It All Happens at Once: Understanding Electoral Behaviour in Second-Order Elections

The second-order election (SOE) model as originally formulated by Reif and Schmitt (1980) suggests that, relative to the preceding first-order election result, turnout is lower in SOEs, government and big parties lose, and small and ideologically extreme parties win. These regularities are not static but dynamic and related to the first-order electoral cycle. These predictions of the SOE model have often been tested using aggregate data. The fact that they are based on individual-level hypotheses has received less attention. The main aim of this article is to restate the micro-level hypotheses for the SOE model and run a rigorous test for the 2004 and 2014 European elections. Using data from the European Election Studies voter surveys, our analysis reveals signs of sincere, but also strategic abstentions in European Parliament elections. Both strategic and sincere motivations are also leading to SOE defection. It all happens at once. cycle; second-order elections; sincere voting; strategic voting; turnout; vote switching Issue


Introduction
Standing on the shoulder of giants, the second-order election (SOE) model was proposed by Reif and Schmitt (1980) in an effort to understand voter motivations and electoral outcomes in the first direct election of the European Parliament (EP) in 1979. Reif and Schmitt (1980) identified this supranational election as another case of a 'low stimulus election.' The roots of this stream of research go back to the US electoral context and here in particular to efforts to explain the typical losses of the presidential party in midterm elections (Campbell, 1966;Campbell, Converse, Miller, & Stokes, 1960). Beyond the US, there are other elections that do not directly or indirectly fill the prime executive post of a polity, be it that of a president, a prime minister or a chancellor. Among them are midterm elections in Latin America (see among others, Erikson & Filippov, 2001;Remmer & Gélineau, 2003;Thorlakson, 2015) but also all sorts of subnational elections-such as state elections in Germany (Dinkel, 1977(Dinkel, , 1978 or provincial elections in Canada (Erikson & Filippov, 2001).
In their effort to understand EP election outcomes and their difference to national first-order elections (FOEs) and over time, Reif and Schmitt (1980, 9-15) propose altogether six dimensions of variability. Among them, the 'less-at-stake' dimension is the most important. Based on this, the skeleton version of the SOE model suggests that, compared to both the preceding and the subsequent FOE result: (1) turnout is lower in SOEs; (2) government parties loose; (3) big parties loose too; while (4) small parties win; and (5) ideologically extreme parties and protest parties win. Moreover, these regularities are not static but dynamic-they are likely to be inflated and deflated again as a function of the firstorder national electoral cycle. Most important here is that (6) government parties' losses are greater the closer a SOE is located around the midterm of the first-order electoral cycle (Reif, 1984(Reif, , 1997Reif & Schmitt, 1980; but see also Stimson, 1976).
Most of the studies that followed the initial work of Reif and Schmitt focused on testing the aggregatelevel predictions of the SOE model. These hypotheses have mostly been corroborated (among others, Freire, 2004;Hix & Marsh, 2011;Norris, 1997;Reif, 1984Reif, , 1997Schmitt, 2005;Schmitt & Teperoglou, 2015;Schmitt & Toygür, 2016;Teperoglou, 2010;van der Brug & van der Eijk, 2007;van der Eijk, Franklin, & Marsh, 1996). However, aggregate electoral outcomes do not themselves speak about the micro-level processes causing them. Electoral results are silent about the determinants of electoral behaviour. In their article, Reif and Schmitt (1980) were not ignorant about individual-level processes, but their respective hypotheses have received less attention. This is not to say that nothing had been done in that regard: Some scholars started to uncover the micro-foundations of the SOE model (see in particular Carrubba & Timpone, 2005;Clark & Rohrschneider, 2009;Hobolt & Spoon, 2012;Hobolt, Spoon, & Tilley, 2009;Hobolt & Wittrock, 2011;Magalhães, 2016;Schmitt, Sanz, & Braun, 2009;Schmitt & Teperoglou, 2015, 2017Weber, 2011). However, a comprehensive test of the micro-level mechanisms of the SOE model is still lacking, resulting in the fact that we do not 'have a good understanding of the phenomenon' (Golder, Lago, Blais, Gidengil, & Gschwend, 2017). Against this background, we aim to contribute a succinct theoretical model for the comparative study of voting behaviour in FOEs and SOEs, and to test this model for two EP elections in all EU member countries and in very different political environments. In doing so, we provide a micro-model of inter-election voting patterns and study the most relevant of themabstaining and defecting in SOE-at the occasion of the 2004 and 2014 EP elections.

A Sketch of a Micro-Model of SOE Voting
We start from the observation that the behaviour of voters in consecutive first-and SOEs is far from independent. This is to say that some voters may support party a in election B because they have supported party b in election A (and because the two elections differ in importance, among other things). In order to sort this out more systematically, it might be useful to begin with a distinction of the behavioural alternatives a voter faces in consecutive FOEs and SOEs. There are five of them. A voter might: (1) vote for the same party in both elections; (2) defect from her or his FOE choice in the SOE; (3) abstain in the SOE while having voted in the FOE; (4) vote in the SOE while having abstained in the FOE; and (5) abstain in both. We claim that the second and the third of these behavioural alternatives are the most informative when analysing electoral patterns in EP elections because they are at the base of often markedly different results between the two types of elections. Of course, this does not mean that the same voting decision-such as supporting the same party or abstaining in two successive electionsmust originate in identical sets of factors causing them. We just propose that we are more likely to unearth characteristic motivational differences when we focus on different decisions of a voter-such as defecting from the earlier choice or abstaining although having voted previously.
It is the early insight of Angus Campbell (1966) that a certain group of voters, he calls them 'peripheral,' are likely to abstain in low stimulus elections due to a lack of mobilisation. These voters are characterised by little political interest and partisanship. As a result, this group of the electorate needs a particular stimulus to participate in an election, while others need less prompting because they either identify with one of the parties or habitually turn out on election day (Franklin, 2004;Schmitt & Mannheimer, 1991). Regarding the motivations of abstention, the conclusions from previous studies are inconsistent. Some claim that abstention is motivated by Euroscepticism (the so-called voluntary Euro-abstention; see Blondel, Sinnott, & Svensson, 1998;Wessels & Franklin, 2009). Others argue that SOE abstention reflects primarily a lack of politicisation and electoral mobilisation (e.g., Franklin & Hobolt, 2011;Schmitt & Mannheimer, 1991;Schmitt & van der Eijk, 2007). The latter explanation might have minor but significant repercussions for subsequent national FOEs (see for turnout Franklin & Hobolt, 2011; this has also been argued for vote choice by Dinas & Riera, 2018).
Some voluntary abstainers are guided by sincere motivations. This can happen as a result of a certain scarcity of arena-specific policy positions on offer (i.e., when no viable Eurosceptic parties are running; see van der Eijk & Franklin, 1996;Schmitt & van der Eijk, 2001). Other voluntary abstainers follow more instrumental motivations and use abstention as a form of protest-against governmental policies more narrowly conceived or, more fundamentally, against the political supply more generally. Such strategic abstentions are expected to increase around the midterm of the FOE cycle, when government approval tends to be plummeting (Gélineau & Remmer, 2006;Mughan, 1986;Stimson, 1976).
While sincere and strategic motivations may contribute to differential abstentions, they are at the same time the two main mechanisms responsible for interelection vote switching. 'Sincere voting' is when citizens vote for the party (or candidate) that is most attractive to them, either by way of socialisation and habit (party identification), policy considerations (position issues or valence issues and political competence), or candidate traits. By way of contrast, a 'strategic vote'-whatever its motivation might be-is indicated by the support of another than the first-preference choice option (much of the relevant literature here is initiated by Cox, 1997). Sincere and strategic voting is, again, not a behavioural constant but a variable. Voters may change their mode of operation from sincere (in one electoral arena) to strategic (in another) and vice versa. Reif and Schmitt (1980) distinguish two versions of sincere switching. The first one is switching from a strategic FOE to a sincere SOE vote choice. This can be motivated by the desire of voters to express their true first preference (in any kind of election) even if the chosen party is not expected to gain representation (i.e., when the first preference is a small party). This version of switching is facilitated by the fact that the likely political consequences of such a choice are limited in an EP election because there is less at stake. The second type of sincere vote switching occurs when SOE defectors sincerely support another than their FOE party due to arenaspecific issues and policies. In the case of EP elections, voters may prefer one party on European and another on domestic politics and sincerely support different parties in the two electoral arenas. Running somewhat against the original SOE model, this idea is substantiated by a growing literature focusing on European policy and polity issues as explanatory factors of vote choices in EP elections (e.g., Clark & Rohrschneider, 2009;de Vries, van der Brug, van Egmond, & van der Eijk, 2011;Hix & Marsh, 2007;Hobolt & de Vries, 2016;Hobolt & Spoon, 2012;Hobolt et al., 2009;Hobolt & Wittrock, 2011).
The analysis of the different electoral patterns between FOE and SOE and in particular vote switching in EP elections revealed a specific type of strategic voting, the so-called protest voting or voting with the boot (Reif & Schmitt, 1980;van der Eijk et al., 1996). Strategic defec-tors in EP elections signal discontent at the occasion of a SOE with their government's performance in the firstorder electoral arena. These signals can be issued in a harder and a softer form. The harder version is actual defection (former FOE government voters support an opposition party in a subsequent SOE); the softer version manifests itself in differential abstention (former FOE government voters abstain in a subsequent SOE). This being said, we note that (strategic) protest voting is likely to boost around the midterm of the first-order electoral cycle. During the 'honeymoon' period shortly after a FOE, government parties will receive greater or near identical support in an EP election (e.g., Reif & Schmitt, 1980). Regarding the later term of the national electoral cycle, some argue (e.g., van der Eijk & Franklin, 1996) that the closer a SOE falls to the next FOE the more likely voters are to opt for 'strategic defection,' while the original statement of the model expects a certain recovery of support for national government parties.

Micro-Level Hypotheses about Inter-Election Voting Patterns
Considering that we analyse voter data from half a hundred different electoral contexts (23 in 2004, and 28 in 2014), we expect a multitude of inter-election voting patterns. While some SOE abstainers might lack mobilisation, others follow sincere and still others strategic motives in abstaining. SOE vote switchers are expected to do much the same: Compared to their previous FOE vote choice, some switch due to sincere and others due to strategic motives. These are our five main hypotheses about inter-election voting patterns between first-and SOEs (see Figure 1).
Four of them are conditional upon contextual factors; for those we specify a couple of additional expectations about contextual interactions. Specifying direct and interaction effects, the following two figures display those hypotheses in the form of path diagrams for each of the dependent variables.
Starting with differential abstentions (Figure 2), our first hypothesis (H1) predicts that SOE abstentions can in part be explained by the lack of mobilisation: Those with no partisanship and no interest in the electoral cam- paign are more likely to abstain than others. The second hypothesis (H2) predicts that abstention has also sincere roots: EU sceptics are more likely to abstain in EP elections than others. The third hypothesis (H3) specifies strategic motivations to abstain and refers to FOE government voters who are dissatisfied with its performance. In accordance with most recent scholarship, we expect those to be more likely to abstain than others. This effect should be moderated by the distance of the EP election to the midterm of the national electoral cycle when government popularity is plunging quite generally (H3.1). Moreover, government dissatisfaction is expected to have a stronger effect on abstention when there is only one party in government (rather than a gov-ernment coalition of several parties), as the responsibility for government policies is clearer then (H3.2). The second set of hypotheses addresses vote switching between the previous FOE and the EP election under study ( Figure 3). Our first defection hypothesis predicts that voters with a first preference for a small party (i.e., if their party with the highest probability-to-vote score is small) are more likely to defect because there is less at stake in SOEs (H4). This sincere defection mechanism is expected to be moderated by two trigger variables. One is the ideological distance to the most preferred party in terms of left and right: We expect that sincere defection from a FOE vote choice is the likelier the smaller this distance is (H4.1).  between self and party regarding support for European integration: Here as well we expect that sincere defection from a FOE vote choice is the likelier the smaller this distance is (H4.2). We note in passing here that ideological positions enter the equation not as absolute values (like a '4' indicating a centre-left position of a citizen) but as the distance between that '4' and the perceived position of the most preferred party (which may stand at '5,' in which case the distance would be '1' and very small). This procedure closely follows the seminal work of Downs (1957), who suggests that voting decisions of citizens are guided by the distance between the ideological position of a citizen and his or her perception of the ideological position of relevant choice options. We come back to this operationalisation in the Supplementary File to the article. Finally, the defection hypothesis is of a more strategic nature. Here we predict that former government voters who are dissatisfied with government performance are more likely to switch (H5). This effect should be moderated by the shrinking distance to midterm (H5.1), and by the clarity of responsibility for government policies (H5.2). These last two 'strategic interactions' are identical to those put forward with respect to differential abstention.

Data and Case Selection
In order to test these micro-level hypotheses, we are analysing post-electoral voter surveys of the European Election Study (EES; for further details see www.europeanelectionstudies.net). As we aim at a comprehensive test of the micro-level mechanisms of SOE abstention and vote switching, two different waves of these voter surveys are analysed, namely the EES 2004 and the EES 2014. Among a much larger set of variables in each study, these surveys contain a largely identical set of indicators and are separated by a 10 years' period of turbulent political and economic change. As a result, these two studies allow for testing the SOEs model in vastly different contexts. The 2004 EES was conducted when EU membership was expanded to include eight new member countries from Central and Eastern Europe (plus Cyprus and Malta); this made it possible to study the determinants of the vote for the new Eastern EU citizens for the first time. Ten years later, the 2014 EES study was conducted in a completely different environment. The shape of the EU had changed dramatically during these ten years. At a most general level, the EUand the EP within it-gained greater legislative powers in many policy domains (Hooghe & Marks, 2001;Schmitt & Toygür, 2016). From the Maastricht Treaty in 1992 onwards, Europe has been moving into its postfunctionalist phase, which is characterized by a constraining dissensus over issues of European integration rather than the permissive consensus of the past (Hooghe & Marks, 2009;Hutter & Grande, 2014). More specifically, the 2014 EP elections are held in a very different sociopolitical context, provoked by the turmoil in some of the EU member states following the global economic crisis (Braun & Tausendpfund, 2014;Schmitt & Teperoglou, 2015;Talving, 2017). In the words of Reif and Schmitt's (1980) original statement of the SOE model, the specific arena dimension-one of the six dimensions that are at the base of differences between FOE and SOE election outcomes-has changed dramatically between the elections of 2004 and 2014. Despite these fundamental changes, studies of aggregate election results conclude that the character of the 2014 EP elections still follows the SOE predictions (see Schmitt & Teperoglou, 2015;Schmitt & Toygür, 2016). In the rest of this article we will investigate whether this also holds for the micro-level mechanisms governing electoral behaviour as specified by the SOE model.

Dependent Variables
With regard to both participation and party choice, we analyse vote recall (measured at one point in time) about voting decisions in successive elections (conducted at different points in time). Their validity is less than perfect because memory is fading with growing distance to the event, and wishful thinking coloured by the opinion climate of the day is likely to take over (e.g., van der Eijk & Niemöller, 1979;Waldahl & Aardal, 1982, 2000. Panel data would certainly be preferable here, but they are not available for 2004 and efforts to collect them in 2014 were less than successful. There is an additional complication regarding the determination of inter-election vote patterns. It originates in the fluidity of party systems, which is particularly aggravated in the new member countries from Eastern Europe. The problem arises when choice options in one election are no longer available in the next because parties in between have split or merged or simply disappeared. The methodological issues associated with the recall of the vote in multiple elections (in one survey) must be expected to lead to an overestimation of stable voters. Our painstaking recoding of vote patterns in fluid party systems avoids the overestimation of vote switchers as far as possible.
What we know about the over-reporting of electoral participation suggests, in addition, that the relatively small number of abstainers our surveys identify are real abstainers, while many declared voters are true non-voters (e.g., Bernstein, Chadha, & Montjoy, 2001;Traugott & Katosh, 1979). In sum, there are reasons to believe that the two categories of our dependent variable we are most interested in-SOE defectors and SOE abstainersare rather crisp and clean, while stable voters-our reference category as presented above-are most likely to be overestimated. wards the stability of inter-election voting patterns, we find an astounding increase in the percentage of citizens who abstained in both FOE and SOE in 2014 (33% compared to 21% in 2004). This discrepancy could be a true reflection of the reality in the two election years, but it could also be a result of different sampling strategies. We will not dwell on this, however, as we are interested here in the determination of inter-election electoral behaviour rather than in its distribution. We also find a modest decrease in the percentage of differential abstainers between the two EP elections (26% in 2004 and 19% in 2014). The number of 'accidental' SOE voters who abstained in the preceding FOE is equally small (5% in both elections). These are the most remarkable descriptive findings. Based on these patterns of inter-election voting behaviour, we construct the two dependent variables of this article, i.e., differential abstention and defection. Both of them will take the value of '0' for stable voters. Differential abstainers are identified by the value of '1' when respondents voted in the previous national election but abstained in the EP election. Defectors are coded '1' if the respondents chose a different party in the national FOE as compared to the EP election.

Independent Variables
We consider just a few independent or predictor variables at the individual level. In the case of electoral participation, these are citizens' party identification and their interest in the EP campaign, their evaluation of EU membership, and the (dis-)satisfaction with the national government of former government voters. With regard to party choice, we consider just two predictors: Whether or not the first-preference party of a respondent is small and whether the respondent voted for the incumbent national government in the previous FOE and has been dissatisfied with its performance since. These independent variables are not meant to do all the explanatory work by themselves. Rather, they often become effective only in conjunction with what we call trigger (i.e., moderating) variables. One of these trigger variables is whether a country is run by a single party government. The other trigger variable is the location of the EP election in the national electoral cycle. Finally, with regard to the pre-diction of vote choice, we expect the size of the first preference party to interact with the left-right distance between the respondent and that party, and the EU integration distance of the respondent and that party. The details on question wording, variable recoding and descriptive statistics for the dependent, the independent and the trigger variables of our study are available in the Supplementary File of the article.

Findings
The determinants of differential abstention and defection for the 2004 and the 2014 EP elections are reported in Table 2. The table reports the three SOE mechanisms as presented above, namely mobilization first, sincere motivations second, and strategic motivations third. Due to the hierarchical structure of our data, we test our individual and system level hypotheses with multi-level logistic regression models. Hierarchical multi-level models were specified with fixed and random effects. Fixed effects were used in accordance with our hypotheses, while random (country) intercepts account for the country clustering of our data. The quantitative variables involved in interactions were centred at their means to ease the interpretation of additive effects. In order to make the interpretation of interaction effects more accessible, we also present them graphically as marginal effect plots in Figures 4 and 5.

Differential Abstention
Our first hypothesis predicts that the decision not to cast a vote in an EP election (although having voted in the previous FOE) is a result of the characteristic lack of mobilisation in these elections, among other things. We find H1 corroborated for both elections under study. A low interest in the EP election campaign and the lack of partisanship are strongly increasing the likelihood of abstaining both in 2004 and 2014. Having said that, we note that the effect of low interest in the EP election campaign seems to be higher in 2004 than in 2014, while the reverse holds regarding partisanship.
But the lack of mobilization is not the only factor that contributes to the likelihood of abstention in these two EP elections, there are also signs of sincere non-voting. Notes: Entries are multilevel logistic regression coefficients (standard errors in parentheses); * p < 0.1, ** p < 0.05, *** p < 0.01. Sources: EES 2004 (Schmitt, Bartolini, et al., 2009) and EES 2014 .
Euroscepticism (i.e., negative evaluations of EU membership) increases the likelihood of abstaining both in 2004 and somewhat stronger in 2014, even if controlled for mobilisation effects. This confirms H2. Therefore, low levels of electoral participation in EP elections are associated with critical stances towards the EU. We continue with our hypothesis regarding strategic abstention in EP elections. The findings in Table 2 reveal that voters who previously supported a national government party might have abstained in EP elections in order to signal discontent with its performance. This kind of strategic abstention in EP election is triggered by the timing of an EP election in the first-order national electoral cycle. According to this, voters of a national government party who are dissatisfied with government performance are more likely to abstain around the midterm of the national electoral cycle (H2.1). Finally, H2.2 predicts that strategic abstention is also triggered by the clarity of government responsibility. In coalition governments with less clarity of government responsibility, strategic abstention is more likely to happen. This is only confirmed for the 2014 EP election.  (Schmitt, Bartolini, et al., 2009) and EES 2014.

Defection in EP Elections
Moving on to the results of sincere SOE vote switching, we find that perhaps the most important factor of the original SOE model, the fact that there is 'less at stake' in these EP elections, is confirmed at the micro level. We hypothesized that voters whose first preference party is small will show higher probabilities to switch their votes from the previous FOE to the EP election under study (H4). The results in Table 2 confirm this hypothesis; the coefficient of a first preference for a small party is positive and very substantial for both the 2004 and the 2014 EP elections.
In addition, we hypothesized that the left-right distance to the most preferred and small party is one of the triggers for defecting (H4.1). Because there is less at stake, supporters of small parties might take the EP election as an opportunity of 'voting with the heart' (van der Eijk et al., 1996). The results as presented in Table 2 seem to refute this presumption at first sight. In both elections, we find that the larger the left-right distance to the most preferred party is the likelier it is to defect from the FOE vote choice. But we must not forget here that we are still talking about distances to the most preferred party: those who are ideologically very close to it might not have defected previously to begin with. Rather, it seems that those with some distance have abandoned their first preference in the previous FOE, and that these voters are 'returning home' when less is at stake in the subsequent SOE. In essence, then, our hypothesis H4.1 is not falsified, but nicely specified by our analysis. What about the proposed trigger of support of or opposition to EU integration (H4.2)? For the 2004 election we find that the distance on EU integration does not significantly moderate the voting decision of small party supporters. Only in 2014, when the politicisation of EU integration has increased, we find the same phenomenon as for the left-right dimension: Small party supporters return to their first preference in a SOE when they are at some distance to it on EU integration matters.

DefecƟon 2014
Coefficient for EU distance Figure 5. The marginal effects of the 'less at stake' mechanism as moderated by the left-right distance (first row) and European integration distance (second row) on defection in the EP elections of 2004 and 2014. Source: EES 2004 (Schmitt, Bartolini, et al., 2009) and EES 2014.
Are then SOE defections from the previous FOE choice strategically motivated? Signalling discontent with the incumbent government not only motivates differential abstentions from former government voters, as we have shown previously. It is at least equally important as a mechanism for defecting from the previous FOE vote choice. Both in 2004 and in 2014 dissatisfaction with the government party respondents previously supported significantly contributes to their probability of defection in the EP election. These signs of 'voting with the boot' (van der Eijk et al., 1996) confirm H5. If it comes to interactions, our measure of protest voting is moderated by the FOE electoral cycle only in 2014. Only in these elections does the decreasing distance to midterm significantly increase the contribution of dissatisfaction to the probability of defecting from the party voted in the previous FOE. Therefore, the expectations formulated in H5.1 are only partly confirmed. One explanation for this could be that in 2004 there was not yet much of a cyclical regularity in the political orientations of citizens in the then eight new Eastern member countries (e.g., among others Marsh, 2007;Schmitt, 2005). Finally, our hypothesis that strategic signalling might be especially intense in countries with single party national governments (H5.2) is not confirmed in either election. The clarity of responsibility for government policies is not moderating the likelihood of defection of dissatisfied FOE government voters.

Concluding Remarks and Perspectives
The formulation of the aggregate hypotheses of the SOE model as originally proposed by Reif and Schmitt (1980) aims to explain the differences in the electoral results between FOEs and SOEs. While likely reasons for these characteristic differences in the outcomes of these elections were suggested early on, the easier-to-grasp macrolevel predictions of the model have long been in the centre of scholarly testing. Less attention has been given to the micro-foundations of the SOE model, that is, the hypotheses about the motivations and intentions of individ-ual voters that drive their behaviour in a SOE-relative to what they have done in the preceding FOE. In this article, we summarise the hypotheses underlying the SOE model and subject them to a rigorous empirical examination using the data from the EES 2004 and 2014 post-electoral voter surveys.
At a conceptual level, we started by distinguishing the behavioural alternatives that present themselves to the citizens at a SOE following a FOE and identified two of them which we claim to be critically important for the empirical analysis of inter-election voting. These are differential abstention and SOE defection. In the empirical part of the study, we analysed the mechanisms for both participation (differential abstention) and vote choices (SOE defection)-something has hardly ever been done in the scholarship available so far. Our main finding is that it all happens at once: Mobilization as well as strategic and sincere mechanisms affect electoral behaviour at different levels of a multi-level electoral system. Differential abstention is mostly due to a lack of mobilization. Moreover, in 2014 (but not in 2004) we also found support for our hypothesis on sincere abstentions: those who are 'opposing Europe' (Szczerbiak & Taggart, 2008) were significantly more likely to abstain in the EP election while having voted in the previous FOE. In addition, our analyses also identified signs of strategic abstentions. Dissatisfied former government voters were more likely to abstain, particularly in EP elections close to the midterm of the national electoral cycle. Turning to voter motivations for defection in EP elections, we mostly found sincere mechanisms at play. Voters whose first preference is a small party were found more likely to defect than others, and first preference voting means sincere voting. Moreover, our trigger variables were shedding some additional light on the motivations of the defectors: the left-right distance as well as the EU integration distance (in 2014) to the first preference party was found to significantly moderate the likelihood of voters with first preference for a small party to defect. But there are also indications that SOE defection is of a more strategic nature. Dissatisfied former government voters were found to be more likely to defect than others. This strategic defection is moderated by the distance to the midterm of the electoral cycle (in 2014, but not in 2004); defections increase with decreasing distance to first-order midterm. However, our second trigger variable for strategic defection-clarity of responsibility for government policies-was not found to be moderating defection in either election.
So we know that mobilisation and sincere and strategic factors are all playing an important role in our understanding of differential abstention and defection: it all happens at once. We might therefore conclude that the decision to participate in an EP election is a more multi-layered phenomenon than often portrayed, combining both motivations from the national (domestic) and the European political arena. This has been established in two largely different electoral environments-the enlargement election of 2004 and the post-crisis election of 2014-and we therefore are confident that these findings are reliable and can be found again and again.
Our results are important for the current and future understanding of electoral behaviour of political actors in multi-level electoral settings. We claim that they can be generalised to previous and future EP elections as well as other SOE at sub-national levels, like Canadian provincial elections, Spanish regional elections, German state elections, and so on. Multiple levels of a multi-layered electoral systems are not isolated from one another, they are permeable and interwoven. This openness comes at a price: As we have shown in this article, voters take their behavioural cues from different levels. Moreover, political parties present themselves more or less uniformly at different levels in order to retain their credibility (which means among other things to assume compatible policy positions, see e.g., Braun & Schmitt, 2018).
It is often said that the EU is a moving target; no other SOE-specific arena undergoes as rapid and profound institutional and procedural changes than the EU does. Will future EP elections with perhaps even starker changes in this 'specific arena dimension'-like the 2019 election with its fundamental challenge of the EU by populist parties on the right-still fit that picture? We claim that they will, provided that citizens still perceive that there is 'less at stake' in the EU electoral arena.