Second-order effects or ideational rifts? Explaining outcomes of European elections in an era of populist politics

Abstract This article seeks to enhance our understanding of the European Parliament (EP) elections in an era of populist and anti-European Union (EU) politics. Specifically, it aims to evaluate both the conventional second-order elections theory as well as an alternative approach that regards EP elections as an arena for conflict between liberal-democratic Europeanism and populist, extremist and euroskeptic alternatives. It does so by deriving a series of hypotheses from both approaches and testing these with party-level data from all EU member states in the context of 2019 EP elections. Our results challenge both explanations. Party size is a robust predictor of electoral performance in EP elections, and its effect is moderated by electoral system design. While large parties lost votes across the EU, their losses were more pronounced in countries where national legislatures are elected under plurality or mixed systems. We find no evidence of incumbent losses or electoral cycle effects. Party-level populism, extremism and euroskepticism did not systematically predict electoral performance but party ideology appears to have moderated the effects of incumbency and party size. Incumbency was associated with vote gain among populist and far-right parties but not other parties, and the effect of size also varied across party ideologies. In sum, these results suggest that vote fragmentation in the 2019 EP elections is partly explained by electoral system design, while it was not driven by the desire to punish political incumbents. Populist and far-right parties in power appear to be particularly immune to punishing behavior often associated with EP elections.


Introduction
The rise of populism has arguably been one of the most important developments in European politics over the last two decades. Defined as a set of ideas that emphasizes antagonism between the people and the elite and privileges the will of the people, populism is a thin ideology embraced by diverse political actors across the political spectrum (Mudde, 2004;Mueller, 2016). Since the turn of the century, electoral support for populist parties in Europe has more than tripled, and the number of Europeans who live in a country that has a populist party in government increased tenfold between 1998 and 2018 (Lewis et al., 2018). While the success of populist actors is linked to the multiple crises that have shaken European politics (Kriesi and Pappas, 2015), it is also attributed to broad structural changes in European societies (Betz, 1994;Mair 2013;Norris and Inglehart, 2019). This implies that populism may be the new normal, as opposed to a temporary crisis-induced deviation from liberal democracy.
The populist tide has not spared European Union (EU) politics. Because of the symbiotic relationship between populism and euroskepticism , the rise of populist parties has the potential to undermine the European integration project. The 2014 elections to the European Parliament (EP) constituted a 'great leap forward' for the populists who won a quarter of all seats (Luo, 2017;Martín-Cubas et al., 2019) while traditional liberal, conservative and social-democratic parties suffered losses. The astonishing success of euroskeptics in these elections lent support to the argument that European integration has become politicized, constituting an important new divide in European societies (Hooghe and Marks, 2009;Kriesi, 2016). In the 2019 European contests, populists and euroskeptics largely held the ground gained in 2014. The European Parliament became more fragmented than ever before, with the two largest political groups for the first time controlling less than half of the seats. However, populist performance varied by ideology. While right-wing populists made marked gains, including in the large member states such as Italy, France and the UK, left-wing populists registered significant losses (Mudde, 2019). The dynamics of electoral support for populist, euroskeptic and extremist groups also varied greatly across member states. Thus, the picture that emerges from the 2019 elections is one of complexity and fragmentation (Bolin et al., 2019).
These developments have intensified debates about what EP elections are and how they work (van der Brug et al., 2016;Nielsen and Franklin, 2017;Schmitt and Teperoglou, 2019). The conventional understanding is that EP elections are second-order national contests that revolve around national political concerns (Reif and Schmitt, 1980). This view is challenged by the large and diverse literature on the salience of European integration and the prevalence of EU issue voting (e.g. Carrubba and Timpone, 2005;Hobolt, 2015;Hobolt and de Vries, 2016;van Elsas et al., 2019;Angelucci et al., 2020), as well as theoretical contributions emphasizing the politicization of European integration (Hooghe and Marks, 2009;Kriesi, 2016). More recently, the rise of populism, extremism and euroskepticism has been regarded as a death bell to the permissive consensus that characterized European integration for many decades. In this context, a re-assessment of existing theoretical models of EP elections is in order.
This article seeks to enhance our understanding of EP elections in the era of populist and anti-EU politics. Specifically, we aim to evaluate both the familiar second-order theory and a rival view that regards EP elections as an independent, EU-wide arena for contestation between antagonistic political ideas in light of evidence from the most recent EP elections. We derive empirical predictions from both approaches and test these with party-level data from all 28 EU member states in the context of the 2019 EP elections. Our results challenge both explanations. While we find no systematic evidence of voters punishing political incumbents, party size emerges as a consistent negative predictor of electoral performance in EP elections relative to preceding national elections. The effect of party size is moderated by the institutional setting: large party losses are more pronounced in countries that use plurality or mixed electoral systems in national elections compared to countries that use proportional representation in both national and EP elections. Party-level populism, extremism and euroskepticism did not systematically predict performance in EP elections but party ideology appears to have moderated the effects of incumbency and party size.

Second-order effects or ideational rifts?
Does the second-order elections theory that was developed in response to the first direct elections to the EP in 1979 still apply four decades later? Much has changed in the realm of European integration, as well as in party politics. With a societal consensus on the desirability of European integration now an increasingly distant memory, and liberal democracy under attack by various challengers, have EP elections become an independent arena on which the battle between liberal-democratic Europeanism and opposing ideologies is fought out? This section revisits the familiar second-order thesis before outlining an alternative explanation of electoral outcomes that focuses on ideational party characteristics.

Second-order national elections
The second-order national elections thesis, formulated by Reif and Schmitt (1980), has dominated the literature on EP elections for nearly four decades. The theory argues that because EP elections do not lead to government formation at the national level, they are viewed as less important by voters, parties and the media. The main symptom of the lesser importance of these elections is low turnout. Furthermore, these elections are national, not European: they are simultaneous nationally-organized contests between national parties, revolving around domestic issues and cleavages. When EP elections produce results that differ from those of firstorder elections, this happens because EP elections take place in different phases of the first-order election cycle and because the lesser importance of these elections has implications for the behavior of both voters and parties.
A key behavioral implication of the second-order setting is that voters are less likely to act strategically, i.e. they are less concerned with the expected value of the election outcome. Thus, second-order elections are associated with an increased prevalence of sincere and protest votingvoters are more likely to vote with the 'heart' and the 'boot' than with the 'head' (Oppenhuis et al., 1996). Sincere voting means 'supporting parties less relevant to government formation but closer to the voters' ideal positions' (Hix and Marsh, 2007, p. 497). While a closer correspondence of the vote to the true preferences of citizens may be considered desirable from the perspective of democracy, the SOE approach associates sincere voting with affective behavior which translates into electoral gains for small and fringe parties, and losses for large mainstream parties. Protest voting means that voters use their vote to send signals to political incumbents (Hix and Marsh, 2007). This means withholding support from government parties, either by abstention or by switching vote to the opposition. Because dissatisfaction with governments tends to build over time, the performance of political incumbents depends on the phase of the first-order election cycle (Reif and Schmitt, 1980;Reif 1984). Incumbent losses are expected to be most pronounced in the middle of the national electoral cycle, as opposed to soon after or immediately preceding first-order elections.
In sum, second-order national elections are associated with a clear set of aggregate-level symptoms, including low turnout, electoral losses of government parties and large parties, electoral gains of opposition parties and small parties, and variation in the magnitude of incumbent losses depending on the phase of the electoral cycle. The predictions of the SOE model have been extensively and systematically tested, and mostly corroborated, with aggregate EP election data spanning four decades (e.g. Reif and Schmitt, 1980;Marsh, 1998;Schmitt, 2005, Hix and Marsh, 2007Schmitt and Toygür, 2016). Individual-level studies that confirm voters' pre-occupation with national issues and concerns, as well as proclivity to punish political incumbents, lend additional support to the SOE thesis (e.g. Schmitt et al., 2008;Clark and Rohrschneider, 2009;Hobolt and Wittrock, 2011).
Altogether, the SOE model yields the following predictions about the drivers of aggregate outcomes in European elections: Hypothesis 1: Government parties lose votes compared to preceding national elections ('incumbent losses'); Hypothesis 2: Large parties lose votes compared to preceding national elections ('large party losses'); Hypothesis 3: Electoral losses of government parties are more pronounced when EP elections occur in the middle of the first-order electoral cycle ('cycle effects').
Responding to recent calls to pay more attention to possible electoral system effects in the study of voting behavior in EP elections (Farrell and Scully, 2005;Prosser, 2016), we add an additional expectation that the effect of party size on vote gains or losses is modified by the type of electoral system used in national elections. While EP elections are held under a system of proportional representation (PR), some EU member states use non-PR systems, including plurality, two-round or mixed systems in national elections. Because plurality or mixed systems may prompt voters to strategically vote for larger parties or disincentivize small parties to form or run (Duverger, 1954;Blais and Carty, 1991), a fair comparison between vote shares at national and EU elections should take into account whether national and EP elections are held under broadly similar rules (PR) or not. Thus, we include the following auxiliary hypothesis: Hypothesis 2a: The negative effect of party size on electoral performance in EP elections, relative to preceding national elections, is stronger in countries where national legislatures are elected under a plurality or mixed system compared to countries that use PR in both elections ('electoral system effects').
Finally, in line with our focus on evaluating theories in the context of allegedly growing ideological contestation, we include the expectation that the effects of the SOE variables are stable across party ideologies: Hypothesis 4: Effects in Hypotheses 1-3 do not vary across ideational party characteristics such as populism, extremism and euroskepticism ('SOE effects stable across party ideology').

EP elections as a sui generis arena for political contestation
Recent developments in European integration have cast serious doubts on the continued validity of the argument that EP elections are merely an additional, secondary arena for domestic political contestation. With the expansion of EU powers and policy responsibilities, the permissive consensus (Lindberg and Scheingold, 1970) that characterized public attitudes towards European integration for decades has been replaced by a 'constraining dissensus' (Hooghe and Marks, 2009). Public attitudes towards the EU have become increasingly well defined as well as polarized, and political entrepreneurs keen to profit from popular dissent have deliberately cultivated conflict over Europe, linking it to pre-existing political cleavages and framing integration in ways they consider electorally advantageous (Kriesi, 2016;Braun, Popa, and Schmitt, 2019). As a result, European integration has become politicized: it constitutes a new political divide that structures public opinion and party competition, shaping voting behavior in elections and referenda (Hix, 1999;Marks and Steenbergen, 2004;Kriesi, 2007Kriesi, , 2016Hobolt, 2009;Hooghe and Marks, 2009). The numerous challenges that Europeans have experienced over the past decade, including the eurozone crisis, the refugee crisis, terrorism, geopolitical tensions and Brexit, have served as catalysts for the politicization of integration (Kriesi and Pappas, 2015;Hutter and Kriesi, 2019). In sum, while the EU could be seen as a 'polity without politics' in the past (Schmidt, 2006), many scholars believe that such a characterization is no longer justified.
Both the extent and the structure of political conflict over Europe have been extensively debated. In particular, the discussion has focused on how the pro-and anti-EU divide is related to the conventional left-right divide (Hix, 1994;Marks and Steenbergen, 2004), as well as to the Green-Alternative-Libertarian and Traditional-Authoritarian-Nationalist (GAL-TAN) axis (Hooghe et al., 2002;Marks et al., 2006). The relationship between the EU divide and other political cleavages appears to vary greatly across regions, individual member states and time (Hooghe and Marks, 2009). This is the case because the politicization of European integration is embedded in diverse national political conflict structures and driven by diverging experiences of the member states with recent crises (Kriesi, 2007(Kriesi, , 2016. However, broadly speaking, euroskepticism has a symbiotic relationship with extremism of both the far-right and the far-left variety, as well as with populism (e.g. Rooduijn and van Kessel, 2019).
The question of what the politicization of European integration implies for second-order elections theory requires careful consideration. While the politicization argument challenges the view that European issues do not matter to voters, it does not refute the primacy of the first-order arena or the notion that EP elections simply mirror first-order dynamics. As pointed out by Kriesi (2016, 32), national politics are the crucial arena for the politicization of European integration due to the 'weakness of the partisan channel of representation at the European level'. Political entrepreneurs who see electoral advantages in politicizing integration will focus their efforts, above all, on the national level. Politicization, thus, means incorporating the issue of European integration into national politics but does not suggest that EP contests become important in their own right.
A more fundamental challenge to the SOE theory would assert that EP elections constitute a sui generis arena for political contestation. Such an assertion calls for demonstrating that aggregate outcomes of EP elections cannot be reduced to a combination of party performance in firstorder elections and the erosion of incumbent support as a function of 'political time'. On the party level, this would mean that party gains and losses in EP elections relative to preceding national elections vary as a function of ideology, programmatic differences or policy positions, or reflect pan-European political dynamics. On the individual level, vote should be driven by attitudes towards European integration, support for EU institutions, or positions on EU issues, rather than evaluations of the national government or other domestic preferences. Given the recent importance of European issues at the national level it might be hard to demonstrate that EP elections that evince conflict over Europe are not simply showing their SOE character at a time of national politicization of European issues.
Still, existing studies lend some support to these expectations. The extensive literature on EU issue voting suggests that EU attitudes are an important determinant of vote choice in EP elections (Carrubba and Timpone, 2005; van Spanje and de Vreese, 2011;Hobolt, 2015;Hobolt and de Vries, 2016;van Elsas et al., 2019). Issue voting in EP elections is enhanced by political campaigns that inform and persuade voters about candidates, policies and performance (Beach et al., 2018), as well as by the availability of information on European integration (Hobolt and Wittrock, 2011). Thus, information and campaigns make EP elections less second-order, enabling voters to become more aware of their own and the parties' EU attitudes and rely on these attitudes as opposed to national heuristics (Beach et al., 2018). 1 On the aggregate level, attempts to identify 'European effects' in EP elections distinct from second-order effects have focused on pan-European shifts in the behavior of voters towards or away from particular party families (Hix and Marsh, 2011). The argument is that comparatively strong or poor performance of a particular party family in EP elections, independent from government status and size, would occur as the result of European-wide policy preferences at a particular time point, or because voters hold one level of government more responsible for the given area than the other (Hix and Marsh, 2011, 10). Analyzing aggregate data from seven EP elections, Hix and Marsh identify several such pan-European swingse.g. for green parties, away from socialist partiesand argue that such reactions to common policy concerns could be regarded as 'the first step in the evolution of European Parliament elections into genuine European-wide votes about the direction of the EU policy agenda' (ibid.).
Our analysis builds on Hix and Marsh (2011) in focusing on ideational party characteristics and the differential performance of party families across different types of elections. Instead of conventional party families, we focus on what is arguably the most prominent ideological fault line in current European politicsthe clash between liberal democratic Europeanism, on the one hand, and a populist-nationalist-sovereignist opposition that seeks to restore autonomy, identity and control, on the other. Building on the notion of pan-European swings, we argue that systematic EU-wide electoral gains or losses of populists, extremists and euroskeptics in EP elections, controlling for government status, party size and electoral cycle effects, would challenge the second-order elections theory, while lending tentative support to the argument that EP elections are, at least partly, a sui generis political arena. 2 In sum, the argument about the importance of ideational rifts yields the following predictions about aggregate outcomes in European elections: Hypothesis 5: In comparison to other parties, Euroskeptic parties gain or lose significantly more votes in EP elections compared to preceding national elections ('euroskeptic gains/losses'); Hypothesis 6a: In comparison to other parties, far-right parties gain or lose significantly more votes in EP elections compared to preceding national elections ('far-right gains/losses'); Hypothesis 6b: In comparison to other parties, far-left parties gain or lose significantly more votes in EP elections compared to preceding national elections ('far-left gains/losses'); Hypothesis 7: In comparison to other parties, populist parties gain or lose significantly more votes in EP elections compared to preceding national elections ('populist gains/losses').
In order to rule out any confounding of ideational effects with second-order effects, we also posit the following hypothesis: Hypothesis 8: Effects in Hypotheses 5-7 do not vary across incumbency, party size and the phase of the electoral cycle. ('Ideological effects stable over SOE variables').

Data and methods
We constructed a database covering all parties and electoral alliances in 28 EU member states that (a) competed in both the 2019 EP election as well as the preceding national parliamentary election and (b) obtained at least 2% of the vote in the national election. Because our dependent variable is defined as difference in vote shares obtained in EP and national election, parties that competed in only one election but not the other were dropped from the analysis. Cases involving membership in electoral alliances were examined one-by-one to determine if the case can be meaningfully included in the analysis or must be coded as missing. When parties competed separately in one election and as part of an alliance in the other election, vote shares of individual parties were added to calculate the vote difference. However, cases with complex shifts of alliance membership had to be dropped from the analysis because it was not possible to calculate the dependent variable. The final number of parties and electoral alliances included in the analysis is 192.
The dependent variable, difference in vote shares, is calculated by deducting the vote share obtained by the party in national elections from its vote share in the 2019 EP election. Information about parties' electoral performance is obtained from the ParlGov database (Döring and Manow, 2019) which covers all EU and most OECD democracies. On average, 2 A possible objection to this argument is that the gains or losses of populists, extremists and euroskeptics in EP elections, relative to national elections, could be due to a 'time effect'i.e. some development or event that occurs before EP elections that sparks a vote for these parties (which would also have been expressed on a different electoral arena had it been available). While this is valid point that deserves to be examined further with data spanning multiple rounds of national and EP elections, we note that the idea of systematic EU-wide vote swings in response to some development or event are not consistent with the original SOE depiction of EP elections as a predominantly 'national affair.' the parties included in our analysis lost 0.8 percentage points of the vote compared to preceding national elections: the values of the variable range from −33.4 (the Conservative Party in the UK) to +19.5 (Unity in Latvia).
The analysis employs separate dummy variables for populist, far-right, far-left and euroskeptic parties, as well as a general populist-extremist-euroskeptic (PEES) dummy for parties that belong to at least one of these categories. Our classification is based on the PopuList database, a resource providing an overview of populist, far-right, far-left and eurosceptic parties in Europe . This classification of parties is based on widely-accepted definitions developed by leading experts in the field. Thus, populism is defined as an endorsement of the idea that society is separated into two homogeneous and antagonistic groups, 'the pure people' and 'the corrupt elite' and that politics should be an expression of the 'general will' of the people (Mudde, 2004). The definition of far-right parties combines a subscription to a nativist ideology that regards nonnative elements as fundamentally threatening with an endorsement of authoritarianism, defined as a belief in a strictly ordered society (Mudde, 2007). The definition of far-left employed by the PopuList builds on March (2012) and includes parties that reject capitalism and advocate alternative economic and power structures as well as redistribution of resources from existing political elites. Finally, consistently with Taggart and Szczerbiak (2004), euroskepticism is defined as entailing both the soft and hard varieties, i.e. including parties that express either contingent or qualified opposition or outright and unqualified opposition to the process of European integration (see . The coding of parties in the PopuList has been peer-reviewed by more than 30 academics. The PopuList is supported by the Amsterdam Institute for Social Science Research, The Guardian and the ECPR Standing Group on Extremism and Democracy. Based on the PopuList data, out of the 192 parties considered, we classify 41 parties as populist, 23 as far-right, 22 as far-left and 52 as euroskeptic. The categories partially overlap: several parties are simultaneously extremist, populist and anti-EU (the implications of this overlap for our regression analysis will be explained below). Among all EU member states, Italy and Slovakia had the largest number of populist parties (4) We include in the analysis a dichotomous variable for incumbency, coded as 1 for parties that belonged to the national government at the time of the 2019 EP elections, and as 0 otherwise. The coding is based on information available in the ParlGov database. Of the 192 parties included in the analysis, 59 (30.7%) were in government at the time of EP elections. Of the 41 populist parties in our sample, 12 were in government. Of the 23 far-right parties, 6 were political incumbents. Only one far-left party (out of 22) was in government, while 9 out of the 52 euroskeptic parties held executive power.
We follow previous studies (e.g. Hix and Marsh 2007) in using vote share in national elections as a proxy for party size. The variable ranges from 2% to 55%, with a mean value of 13.1%. In order to test the hypothesis that the effect of party size on vote difference is moderated by the electoral system, we include a dummy variable that is coded 1 for countries that used proportional representation in preceding national elections and 0 for countries where national elections were held under a plurality or mixed system (UK, France, Hungary, Italy, Lithuania).
Finally, we include a variable for the electoral cycle, operationalized as the percentage of the first-order electoral cycle completed by the time of the 2019 EP elections. The observed values of the variable range from 1.9% in Spain to 100% in Belgium, with a mean value of 48.5%. 3 Because the expected effect of the electoral cycle on party performance is curvilinear, we follow previous studies (e.g. Marsh 1998) in adding the square of the variable to our equations.
To test our hypotheses, we regress the difference in party vote share in EP and preceding national elections on populism, right-wing and left-wing extremism, euroskepticism, party size, incumbency, electoral system, electoral cycle, electoral cycle squared and a number of interaction terms. To account for the nested structure of our data, we estimate multi-level regression models with parties (192) nested in countries (28).

Results
The results of the multi-level regression models are reported in Table 1. The first model includes the variables central to the SOE theory: incumbency, party size, the two measures of electoral cycle and an interaction term that isolates cycle effects for parties in government. It also includes the dummy for PR systems and an interaction term between party size and the electoral system. To confirm the SOE theory, we would need to find statistically significant negative incumbency and size effects, as well as a U-shaped effect of the electoral cycle on the electoral performance of government parties. These expectations are only partly confirmed in Model 1 in Table 1. As expected in Hypothesis 2, party size has a statistically significant negative effect on the dependent variable, suggesting that the higher a party's vote share in national elections, the greater its losses in EP elections. Predictive margins (Table A1 in Appendix) indicate that the smallest party in the sample has a likelihood of increasing its vote share by 2.1 percentage points from one election to another, while the largest party is likely to lose 11.5 percentage points. The interaction term for party size and the national electoral system has a statistically significant effect, confirming the expectation (Hypothesis 2a) that large parties suffer less in countries where national elections are held under proportional rule (Figure 1). Predicted vote difference (Table A1 in Appendix) varies from 1.8 percentage points for the smallest party to −9.6 for the largest party in countries that use PR in both national and EP elections, while the variation is much larger (from 3.3 to −20.5 percentage points) in countries where national elections are conducted under a plurality or mixed system. To account for the possibility that party size effect could be at least partly mechanical, considering that larger parties have a greater range of possible negative values in the dependent variable than smaller parties, we conduct an additional robustness check, running the model with a relative, as opposed to the absolute percentage change in vote shares as the dependent variable. The results show that party size remains a consistent predictor of electoral performance across different model specifications (Table 1b in Online Appendix).
Aside from party size effects, key expectations associated with the SOE model are not supported by the results of our analysis. The results in Model 1 in Table 1 suggest that incumbency and electoral cycle were not statistically significant predictors of the difference in party performance between the 2019 EP elections and preceding national elections. This does not allow us to confirm Hypothesis 1 and Hypothesis 3. These findings challenge the SOE expectation that voters use EP elections to send signals to parties in government while corroborating the argument that in EP elections, voters abandon large parties in favor of small and marginal political actors.
Turning to the proposition that party performance in EP elections, relative to preceding national elections, varies as a function of party ideology, we begin by examining descriptive results. The average vote share of PEES parties in the 2019 EP contests was 2.3 percentage points lower than in the preceding national election. The electoral performance on non-PEES parties, at the same time, remained virtually unchanged at −0.1 percentage points. Populist parties saw their vote share drop by 2.9 percentage points on averagefrom 15.6% of votes in national to 12.7% in EP elections. In total, 33 out of the 41 parties were classified as populist in our sample lost votes between the two elections, with the Italian Five Star Movement losing and the North League winning the most (−16.2 and +16.9 percentage points, respectively). With an average vote share of 12.9% in national and 10.7% in EP elections, euroskeptic parties lost 2.2 percentage points from one election to another. The largest losses in this group fell upon the Conservatives in the UK and the largest gains on the North League in Italy (−33.4 and +16.9 percentage points, respectively). Among extremists, far-left parties did worse than the far-right. The former lost 1.7 percentage points of the vote (from 9.9 in national to 8.2% in EP elections). The largest decrease occurred for SYRIZA in Greece and the largest increase for the Socialist People's Party in Denmark (−11.7 and + 9 percentage points, respectively). The far-right mostly held its ground, as its losses amounted to only 0.8 percentage points (from 14.9% of the vote in national to 14.1% in EP elections). The heaviest losses were suffered by Jobbik in Hungary while the biggest winner was the Italian North League (−12.8 and +16.9 percentage points, respectively). In sum, descriptive results show that PEES parties performed worse in EP elections than in national elections. The fact that many PEES parties were also large government parties, however, highlights the need for multivariate analysis.
We continue our exploration of the effects of party ideology by estimating a series of regression models that control for the effects of incumbency, party size and the electoral cycle. As populism, extremism and euroskepticism often coincide at the party level, our four measures of party ideology cannot be included in the same model. Instead, we report the results of four models, each including Notes: Entries are coefficients from multi-level regression models with standard errors in parentheses. *** P < 0.01, ** P < 0.05, * P < 0.10.
the baseline SOE variables as well as one of the ideational markers (Models 2-5 in Table 1). We find that party-based populism, euroskepticism and extremism do not have statistically significant effects on electoral performance in EP elections when incumbency, party size and the electoral cycle are controlled for. Our results, thus, do not lend support to Hypotheses 5-7. Next, we turn to assess Hypothesis 4 which posits that the effects of incumbency, size and cycle do not vary across ideational party characteristics, as well as the related hypothesis that the effect of party ideology is stable across values of the SOE variables (Hypothesis 8). We estimate a series of regression models that interact incumbency, size and cycle with PEES, our umbrella category for populists, extremists and euroskeptics (Table A2 in Appendix). The effect of the interaction term between the PEES dummy and incumbency status is positive but not statistically significant (Model 2 in Table A2 in Appendix). When we repeat the analysis for populist, extremist and euroskeptic parties separately (Table A3 in Appendix), we find statistically significant positive effects for interaction terms between populism and incumbency as well as a far-right ideology and incumbency ( Figure 2). In other words, populist and far-right parties in government did significantly better in EP elections relative to preceding national elections, compared to populist and far-right parties in opposition. For parties that are not populist or far-right, electoral success did not depend on incumbency status. These results challenge the expectation that the effect of party ideology is stable across SOE variables and vice versa (Hypothesis 4, Hypothesis 8).
Party size affects the electoral performance of PEES and non-PEES parties in a roughly similar manner, as the respective interaction term is not statistically significant (Model 3 in Table A2 in Appendix). When repeating the analysis for populist, right-wing and left-wing extremist and euroskeptic parties separately (Table A4 in Appendix), however, we find that interaction terms for populism and party size and for far-right ideology and party size have statistically significant positive effects (Figure 3). The negative effect of party size on electoral performance is much weaker for populist parties than for other parties. Among far-right parties, party size is associated with an increased vote share in EP electionsa finding that defies the second-order logic. These results suggest that second-order effects are not stable across party ideologies, challenging Hypothesis 4 and Hypothesis 8.
Finally, we test the proposition that cycle effects are independent of party ideational characteristics. To test this, we estimate a model (Model 4 in Table A2 in Appendix) with a three-way interaction between incumbency, PEES status and the electoral cycle. The results indicate that while government parties score a little lower on the dependent variable, their electoral Notes: Entries are average marginal effects with 95% confidence intervals.
performance is not statistically different from that of opposition parties. These tendencies do not vary across the electoral cycle, as the SOE model would suggest, nor across party ideologies (see also Table 2b in Online Appendix). This is in line with Hypothesis 4 and Hypothesis 8.
In sum, our results lend clear and consistent support to Hypothesis 2 which posits that party size is negatively associated with electoral gains in EP elections. The effect of size is remarkably stable across our models. The negative effect of party size is particularly pronounced in countries where national legislatures are elected under a plurality or mixed electoral system (Hypothesis 2a). We find no support to the proposition at the core of the SOE theory according to which government parties lose votes in EP elections. This does not allow us to confirm Hypothesis 1. Hypothesis 3 is also not confirmed because our analysis did not reveal any electoral cycle effects. Our analysis yields mixed results with regard to the effects of party ideology. While partybased populism, extremism and euroskepticism are not statistically significant predictors of electoral performance in EP elections when party size, incumbency and cycle are controlled for, party ideologies appear to moderate second-order effects. Specifically, incumbency and party size effects on electoral success seem to be moderated by populism and right-wing extremism. These results do not allow us to confirm either the hypotheses positing ideological effects (Hypotheses 5-7) or the expectation that SOE effects are stable across party ideologies and vice versa (Hypothesis 4, Hypothesis 8).  Notes: Entries are average adjusted predictions with 95% confidence intervals.

Discussion
This article set out to explore the question of whether EP elections can still be characterized as second-order national contests in the context of the growing prominence of populist, euroskeptic and extremist challenges to liberal-democratic Europeanism. While the symptoms of the second-order nature of EP elections have been firmly established in the literature, determining what kind of aggregate-level electoral outcomes would challenge the second-order theory required more effort. Demonstrating that European issues play a significant role in EP elections does not constitute a sufficient basis for rejecting the claim that EP elections are second-order contestsunless it is shown that European issues did not matter in preceding first-order elections. In other words, the SOE theory expects EP elections to echo national elections, regardless of what issues were or were not salient in these elections. If the politicization of European integration occurs, above all, on the national arena, and is driven by national political actors, as the politicization literature unequivocally argues, EP elections may well remain second-order even if they have 'European content'. Refuting the second-order argument thus requires demonstrating that EP elections do not simply mirror the political realities of the first-order elections but constitute, at least to an extent, a sui generis arena for political contestation. Such an assertion calls for demonstrating that aggregate outcomes of EP elections cannot be reduced to a combination of party performance in first-order elections and the erosion of incumbent support as a function of 'political time'. A finding that there are EU-wide swings in electoral support for groups of parties distinguished by ideational characteristics would lend tentative support to the sui generis argument.
This article tested the ideal-typical second-order and sui generis models of EP elections with party-level data from the 2019 EP elections as well as preceding national elections. Our empirical test focused on establishing the importance of second-order effects (incumbency, party size and the electoral cycle) as well as ideational party characteristics (populism, extremism and euroskepticism) in driving differences in party vote shares across the two elections. Our results challenge the second-order theory in that we find no evidence of incumbency or electoral cycle effects: in the 2019 EP contests, voters across Europe did not appear to punish political incumbents, irrespective of the timing of the EP election in the national electoral cycle. Our findings, however, corroborate the expectation, central to the SOE approach, that party size is negatively associated with electoral performance in European elections compared to preceding national elections. The fact that this finding is robust across different operationalizations of the dependent variable suggests that the negative effect of party size is substantive rather than just mechanical. Finally, this study went beyond conventional tests of the SOE model by demonstrating that the effect of party size on electoral performance varies systematically across electoral systems, with large parties losing more votes in countries where national elections are held under plurality or mixed systems as opposed to proportional representation. The robustness of this effect strongly suggests that future tests of the second-order model that measure electoral performance in terms of difference in party vote shares across national and EP elections should control for electoral system effects.
Our results do not support the conjecture that electoral gains and losses in EP elections are driven by ideational party characteristics. While populists, extremists and euroskeptics, on average, lost votes compared to preceding national elections, party ideologies did not systematically predict electoral performance when conventional second-order variables were controlled for. These results challenge the proposition that EP elections constitute a privileged arena on which the conflict between the liberal-democratic pro-integrationist mainstream and a populist-extremist-euroskeptic opposition is played out. Instead, these findings appear to be more consistent with the argument, central to the politicization literature, that conflict over European integration is articulated and waged, above all, on the national political arena.
However, our results suggest that party ideologies matter because they moderate the effects of party size and incumbency on party performance in EP elections. The finding that large populist parties lose, on average, a smaller share of the vote than large non-populist parties, while among far-right parties, party size is positively associated with electoral gains, challenges the second-order model. These findings imply that the voters of populist and far-right parties are, for whatever reason, less affected or unaffected by the prevalent impulse to abandon large parties in EP elections. They also suggest that the shift of populist and far-right actors from the fringes of national politics to the mainstream and from opposition to government, if continued, is likely to have significant implications for the composition of the European Parliament and possibly for the future of European integration. Thus, it appears that one cannot count on the party size effectthe only mechanism of second-order model that systematically worked in 2019 according to our resultsto serve as a check on the rise of populists and right-wing extremists.
Our results point to four promising directions for future research on European elections. First, they suggest that such research should pay closer attention to how differences in electoral system design, both across countries and across different types of elections, influence voter behavior and the fortunes of political parties. In doing so, future studies should take into account both mechanical as well as psychological effects of electoral systems. Second, more research is needed to understand the causes of vote fragmentation which this and other studies have identified as the dominant trend in the 2019 European elections. While voters across Europe challenged the political status quo by abandoning parties they voted for in national elections, there appears to be no single explanation for why they did so. Overall, fragmentation of the vote with no single underlying driver is consistent with the literature that emphasizes the heterogeneity of voters, motives and contexts in a complex multi-level political system (Hooghe and Marks, 2009). However, the loss of structure may also be interpreted as being consistent with the notion of sincere voting, associated with the second-order thesis. Third, aggregate-level tests of the proposition that EP elections are sui generis, with vote driven by different factors than in national elections, are limited in many ways, notably by their inability to rule out 'time effects' that may lead voters to change their preferences. Case studies focusing on countries that have held national and EP elections on the same day or in close temporal proximity to one another, along with studies employing survey data about vote intention in hypothetical simultaneous elections, could cast more light on the issue. Finally, our results point to the need to better understand the role of party ideologies and ideological conflicts in EP elections. Instead of regarding the SOE model and ideational explanations as contending alternatives, future studies may want to examine whether, how and with what consequences ideologies and second-order effects interact. Such an agenda appears promising, given that both the SOE model and the expanding body of knowledge on populism, extremism and euroskepticism share a common focus on popular discontent and its electoral manifestations.
Funding. The research has been funded by the European Commission (grant agreement 822682) and by the Estonian Research Council (project IUT20-39).
Cite this article: Ehin P, Talving L (2020). Second-order effects or ideational rifts? Explaining outcomes of European elections in an era of populist politics.