Moving beyond the second-order election model?

ABSTRACT
 In this introduction to the sixth annual review of regional elections we identify three generations of regional election studies that have applied the second-order election (SOE) model. First-generation literature finds that regional authority, territorial cleavages, and election (non-)simultaneity explain territorial heterogeneity in SOE-effects because they affect ‘what is at stake’ in a regional election. A ‘stake-based’ approach also underlies a second-generation literature that finds that voters with strong regional identities and who find regional government important are more likely to make distinct party choices in the regional electoral arena. Third-generation research adopts a multilevel electoral system perspective and considers the impact of political-institutional variables on the extent of horizontal and vertical top-down and bottom-up spill-over between regional and national electoral arenas. Four election articles and four election reports make important contributions to the three generations of literature and thereby reveal that these generations of regional election scholarship remain highly relevant.

departamentos and one capital city in Peru, and 41 județe and one capital city in Romania. The total of six annual reviews collectively covers elections held in a total of 1,020 sub-state units in 42 countries which include a total population of approximately 3.29 billion people (Schakel and Romanova 2018;2019;. The main objective of the annual review is to systematically and comparatively report on regional elections across the globe. Through the accumulation of annual reviews, we seek to increase our understanding of the factors that drive regional voters, regional election outcomes, and regional electoral dynamics (Schakel and Romanova 2018, 233-236). In each introduction, we draw comparative lessons and discuss the implications of the findings for our understanding of regional elections. With the series of introductions to the annual reviews, we aim to identify several crucial topics for the understanding of regional election outcomes. Hence, we focus on one or more themes derived from our own reading while adopting a multilevel election system perspective. A multilevel election system perspective brings together nationwide and regional elections and considers the vertical and horizontal interactions between and the integration of national and regional electoral arenas (Schakel and Romanova 2020).
Many regional election studies, including the articles and reports of the current and previous annual reviews of regional elections, scrutinize election results according to the second-order election (SOE) model. The main assumption underlying the SOE-model is that regional contests are perceived by voters and parties to be not important or less important than national elections. As a result, voters do not bother to turn out, and those who do use the regional election to voice their discontent by voting against parties in national government and rewarding opposition, small and new parties. In this introduction we review the regional election literature that has applied the SOE-model. We identify three generations of regional election studies. First-generation literature finds that regional authority, territorial cleavages, and election (non-)simultaneity explain territorial heterogeneity in SOE-effects because they affect 'what is at stake' in a regional election. A 'stake-based' approach is verified by the second-generation literature that finds that voters with strong regional identities and who find regional government important are more likely to make distinct party choices in the regional electoral arena. Third-generation research adopts a multilevel electoral system perspective and considers the impact of political-institutional variables on the extent of horizontal and vertical top-down and bottom-up spill-over between regional and national electoral arenas. By adopting the term 'generation', we do not intend to imply that one body of literature supersedes another. The term 'generation' refers to a temporal order of three bodies of literatures that developed within an emerging scholarship on regional elections. All three generations of regional election scholarship and the second-order election model are still relevant as the four election articles and four election reports included in this special issue attest to. A review of regional election research according to three generations helps us to highlight the contributions of the election articles and election reports included in this sixth annual review.
In the next section, we discuss the baseline SOE-model, and we review the first-generation literature. We then proceed with a discussion on the second and third generations of regional election studies in the third and fourth sections. In these discussions we identify the main findings and contributions of these literatures regarding increasing our understanding of regional voting. In the final section, we reflect on the question to what extent regional election research has moved beyond the second-order election model.
The second-order election model applied to regional elections Reif and Schmitt (1980) introduced the SOE-model to explain the outcomes of the first direct election to the European Parliament in 1979. They observed that turnout was low in comparison to participation rates in previously held national elections and that parties in national government lost vote share whereas opposition, small, and new parties won vote share. In addition, these SOE-effects were stronger at the midpoint in the national election cycle. In their seminal article, they anticipated that the SOE-model can be applied to a wide range of other types of elections including by-elections, elections to a second chamber, municipal elections, and 'various sorts of regional elections' (Reif and Schmitt 1980, 8).
This 'invitation' was taken up by scholars to identify whether regional elections are second-order (Schakel and Jeffery 2013). A Google Scholar search (on 16 June 2023) with the terms 'regional election' and 'second order' reveals an increasing number of publications over time: 7 publications until 1999, 22 for 2000-2004; 73 for 2005-2009; 138 for 2010-2014; 194 for 2015-2019; and 137 for 2020-2022. The second-order election model is clearly widely referred to in analyses of regional election outcomes. Most research confirms that turnout in regional elections is lower compared to first-order national elections and that parties in national government perform less well than opposition, small and new parties. In addition, these effects tend to vary depending on the timing of the regional election in the national election cycle although the strongest impact is not always observed at mid-term (Dandoy and Schakel 2013;Jeffery and Hough 2006;Schakel 2017;Schmitt and Teperoglou 2017, 58-59). The omnipresence of SOE-effects across a wide variety of regional elections clearly indicates that the SOE-model is key to understanding regional voting.
The basic core of the SOE-model stipulates that SOE-effects occur because there is 'less at stake' in a SOE (Reif and Schmitt 1980). As a result, voters do not bother to turn out, and those who do are more concerned with voicing their discontent with national government performance than expressing their opinions regarding regional politics (Lau 1985). In other words, SOEeffects appear when a significant share of voters perceive the SOE to be trivial. SOE-effects are bound to appear in non-national elections, i.e. elections that do not produce a national executive or select a national president because, for most countries and regions, one may assume that voters consider national government to be the most important government tier which has the largest impact on their lives. Finding SOE-effects may support a SOE-model interpretation but does not exclude the possibility that these effects are (co-)produced by other causal mechanisms. For example, SOE-effects may be caused by voters returning to their 'sincere vote' and support small and opposition parties in the SOE after having voted 'strategically' in the first-order election for a party that takes up policy positions that are more distant from the policy preferences of voters, but which has a higher chance of being included in the national government (Schmitt et al. 2020).
The research objectives of Reif and Schmitt (1980) did not require them to develop a fully fleshed out theory that is applicable to all kinds of SOEs. Reif and Schmitt (1980, 10) were fully aware that an 'important aspect of secondorder elections is the political and institutional circumstance of the respective political arena: parties, platforms, candidates, the policy-areas and positions of control that are at stake'. Regional councils may take decisions in important matters, and the selection of the executive is dependent on the election outcome in many regions. In addition, differences between national and regional party systems may limit the opportunities for voters to signal their discontent by supporting national opposition parties. Furthermore, parties in national government may lose less vote share when they are also in power at the regional level. As the Google search above indicates, it took two decades before the first studies appeared that explored the 'usefulness' of the SOE-model to make sense of regional election outcomes. Table 1 displays a categorization of the literature in three 'generations'. Firstgeneration literature sets out to assess the applicability of the SOE-model to Simultaneity between regional and other types of elections State of the regional/ national economy Factors inducing vertical and horizontal electoral spillover regional elections and has found that regional authority, territorial cleavages, and simultaneity between elections are important variables that impact the magnitude of SOE-effects. Second-generation regional election research aims to assess what drives 'regional voters' and has found that regional voting happens when voters attribute policy outcome responsibility to regional government. This literature also focuses on other factors that may explain national and regional government popularity, most prominently the state of the national and/or regional economy. A multilevel perspective puts the multilevel voter central and conceives SOE-effects to be one out of three possible types of electoral spill-over. The main aim of this literature is to identify the politicalinstitutional variables that affect the incentives for parties and voters to consider the politics of other electoral arenas. In the remainder of this section, we will discuss first-generation research, and second-and third-generation studies will be discussed in the two following sections. Regional governments vary widely regarding their competences Loughlin, Hendriks, and Lidström 2011) and this greatly affects how much is at stake in an election. A dominant hypothesis is that more significant decision-making powers decrease second-order voting. When regional elections are sub-ordinate to national electoral arenas, then parties are unlikely to differentiate their campaign strategies across regions (Jeffery and Hough 2009). When regions decide over policies voters care about, then regional branches of statewide parties are induced to tailor their electoral strategies towards the specific demands of regional electorates (Hopkin 2009;Maddens and Libbrecht 2009). This, in turn, may lead to the mobilization of different issues across the regions, resulting in variation in voter and party alignments (Thorlakson 2007). These 'pressures' may lead statewide parties to decentralize organizational powers internally from the national centre to the regional branches as to enable them to respond effectively to region specific demands (Detterbeck and Hepburn 2018;Fabre and Méndez-Lago 2009;León-Alfonso 2007;Thorlakson 2009).
Another variable that may decrease second-order voting is the presence of territorial cleavages. If regional elections are held in areas with distinctive territorial identities, voters are more likely to base their vote choice on regional factors instead of factors arising from the first-order national electoral arena (Jeffery and Hough 2009, 223). Voters may be motivated to vote based on ideological criteria in national elections whereas their regional vote choice is driven by regional interests (Jeffery and Hough 2006, 250). Furthermore, electorally successful regional parties induce statewide parties to differentiate their strategies electorally where territorial voting is prominent (Alonso, Cabeza, and Gomez 2015).
Applying the SOE-model to regional elections has induced scholars to see how far the timing of the regional election in the national election cycle impacts SOE-effects. Parties in national government are expected to lose more vote shareand opposition, small, and new parties are expected to win more vote sharethe further away in time a regional election is held from the previous and next national election (Jeffery and Hough 2001;2003). In addition to relative timing of a regional election in the national election cycle, scholars have focused on simultaneity between regional and local and other regional elections. The assumption is that election simultaneity induces candidates, media, and parties from the statewide electoral arena to be involved in the regional election campaign ( Van der Eijk, Franklin, and Marsh 1996). This increased involvement creates an approximation of a first-order poll, whereby SOE-effects are increased because the SOE is given collective nationwide reach and resonance (Jeffery and Hough 2006, 249), and because dissatisfied voters are easier to mobilize to vote in a non-important election (Lau 1985). However, regional election outcomes tend to mirror national election results when a regional election is held close to, or concurrently with, a national election (Romanova 2014).
A 'stake-based' approach also fares well in explaining turnout in regional elections and territorial identities and political decentralization increase participation rates (Henderson and McEwen 2010;. Furthermore, turnout increases when a regional election is held simultaneously with local, other regional, or national elections, and it declines when another type of election has been held just before the regional election (Schakel and Dandoy 2014). The combined impact of regional authority, territorial cleavages, and electoral timing suggests variation in SOE-effects across the statewide territory. When decentralization is asymmetrical, strong territorial identities are held by minorities of the statewide population, and regions have independent electoral cycles from the national electoral arena and from each other, one may expect the SOE-model to have different explanatory value territories (Amat, Jurado, and León 2020, 279-281;Jeffery and Hough 2009, 223). Asymmetry in regional authority has increased (Hooghe and Marks 2016), regionalist parties have become electorally stronger (Massetti and Schakel 2017), and the linkages between national and regional electoral cycles have become more varied (Schakel and Dandoy 2014;Schakel and Verdoes 2023). Hence, one may also expect the SOE-model to have different explanatory value time.
First-generation regional election research has been crucial to uncover territorial variation in SOE-effects. These 'territorial effects' induced Jeffery and Wincott (2010, 179) to urge scholars to study regional elections 'on their own terms' which might 'generate a different or at least a more nuanced picture'. Studying regional elections 'on their own terms' does not always provide for new insights as is exemplified by the election report on Croatia. Glaurdić and Fel (2023; this issue) reveal that the experiences towards the 1991-1995 War of Independence of Croatia can explain support for the two main parties in both regional and national elections. Voters in waraffected areas reward candidates for their military service whereas these candidates are penalized in areas that were not affected by the war. The centre-right party HDZ (Croatian Democratic Union; Hrvatska demokratska zajednica) took Croatia into independence and led it during the war, and this party received higher vote shares in municipalities with a higher number of citizens with disabilities caused by war operations. The centreleft SDP (Social Democratic Party of Croatia; Socijaldemokratska partija Hrvatske) and its allies, on the other hand, obtained significantly lower vote shares in these municipalities (Glaurdić and Fel 2023).
However, studying regional elections 'on their own terms' is often worthwhile because it may, to the very least, draw scholarly attention to territorial heterogeneity in the vote. For example, Székely (2023; this issue) analyses Romanian regional elections which have been previously described as 'barometer' elections. A barometer election is a regional election which reflects the popularity of statewide parties based on their (perceived) performance in the national electoral arena, but which does not result in SOE-effects (Schakel and Romanova 2018, 243-244). Rather, these regional elections may function as reliable predictors of the outcomes of an upcoming national election. In the case of Romania, the barometer status of regional elections can be in large part ascribed to electoral timing: regional elections are held together with local elections and six months before a parliamentary election (Dragoman and Zamfira 2018).
Adopting a longitudinal perspective, Székely (2023) observes that the same party or bloc has been dominating both regional and national elections held in 17 out of a total of 42 counties ( județe) between 2004 and 2020. The electoral strongholds appear to align with a historical-territorial cleavage: leftleaning counties can be found in the 'Old Kingdom of Romania' whereas right-leaning counties are located in Transylvania. Shifts in electoral strongholds have been regularly changing in one-fourth of the counties whereas they occur rarely in the remaining one-third of the counties. Hence, even though party system nationalization in Romania is high, territorial cleavages set limits to the extent of nationalization in both regional and national elections. Furthermore, the observations from Romania (and Croatia) are important because they reveal that many party systems that appear at the statewide level are likely the result of the aggregation of various different party systems at the regional level, rather than the sum of similar regional party systems (Schakel and Romanova 2020).
The election report on the 2021 French regional elections is an example of how first-generation research can offer a 'nuanced or different picture'. La République en Marche, the party of President Emmanuel Macron, lost significant vote share compared to the presidential elections held in 2017. A vote share drop from 23.8% to 10.5% combined with an exceptional low turnout of 33.3% can easily be interpreted as the result of a typical SOE. However, Gougou (2023; this issue) argues that a candidate-centred politics model provides for a better fit. The starting point of the candidate-centered politics model is party dealignment, i.e. the declining importance of parties and partisanship for vote choice. Instead, voters take performance evaluations as the basis for their vote choice. As a result, fewer voters vote for the same party across elections, incumbent re-elections rates are high, and divided government between national and subnational governments is more frequent (Gougou 2023;this issue). For the first time since the very first regional elections of 1986, all incumbent regional presidents were reelected in 2021 in France. The regional presidency shifted hands regularly in previously held regional elections (Escolana, Labouret, and Vieira 2013;Gougou and Labouret 2010;Jerôme and Jerôme-Speziari 2000;Shields 2018). In addition, split-ticket voting between regional and simultaneously held departmental elections reached much higher levels in 2021 when compared to the concurrent elections in 2015 (Gougou 2023;this issue). Although the empirical evidence suggests a better fit with candidate-centered politics model, a SOE-interpretation is not disconfirmed.
Furthermore, the election article on Italy is an example of how first-generation research can offer a 'nuanced or different picture' of electoral behaviour which is difficult to achieve when one focuses on national elections. A focus on regions within countries often meet the requirements of a most-similarsystem-design whereby the national political-institutional context is common and shared, and can therefore be treated as constants (Giraudy, Moncada, and Snyder 2021;Przeworski and Teune 1970;Snyder 2001). Basile (2023; this issue) sets out to study the impact of territorial inequalities on turnout in Italian regions since 2003, and she hypothesizes that economic development and social capital increases turnout whereas institutional quality has no impact. Italian regions vary widely regarding their socio-economic development and quality of institutions, but their electoral systems share key features relevant for turnout, and regional authority after the 2001 constitutional reform is similar, especially for the fifteen ordinary statute regions (Basile 2023;this issue). This research design enables Basile (2023; this issue, Table 1 and Figure 6) to isolate an interaction effect which exposes that economic capitalmeasured by regional wealth and employmentincreases turnout but not when social capitalmeasured by the percentage of people aged 14 and over that have performed free activities for voluntary associations or groups in the last 12 monthsis low.
Regional elections on 'their own terms': 'First-order' regional versus 'second-order' national voting First-generation regional election research converges on the observation that territorial variation in SOE-effects can be best understood as the result of the impacts of both national and regional factors (Bolgherini and Grimaldi 2017;Bolgherini, Grimaldi, and Paparo 2021;Henderson and Romanova 2016;Schakel and Verdoes 2023;Tronconi and Roux 2009;Trystan, Scully, and Jones 2003). Most likely, regional election outcomes are produced by some voters basing their vote choice primarily on national factors whereas other voters mostly rely on regional cues. Second-generation regional election research sets out to assess which voters conceive regional elections as 'first-order' or 'second-order' and under which circumstances regional elections tend to have more or fewer 'first-order' and 'second-order' voters (Cabeza and Scantamburlo 2021;Cutler 2008;Ingram 2003). Second-generation literature frequently exploits election survey data which enables scholars to uncover individual level causes for regional voting. Although secondgeneration literature ambitions to scrutinize first-order, regional voting, i.e. voting based on regional instead of national arena specific factors, a stakebased approach is still dominant (Table 1).
Regional voting is first and foremost thought to depend on the extent to which voters 'care' about regional government. Citizens who have strong regional identities and citizens who prefer decentralization or independence for their region are more likely to turn out and to make distinct vote choices in regional elections (Cabeza 2018;Henderson and McEwen 2015;Liñeira 2011;Thorlakson 2015). In addition, voters who attribute significant power and responsibility to their regional governments tend to evaluate the performance of parties in the regional executive and reward or punish them accordingly, also in less powerful regions (Johns et al. 2009;Lindstam 2019;Linek and Škvrňák 2022;Wyn Jones and Scully 2006). Decentralization may not always increase regional voting since it can blur responsibility attribution by fragmenting power between regional and national governments (León 2011;. Voters may use national policy outcomes to evaluate regional incumbents when responsibilities overlap, but voters are more likely to evaluate regional incumbents on the basis of regional policy outcomes when there is a clear division of responsibilities (Amat, Jurado, and León 2020;Bosch 2016;Leon 2018).
Second-generation literature confirms SOE-voting at the individual level including the impact of the relative timing of a regional election vis-à-vis the national election (Müller 2018). Many European countries have fixed electoral cycles and the temporal distance to national elections varies less much compared to countries where regions set their own election dates. Yet, SOEeffects vary significantly between regional elections. Furthermore, SOEeffects also vary significantly in regions with an independent election cycle. This has induced second-generation scholars to focus on the state of the regional and/or national economy that impacts the extent to which voters hold their regional or national governments accountable when they cast their regional vote (Table 1;  also proposed a valence model of voting that posits that perceived government performance is an important factor influencing vote choice in regional elections (Cabeza 2018;Curtice 2011;Scully and Jones 2012). The election article on Australia provides further support for a valence model, the election article on the UK develops the valence model further, and the election article on Canada illustrates a major challenge scholars face when providing empirical evidence for valence voting.
Evidence for a valence model of voting is suggested by Smith et al. (2023; this issue) who analyse voting intentions for the regional incumbent in six states and two territories in Australia. Respondents were asked to evaluate how well their state/territory and federal governments handled the COVID-19 pandemic. Two thirds of the respondents indicated that both tiers handled the pandemic well, whereas 7.5% of the respondents thought that neither handled the crisis well. Almost a quarter of the respondents provided differential evaluations: 10.4/13.9% indicated that the national/regional government performed well but the regional/national government did not (Smith et al. 2023;this issue). A logistic regression model reveals that respondents who thought their regional government had handled the COVID-19 crisis very well were 13 times more likely to vote for the incumbent party than citizens who thought that the pandemic had not at all been handled well. In stark contrast, evaluations on crisis management of the national government did not have an impact and vote intentions for a LNP (Liberal National Party Coalition) incumbentthe party in national governmentat the regional level were only slightly reduced indicating a minor SOEeffect (Smith et al. 2023; this issue). Griffiths et al. (2023;this issue) further develop the valence model of voting by pointing out that, in addition to evaluating competence in regional office, voters with strong regional identities also assess the extent to which parties defends territorial interests. At the time of the first devolved elections in 1999, the Labour Party was dominant in both Scotland and Wales. As of 2007, and until the latest elections of 2021, the Labour party in Scotland was outpolled by the Scottish National Party (SNP) whereas in Wales, the Labour Party continues its dominance and remains the largest party both in terms of vote share and seats in every devolved and UK-wide election since 1922. These striking different trajectories materialized even when Scottish and Welsh identities remained relatively stable ( Table 3 in Griffiths et al. 2023; this issue). Griffiths et al. (2023; this issue) explain the diverging Scottish trajectory by changing perceptions among those who identify as Scottish by the ability of Scottish Labour to defend Scottish interest. In 1999, around 50% of the respondents who identified themselves as 'Scottish' trusted the SNP to stand up for Scotland's interest 'just about always' or 'most of the time' whereas this percentage for Scottish Labour was about 70%. In 2021, the percentages of 'Scottish' respondents who answered the question whether a party stands up for Scotland's interests 'a great deal' or 'quite a lot' was about 90% for the SNP and about 20% for Scottish Labour. Similar but smaller shifts in percentages can be observed for respondents who identify themselves as 'more Scottish than British' (Figure 4 in Griffiths et al. 2023;this issue). Hence, party competition has significantly altered the link between regional identity and vote choice because Scottish Labour lost its appeal to Scottish identifiers as a party that stands up for Scottish interests. Griffiths et al. (2023;this issue) argue that the 2014 Scottish independence referendum campaign has been decisive. Labour cooperated closely with the Conservatives as part of the 'Better Together' campaign which supported the union and, by extension, British identity.
A major challenge for providing empirical evidence for a valence model of voting is to disentangle the impact of competence evaluations on vote choice from the impact of party preferences and regional identities. Competence evaluations appear to strongly correlate with party preferences and regional identities. Those who prefer the party in office tend to have positive evaluations of party performance, whereas voters who align with the party in opposition tend to have negative views on party performance (Henderson and McMillan 2022;León and Orriols 2019;Rico and Liñeira 2018;Schonhage and Geys 2021;Toubeau and Wagner 2018). This also applies to the impact of the COVID-19 crisis on regional voting as is exemplified by Stephenson and Harell's (2023; this issue) analysis of five provincial election studies in Canada. They set out to assess the impact of satisfaction of the handling of the COVID-19 pandemic by the provincial government on voting for the provincial incumbent. These evaluations appear to be driven by an incumbent party bias, and out-partisans and non-partisans are substantially less satisfied (Figure 2 in Stephenson and Harell 2023;this issue). A logistic regression model confirms a positive correlation between pandemic management evaluations and voting for the provincial incumbent, but this relationship disappears once the model includes partisanship, party competence, and incumbent premier evaluations. Although confirming a valence model of regional votingpartisanship, party issue handling, and feelings towards the incumbent premier, drive incumbent voting in all five provinces ( Figure 5 in Stephenson and Harell 2023; this issue)positive pandemic management perceptions did not make out-partisans and non-partisans more inclined to vote for the provincial incumbent (Stephenson and Harell 2023; this issue).
First-generation regional election scholarship has further refined the theoretical foundations of the SOE-model and, thereby, has significantly increased our understanding of regional election outcomes. Finding significant territorial heterogeneity in SOE-effects has induced scholars to study regional elections 'on their own terms' and this research has advanced our understanding of the drivers of regional voting. The SOE-model still permeates second-generation regional election research, which seeks to understand for whom regional elections are 'first-order' or 'second-order'. A multilevel election perspective reveals the boundaries of the analytical reach of the SOE-model, as well as the research that studies regional elections on 'their own terms'.

Regional Elections in a Multilevel Electoral System
A multilevel electoral framework considers both nationwide and regional elections and takes into account the vertical and horizontal interactions between as well as the integration of national and regional electoral arenas (Schakel 2018;Schakel and Romanova 2020). Figure 1 displays a hypothetical multilevel electoral system of a country with three regions. There are two consecutive national elections, and three regions hold their election in between the two national elections. In a multilevel system, electoral behaviour and outcomes can be compared between electoral arenas at the national and regional levels and between regions. These comparisons often also involve a temporal dimension depending on the extent of simultaneity between elections (Table 1). A multilevel electoral system perspective enables the conceptualization of three different kinds of electoral spill-over, i.e. factors arising from one electoral arena that impact voting behaviour and outcomes in another electoral arena (Schakel and Romanova 2021). The three different types of electoral spill-over are indicated by different arrows: vertical topdown (striped-dotted arrows), vertical bottom-up (striped arrows), and horizontal (dotted arrows).
Vertical electoral spill-over occurs when developments in the statewide electoral arena impact on processes in regional electoral arenas (top-down Figure 1. A multilevel election system perspective. spill-over) or vice versa (bottom-up spill-over). Horizontal spill-over occurs when developments in one regional electoral arena impacts processes in another regional electoral arena. A multilevel election perspective is important because regional election outcomes cannot be fully understood without taking into account the possibility of horizontal and vertical spillover from other electoral arenas (Golder et al. 2019, 3). The SOE-model is concerned with vertical top-down spill-over because it explains regional election outcomes by the governmental status of parties in the national electoral arenae.g. parties in government lose whereas parties in opposition win vote share. By frequently adopting the SOE-model, scholarship on regional elections may overlook the importance of horizontal spill-over between regional electoral arenas. An example of horizontal electoral spill-over is the widening support of regionalist parties in Spain, which were first successful in the historic communities (Basque Country, Catalonia, and Galicia), and which triggered the establishment of regionalist parties in the other communities (Pallarés and Keating 2003;Swenden and Maddens 2009, 8-9).
Similarly, what happens in regional electoral arenas might be important for the outcomes in national elections. An example of vertical bottom-up spill-over is when parties are first electorally successful at the regional level and later also at the national level (Farrer 2015;Massetti and Schakel 2017;Spoon and Jones 2015). Another example is when particular government coalitions are first tried out in one or more regions before being formed at the national level (Bäck et al. 2013). As we have noted before, there is relatively abundant research that uncovers vertical top-down spill-over, but there is an increasing body of evidence that reveals vertical bottom-up and horizontal spill-over (Schakel and Romanova 2021). A multilevel electoral system perspective applied to individual-level voting invites scholars to compare whether vote motivations are similar or different for electoral arenas at different territorial scales (Table 1). For example, turnout rates are found to be dependent on the importance voters ascribe to tiers of government and on their attachments to various territorial scales (Golder et al. 2019, 67-85). Third-generation research on regional elections has put the spotlight on the impact of institutions that were largely absent or included as control variables in statistical models in first-generation and second-generation regional election research (Table 1). For example, electoral systems have a huge impact on the vote and electoral rules may vary between national and regional elections as well as between regions. Thereby, voters and parties are confronted with differential incentives to behave tactically. In order to understand the multilevel voter, i.e. why voters switch their party preference between national and regional elections, one needs to take the differences between national and regional electoral rules into account (Golder et al. 2019, 35-36;Selb 2006;Schakel and Verdoes 2023).
A major empirical question is the extent to which election outcomes are produced by vertical top-down and horizontal spill-over in addition to the impacts of factors arising from the electoral arena itself. The relevance of electoral spill-over is highly dependent on the extent to which national and regional party systems are nationalized and on the presence of institutions that affect the incentives for parties and voters to consider the politics of other electoral arenas. Haman and Školník's (2023; this issue) report on regional elections in Peru describes an 'extreme case' where vertical and horizontal electoral spill-over is minimal. Voting is compulsory and the shares of invalid (blank plus void) votes are very high for both regional and national elections. It ranges between 12.1 and 18.5% for regional elections held during 2002-2018, between 23.1 and 35.0% for Congressional elections held during 2001-2016, and between 12.1 and 18.1% for the first round in Presidential elections held during 2001-2016 (Table 1 in Haman and Školník 2023;this issue). Hence, many voters do not know which party to vote for or they are not satisfied with the supply of parties and candidates, no matter the type of election. Regional movements attract more than 30% of the regional vote in 21 departments and deliver the regional governor in 15 out of a total of 24 departments (Table 3 in Haman and Školník 2023, this issue). Regional movements are not allowed to participate in national elections and, thereby, vertical bottom-up electoral spill-over is severely limited. Each of the 15 governors were elected on a ticket from a different regional movement, three of which were established one year before the regional election, three participated in one, six in participated in two, and three participated in three previously held regional elections. The level of party institutionalization is also low at the statewide level and parties, coalitions of parties, and party names regularly change between elections ( Table 2 in Haman and Školník 2023; this issue). Hence, voters face enormous challenges to evaluate the performance of candidates and parties from their own regional as well as from other electoral arenas.
Multilevel voting occurs when voters realize that they are casting votes in different arenas in a multilevel system of governance (Golder et al. 2019, 93). Hence, multilevel voting requires that a regional vote choice is (also) based on cues originating from another electoral arena. Strategic balancing is an example of multilevel voting, and it is an option for voters who can vote in elections in regions with shared rule, i.e. regions that co-exercise authority in the country as a whole together with other regions and the national government. There are two ways in which regions may have shared rule: through regional representation in the national parliament, and through intergovernmental meetings between representatives from regional and national executives . In several countries, regional elections determine or impact the constellation of an upper chamber of parliament. Examples are the Dutch Eerste Kamer, the French Sénat, the German Bundesrat, the Spanish Senado, and the Swiss Ständerat. These upper chambers are based on regional representation, consist of representatives or delegates from regional parliaments or governments, and can influence or veto national legislation. Shared rule provides regional voters with an opportunity to strategically balance the different institutions and actors that are involved in the policymaking process (Golder et al. 2019, 93-97;Schakel and Jeffery 2013). For example, a left-leaning national government can be balanced by a rightleaning upper chamber of parliament, and by right leaning regional ministers who meet with national ministers to develop and implement national policy.

Conclusion
To what extent has regional election research moved beyond the secondorder election (SOE-)model? First-generation and second-generation literature often explicitly takes the SOE-model as a starting point. First-generation literature applies the SOE-model to regional elections and finds that regional authority, territorial cleavages, and election (non-)simultaneity, limit SOEeffects. A second-generation literature sets out to find out for whom and when regional elections are 'first-order' instead of 'second-order'. A 'stakebased' approach also underlies this literature which finds that first-order regional voting is prominent among voters with strong regional identities and those who find regional government important. Third-generation research adopts a multilevel electoral system perspective and considers horizontal and vertical electoral spill-over between regions and between regional and national electoral arenas. This literature moves beyond the SOE-model by highlighting that regional elections are impacted by both vertical top-downincluding SOE-effectsas well as horizontal electoral spill-over, and that national elections are affected by vertical bottom-up electoral spill-over from regional electoral arenas.
In our literature review, we differentiated between three generations of regional election research to highlight the different timing of these literatures and we would like to emphasize that these literatures do not supersede each other. On the contrary, as attested by the election articles and reports featuring in this sixth annual review of regional elections, the first two-generations of regional election studies have not lost their relevance.
The main contribution of first-generation literature is to draw attention to territorial heterogeneity in second-order election effects which induced scholars to study regional elections 'on their own terms' and to develop a 'different or more nuanced picture'. A focus on regional elections does not necessarily mean that one finds different territorial effects from those that can be observed by looking at national election outcomes, as is exemplified by the election report on Croatia (Glaurdić and Fel 2023;this issue). However, putting regional elections centre stage, to the very least, draws scholarly attention to territorial heterogeneity in the vote as is well illustrated by the Romanian election report (Székely 2023;this issue). The election report on France (Gougou 2023; this issue) offers a 'candidate-centered politics model' as an alternative to the second-order election model which helps to increase our understanding of regional election outcomes. The Italian election article (Basile 2023; this issue) offers an example of how regional election research may offer insights which are difficult to gain when one focuses on national elections because of the methodological advantages associated with a comparison of regions within a country.
One of the main contributions of the second-generation literature is the development of a valence-voting model which helps to understand when and for whom perceived government performance is an important factor influencing vote choice in regional elections. The election article on Australia (Smith and Brown 2023;this issue) illustrates the explanatory value of the valence-voting model by revealing the impact of voters' evaluations of how regional and national government handled the COVID-19 crisis on regional vote choice. The election article on the UK (Griffiths et al. 2023; this issue) further develops the valence model of voting by pointing out that, in addition to evaluating competence in regional office, voters with strong regional identities also assess the extent to which parties defends territorial interests. The election article on Canada (Stephenson and Harell 2023; this issue) reveals that one of the main challenges for uncovering valence voting is to disentangle the impact of competence evaluations on vote choice from the impacts of party preferences and regional identities.
The main contribution of the third-generation literature is to point out that regional and national electoral arenas are interconnected and that election outcomes are impacted by horizontal and vertical spill-over between electoral arenas. One of the main challenges of regional election research is to identify the political-institutional variables that induce parties and voters to focus on other electoral arenas than the one in which an election is taking place. In addition, analysing elections in a multilevel electoral system requires a methodological toolkit that can track various types of horizontal and vertical electoral spill-over and, at the same time, is sensitive to territorial differentiation. Regional and national party systems are integrated and nationalized to very different degrees, but, as is illustrated by the election report on Peru (Haman and Školník 2023; this issue), even in the extreme cases of 'disconnection' some interlinkages between regional and national electoral arenas will remain because they are part of the same political system. Considering that regional and national party systems are integrated and nationalized to a moderate or high extent in most countries, one may expect that the SOE-model will remain prominent in regional election research.

Disclosure Statement
No potential conflict of interest was reported by the author(s).