attitudes toward European integration, on the eli-te and mass levels, show the opposieli-te. There is no congruence. Moreover, according to the β-coefficient, the relationship is the other way around. When the public is more polarized, chances are that the elite is less polarized. The cohesion between public and elite polarization of attitudes toward European integration is less strong though, with a Pearson's R of -0,31 (p<0,05).

That the negative correlation is also quite weak, is probably because of the lack of variation on the side of elite polarization of preferences toward European inte-gration. Most of the countries have an elite polarization value between 0 and 1.

ATAT AT AT BE BG BG CY CY CY DE DE DE DE DK DK DK DK EE EE EE ES ES ES ES FI FIFIFRFI FR FR FR GR GR GR GR HR HU HU HU IE IE IE IE IT IT IT IT LT LT LU LU LU LU LV LV LV MT MT NL NL NL NL PLPL PL PT PT PT PT RO RO SE SE SE SE SI SI SI SK SK SK UK UK UK UK 1. 5 2 2. 5 3 3. 5 4 Mass polarizat tion lef t right 0 1 2 3 4 5 Elite polarization left right

Country β = 0,22 AT AT AT BE BG CY CY DE DE DE DK DK DK EE EEES ES ES FI FI FI FRFR FR FR GR GR GR HU HU IE IE IE IT IT IT LT LU LU LU LV LV MT NL NL NL PL PL PT PT PT RO SE SE SE SI SI SK SK UK UK UK 2 2. 5 3 3. 5 4 1. 5 Mass polarizat ion E U 0 1 2 3 4 5 Elite Polarization EU Country β = -0,23

This seems to confirm the earlier observed problem with European integration politics. There is a lack of viable voting alternatives for the relatively polarized public. At the same time, ideologically the public and elite seem to be congruent.

Summarizing, figure 3 to 5 showed the degrees of polarization in Europe. While ideo-logical polarization seems to be happening, both among the public and political elites, the same can't be said for polarization on EU integration. Although the people have polarized, political parties did not polarize.

The rest of the results will focus on the question whether polarization is a top-down or a bottom-up proces.

Responsiveness and leadership hypotheses for European integration

The European integration polarization mechanisms will be discussed first. Since the-re isn't even congruence of polarization on this theme, it is hardly expected that the responsiveness or leadership model could explain levels of polarization.

As is seen in table 1, the leadership model does show a significant coefficient for elite polarization. So for polarization of preferences toward European integration, there is a significant leadership effect. Whenever the political elites polarize, this has a polarizing effect on the public. The democratic leadership model can explain 27 percent of the variance of public polarization of attitudes toward European integrati-on. There are no significant effects found in the responsiveness model and the R-squared is only 0,06. Political elites are not responsive to changing levels of polariza-tion in the public.

Table 1: responsiveness and leadership models EU

Responsiveness model: Dependent variable: elite

polari-zation t=0

Leadership Model 1: Dependent variable: mass pola-rization t=0 Mass polarization t-1 0,25 (0,14) 0,05 (0,09) Elite polarization t-1 -0,18
 (0,32) 0,34* (0,15) Dummy former communist

countries -0,30 (0,37) 0,17 (0,16) Disproportionality -0,04 (0,04) 0,02 (0,02) Table 1: responsiveness and

*p<0,10, **p<0,05, ***p<0,01

For polarization of attitudes toward European integration, the responsiveness hypo-thesis can be rejected. The democratic leadership hypohypo-thesis can be accepted, since the coefficient for elite polarization is positive and the model can account for 27% of the variation of mass polarization levels.

Responsiveness and democratic leadership hypotheses for ideology

Since there was congruence observed for ideological polarization in the first place, ideological polarization provides a better case for investigating who is cueing whom in the context of polarization. The results for the responsiveness and the leadership models are in table 2.

GDP t-1 0,00
 (0,00) 0,00
 (0,00) Economic growth t-1 -5,12
 (10,53) 0,01 (0,04) Unemployment t-1 5,14 (10,53) -0,03 (0,04) Constant -0,87 (0,92) 1,83*** (0,42) N 36 61 Adjusted R-squared 0,06 0,27 Responsiveness model: Dependent variable: elite

polari-zation t=0

Leadership Model 1: Dependent variable: mass pola-rization t=0 Table 1: responsiveness and

leadership models EU

Table 2: responsiveness and leadership models L/R

Responsiveness model: Dependent variable: elite

polari-zation t=0

Leadership Model 1: Dependent variable: mass pola-rization t=0 Mass polarization t-1 0,52** (0,24) -0,18 (0,14) Elite polarization t-1 0,74***
 (0,12) 0,25*** (0,07) Dummy former communist

countries -0,29 (0,24) 0,18 (0,14) Disproportionality 0,04 (0,02) 0,01 (0,02) Table 2: responsiveness and

*p<0,10, **p<0,05, ***p<0,01

The responsiveness model shows significant coefficients for mass and elite polariza-tion, meaning that the polarization of the elite is an effect of polarization of the mass and of the elite at t-1. That there is a positive effect from the elite at t-1 on the elite at t=0 is no surprise. That the mass polarization variable has a significant, positive coef-ficient, shows that the political elites are also influenced by levels of polarization among the public.

At the same time, the democratic leadership model also shows a significant positive coefficient for elite polarization on mass polarization. When comparing the two models, the results don't unequivocally point toward one model being the better one, as was the case when comparing the two models in the context of European in-tegration. Both hypotheses, the responsiveness and the democratic leadership hypo-thesis can be accepted in the case of ideological polarization.

Polarization bottom-up or top down?

The main question of this research was: is polarization a bottom-up or a top-down proces? And does that work differently for ideology and preferences concerning Eu-ropean integration?

When looking at table 2, the adjusted R-squared of the responsiveness model is higher than the adjusted R-squared of the democratic leadership model, 0,59 and 0,41 respectively. The coefficients in the responsiveness model for mass polarization

GDP t-1 0,00
 (0,00) 0,00
 (0,00) Economic growth t-1 0,02
 (0,06) 0,02 (0,03) Unemployment t-1 -0,02 (0,05) -0,01 (0,03) Constant -0,82 (0,61) 1,77*** (0,35) N 60 61 Adjusted R-squared 0,59 0,41 Responsiveness model: Dependent variable: elite

polari-zation t=0

Leadership Model 1: Dependent variable: mass pola-rization t=0 Table 2: responsiveness and

on elite polarization is also higher than the coefficient for elite polarization on mass polarization in the democratic leadership model. So the actual effect within the mo-dels is higher for the responsiveness model. Combining the information from the coefficients and, more importantly, the R-squared values, is leading me to conclude that the responsiveness model is a better model for explaining polarization. Con-cerning ideological polarization, the process is mainly bottom-up.

That does not mean there are no top-down effects. The democratic leadership model also worked, so the leadership effects should not be discarded. This should rather be interpreted as both models working, the elite is responsive to the public and the public is influenced by the elites. Only the effect from the public on the elites is greater than vice versa.

For polarization of attitudes toward European integration this functioned diffe-rently. There are no responsiveness effects found in this study, while leadership ef-fects were found. In the context of European integration, polarization of the public does not affect political elites, but when political elites polarize, this has a polarizing effect on the public.

The leadership effects of polarization of preferences toward European integra-tion are smaller than the effects of ideological polarizaintegra-tion though. The democratic leadership model could only account for 27% of the variation in mass polarization concerning European integration, while the democratic leadership model for ideology could explain 41% of the variation in polarization. So, in the context of ideological po-larization, both the democratic leadership model and the responsiveness model are better at explaining levels of polarization than both models are at explaining polariza-tion in the context European integrapolariza-tion. Therefore hypothesis 3, the causal mecha-nism behind congruence works better for ideologic polarization than for polarization of preferences toward European integration, is also accepted.

Institutional settings: number of parties

To test whether the number of parties matters, the responsiveness model is tested again. Since the median value is 4,15, the model is tested once including all observa-tions with less 4,15 effective political parties and once including all observaobserva-tions with more than 4,15 effective political parties. The results are shown in table 3.

*p<0,10, **p<0,05, ***p<0,01

Interestingly, levels of elite polarization are highly dependable of previous level of eli-te polarization in party syseli-tems with 4,15 or less political parties. There is no signifi-cant effect of mass polarization on elite polarization in these smaller party systems. This could be problematic, because when the electorate is getting more polarized the more extreme ideological positions are not represented within parliaments. The func-tion of political parties as channels of expression is impeded.

For the bigger party systems, the political elites are responsive to increasing levels of polarization in the public. But this model, and that has to be taken into ac-count, can only explain 22% of the variation in elite polarization. So there is an effect from mass on elite for bigger party systems, but there are still a lot of other factors which also influence the levels of elite polarization. For bigger party systems, the le-vels of elite polarization are not dependent on previous lele-vels of elite polarization.

Table 3: Number of parties Model 1: number of

parties<4,15
 Dependent variable: elite polari-zation t=0

Model 1: number of parties>4,15
 Dependent variable: elite polari-zation t=0 Mass polarization t-1 0,53 (0,34) 0,78* (0,35) Elite polarization t-1 0,90*** (0,14) 0,07
 (0,24) Dummy former communist

countries 0,29 (0,31) -0,33 (0,37) GDP t-1 0,00
 (0,00) 0,00
 (0,00) Economic growth t-1 10,07
 (28,09) 0,01
 (0,05) Unemployment t-1 -10,03 (28,10) -0,02 (0,04) Constant -1,21 (0,89) -0,66 (0,89) N 30 30 Adjusted R-squared 0,73 0,22

Concluding, when talking about party system size in the context of polarization, the size matters. From a normative point of view, bigger is better, since political elites are quicker to respond to levels of polarization in the public. When the public becomes polarized, it is probably a good thing that this is also translated to politics and that the people on the extremes have a voice. When institutional settings make it hard for new parties to arise and the party system is smaller, it is not clear what happens to the people on the extremes. What do you do if you can't channel your expression through political parties? An advantage of more fragmented party systems could the-refore be the representation of polarized attitudes within parliaments, involving the people with more extreme attitudes in the decision making processes of the govern-ment.

Part VI: Conclusion and discussion

There is one important methodological constraint which has to be taken into account. When investigating the causal mechanisms using a time lag, the time lag for this re-search is quite large. For the responsiveness model, levels of elite polarization are compared to previous levels of mass and elite polarization, but these previous levels of polarization are measured five yeas in advance. In five years a lot can happen which could alter the levels of mass and elite polarization. It would have been better if indicators of mass level polarization were available with a time lag of just a year. Un-fortunately, election studies are usually only held alongside elections. So, it is quite hard to overcome these limitations when investigating causal nature of polarization using election studies. But despite these limitations, conclusions can still be drawn from this research.

First of all, journalists, politicians and political scientists are right to increasingly focus on the phenomenon of polarization. The data showed that ideological polarization is happening on both the elite and mass level in Europe. Polarization of attitudes to-ward European integration is also happening within the public, but political parties remain relatively homogenous in their support for European integration.

The lack of congruence of the polarization of attitudes toward European inte-gration points to a democratic problem within this policy field. The variation of attitu-des within the public toward European integration is not represented in European parliaments. Political parties do not fulfill their role as channels of expression within this policy field. On top of that, no responsiveness effects have been found, while some leadership effects were found. So, political parties don't seem to respond at all to changes in the public, while the public responds a little to changes among political elites.

The note that European integration is an elite driven project (Gabel & Scheve, 2007: 1016) seems to be supported by these findings. The elite remains homoge-nous, while the public is not. These finding also confirm Toshkov's (2011: 185-186) remarks, claiming the electoral connection has been lost. This means that if voters can't show dissatisfaction about European integration through the democratic proces,

their opinions are not represented. This might fuel anti EU sentiments, because you could believe your opinions are not taken into account in the decision making proces. For political parties to behave more democratically, they should adopt more diverse positions on EU integration, providing the electorate with actual policy alternatives in this field.

Although the results aren't positive from a normative perspective, the small number of observations should be taken into account. Because of missing data in the EES, for the responsiveness model in the context of EU integration, the total number of observations was only 36. This is not enough to really make substantive claims about the responsiveness of the political parties to changes of attitudes on European integration in the public. At the same time, only eight elections of the total of sixty-one, had an elite polarization value of over 1. With this data in hand, you can easily claim that there is a lack of diversity of elite attitudes. So the dataset is large enough to describe levels of polarization, but not really large enough to make claims about the responsiveness model on this theme.

The total number of observations on this theme for the democratic leadership model was 61, which makes is easier to conclude that there are in fact some leader-ship effects.

For ideological polarization the results paint a more positive picture. Although leader-ship effects and responsiveness effects were found, the responsiveness model pro-ved to be the dominant model. The coefficient for mass polarization on elite polariza-tion was higher than vice versa. The R-squared value of the responsiveness model was also higher. This means that the effect from mass polarization on elite polarizati-on is strpolarizati-onger than the other way around. The higher R-squared value means that the responsiveness model is better at explaining the variance of elite level polarizati-on than the democratic leadership model is at explaining the variance of mass level polarization.

The dominance of the responsiveness model does not mean there are no lea-dership effects. This should rather be interpreted in the same manner as Steenber-gen et al. (2007) interpreted their "who is cueing whom?" research results. The natu-re of the natu-relation between the mass and the public seems to be natu-reciprocal, as is

visu-alized in figure 6. What other factors cause mass polarization was not part of this re-search. Therefore, this is still a question mark.

As is made visible in figure 6, mass polarization, when happening, causes elite pola-rization. So the electoral turnover mechanism and the rational anticipation seem to be working for ideological polarization. Elite polarization in turn has a polarizing effect on the public, so the democratic leadership mechanism is also at work. The line from elite to mass polarization is dotted, because this effect is less strong than the effect in the opposite direction.

Although leadership effects in the context of polarization are negative, they are only negative if there is no responsiveness. Evaluations and ideological positioning are a product of the individual's own believes and that person's surroundings (Druckman & Lupian, 2000: 6). Political leaders are, of course, one of the sources for individuals to gain new information from. Although, in the context of polarization it is preferred for leadership effects to be absent, this is also utopian thinking. As long as the political elites are mainly responsive to changes in the public, the observed leadership effects aren't a threat to the democratic proces. On top of that, this data suggest that the ori-gin of ideological polarization lies in the public.

The final part of the research revolved around the number of effective political par-ties. It seems that democratic systems with more effective parties are more responsi-ve to changes in the electorate. When there are more parties, it is more likely that a larger part of the variation of ideological positions can be represented. In party sys-tems with only three or four parties, only three or four ideological positions can be re-presented. Systems with more parties are better able to do justice to the difference in ideological positions in the public. Within smaller party systems no significant relation

has be found between mass polarization and elite polarization. Although fragmentati-on of party system can be seen as a negative attribute of party systems, because it can lead to less stable governments (Taylor & Herman, 1971: 37), the larger party system do more justice to the variation of ideological positions within the public and the larger party system have proven to be more responsive.

Part VII: References

• Abramowitz, Alan I. 2010. The Disappearing Center: Engaged Citizens, Polarizati-on, and American Democracy. New Haven, CT: Yale University Press.

• Abramowitz, A & Saunders, K (2008) Is polarization a myth? Journal of politics 20(2): 542-555.

• Baldassarri, D & Gelman A (2008) Partisans without Constraint: Political Polarizati-on and Trends in American Public OpiniPolarizati-on. American Journal of Sociology 114(2): 408-446.

• Burstein, P (2003) The Impact of Public Opinion on Policy: A Review and a Agenda. Political Research Quarterly 56(1): 29-40.

• Dahl, RA (1998) On Democracy. New Haven: Yale University Press.

• Dalton, RJ (2008) The Quantity and Quality of Party Systems: Party System Polari-zation, It's Measurement and It's Consequences. Comparative Political Studies 41(7): 899-920.

• DiMaggio P, Evans J, Bryson B. 1996. Have Americans’ social attitudes become more polarized? American Journal of Sociology 102:690–755

• Downs, A (1957) An Economic Theory of Political Action in a Democracy. Journal of Political Economy. 65(2): 135-150.

• Druckman JN & Lupia (2000) Preference Formation. Annual Review of Political Science. 3: 1-24.

• Druckman, JN, Peterson, E & Slothuus, R (2013) How Elite Partisan Polarization Affects Public Opinion Formation. American Political Science Review 107(1): 57-79. • Eichenberg, RC & Dalton, RJ (1993) Europeans and the European Community:

The Dynamics of Public Support for European Integration. International Organizati-ons 47(4): 507-534.

• Ezrow, L (2007) The Variance Matters: How Party Systems Represent the Prefe-rences of Voters. The Journal of Politics. 91(1): 182-192.

• Ezrow, L, Tavits, M & Homola J (2014) Voter Polarization, Strength of Partisanship, and Support for Extremist Parties. Comparative Political Studies. 47(11):

1558-1583.

• Fiorina, M, Abrahams, S & Pope, J (2006) Culture War? The myth of a polarized America. New York: Pearson Longmann (digitale versie: http://www.uvm.edu/~dgu-ber/POLS125/articles/fiorina2.pdf

• Fiorina, M, Abrahams, S & Pope, J (2008) Polarization in the American public: mis-conceptions and misreadings. The Journal of Politics 70(2): 556-560.

• Gabel M & Scheve K (2007) Estimating the Effect of Elite Communications on Pu-blic Opinion Using Instrumental Variables. American Journal of Political Science. 54(1):1013-1028.

• Gallagher (1991) The Variance Matters: How Party Systems Represent the Prefe-rences of Voters. Electoral Studies. 10(1): 33-51.

• Golder, M & Stramski, J (2010) Ideological Congruence and electoral institutions. American Journal of Political Science 54(1): 90-106.

• Granger, CWJ (1980) Testing for Causality: A Personal Viewpoint. Journal of Eco-nomic dynamics and Control 2: 329-352.

• Grofman, B (2004) Downs and the Two Party Convergence. Annual Review of Poli-tical Science 7: 25-46.

• Hakhverdian, A (2012) The Causal Flow between Public Opinion and Policy:

Government Responsiveness, Leadership, or Counter Movement? West European Politics. 35(6): 1386-1406.

• Hakhverdian, A & Mayne, Q (2016) Ideological Congruence and Citizen Satisfacti-on: Evidence From 25 Advanced Democracies. Comparative Political Studies 50(6): 822-849.

• Hetherington, MJ (2001) Resurgent Mass Partisanship: The Role of Elite Polarizati-on. American Political Science Review 95(3): 619-631.

• Hibbs, DA (1977) Political Parties and Macroeconomic Policy. The American Politi-cal Science Review. 71(4): 1467-1487.

• Hobolt, SB & Klemmensen, R (2008) Government Responsiveness and Political Competition in Comparative Perspective. Comparative Political Studies. 41(3): 309-337.

• Huber, JD & Powel, GB (1994) Congruence Between Citizens and Policymakers in Two Visions of Liberal Democracy. World Politics 46(3): 291-326.

• Iversen, T & Soskice, D (2015) Information, Inequality, and Mass Polarization: Ideo-logy in Advanced Democracies. Comparative Political Studies 48(3): 1781-1830. • Iyengar, S, Yphtach, L, Levendusky, M, Malhorta, N & Westwood, SJ (2019) The

Origines and Consequences of Affective Polarization in the United States. Annual Review of Political Science. 22(7): 1-18.

• Iyengar, S & Westwood, SJ (2015) Fear and Loathing across Party Lines: New Evi-dence on Group Polarization. American Journal of Political Science 59(3): 690-707. • Jacob, LR & Shapiro, RY (1994) Studying Substantive Democracy. PS: Political

Science and Politics. 27(1): 9-17.

• Klar, S, Krupnikov, Y & Ryan JB (2018) Affective Polarization or Partisan Disdain? Untangling a Dislike for the Opposing Party from a Dislike of Partisanship. Public Opinion Quarterly. 82(2): 379-390.

• Laako, M & Taagepera, R (1979) "Affective" number of parties: A measure with ap-plication to Western Europe. Comparative Political Studies. 12(1): 3-27.

• Layman, GC & Carsey, TM (2002) Party Polarization and "Conflict Extension" in the American Electorate. American Journal of Political Science. 46(4): 786-802.

• Layman, GC, Carsey TM & Horowitz, JM (2006) Party Polarization in American Po-litics: Characteristics, Causes and Consequences. Annual Review of political Sci-ence 9: 83-110.

• Lelkes, Y (2016) Mass Polarization: Manifestation and Measurement. Public Opini-on Quarterly. 80: 392-410.

• Levendusky, MS (2010) Clearer Cues, More Consistent Voters: A Benefit of Elite Polarization. Political Behavior 32: 111-131.

• Mair, P (2008) Electoral Volatility and the Dutch Party System: A Comparative per-spective. Acta Politica 43: 235-253.

• Mason, L (2015) ‘I Disrespectfully Agree’: The Differential Effects of Partisan Sor-ting on Social and Issue Polarization. American Journal of Political Science. 59: 128–145.

• McCarthy, N, Poole, KT & Rosenthal, H (2006) Polarized America: The dance of Ideology and Unequal Riches. Cambridge: The MIT Press.

• Powell, GB (2013) Representation in Context: Election Laws and Ideological Con-gruence Between Citizens and Governments. Perspectives on Politics 11(1): 9-21. • Ray L (2003) When Parties Matter: The Conditional Influence of Party Positions on

Voter Opinions about European Integration. The Journal of Politics 65(4): 978-994. • Rogowski, JC & Sutherland, JL (2016) How Ideology Fuels Affective Polarization.

Political Behavior 38: 485-508.

• Sartori, G (1976). Parties and party systems. New York: Cambridge University Press.

• Sims, CA (1972) Money, Income and Causality. The American Economic Review. 62(4): 540-552.

• Steenbergen, MR, Edwards, EE & De Vries, CE (2007) Who is Cueing Whom? Mass-Elite Linkages and the Future of European Integration. European Union Poli-tics 8(1): 13-35.

• Stimson, JA, Mackuen, MB & Erikson RS (1995) Dynamic Representation. The American Political Science Review. 89(3): 543-565.

• Taagepera, R & Grofman, B (2003) Mapping the indices of seats-voters disproporti-onality and inter-election volatility. Party Politics 9(6): 659-677.

• Taylor, M & Herman, VM (1971) Party Systems and Government Stability. The American Political Science Review 65(1): 28-37.

• Toshkov, D (2011) Public opinion and policy output in the European Union: A lost relationship. European Union Politics 12(2): 169-191.

Internet sources:

• CNBC (2018) Germany has difficulties and polarization that we haven't seen before for decades: Merkel tells Davos. https://www.cnbc.com/2018/01/24/germany-has- difficulties-and-polarization-that-we-havent-seen-for-decades-merkel-tells-davos-.html (last checked June 19, 2019)

• European parliament (2019) http://www.europarl.europa.eu/doceo/document/ TA-8-2018-0029_EN.html (last checked June 19, 2019)

• Financial Times (2018) US midterms: What we learned from the election in charts. https://www.ft.com/content/aa8ea100-e27b-11e8-a6e5-792428919cee

• The Guardian (2019) The death of consensus: how conflict came back to politics. https://www.theguardian.com/politics/2018/sep/20/the-death-of-consensus-how-conflict-came-back-to-politics (last checked June 19, 2019)

• Meeus, TJ (2019) Özdil, Meyer en Rutte III: Den Haag in het tijdperk van de polari-satiezwendel. NRC. Available at: https://www.nrc.nl/nieuws/2019/06/01/ozdil-meyer-en-rutte-iii-den-haag-in-het-tijdperk-van-de-polarisatiezwendel-a3962242 (last checked June 19, 2019)

• NOS (2019) Rutte over polarisatie: kloven passen niet bij Nederland. https://nos.nl/ artikel/2278122-rutte-over-polarisatie-kloven-passen-niet-bij-nederland.html (last checked June 19, 2019)

• Parliaments and governments database: http://www.parlgov.org (last checked June 19, 2019)

• Web of Knowledge database: http://wcs.webofknowledge.com/RA/analyze.do?pro- duct=WOS&SID=C4IfDoz31sj6TPr8UUx&field=PY_PublicationYear_PublicationYe-ar_en&yearSort=true(last checked June 19, 2019)