From polarization of the public to polarization of the electorate: European Parliament elections as the preferred race for ideologues

This study examines the effect of voters’ ideological extremism on turnout in European national and European Parliament elections. Using data from recent European Election Studies, the Comparative Study of Electoral Systems, and other national election studies, we find that, relative to centrists, ideological extremists (measured by self-placement on the left–right scales) are more likely to vote in European Parliament elections (2014 and 2019) but not national elections. We argue that these differences stem from the fact that European Parliament elections are second-order races. The results help to explain why the European Parliament has become more polarized, even in the absence of significant changes in overall attitudes among the European public, and why extreme parties have been more successful in recent European Parliament than national elections.


Introduction
Voter turnout in European Parliament (EP) elections is notoriously low compared with national elections, a fact oft lamented by commentators and scholars of European democracy (Follesdal and Hix, 2006;Ford, 2014)-and one which may have long-term effects on the likelihood of participation in national elections (Hobolt and Franklin, 2011). Low turnout and different voting patterns to those in national elections have led scholars to refer to EP elections as second-order national contests (Hix and Marsh, 2007;Reif and Schmitt, 1980). Many voters who regularly participate in national elections simply decide to remain on the sidelines during EP elections. But at the same time, commentators have noted that countries that do experience higher turnout tend to see the mobilization among voters on the ideological extremes. And while some have even feared that higher mobilization among anti-EU extremists has the potential to drown out a silent pro-EU majority, others have pointed out that mobilising turnout at the EP elections would likely help moderate parties. 1 So just what is the link between voter ideology and participation in elections? Do ideological extremists and moderates mobilize at different rates in EP and national elections? And, if so, why? To answer these questions, we posit a theory that links individual voter-level ideology with turnout and we explore this relationship in EP and European national elections. We argue that turnout among moderate and extreme voters is driven by different processes, leading them to participate at different rates in national-level and European-level contests.
The second-order nature of EP elections means that the calculus of voting in these contests differs from national elections. Moreover, we argue that the calculus is different for ideological moderates and extremists. Second-order elections are viewed as less important and, thus, voters of all ideological stripes experience lower social pressure to vote in these elections compared with national contests. But for some voters-namely, ideological extremists-the negative impact of lower social pressure on turnout is offset by increases in other motivations to vote. The ability to cast an instrumental and expressive vote can motivate extremists to participate in EP elections compared with national elections. Moderates, in contrast, do not experience an increase in instrumental and expressive motivations in EP elections compared with national elections. Over time, this process can lead ideologues to experience electoral gains in EP elections, reinforcing their willingness to participate.
Empirically, we test our argument using data from four recent European Election Studies (EES; for the years 2004, 2009, 2014 and 2019) as well as 17 comparable national election studies (14 studies from the Comparative Study of Electoral Systems Modules, CSES, 4 and 5, supplemented with three additional national studies). 2 We demonstrate that in recent EP elections voters holding extreme ideological attitudes, measured by selfplacement on a left-right ideological scale, are more likely to vote relative to moderates. Higher participation in EP elections among ideologues started in Western European countries after the economic crisis. In contrast, ideologues have not become more likely to participate in national races during this same time period. In contrast, our results show that ideologues are less likely (relative to moderates) to participate in national elections across Western Europe.
Our argument and results have significant implications for European politics and elections, as well as for studies of turnout outside the European context. Differential participation rates across the ideological spectrum may explain, at least in part, the relative success of ideologically extreme parties in EP elections compared with national contests (see Spoon, 2011), offering a new explanation for their success. Even if the views of the European public remain stable over time, we may still observe changes in the ideological composition of the EP parliament if different ideological groups turn out at different rates. Furthermore, our findings suggest that scholars ought to revisit the role of ideology and issue salience in turnout for other second-order elections. These might include US midterm elections, or state-level elections in other federal countries such as Austria, Belgium, Canada, Germany, and Switzerland. Finally, our analyses show heterogeneity between EU15 and Central and Eastern European (CEE) countries. This result corroborates previous accounts of politics in CEE countries (see e.g. Savage, 2014;Tavits and Letki, 2009), and calls for greater caution when considering political processes across EU member states.

Voter turnout, ideology, and European elections
Voter ideology is a key theoretical and empirical variable in the literature explaining individual voter turnout (e.g. Adams and Merrill, 2003;Adams et al., 2006;Downs, 1957;Glaeser et al., 2005;Hinich and Ordeshook, 1969;Peress, 2011;Plane and Gerhtenson, 2004). Beginning with Downs (1957), research linking ideology to turnout has argued that voters who are ideologically distant to the parties competing in a political system, or otherwise indifferent to the choices available to them, are less likely to participate in elections. These models of individual turnout, all developed in the context of the American two party system, tend to follow a Downsian logic. They examine a voter's perception of ideological differences between candidates or parties (e.g. Adams and Merrill, 2003;Plane and Gerhtenson, 2004). A voter holding extreme views may see two or more relatively centrist candidates (or parties) as similar, and thus be less likely to participate; Adams et al. (2006) refer to this as alienation. But similarly, a centrist voter could feel indifferent between two polarized parties, also leading to abstention. These US-based studies find that voters expressing support for the positions of a particular candidate or party-those who are neither alienated nor indifferent-are more likely to vote.
The link between turnout and voter ideology-and ideological extremism in particular -has not received the same attention in the comparative and European literature. Studies on European elections have found that citizens holding particular policy attitudes may be more likely to vote than others, especially if those issues lead them to engage with politics. There is evidence suggesting that voters who express support for European integration are more likely to participate in EP elections (see e.g. Hobolt et al., 2009). Echoing findings from the US that voters tend to be more conservative than non-voters (Leighley and Nagler, 2014), Gallego (2015) finds significant unequal participation in comparative context, but also that institutional and party system variables may mitigate turnout inequality between the left and the right (see also Gallego, 2010). But most models of individual-level voter turnout do not examine the impact of voter attitudes, either with respect to particular issues or to general ideology. 3 A comprehensive meta-analysis of 90 peer-reviewed studies of individual-level voter turnout published in top journals between 2000 and 2010 found that only five of the 90 studies included an ideological selfplacement variable (Smets and van Ham, 2013). The variables most commonly included across these 90 studies were age, gender, income, education, race, marital status, political interest, and party identification.
Following Down's (1957) paradox of voting, a large tradition of scholarship attributes the act of voting to a sense of civic duty (see e.g. Blais, 2000;Riker and Ordeshook, 1968). Personal and socio-demographic characteristics gained the spotlight in this scholarship because of the ability of these variables to induce and foster this sense of duty. Scholars have uncovered the role of socioeconomic characteristics, such as education, income, and employment status, as determinants of the decision to turnout (see e.g. Abramson and Aldrich, 1982;Sondheimer and Green, 2010;Verba et al., 1995). Experimental evidence then confirmed the relevance of social pressures (and hence socioeconomic factors) for turnout (see Gerber et al., 2008).
Beyond the spatial tradition and civic duty motivations, voters may decide to vote because they think that they can influence the electoral result-even if the Downsian math makes it unlikely-or because they care about issues and wish to express their views, even in the absence of influence. In EP elections, which are characterized by lower turnout and, in some cases, more proportional electoral laws, voters may feel better able to influence the outcome; and they may have more, and more prominent, parties to vote for whose position reflects their own. In such a setting, where more parties exist with the possibility to perform well, expressive motivations might also justify individuals' participation (see Brennan and Buchanan, 1984;Brennan and Lomasky, 1993;Hamlin and Jennings, 2011).

A theory of voter turnout in EP elections
We posit a theory of participation in EP elections premised on the fact that EP elections are second-order in nature (Hix and Marsh, 2007;Reif and Schmitt, 1980), with significant consequences for voter behavior and national-level party systems (Dinas and Riera, 2018;Schulte-Cloos, 2018). We argue that the effects of these second-order contests impact voter turnout. Our argument is built around three core ideas: (1) social pressure to vote is lower in EP contests than in national contests due to the former's second-order nature; (2) increases in expressive and instrumental motivations to vote can offset the negative impact of lower social pressure; and (3) extremists are relatively more motivated by the ability to cast an instrumental and expressive vote than moderates at EP elections, again due to their second-order nature. Thus, at EP elections, a decline in social pressure is offset by other motivations for extremists, but not for moderates.
While social pressure may drive voters of all ideological stripes to the polls at national parliamentary elections (see e.g. Blais and St-Vincent, 2011), social pressure to vote in EP elections is relatively low (Moeller et al., 2018). Citizens are less likely to pay attention to, or care about, who votes in EP elections since these contests are perceived to be less relevant than national contests. In other words, people feel less social pressure to vote merely because their neighbors do so. Voters most motivated to vote by social pressure and who have few other reasons to go the polls, will be more likely to abstain in EP elections relative to national elections.
When social pressures decrease, other motivations gain in prominence and can make a difference in who turns out and who abstains at the election polls. Such concerns include the ability to vote for a party expressing ideas closer to one's own, the chance to cast a protest vote, or the chance to support a party that generally does not perform well in highly salient national contests (see e.g. Hobolt and Spoon, 2012;Marsh, 1998;Weber, 2011). We argue that in EP elections these expressive and instrumental concerns are more likely to drive ideologues to turn out than moderates. 4 Moderates already have the opportunity to vote for parties that express their points of view and capture a significant vote share during more salient national contests. There are no new or additional expressive or instrumental motivations driving them to participate during EP elections. In contrast, because parties tend to emphasize issues that are salient to extremists in EP elections (see Hobolt and De Vries, 2015), voters at the extremes of the political spectrum may find viable parties that explicitly take a position reflective of their own (e.g. parties that focus specifically on the issue of EU integration in a very negative or very positive manner). When ideological position acts as a proxy for issue salience (Rabinowitz and Macdonald, 1989), the ability to cast a vote for a party that emphasizes issues and expresses positions that matter more to extreme voters should motivate them to participate.
Furthermore, given the second-order nature of the EP elections and the relative success of more extreme parties in these contests, extremists supporting these parties become more likely to perceive themselves as winners. Plescia (2019) shows that individuals feel like winners when the party they support experiences a simple improvement in vote share, or a first entry into a parliament, not only when their preferred party wins an election outright, as previous studies have suggested (see e.g. Anderson and Guillory, 1997;Anderson and LoTempio, 2002;Bowler and Donovan, 2002). Given the success of smaller, more extreme parties, EP elections become an ideal setting for ideologues to perceive themselves as winners.
In sum, we argue that at an EP election time, ideologues-motivated by instrumental and expressive concerns and finally perceiving the parties that they support as having a better chance at performing well-are more likely to turn out to vote than moderates. Ideologues, unlike moderates, can cast a ballot for a (non-mainstream) party that expresses their views and they have the chance to feel like a winner. In many national contests, the extremists may instead feel alienated and at a disadvantage compared with centrist voters. While all voters may feel a decline in social pressure to vote in EP elections, in contrast to moderates, for extremists this decline is offset by an increase in instrumental or expressive concerns.

H1:
Ideologically extreme voters are more likely to vote in EP elections than moderates, but not in national elections.
We believe that this hypothesis could be driven by two parallel and related mechanisms, both of which provide ideologically extreme voters with opportunities to cast ballots for parties that align with their views in EP elections, similar to moderate voters in national elections. The first mechanism is the supply of extremist parties participating in EP elections, and the second is the proportionality of EP elections. More options on the extremes of the ideological spectrum may draw voters holding more extreme views to the polls. Likewise, higher levels of proportionality give smaller, often more extreme parties opportunities to win seats. Voters holding extreme views, perhaps looking to vote expressively and hoping that the party they support has a chance to gain seats, may be more likely to go to the polls when the supply of extreme parties increases. Thus, we examine the institutional mechanisms that underpin our theoretical argument.
While research has demonstrated that proportional systems generally have higher turnout, the findings focus on the aggregate, country-level turnout (e.g. Blais and Carty, 1990). They do not examine how these system-level variables impact individual voters differently depending upon their ideology. Here, we argue that the effects of these system-level variables should be particularly noticeable among ideologically extreme voters during EP elections. In these contests the micro-level arguments underpinning the relationship between the system-level variables and aggregate turnout are most likely strongest. Namely, ideologically extreme voters are more likely to experience increased opportunities for expressive voting, to support a party that can do well, and a chance to feel like a winner.
H2a: Ideologically extreme voters are more likely to vote in EP elections than moderate voters especially in countries with a higher supply of extremism in EP elections.
H2b: Ideologically extreme voters are more likely to vote in EP elections than moderate voters especially in countries where the EP election proportionality is higher.

Data and methods
To test our hypotheses, we first employ survey data from the European Election Studies (EES) from 2004 to 2019. The main dependent variable is whether an individual reported turning out to vote or not. 5 Given the binary nature of our operationalization, we apply a logit model with standard errors clustered at the country level. To test the role of ideological extremism on turnout, we use the self-reported placement on the left-right scale. In the EES, respondents can place themselves on a scale ranging from 0 (extreme left) to 10 (extreme right). We code a dummy variable for ideological 'extremism' whenever the respondent indicated a value less than or equal to 2, or greater than or equal to 8. This approach allows us to capture ideologues on both ends of the political spectrum. We also test the robustness of our findings using a continuous variable for ideological extremism. The latter is derived by centering the 0-10 left-right scale on zero so that the scale runs from −5 (extreme left) to +5 (extreme right). We then take the absolute value of this re-scaled measure which allows us to pool together those that hold more extreme views, on the right or left of the ideological scale. The Online appendix includes this robustness check.
We test the effect of extremism for every election on two main samples: (a) all EU member states at the time of the election; and (b) the group of EU15 countries (Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, the Netherlands, Portugal, Spain, Sweden, and the UK). Previous studies found that the logic underpinning the left-right political divide in Western Europe does not always apply in CEE countries (see Savage, 2014). Tavits and Letki (2009) also report that expected policy outcomes are not consistent with the ideology of governments in CEE countries, and government spending does not follow expectations based on the left-or right-leanings of the government. These results call for greater caution when pooling together EU member states and lead us to take into account possible heterogeneity between EU15 and CEE states.
The models control for other relevant variables such as attitudinal, socio-economic, and individual features of respondents. For attitudinal controls, we include political interest (categories on a four-point scale, where higher values indicate lower political interest); the general left-right ideological position of the respondent; support for Europe (on an 11-point scale, where higher values indicate stronger support for an ever-closer union); and perceptions of economic performance (respondents indicate if they perceive a worsening of the economy on a four-point scale). We then control for party attachment, given its relevance for political mobilization. To capture socio-economic characteristics, we include variables for unemployment and perceived social class. 6 Finally, we consider other characteristics such as gender and age. These cover the variables most commonly found in studies of individual-level turnout (Smets and van Ham, 2013) and which are readily available across the datasets that we use. The distributions of the variables that we use can be found in the Online appendix.
Our theory suggests that there are differences in the relationship between extremism and turnout in EP and national elections. Specifically, there ought to be no relationship between ideological extremism and turnout in national elections. We replicate our EP models as closely as possible, given the availability of data, for a series of countries using recent national election studies data both from the CSES and other sources (the full list is provided in the first endnote). 7 We start with an analysis of EU15 countries and then replicate the models for all cases. We replicate our EP models as closely as possible, however, some changes were necessary. Specifically, we could not include a variable for perceived socio-economic status, instead, we measured status through income categories (treating the variable as continuous). We also had to remove other covariates due to large amounts of missing data (for instance in the case of employment data). However, including socio-economic status measured through income is likely to proxy for other important factors that are highly correlated with household income. All models are estimated with clustered standard errors based on countries. As before, we replicated the analysis using a multilevel model approach with country effects (see the Online appendix).
Finally, we turn to an analysis of the mechanisms we identify as drivers for the mobilization of ideologues: (a) the proportionality of the EP electoral law in each country, which influences the ease with which parties can translate voter support into seats; and (b) the supply of extremism in the party system. To measure both the proportionality of the electoral system and system-level supply of extremism, we primarily use data based on expert surveys from the ParlGov data set (Döring and Manow, 2019), but we also consider perceived supply of extremism using the responses in the EES.
To measure the supply of extremism, we calculate a modified version of Dalton's (2008) polarization measure. This captures the standard deviation of party positions among those parties that participated in the EP election and have been listed in the ParlGov dataset. The presence of parties holding more extreme positions increases the standard deviation, but because we weight by the number of parties, more parties expressing the same amount of extremism has little impact on the measure.
Our choice for this operationalization results from considerations regarding the tradeoffs with other alternatives. For instance, measuring the raw number of ideologically extreme parties: (a) would have implied a subjective evaluation on which parties should be considered ideologically extreme; and (b) would have given equal weights to, for instance, an extreme right-wing political party and a conservative party that had obtained a score just over the chosen threshold on the ideological scale. It would also give extra weight to scenarios in which there were multiple parties expressing similar positions, even though the supply of extremism would have been the same as when one larger party expressed that position. 8 An alternative approach would have been to consider the range between the two most extreme parties. But this measure is determined only by two parties rather than all parties in a multi-party systems. Moreover, a situation in which the range is determined by a single very extreme political party, when the others are all scattered around the center, is different from one in which all political parties tend to be distant from another. Our measure is the only one that allows us to account for the positions of all parties in multiparty systems when assessing the supply of extremism. 9 To test our hypotheses regarding the mechanisms, we interact our measures of systemlevel supply of extremism and proportionality with our dummy for ideological extremism in a multilevel context. We then estimate the marginal predicted mean for the effect of extremism in polarized versus non-polarized systems, and in proportional versus disproportional ones. Finally, we estimate the difference in these quantities for both the 2019 and 2014 EP elections-namely, the elections where we find an effect for ideological extremism.
We also consider a second, individual-level measure of polarization: perceived polarization. This is measured as the average absolute distance between the position that each EES respondent assigns to parties in their political system and the center of the system (measured at the level of the respondent). Given the individual nature of this second measure, however, the approach cannot be the same as with a system-level variable. We argue that there is a close relationship between one's own ideological position and the perceived positions of political parties. It is reasonable to assume that an individual voter sees the system as more or less polarized because they are more extreme themselves. We therefore look at both the effect of extremism on the perception of polarization, and at the interaction between ideological placements and perceived polarization (see the Online appendix).
Additional analyses also replicate the findings on the role of ideological extremism on turnout in EP election by including a control for those countries that held the EP and national election in the same day/weekend as the EP election. 10 Our main result, that ideologues can participate in greater numbers during second-order elections, is robust to this further specification and to the use of a multilevel model with country-level random effects. The Online appendix includes an analysis of different motivations that drive voters to participate in EP elections (from the EES 2014 Edition that includes an item on the motivations for turnout). Lacking a corresponding variable in the context of national election studies, we cannot compare the motivations across types of elections. But we can explore the relationship between post hoc justifications for voting and ideological extremism for those who voted. The results are reported in the Online appendix and show that expressive considerations are indeed associated with extremism.

Do ideologues mobilize in EP elections?
In Table 1, we present our main results for the four EP elections held since 2004. The results show that respondents who place themselves on the extremes of the ideological spectrum have been more likely to turn out in the two most recent EP elections. Predicted probabilities of turnout calculated based on the models for 2009-2019 show that ideological extremists are between roughly 2 and 3.5 percentage points more likely to turn out than moderates. This effect is approximately the same as the difference in turnout between respondents who are on average more Euroskeptic and those who are more Europhilic, based on the same model. This holds true in both Western and Eastern Europe, even when considering traditional determinants of participation, such as interest in politics, social class, employment status, age, and party attachment. In the earlier elections, the coefficient on extremism is positive, but not statistically significant. This result may be indicative of a process of learning (see Bendor et al., 2003). At the same time, other potential explanations, such as a change in the electoral environment following the financial crisis, may also be at play. This explanation appears particularly plausible since Western European countries experienced a rise in the importance of extremist ideology in 2009, while the process appears to have diffused to CEE countries only in later elections.
The control variables perform as expected and, in particular, we confirm that Europhiles are more likely to vote on average. This result, consistent with previous literature, highlights even further the role that issues and ideologies can play in mobilizing voters within second-order elections. It is also interesting that the coefficient for left-right placement does not exhibit a homogeneous effect across election years. In some years the sign is positive (indicating that right-wing voters are more likely to participate); while in some years it is negative (indicating the opposite effect). Significance at the 95% is not reached in any election. Voters with higher social status, older citizens, voters with a high interest in politics, and party members or affiliates are instead always more likely to cast a vote in EP elections.
Thus, our results confirm well-known findings from existing literature, but also show that, especially in recent EP elections, ideological extremists are more likely to go to the polls. The Online appendix shows that our result is independent of the way in which we   We now turn to an analysis of domestic-level races to investigate whether the importance of ideological extremism is limited to EP elections, as our theory would suggest, or whether the same effect can be found for first-order elections, too.

Do ideologues mobilize in national elections?
Our argument is not only that ideological extremists are more likely to vote in EP elections than moderates, but also that extremism should not matter for turnout in national elections. We again identify participation as a dummy, and operationalize extremist voters as those who self-reported their ideological position equal or below 2; or equal or above 8 on an 11-point scale. Table 2 shows parameter estimates for logit models in national elections and presents evidence of the effect of extremism. The samples pool together as many countries as possible from the CSES as well as additional national election studies, replicating (as closely as possible) the analyses reported for the EP elections. As before, we estimate a model for Western European countries and then include CEE member states in the sample. The results confirm our hypothesis that ideologues are not more likely to participate in national elections-unlike in EP races. Even more, when only considering Western European countries, we observe a negative (albeit insignificant) coefficient: this suggests that ideologues may be less likely to turnout. This points to the exceptionalism of EP elections when considering the role of ideology in turnout. Second-order elections mobilize ideologues, while their participation is offset by the turnout of moderates in national elections. Concerning control variables, we observe that socio-economic status and economic grievances are the strongest determinants of electoral participation. This may be due to country-level heterogenous effects of the other covariates.
As in the case of EP elections, we perform robustness checks to investigate the sensitivity of our results. In particular, we estimate the same models using a continuous (rather than dichotomous) variable for extremism and the results remain unchanged. Then, we apply a multilevel approach with country random effects. This analysis highlighted some interesting findings (reported in the Online appendix): while EU15 countries still show a negative (but insignificant) effect, the inclusion of CEE member states turns the effect positive and significant. This apparently odd result may, however, help explain why CEE countries seemed to moderate the effect of extremism at the EP level. If, in some of these countries, ideologues are already more likely to participate in national elections, second-order races may be relatively less mobilizing for them. Beyond the heterogeneity of CEE countries, these analyses confirm that-at the very least in EU15 countries-ideologues do not participate more than moderates. If extremism has any effect in EU15 countries, it is most likely negative.
The next section tests two mechanisms that could drive these results, namely, the proportionality of the electoral system at EP elections, and the supply of extremism among the parties that participate, both of which affect opportunities for expressive and instrumental voting among extremists.

What drives ideologues to the polls in EP elections? Evaluating the possible mechanisms
Our previous results show that EP elections have become an arena in which more extreme voters across European electorates are mobilized, which may explain why the EP has become increasingly polarized. We have also demonstrated that this effect is not simply due to a general increase in the participation of extreme voters, as in national elections we fail to uncover the same result. We now turn to testing the two explanations that we have argued can motivate this phenomenon-the supply of extremism (measured by party system polarization) and proportionality. First, we have claimed that second-order elections with more polarized political parties offer fertile ground for the mobilization of ideologues because ideological voters can better identify parties that hold views that are similar to their own. Ideologues are therefore presented with an opportunity to cast a sincere vote. The second mechanism we identify is the (dis-)proportionality of the electoral system. We argue that when the system is more proportional, parties supported by ideologically extreme voters can more easily obtain seats-hence increasing the perceived representation of these voters and the sense that they have a chance to win. This increases the perception of political voice, in stark contrast to first-order contests. Therefore, we expect more proportional systems to foster the mobilization of ideologues. Figure 1 shows the levels of party system polarization and disproportionality by country for the 2014 and 2019 EP elections.
To test the validity of these claims, we replicate our results from Table 1 but we interact our dummy for extremism with party system polarization and electoral system disproportionality in both the 2019 and 2014 EP elections. The results are presented in Table 3: We seek to compare and contrast the effect of ideological extremism (versus moderation) in polarized and in non-polarized systems (or under disproportional and proportional electoral laws). We then calculate the marginal predicted effect for the interactions, obtaining eight values for each election. These values are the result of the following scenarios: moderate voters participating in non-polarized systems; moderate voters participating in polarized systems; ideological voters participating in non-polarized systems; ideological voters participating in polarized systems; moderate voters participating in  proportional systems; moderate voters participating in disproportional systems; ideological voters participating in proportional systems; ideological voters participating in disproportional systems. Table 4 shows the estimated probability of turnout for these scenarios. We evaluate the difference in the probability of turnout between ideologues and moderates in polarized and non-polarized systems. We then replicate the same analysis examining disproportionality instead of polarization. Results are graphically shown in Figure 2. While the differences shown in Figure 1 can be gleaned from Table 4, it allows us to visualize them more easily. In 2019, we observe that in a polarized system the difference in the marginal mean effect of ideological extremism versus moderation is 0.029, with a standard error of 0.010 (hence significant at the 99% of confidence). This implies that when the party system is polarized, extremists are significantly more likely to participate than moderates. This effect does not hold in non-polarized contexts. In these contexts, the difference between extremists and non-ideological voters is negative and not statistically significant, implying that we observe an effect in the expected direction for polarization. The difference of these differences, 0.038, also approaches, but does not quite reach, statistical significance at the 10% level (p = 0.122). Moving on to examine the effect of electoral disproportionality, we observe that in highly disproportional systems the difference between ideologues and other voters is slightly positive but non-significant, as we would expect. In contrast, when we consider strongly proportional systems, the effect is both positive and significant at the 95% level (with an estimated difference of 0.024 and a standard error of 0.012), again as expected. However, the difference of the differences is not statistically significant. Therefore, our findings follow expectations, but are somewhat weaker, when considering electoral disproportionality compared with party system polarization. Estimates derive from the multilevel models in Table 3. The dependent variable is turnout at EP elections (reported). Survey data is unweighted.
Next, we replicate the analysis above for the 2014 election. Here, the evidence appears mixed. In both polarized and non-polarized systems ideologues are more likely to participate than moderates. There is a positive and statistically significant difference when the electoral law is highly disproportional, but not when there is proportionality, contrary to our expectations. This mixed evidence for the mechanisms that can drive ideologues to the polls during EP races may indicate that other conditions also contribute to the effect we report for ideology. We evaluated the number of parties (and of extreme parties) as well, but find similarly inconclusive evidence in these two recent elections (results are reported in the Online appendix). While learning processes could explain why these effects are stronger in 2019, we speculate that party strategies might also come into play. In particular, political parties might emphasize certain issues over others during EP elections, with the intention to test campaign strategies for national races. This would not necessarily result in a more polarized party system, but could nonetheless drive ideologues to the polls.

Conclusion
We have demonstrated that forces driving turnout at recent EP elections are different than those that impact participation in national elections. In 2014 and 2019 (and in 2009 for  Table 4. EU15 countries), voters holding extremist views were more likely than moderates to vote in second-order EP elections, but not in national contests. We argue that we can understand these patterns by examining the different incentives for participating in national parliamentary elections and second-order EP elections. In the former, social pressure drives voters of all ideological stripes to participate. In the latter, when social pressure is reduced, extremists perceive a greater opportunity to cast a vote for a party that represents their views, and a higher likelihood to contribute to their preferred party's meaningful gains. To test these mechanisms, we examined whether extremists behaved differently compared to moderates depending on the supply of extremism and disproportionality of the party system in EP elections, and found some supporting evidence in the case of the most recent election.
Our findings have implications that extend significantly beyond elections to the EP. They suggest that the literature on voter turnout must more carefully consider voter ideological extremity. Ideology may matter in a directional rather than Downsian spatial manner. Additionally, more attention must be given to the potential effects of ideology in different types of elections. Ideological voters may participate at higher rates in all second-order elections, not just those to the EP. Such elections could include midterm elections to the US Congress, by-elections in the UK, state elections in Germany, and others. Likewise, future research should examine why we find a greater impact of ideology in the two most recent EP elections compared with earlier elections. It may be related to electoral politics following the economic and financial crisis. To the best of our knowledge, the differential effect of ideology in these elections compared with first-order elections has not been examined.
Lastly, variation in turnout could play out differently across country and issue. In this article, we have examined the impact of extremism on a general left-right dimension, yet, future studies could examine the effect of extremism on many other dimensions including support or opposition to European integration, cultural and post-material issues, and economic issues. Different forms of ideological extremism may affect turnout differently depending upon the country, and specifically the party system within a particular country. This conclusion appears particularly plausible given the heterogeneity that we find across EU15 and CEE countries.

Supplemental material
Supplemental material for this article is available online.  (2015), Slovakia (2016), and Sweden (2014). We consider turnout for the election of the countries' lower chamber of parliament. For France, due to CSES data availability, we consider turnout in the first round of the presidential election. We then added additional cases for which we could access data and follow an English-language codebook, i.e. Belgium (2014), the Netherlands (2017), and the UK (2015).
In the case of the UK, the 2015 British Election Study included an item for validated turnout, which we use as our dependent variable. 3. Of course, socio-economic and demographic variables are correlated with attitudes, but by no means perfectly. A recent study on turnout in European elections does include left-right ideology and its squared term, but only as a control variable (Marquart et al., 2020). 4. Although similar variables may determine both, and they are clearly related, we argue it is important not to conflate turnout with vote choice. A voter could experience expressive motivations to participate in an election because they feel close to a party, but ultimately decide to vote for a different, more distant party due to other, possibly strategic, considerations. We therefore prefer to focus on the decision to turn out, and leave the question of vote choice for future research. Previous results that demonstrate that ideological extremists who turn out often cast ballots differently and in an expressive manner at EP elections (Camatarri and Zucchini, 2019) help to justify this decision. Extreme parties, knowing this, may engage in greater mobilization efforts at an EP election time to turn out these voters. 5. We can only test our theory on reported turnout. While in some countries it is possible to obtain data for validated turnout (e.g. the UK), EP election surveys only offer information on reported voting. However, we have theoretical reasons to believe that this problem does not significantly affect our results. The nature of second-order elections-where people participate less on average-means that voters should face lower social desirability bias. Fewer voters ought to falsely report turnout compared to national elections. This is true for all voters, and we have no theoretical reasons to expect extremists to be more inclined to falsely report voting than moderates. If anything, the opposite may well be true-that moderates falsely report turnout at higher levels than extremists. Therefore, if using reported turnout had any effect on our estimates, these effects should go in the opposite direction of what we suggest in this study. 6. Given the difficulty to capture different educational levels across countries, the inclusion of perceived social class and employment status also captures, at least to some extent, the possible effect that education has on an individual's life course.