Presidential Elections and European Party Systems (1848–2020)

Abstract

Political scientists have long been concerned that a popularly elected presidency may destabilize competitive party systems. We develop and test a new theory holding that the impact is more immediate and severe than previously assumed. Coexisting legislative and presidential coalitions first and foremost impede the evolution of predictable party interactions at the executive level, which is the domain of the cabinet. This quality has become accessible for comparative research thanks to the concept of party system closure. Using a new dataset for all European democracies since 1848, we show that presidential elections undermine party system closure in two ways: (1) by confounding patterns of government formation (notably under powerful presidencies) and (2) by disturbing electoral/legislative politics, which in turn affect executive politics. The former, direct effect emerges as dominant from a series of panel analyses and case illustrations. These findings have important implications for current problems of constitutional design and institutional reform.

Introduction

Popular presidential elections are fashionable. Of the thirty-six European republics, two-thirds require voters to elect the head of state. In the last decade, this group was joined by Czechia (2013), Turkey (2014), and Moldova (2016), and talks also took place in Kosovo (2014), Latvia (2017), Estonia (2017), and Armenia (2021). At a time when representative democracy is considered to be in crisis (Mair 2013) and more direct forms of democracy seem to offer a solution, it is not surprising to hear calls from populist parties for the introduction of presidential elections in, for example, Germany and Italy.

A personified representation of the people in the highest office has strong normative appeal. However, the implications of such an institutional choice for the political environment are another matter. After Linz (1990) famously proclaimed the ‘perils of presidentialism’, much scholarship has studied the relationship between presidential elections and concepts such as electoral volatility, fragmentation, and party institutionalization. The resulting picture is complex. While some studies confirmed the destabilizing impact of presidentialism on party systems brought about by ‘coattail’ effects, growing factionalism, and personalization of politics (Epperly 2011; Filippov, Ordeshook, and Shvetsova 1999; Golder 2006; Samuels 2002; Stepan and Skach 1993), others found that popular presidential elections can also further consensus and stabilization by ‘presidentializing’ intricate patterns of party competition (Mainwaring and Shugart 1997; Pasquino 1997; Samuels and Shugart 2010; Shugart and Carey 1992; Taras 1997).

While the evolution of the literature has yielded a much-refined understanding of institutional effects on party systems, we hold that Linz's initial warning remains highly relevant. The ‘perils’ of presidential elections materialize more vividly when the focus is shifted from electoral and legislative politics – which occupy most of the literature – to executive politics, referring to party control of the cabinet. The executive sphere stands out because the logic of cabinet formation naturally forces party systems of any complexity into simple patterns: government v. opposition, insiders v. outsiders. Presidential elections tend to destabilize this natural logic by overlaying it with conflicting interests and competing alignments. Accordingly, a tradition mainly organized around the ‘principal-agent’ theory has studied events of presidential interference, such as ministerial appointments or cabinet termination (Fernandes and Magalhães 2016; Schleiter and Morgan-Jones 2009a; Schleiter and Morgan-Jones 2009b; Schleiter and Morgan-Jones 2010; Sedelius and Ekman 2010; Shugart and Carey 1992). But do these events accrue due to a structural impact on executive politics comparable to manifestations of electoral volatility or legislative fragmentation? This question is an important desideratum: if popular presidential elections were to destabilize the executive arena without a detour through electoral and legislative politics, this would add an immediate pathway to Linz's ‘perils’.

Such a ‘shortcut’ of presidential interference would be particularly concerning in light of the literature on democratic backsliding, which shows that institutional decline is gradual and cumulative (Bermeo 2016; Levitsky and Ziblatt 2018) – we would not need to observe a pitchfork-wielding president barge into a cabinet meeting to be alarmed. Our study focuses on the system of party interactions in the executive arena – a concept known as party system closure (Mair 1997). Party system closure expresses the extent to which individual party choices build up to form stable and predictable patterns of executive control and thus combines the focus on structural attributes of the party systems literature with a focus on the political agency of the cabinet literature.

Building on this conceptual innovation, we show, theoretically and empirically, how the coexistence of two different electoral coalitions (that is, for legislative and presidential office) undermines the stabilization of the party system at the executive level. Much of the effect is direct, reflecting the strategic interference of presidents and presidential hopefuls over cabinet control. Since such interference requires political clout, we also show how its impact relies on the power of the presidency. Meanwhile, a smaller part of the effect is indirect, as popular presidential elections produce disturbances in electoral and legislative politics, which are then funnelled through to government formation via the chain of accountability.

Our study uses an original dataset of sixty-four European polities between 1848 and 2020. We focus on European politics because that continent offers numerous similar countries (in terms of history, economic development, separate legislative and presidential cycles, etc.). However, they differ significantly with regard to our variables of interest: the mode of election and powers of the head of state, the electoral system for electing the president, electoral fragmentation, and party institutionalization. Analytically speaking, this pattern approximates the logic of a ‘most similar systems design’ (which we will bolster with proper panel estimation and control variables). Moreover, Europe also displays favourable qualities in the temporal dimension, including countries with a long time series (for example, France, Norway, Switzerland, and the UK), as well as various cases where the mode of election and/or the powers of the head of state have changed (for example, Finland, Moldova, Portugal, and Slovakia).

The following section explains our theory of party system closure and illustrates our hypotheses using several cases. We then introduce the dataset and the methods employed in our analysis. In section three, we estimate the different ways (direct and indirect) in which popular presidential elections undermine party system closure. The article concludes by discussing our main findings and their policy implications.

Popular Presidential Elections and Party System Closure

Ever since Linz's (1990) seminal work on the ‘perils of presidentialism’, scholars have widely studied the effects of popular presidential elections on political stability, as reflected in systemic factors such as democratic performance (Elgie 2011), electoral fragmentation (Horowitz and Browne 2005), party institutionalization (Samuels and Shugart 2010), electoral volatility (Epperly 2011), institutional confidence (Ecevit and Karakoç 2017), and political disenchantment (Tavits 2009). The gist of this literature is that electoral and legislative stabilization is subject to regular interference from the presidential arena.

The focus of the ‘perils’ debate has a reflection in the wider literature on party systems. According to Sartori's acclaimed definition, a party system is ‘the system of interactions resulting from inter-party competition’ (1976, 44). Bardi and Mair (2008) explicated that such interactions occur on three functional levels: electoral, legislative, and governmental/executive. The former two levels have been studied extensively, aided by readily available indicators such as volatility or fragmentation. By contrast, the latter level – that of executive party systems – has often been lumped together with the legislative level. This may be justified in pure parliamentary systems, where the two operate in close proximity. However, the situation changes with a popularly elected presidency. In semi-presidential systems and many cases of full presidentialism, executive power is shared between the president and the legislature. Most presidents rule in an environment of multiparty competition for executive power and formal or de facto coalition formation.¹ Therefore, ‘[i]n non-parliamentary systems in particular, the logics of coalition formation and survival can differ greatly between the legislative and the governmental arena’ (Bardi and Mair 2008, 159). In other words, presidential influence may materialize differently, and more profoundly, at the executive level – an arena of high stakes in terms of office and policy, yet remote from electoral control.

The literature provides good reasons to be mindful of these concerns. Research over the last two decades has documented presidential influence in executive processes such as cabinet formation (Kang 2009), cabinet stability (Sedelius and Ekman 2010), cabinet control (Schleiter and Morgan-Jones 2010), prime ministerial appointments (Protsyk 2005), ministerial allocation (Schleiter and Morgan-Jones 2009b), and ministerial turnover (Camerlo and Pérez-Liñan 2015).

The salient point of this literature is its focus on events such as the hiring and firing of politicians or the rise and fall of cabinets. By comparison, our contribution is at a higher level of abstraction: does the myriad of events shaped by presidential intervention accrue to structural change in the executive ‘system of interactions’? This question is central to our purposes because only if the presidential impact was systemic and lasting would the executive sphere qualify as a pathway in the ‘perils’ debate.

A challenge to our endeavour is that the executive dimension of party systems has long eluded conceptualization and measurement. However, this changed with the concept and index of party system closure (Casal Bértoa and Enyedi 2016; Casal Bértoa and Enyedi 2021; Mair 1997). Applying Sartori's idea of a ‘system of interactions’ to the executive level, party system closure is defined as ‘the degree to which competition around governmental office is rigidly and predictably structured’ following ‘predictable relations among a stable set of political parties’ (Casal Bértoa and Enyedi 2021, 5, 20). Where party systems are closed, parties alternate in government in a complete manner, with few new coalitions or parties entering government. Conversely, in open party systems, newer (and unpredicted) party coalitions often access power, usually mingling with parties from the previous government.

Party system closure has three dimensions: access, alternation, and formula. Access expresses whether executive power is reserved for a stable set of parties or is more readily available to outsiders and newcomers. Alternation means that executive control changes from one identifiable party coalition to another and thus refers to predictable relations of parties. Formula captures the extent to which executive coalitions follow familiar, previously established patterns, explicitly linking the stability of parties with the predictability of their relations over time.

A composite index of the three dimensions was developed by Casal Bértoa and Enyedi (2016): the degree to which access to government is closed to newcomers is assessed by the percentage of ministries belonging to former governing parties, the degree to which alternation of executive power is identifiable is captured by the level of ministerial volatility from one cabinet to the next,² and the degree to which a coalition formula is familiar is measured by the percentage of ministries belonging to parties that have governed before in the particular combination of the cabinet concerned.³ The three indicators are then averaged into a party system closure composite index.

A detailed mapping of party system closure across space and time, including a discussion of its validity, is available in Appendix 3.

Conceptually, party system closure occupies the intersection of two research traditions: the literature on executive politics, whose central term is agency (Strøm 2000) and the literature on party systems, which focuses on structure (Weber and Franklin 2018). Party system closure essentially assesses events of political agency against the accrued stock of structure in party competition.

In this regard, closure can be contrasted with other party system dimensions, such as electoral volatility, fragmentation, and party institutionalization. Volatility is largely produced by voters and is only weakly related to party agency. Fragmentation is usually a product of social divisions and the electoral system. Finally, party institutionalization refers to the stability of political parties as organizations rather than how they interact (Casal Bértoa, Enyedi, and Mölder 2023). These conceptual differences are underscored by common examples of closed party systems (for example, Montenegro, characterized by the dominance of the Democratic Party of Socialists, or the Netherlands where, since the First World War, the party system has had a tripolar structure of competition between socialists, liberals, and Christian-democrats) despite low levels of party institutionalization (support for old parties – for example, Socialists – has declined, giving way to entirely new parties – for instance, the Popular Movement of Montenegro and the Dutch Forum for Democracy); and/or high parliamentary/electoral instability (both systems being highly fragmented and volatile); open party systems (for example, multipolar Cyprus and two-blocs Austrian First Republic) despite high party institutionalization (both had the same major parties since the beginning); and/or low fragmentation/volatility (less than three ‘effective’ parties in inter-war Austria, while Cypriot average volatility has been single-digit since independence). Large-N assessment reveals a similar picture.⁴

At the same time, executive politics are not independent of electoral or legislative factors. Rather, the executive level forms the ‘end of the food chain’ in that it depends on processes operating at the preceding levels (Casal Bértoa and Enyedi 2021, 21–2). Like an organic food chain, any effects tend to accumulate in the final ‘consumer’. This is illustrated in Fig. 1, which shows our model of the impact of popular presidential elections on party system closure. Such effects may be both direct and indirect. Indirect effects rely on the electoral and legislative levels as mediators, as popular presidential elections may fragment the party system and further factionalism and personalization at the expense of party institutionalization. These processes are largely known from the literature, summarized in Appendix 1.⁵ By contrast, our main contribution is to show how presidential coalitions of incumbents or candidates directly alter the structure of party competition at the executive level. The direct effect highlights how the ‘perils’ of presidential elections for party systems are more immediate and severe than is commonly assumed.

Figure 1. Theoretical path model of party system closure.

Direct (Executive) Effects

Regarding the novel element of our model, the direct effect represented by arrow ①, Bartolini (1984, 277) noted early on that party interaction in the electoral and legislative arenas might be affected by party interactions during presidential elections because of the ‘double electoral process’ of semi-presidential regimes. While he did not develop this idea much further, two arguments suggest why presidential elections might have a destabilizing effect on party interactions during government formation, reducing its predictability and, consequently, making party systems more open (Mair 1997).

To map out these arguments, let us begin with a (somewhat stylized) description of the ‘ordinary’ operation of parliamentary government. Cabinets generally result from negotiations between a limited number of parties whose leaders strive to secure the benefits of office, policy, and votes (Müller and Strøm). Majorities tend to reflect left-right logic (or some other policy dimension) or mere ad-hoc compromises (Kellam 2015). Importantly, each party caters to a defined portion of the electorate, and legislators are generally subject to party discipline. For these reasons, the number of possible governing coalitions is limited. Office (leaders, portfolios), policy (coalition agreements) and votes (constituencies) cannot be combined at will.

Now, let us add a presidential arena to the model. To win a presidential race, candidates do not need to follow the conventional lines of party competition. Quite the opposite, they might find it useful, or even necessary, to become more inventive. One reason is that popularly elected presidents face the electorate as a whole; another is the ‘winner-takes-all’ logic of the presidential race. Presidential candidates cannot afford to ignore significant segments of the population as easily as a party would. While parties can focus their campaigns on their support base and negotiate cabinet deals after the results are in, presidential candidates need to build their coalitions well ahead of an election. Moreover, since voters (unlike parties) may hold any imaginable combination of policy preferences, they can be approached via a more diverse set of issues. This generates the potential for formal and informal alliances that go beyond the constraints of existing patterns of party competition and cooperation. The decoupling is further facilitated because legislative and presidential elections in Europe are generally held at different times.

Separate electoral arenas are, however, linked in practice, as the principal-agent literature reminds us (Åberg and Sedelius 2020; Samuels and Shugart 2010; Strøm 2000). Presidential candidates do not campaign independently of the party system. Quite the opposite, organizational support is critical, and even personalistic campaigns of outside challengers generally find a party to back them. Since presidential candidates need to appeal to a wider segment of the electorate, broad and unusual coalitions are likely to be formed. These coalitions may include parties that would usually be considered too extreme and/or too small to play a role in executive politics, but ‘one or more of them can plausibly claim to represent the decisive electoral bloc in a close contest and may make demands accordingly’ (Linz 1990, 58) – demands that will include a prize that is otherwise unattainable – participation in government – as a reward for support in the presidential election (Freudenreich 2016). Overall, while both the legislature and the president have a mandate from the same electorate to control cabinet formation (Schleiter and Morgan-Jones 2010), these mandates differ systematically, which opens the party system to a greater variety of potential outcomes.

To give an example, the Social Democratic Party (SDP) of Finland gained executive office in 1937 in coalition with the Agrarian Party (SK), after having been excluded since independence in 1917 – just after SDP agreed to support SK's presidential candidate (Kallio), who only came in third at the elections held two months earlier. Similarly, the 2005 governmental coalition between the Bulgarian Socialist Party (BSP) and the Turkish-minority Movement for Rights and Freedoms (DPS), political enemies since the ethnic persecution of the latter by the Communist precursor of the former in the 1980s, has its roots in the 2001 presidential elections when the DPS provided essential electoral support to the BSP's candidate (Parvanov). More recently, in Poland, the electoral coalition between Law and Justice (PiS), Solidarity Poland, and Poland Together, created for the October 2015 legislative elections, originated in the presidential elections five months earlier, when all of them came together to support PiS’ candidate, Duda.

The abovementioned examples illustrate how a president's popular election can affect party competition patterns, either by forming previously unseen cabinet coalitions or by including political forces previously excluded from government. Moreover, because ‘the [presidential] “majority” generated might not represent a politically more or less homogeneous electorate or a real coalition of parties’ (Linz 1994, 21), and given popularly elected presidents' weaker partisan ties (Samuels and Shugart 2010), these alliances of ‘presidential’ parties tend to be fragile and short-lived since policy differences among ideologically heterogeneous member parties are likely to broaden with time. The ‘ordinary’ operation of parliamentary government, where parties occupying different electoral niches form coalitions based on vote shares and ideological proximity, is thus disturbed as the majoritarian character of presidential elections gives the patterns of party interaction a loose and temporary character. Yet large heterogeneous coalitions are sufficiently durable to have a tangible impact on the complexion of the executive (Grotz and Weber 2012).

Such complications are less relevant in parliamentary republics where, because a qualified majority is usually required, presidents are typically (s)elected either through compromise between the main parliamentary parties (including government and opposition) or by government majority (Strøm 2000). Because this ‘almost compulsory’, collaboration is seen as ad hoc and strictly occasional, the structure of party competition is rarely affected, and the presidential (s)election even tends to exert a ‘positive’ effect in the sense of renewing executive majorities. A clear example can be found in Greece, where all presidents were appointed either with the support of the governing coalition (Tsatsos in 1975, Karamanlis in 1980/1990, Sarzetakis in 1985, Stephanopoulos in 1995) or as a compromise between the two main contending parties (Stephanopoulos in 2000, Papoulias in 2005/2010, Pavlopoulos in 2015). The cartelization of presidential selection reflected a highly stable party system. The same can be said of other parliamentary republics such as Albania, Hungary, Malta, and, until 2014, Turkey.

A second argument, why popular presidential elections can hinder party system closure, derives from the higher status of the office in (semi-)presidential republics compared to parliamentary republics or even monarchies. The very fact of a popular mandate suggests that elected presidents, ‘although initially the choice of a small proportion of the electorate, […] represent a “true and plebiscitary” majority’ (Linz 1994, 21) and that, consequently, they do not need to respect the ‘usual’ process of coalition formation and party cooperation (Bartolini 1984; Duverger 1980). On this basis, popularly elected presidents have often used their greater legitimacy and prestige to stretch their influence into the making and breaking of governments and/or parliamentary coalitions (Kang 2009). For example, as Sedelius and Ekman point out, ‘by publicly criticising the government […] the president can make it more or less impossible for a prime minister to remain in office’ (2010, 523). Since the prime minister's performance is a critical expression of party power (Grotz and Weber 2017; Grotz and Weber 2022), this conclusion concerns the stabilization of parliamentary government as a whole.

A significant event occurred in 1996–7 when Bulgarian president Zhelev refused to invite the ruling BSP to form a new cabinet despite its clear majority in parliament. Instead, he insisted on early assembly elections through media addresses and public speeches (Schleiter and Morgan-Jones 2005). Four years earlier, leaning on his popular mandate, Zhelev had already managed to convince the DPS leaders to join the ex-communist BSP and topple the government formed by his ‘own’ party, the Union of Democratic Forces (Ganev 1999). On both occasions, the process of party system stabilization was affected by the president's intervention. Another clear example is Iceland in the late 1950s, when the Social Democrats first formed a minority government, which was almost immediately followed by the creation of a coalition with the centre-right Independence Party, mainly due to presidential ‘pressures’ (Kristinsoon 1999).

The head of state can also play an active role during government formation in some constitutional monarchies or parliamentary republics (Schleiter and Morgan-Jones 2009a). However, because a monarch or indirectly elected president is expected to stand neutrally above the partisan battle, they cannot claim to be ‘tribunes’ of the partisan battle (Baylis 2007, 89). Thus, in countries like Italy, Germany, or even Spain, the head of state will tend to follow the existing patterns of party interaction, strengthening rather than weakening the system of competition as a whole. This is in clear contrast to (semi-)presidential republics, where popular presidential elections produce unpredictability and instability. For all these reasons, we hypothesize:

H1:

Party system closure will be lower in the presence of a popularly elected presidency.

Institutional Moderators: Presidential Power and the Electoral System

Before we proceed with empirical estimation, this section serves to define the institutional scope conditions of our theory.

First, our argument assumes that an elected president is not purely a figurehead; s/he carries some political clout. Therefore, we expect the effect on party system closure to be stronger when the presidency is powerful.

Once again, a rich literature, inspired by the principal-agent theory, documents negative relationships between presidential power and democratic consolidation (Mainwaring and Shugart 1997; Shugart and Carey 1992), cabinet stability (Fernandes and Magalhães 2016; Protsyk 2005), electoral fragmentation (Clark and Wittrock 2005; Elgie et al. 2014; Golder 2006; Hicken and Stoll 2013), and party institutionalization (Samuels 2002; Samuels and Shugart 2010). Studies of the structure of competition have been fewer. A notable exception is Meleshevich, who concluded that ‘a presidency which occupies a predominant position in a political system and possesses strong political powers is not conducive to the development of a party system’ (2007, 146). This conclusion is based on an association of Russia and Ukraine's strong presidents and weakly institutionalized party systems, in clear contrast to the higher levels of systemic stability in the Baltic states, all with much weaker presidents. The logic is that when presidents, typically ‘outsiders’ without previous governing experience, can unconditionally appoint and/or veto prime ministers and/or ministers, dissolve parliament, and/or veto legislation, party interactions will be under their thumb (Schleiter and Morgen-Jones 2009c). It is no coincidence that notable cases of presidential interference with government formation occurred in countries with rather powerful presidencies (for example, Louis-Napoléon and De Gaulle in France, Hindenburg in Germany, Wałęsa in Poland, Putin in Russia, Kuchma in Ukraine).

More recently, however, Andrews and Bairett (2014) showed that, at least for post-communist Europe, it is not presidential power per se that negatively affects electoral stability but the combination of popular presidential elections and strong presidencies. For example, the Czechoslovak and Hungarian party systems were stable despite having strong but indirectly elected presidents. In this view, presidential power amplifies elite incentives for political coordination. Similarly, at the legislative level, powerful, popularly elected presidencies would foster the formation of new parliamentary groups or the creation of previously unseen parliamentary coalitions to pass relevant legislation, consequently increasing legislative instability. We expect this pattern to be reflected on the executive level.

H2:

The direct negative effect of a popularly elected presidency on party system closure will be more pronounced when the head of state is powerful rather than just a figurehead.

Importantly, while H2 describes an interaction, including presidential power as a control variable will also allow us to isolate the effect that comes with popular election.

Besides the power of the head of state, the direct effect of popular presidential elections is likely to be conditioned by the electoral system. In fact, the operation of the electoral system is not just a moderator but a critical element of our theory. The basic distinction is between majority systems (like France) and plurality/Single Transferable Vote (STV) (like Iceland or Ireland).

In plurality systems, the influence of popular presidential elections on party system closure is most obvious. Electoral plurality in a single-round contest may depend on any number of votes, which may be eventually provided by unusual partners – such as in Finland, where ‘the main changes in […] politics have often been made in connection with [plurality] presidential elections’ (Paloheimo 2001, 93). We thus expect a strong effect in plurality systems.

Regarding majority systems, we need to distinguish between those elections where a majority winner emerged in the first round (as happened in the French Second Republic or the First Portuguese Republic) and those where the winner was elected in a second runoff round (as happened in the French Fifth Republic and the Romanian and Ukrainian elections). Importantly, this distinction is not between different institutions but between different elections with similar institutions.

We expect that the destabilizing effect of presidential elections will be concentrated in those cases where a second round between the two leading candidates is held. The run-off ballot essentially resembles the first and only round under plurality systems in that it will be decisive. The window for popular presidential elections to affect party system closure should open if none of the conventional coalitions succeeds, tempting candidates to vie for the best unconventional coalition.

One clear example of disturbed ideological and historical divisions is Poland in 2005. That year, Poles elected – almost concurrently – a new president and a new legislature. Facing competition from the post-communist Democratic Left Alliance (SLD), the two main post-Solidarity parties, Civic Platform (PO) and PiS, implicitly agreed before the elections that they would form a coalition government. The situation changed when Cimoszewicz, SLD's presidential candidate, withdrew due to a corruption scandal just before the parliamentary elections, which PiS won, beating PO by a close margin, with the populist agrarian Self-Defence (SRP) in third place. Fourteen days later, Tusk (PO's candidate) won the first round of the presidential elections by a similar margin. Both Tusk and Lech Kaczyński, PiS’ candidate and the runner-up, received a letter from Lepper, SRP's candidate, promising each of them his support in exchange for his participation in a future cabinet. In the second presidential debate before the runoff round, Tusk categorically refused Lepper's offer, but Kaczyński did not. The latter became president of Poland, and his twin brother Jarosław arranged a coalition government with the SRP and the League of Polish Families. The result was a blow to the competition structure that had characterized Polish politics since 1989, with fierce opposition between post-communist and post-Solidarity parties and the permanent exclusion of populist or ‘anti-establishment’ forces (Casal Bértoa and Guerra 2018).

While the Polish example stands out in dramaturgy, it is not unusual. Other instances where the composition of the government was affected by the patterns of party collaboration in immediately preceding presidential elections include Croatia (2000), Serbia (2004), Ukraine (2004), Bulgaria (2005), Slovenia (2008), Poland (2015), and France (2017).

Less destabilizing effects will occur in majoritarian elections where one of the presidential candidates has already succeeded in the first round. Initially, this is because minor parties, betting on an eventual second round, will not coordinate as much, presenting their candidacies separately. Thereupon, the abrupt end of the contest will deprive minor – in many occasions ‘anti-establishment’ – parties of their opportunity to ‘blackmail’ the winner, who becomes president by way of their own strength. If the presidential arena produces a winner by ‘conventional’ means, we thus expect less experimentation with ‘unconventional’ solutions in the executive arena. For all these reasons, we expect that:

H3:

The direct negative effect of a popularly elected presidency on party system closure will be more pronounced in elections where the final round was bound to be decisive (either the second round of a majoritarian system or the first/only round under plurality), compared to majoritarian elections that produce a winner in the first round.

To summarize our hypotheses, the popular election of a president will decrease party system closure (relative to constitutional monarchies and parliamentary republics), particularly under powerful presidencies and when electoral systems provide incentives for creative coalition building.

Data and Methods

Our dataset is structured as a panel. The unit of analysis is the legislative period. Each period begins with an election, which produces observable outcomes for electoral fragmentation and party institutionalization and constitutes the players for the following competition for executive power. Therefore, the legislative period provides the most natural frequency with which to measure the structure of the executive party system. The data comprise 735 periods from 64 political systems between 1848 and 2020. This includes all European countries that, at the time concerned, fulfilled three criteria: (1) a score of at least 6 in the Polity IV index,⁶ (2) at least one universal (male) suffrage election, and (3) that governments are not formed at the exclusive pleasure of a monarch. A list of these polities and time periods is available in Appendix 2.

Our dependent variable, party system closure, is operationalized using the original data described in section 2 (Casal Bértoa 2021). Calculations are repeated for each cabinet formed during a legislative period, and their scores are aggregated to the period level. This makes sure that the units are comparable for analysis, while cabinets formed after and between elections are equally included in the procedure, given that they may all be subject to presidential interference. The final variable is formed by averaging over a moving five-period window, typically covering about seventeen years. Following earlier work, this operationalization reflects ‘that party system institutionalization, the latent concept behind closure, is not only determined by the events taking place in a particular year but rather describes more fundamental properties that will only unfold in the longer term’ (Casal Bértoa and Weber 2019, 235) – thus accommodating our theoretical focus on structure as opposed to events. A five-period window is long enough to express patterns of inclusion and exclusion and short enough to register marked changes in single periods.

If the mode of election of the head of state changes in a country, two periods of party competition are distinguished. This prevents patterns from a period without popular presidential elections from having an endogenous influence on a later period where popular presidential elections were held, or vice versa.

In addition, we also replicated the analysis with another dependent variable, the percentage of non-partisan cabinet positions. The two are complementary: while closure expresses the structure of executive party systems, non-partisanship measures the extent to which party control is limited as such. Appendix 9 shows that these dynamics are quite similar.

Our main independent variable is a dummy for the presence (1) or absence (0) of popular presidential elections. The data come from Elgie (2018) and Nohlen and Stöver (2010).

The second institutional variable, presidential power, is based on Siaroff (2003), the most comprehensive source historically. Siaroff identified nine powers of the head of state, from which we use seven.⁷ We identify a ‘figurehead’ by the absence of these powers. In contrast, holding one or more of them defines an ‘executive’ head of state.

Besides its coverage, Siaroff's measure has the advantage of applying a holistic definition of power. Unlike other measures, which focus on constitutional provisions only, Siaroff sought to code the de facto power of the president. The difference between the two is due to political conventions that allow presidents to interpret their constitutional mandate more or less liberally. This focus accommodates our theoretical interest in the ability of the president to influence government formation even when such interference may not have a constitutional basis. At the same time, a de facto measure of power comes with a risk of endogeneity if the role of the head of state in party politics has already influenced the coding. Therefore, we triangulated our findings using a purely constitutional index (Doyle and Elgie 2016), with similar results (see Appendix 10).

Regarding the hypothesized mediators, electoral fragmentation is measured using the effective number of electoral parties (Laakso and Taagepera 1979). Data are from Bormann and Golder (2013) and Casal Bértoa (2021). For party institutionalization, we use the average age of relevant political parties (Tavits 2005).⁸ The variable is logged to express a marginally decreasing effect, as is common in recent work (Schleiter and Voznaya 2018). Data are from Casal Bértoa (2021). We also include five control variables common in the field:

• − age of democracy (years logged), to account for the tendency of party systems to stabilize over time;

• − the electoral system used for legislative elections (dummy for majoritarian system, either fully or partially), since this is our unit of analysis;

• − ethnic fractionalization, as the other standard predictor of political fractionalization (measurement details in Appendix 4);

• − the precursor regime (royal, military, or civilian), which may affect the democratic trajectory (Cheibub, Ghandi, and Vreeland 2010, extended to include pre-Second World War cases);

• − economic performance, which has been shown to affect party system stabilization (Casal Bértoa and Weber 2019). We use GDP growth (in the year preceding an election) as the most comprehensive and widely available economic indicator. Data are from the Gapminder database.

Descriptive statistics for all variables are available in Appendix 5.

An obvious issue with the data structure concerns autocorrelation. Party system closure, electoral fragmentation, and party institutionalization all reflect structural qualities that are not easily thrown off course by single elections. Closure, in particular, is explicitly conceptualized and measured as an accumulated property. Therefore, the error term of the regression equation is correlated over time. We use feasible generalized least squares (FGLS) to estimate the model parameters while allowing for serial correlation within polities.⁹ We also allow for heteroskedastic error across panels because the model may not fit all polities equally well.

Besides avoiding autocorrelation bias, FGLS also comes with two other benefits. First, it helps with a potential endogeneity problem, which unfolds if institutional change involving presidential power or mode of election is a consequence, rather than a cause, of party system closure (Elgie 2011). While there is good reason to doubt such reverse causation historically,¹⁰ to the degree that it does occur, it will be reflected in previous values of party system closure, the quantity consulted by FGLS to condition parameter estimation. Thus, our inferences are actively guarded against causal fallacy.

Second, FGLS also helps with the country dimension of the data, as factoring in past values of the dependent variable ‘provides some “protection” against the effects of unobserved time-constant variables’ (Berrington, Smith, and Sturgis 2006, 23). Nonetheless, it would be unreasonable to assume that FGLS accounts for all polity-level confounders. We will also estimate more conservative models with fixed polity and decade effects to verify that our inferences are not due to under-specification.

Findings

Empirical results will be presented in two steps. First, we estimate the direct and indirect effects of popular presidential elections on party system closure and calculate the total effect. Next, we scrutinize the direct effect (our main contribution) by bringing presidential power and the number of election rounds into the equation.

‘Direct’ and ‘Indirect’ Effects

Table 1 shows the estimated direct effects on party system closure. As can be seen in M1, our first hypothesis receives support from the data. Popular presidential elections have a significant negative coefficient, with party system closure estimated at 4.58 points lower in countries where citizens elect their president.

Table 1. Estimated effects of popular presidential elections

FGLS coefficients with standard errors in parentheses.

N (elections) = 735. Panels (unbalanced, heteroskedastic) = 64.

Six single-observation panels are uninformative.

**Significant at 99 per cent; *significant at 95 per cent; +significant at 90 per cent.

Importantly, this estimate is controlled for electoral fragmentation and party institutionalization. These two variables show the effects expected from the literature. Fragmentation reduces party system closure by 1.07 points for each ‘effective’ electoral party, while institutionalization increases closure by 0.32 points for each 10 per cent increase in average party age.¹¹ Among the control variables, age of democracy and economic growth promote closure as expected; other controls do not have significant direct effects.

Based on the evidence, we can thus conclude that popular presidential elections directly impair party system stabilization at the executive level, controlling for the effects of the electoral/legislative levels. Moreover, as shown in Appendix 7, the effect occurs under different electoral systems.

Moving on to the effects of popular presidential elections on electoral fragmentation and party institutionalization, our analysis supports relevant expectations from the literature: fragmentation is increased by an estimated 0.35 ‘effective’ parties (M2), and party institutionalization is reduced by about 17 per cent (M3)¹².

Popular presidential elections then affect party politics at the executive and electoral/legislative levels. In turn, the electoral/legislative levels affect the executive level in their own right. As was visualized in Fig. 1, presidential elections thus have a direct effect on party system closure and two indirect effects, which are mediated by electoral fragmentation and party institutionalization. Figure 2 reproduces Fig. 1, filling it with the respective coefficients from M1, M2, and M3.

Figure 2. Empirical path model of party system closure.

On this basis, the total effect of presidential elections on party system closure can be calculated as the sum of three quantities: the coefficient of the direct effect, the product of the two coefficients of the fragmentation path, and the product of the two coefficients of the institutionalization path (Baron and Kenny 1986). The multiplication ensures that the coefficients of the three paths are on the same scale. In particular, popular presidential elections increase electoral fragmentation by 0.35, and a one-point change in fragmentation decreases closure by 1.07. Thus, the change in fragmentation induced by presidential elections decreases closure by 0.35 × 1.07 = 0.37 points (95 per cent CI = [−0.72, −0.06]).¹³ For party institutionalization, the effect of popular presidential elections (−0.18) and the effect on party system closure (3.31) yield an overall path of −0.18 × 3.31 = 0.60 points (95 per cent CI = [−0.97, −0.30]). The total effect of popular presidential elections on party system closure is thus estimated as follows:

$${\rm Total}\;{\rm effect} = {-}4.58 + ( {0.35 \times{-}1.07} ) + ( {-0.18 \times 3.31} ) = {-}5.55$$

with 95 per cent CI = [−7.56, −3.62]. To appreciate the size of this effect, let us evaluate it relative to the variability of the dependent variable: popular presidential elections decrease party system closure by 39 per cent of its general standard deviation (14.3) and by 45 per cent on the polity level (12.4). Expressed in terms of a related phenomenon, the effect on closure implies that the odds of parties entering the cabinet for the first time increase by about 40 per cent. For a single institution, these differences appear quite considerable.

Notably, the direct effect of presidential elections (−4.58) accounts for most of the total effect. The executive level (the main focus of our analysis) not only adds to the indirect effects of the electoral/legislative levels discussed in the literature but also dominates the larger model of party system closure.

The same impression also arises from the demanding assumptions of mediation analysis. While the direct effect merely requires control for exposure-outcome confounding, the indirect effects require three additional assumptions: control for mediator-outcome confounding and exposure-mediator confounding, and that no mediator-outcome confounder is itself affected by the exposure (VanderWeele 2015). While our models contain a comprehensive set of controls, this larger set of assumptions is apparently more vulnerable to violations.

One strategy to eliminate omitted controls is to include fixed effects for polity and decade in the regressions. This accounts for institutional, regional, and historical patterns that may confound our estimation strategy. Results in Appendix 6 show that there is reason to be cautious about the dual mediation model. While this test strengthens the direct effect, the indirect effects are weakened (particularly institutionalization). Thus, our study's main contribution, both theoretical and empirical, is the impact of direct presidential interference with executive politics.

Scrutinizing the Direct Effect

Turning to the scope conditions of the direct effect, we first need to recognize that not all presidencies are alike. Some are more powerful than others, which should affect whether parties in parliament will feel urged to consider presidential preferences when forming the executive. Thus, as expected by our second hypothesis, the election of an executive president should have stronger negative effects on closure than the election of a figurehead president.

Second, the use of majority rule for popular presidential elections provides an interesting opportunity to test the plausibility of our theory. If competition for broad presidential majorities indeed spills over into party politics, then the negative effect of presidential elections on party system closure should be stronger in those elections where a second round was held. This is the expectation of our third hypothesis.

In M4 of Table 2, we begin by adding presidential power as a predictor along with the presidential election. As can be seen, this variable, too, has a negative effect on party system closure, but the coefficient is insignificant and clearly weaker than that of popular presidential elections, which remains significant.

Table 2. Refined models of party system closure

Notes: see Table 1.

^a Cases in the base category, parliamentary systems, do not have natural values for the vote share of the leading candidate. The mean value of the observed cases was used for comparison.

However, a proper test of H2 requires an interaction of presidential elections with presidential power. M5 shows that the addition of a multiplicative term indeed produces the expected result. While the negative effect of the popular election of a figurehead president is a mere 0.70 points, the election of an executive president reduces closure by an additional 5.35 points, meaning 6.05 points overall. Thus, the estimate of M1 (−4.58) really was a mix of a strong effect for executive presidents and a much weaker (and insignificant) effect for figureheads. As expected in H2, the negative impact of popular elections is concentrated on those presidents whose power commands respect for their preferences in party politics.

The interaction in M5 can also be interpreted with a focus on the effect of presidential power itself. As indicated by the small and insignificant coefficient of that variable (0.33), power does not affect party system closure if the head of state is not elected by the people. Only in interaction with popular election does power reduce closure, with the aforementioned difference of 5.35 points between a figurehead and an executive president. The impact of presidential power is thus fully contingent on popular election, which supports the argument that it is the competition for presidential majorities that impedes party system stabilization, not a strong presidency per se.

M6 tests H3 by breaking down popular presidential elections into three types: single-round majority, two-round majority, and plurality/STV. As expected, we find that the effect is concentrated on plurality systems and majority elections that went into a second round (−3.63 and −5.74), while the coefficient for majority elections that were decided in the first round is small and insignificant (−0.74). This finding should lend considerable plausibility to our theory because the difference does not just reflect the number of election rounds. The effect occurs in majority elections with two rounds and in plurality elections, which, by definition, have only one round. The two types have in common the knowledge that all relevant actors knew the next round would be the last, bringing coalition-building incentives to the fore. Confirmation of H3 thus illustrates the mechanism through which party system closure is impaired by presidential influence on government formation.

M7, our last model, seeks to dispel possible doubts that the effect for majority systems may be related to reasons why a second round had to be held in the first place rather than to the second round itself. In this case, the true underlying variable would be the performance of the leading candidate: the stronger the first-round winner, the more structured the party alliances are at this time, and the less likely a second round. However, controlling for this variable shows that the alternative explanation carries little weight. The difference between second- and first-round elections remains large and significant. Thus, the institutional explanation prevails over a mere contextual one. Moreover, the coefficient of the vote share of the first-round leader is negative, suggesting that a strong first-round performance does not indicate structured party alliances; rather, it indicates that a less structured system is easily swayed by a strong candidate.

Finally, we explored the interaction of our two scope conditions. Appendix 8 shows that the more powerful the presidency and the more distinct the incentives for coalition-building in the presidential arena, the stronger the negative effect on party system closure.

Conclusion

Institutions have important implications for the way democracy functions. This article studied the institutional effect of the head of state, in particular their mode of election and political power, on patterns of party interaction in executive politics. The executive sphere has been somewhat of a stepchild in the debate about the ‘perils’ of presidentialism despite its relevance for policy formulation, democratic accountability, and regime stability. To remedy this situation, we took a unique theoretical approach, which combined the structural focus of the party systems literature with the agency focus of the cabinet literature. This allowed us to model a distinctively executive dimension of party systems, labelled party system closure.

Following Åberg and Sedelius' recent call ‘for more methodologically sophisticated and empirically sound large-N analysis’ (2020, 1,132–3), we examined the destabilizing impact of popular presidential elections in European party systems since 1848. Our empirical analysis demonstrates two such mechanisms. The first mechanism – clearly the stronger of the two and the article's main contribution – points to a direct link between popular presidential elections and party system closure. In particular, the dual electoral process typical of (semi-) presidential republics, requiring presidential candidates to have a broader electoral base across ideological lines, interferes with the pre-existing structure of party competition, particularly with the stability and predictability of government formation. The second mechanism is indirect. It links party system closure to popular presidential elections through the compromising effects of the latter on electoral fragmentation and party institutionalization.

These findings are highly relevant to the current problems of democracy. The course of democratic decline often begins in the executive sphere (Bermeo 2016). Accordingly, executive stability – and party system closure in particular – have been identified as safeguards of democracy (Alesina et al. 1996; Casal Bértoa 2017; Casal Bértoa and Enyedi 2021, 244–53; Diskin, Diskin, and Hazan 2005). Closed-party systems are more robust to the seizure and ‘aggrandizement’ of executive power than open systems. This is not to deny that party systems can become ‘too closed’ under specific circumstances (Casal Bértoa and Enyedi 2021, 255–7; Mölder, Enyedi, and Casal Bértoa 2022). However, the order of events matters. While democratic backsliding depends on various factors, recent cases in Europe – particularly Hungary, Poland, and Turkey – all saw dominant parties arising out of the turmoil of high volatility. If anything, the initial openness of a party system makes it susceptible to authoritarian tendencies rather than the ensuing closure. In other words, the opportunity to build a resilient executive sphere needs to be seized while democracy is relatively safe. A popularly elected presidency impedes this process, as we have shown, and thus generates inadvertent ‘perils’.

Our findings then have important implications for constitutional design in new democracies. In contrast to the view of semi-presidentialism as the ‘best of both worlds’ (Duverger 1997; Sartori 1997), we identify a critical liability for party system stabilization. New democracies, which already face a multitude of challenges, should refrain from introducing additional fragility through popular presidential elections during their critical early years.

Established democracies have more room for manoeuvre. Institutional reform of the presidency may serve to revive sclerotic party ‘cartels’ in parliamentary government and further democratic accountability (Katz and Mair 2018; Mölder, Enyedi, and Casal Bértoa 2022). Empirically, however, the few cases when systemic cartels were broken (for example, France in 2017 and Austria in 2000) have resulted in intra-party frictions and the weakening of parliamentary government. The very existence of two competing electoral arenas implies that the legislative party system will not seamlessly absorb a new constellation produced by the president. While some innovation may be beneficial, there is no guarantee that the process will come to a halt before entering ‘perilous’ territory.

More pragmatically, if any anticipated benefits of popular presidential elections were found to outweigh the impact demonstrated here, caution advises that the power of the presidential office be allocated sparingly and interpreted conservatively, that the cabinet be given solid anchoring in the legislature, and that electoral systems and party laws be designed to contain fragmentation and foster institutionalization. After all, constitutional design needs to balance the interplay of numerous institutions. Our analysis suggests that the executive party system deserves special attention in this context.