Presidents, Prime Ministers and Legislative Behaviour: The Conditional Effect of Presidential Legislative Powers on Party Unity

Abstract

This article proposes a novel theoretical framework to account for the combined effects of regime type and patterns of executive authority on legislative party unity. We argue that broad presidential legislative powers favour coordination between the president and legislative parties under pure presidentialism, whereas under semi-presidentialism, strong presidents increase the potential for intra-executive conflict, submitting parties to cross-cutting pressures. We expect higher levels of legislative authority to increase party unity under presidentialism but decrease under semi-presidentialism. Moreover, when presidents are endowed with limited legislative authority, semi-presidentialism produces higher levels of party unity than presidentialism, but for sufficiently high levels of legislative authority there should be no difference across regime types. Our analyses of 1,586 pooled observations for 72 democracies from all regions of the world using the V-Dem measure of party cohesion demonstrate that presidential legislative authority, in combination with regime type, is indeed a key predictor of party unity.

Comparative research has argued that disciplined and cohesive party organizations are much less likely to develop under presidentialism than parliamentarism. A crucial reason for these differences lies in the fact that government survival is independent from legislative support in presidential countries. Under presidentialism, party (or coalition) unity has no direct impact on the odds of government survival and, therefore, having the support of a strong parliamentary organization is not a necessary condition for sustaining control over the national executive. Fused power systems create rather distinct incentives because legislators affiliated with the governing party or coalition need to stick with their parties and support the government's agenda to avoid bringing down the government and, in some cases, incurring in electoral losses due to anticipation of elections (Diermeier and Feddersen 1998; Huber 1996; Linz 1990; Mainwaring 1993; Mainwaring and Shugart 1997; Owens 2003; Shugart and Carey 1992).

In addition to separation of survival, presidential countries are characterized by separation of origin: executives and legislatures are elected in separate elections and, most often, respond to distinct constituencies (Samuels and Shugart 2010). Separation of origin implies that legislators and presidents, even when they share the same party affiliation, may not be evaluated by voters on the same national policy issues. In parliamentary countries, in contrast, because the cabinet originates from the legislature and is tied to the same electoral majority, the electoral fortunes of legislators tend to be strongly associated with government performance (Andrews and Bairett 2014; Clark and Wittrock 2005).

These institutional differences between presidentialism and parliamentarism have been assumed to have clear and important implications in terms of government performance. Party unity is a desirable feature of democratic polities because where parties are highly cohesive and disciplined, governments can approve policy at lower cost. Where legislators' behaviour is harder to predict on the basis of their partisan affiliations, the bases of government support are less certain and, thus, chief executives may have incentives to form oversized (and harder to manage) government coalitions or they may rely on ad hoc legislative coalitions by bargaining with individual legislators (Kellam 2015; Meireles 2016). Moreover, internally divided parties lead to less stable and enduring government coalitions (Laver 1999; Martínez-Gallardo 2012).

Although previous research has found evidence that levels of party unity are indeed lower in presidential than in parliamentary countries (Carey 2007; Coman 2015), the literature has neglected the wide institutional variation that exists across countries with elected presidents. In part, this reflects the lack of readily available and comparable cross-national data on party unity. Extant research has focused mostly on a single country or on a small set of countries (Alemán and Navia 2009; Figueiredo and Limongi 2000; Hix 2004; Sieberer 2006). This is unfortunate because not all presidents are created equal, and presidential legislative powers tend to vary widely across countries (Metcalf 2000; Shugart and Carey 1992).

Broad comparative theorizing on polar ideal-types of presidentialism and parliamentarism is unlikely to provide a solid ground for empirical testing. In particular, semi-presidential regimes combine an elected presidency with a prime minister dependent on the confidence of the legislature, and this dual executive authority likely creates incentives and constraints distinct from those observed in pure presidentialism (Elgie 2011). Moreover, presidential powers and executive–legislative relations vary widely both within and across distinct systems of government (Cheibub et al. 2014; Tavits 2009).

The use of presidential power as an explanatory variable has become widespread in recent years to explain different types of outcomes. However, there are few studies that travel both across regions and across regime types (Doyle and Elgie 2016; Shugart and Carey 1992). At the same time, the simple presidential–parliamentary distinction has been abandoned in favour of more refined analyses that combine regime type with the nature and extension of executive powers. As pointed out by Margit Tavits (2008: 12–16), although much of the literature assumes that directly elected presidents are always powerful, there are several institutional constraints and incentives that may affect the president's ability to be politically active in day-to-day government.

Under presidential or semi-presidential constitutions, presidents have different levels and kinds of legislative powers and control different resources and incentives to encourage intra-party legislative cohesion (Åberg and Sedelius 2020: 1127; Owens 2003). Although there exist a number of studies that analyse how presidents' legislative authority shapes legislators' behaviour under pure presidentialism (Alemán and Navia 2009; Figueiredo and Limongi 1999, 2000), there is still a paucity of comparative research looking at variation across countries, government systems and levels of executive authority altogether. By connecting these dots (presidential power, legislative behaviour and a non-binary view of regime type), this article is intended to begin to fill that gap.

The article focuses on the institutional determinants of party unity in countries with popularly elected executives. We propose and test a simple theoretical framework to account for the combined effects of the system of government and patterns of executive authority on the levels of legislative party unity. Starting from José Antonio Cheibub's (2007) argument that parties will have greater incentive to behave in cohesive fashion when the executive branch exerts centralized control over the legislative agenda, we argue that this general expectation takes on different contours under each regime type.

In pure presidential systems, presidents may exert substantial influence over the legislative agenda due to institutional prerogatives conferring a first-mover advantage (e.g. the exclusive right to initiate bills in key areas) and reactive powers that allow them to prevent the enactment of undesired policies (veto powers). When agenda powers are centralized, individual legislators and their parties have greater incentives to cooperate with the executive, to the extent that they obtain access to the policy and office benefits that are central to their own political survival mainly by participating in government. In such context, presidential systems can feature coalition governments similar to those observed in parliamentary systems (Alemán and Navia 2009; Amorim Neto et al. 2003; Cheibub 2007). When presidents lack formal powers to set the legislative agenda, they will have fewer mechanisms with which to shape legislators' votes, whereas legislative parties' influence in policymaking will be less dependent on cooperation with the executive. All else being equal, thus, party unity should be lower when presidents lack significant legislative powers, and higher when they have substantial legislative authority under pure presidentialism.

The picture is more nuanced under semi-presidentialism. A president endowed with substantial legislative powers is unlikely to centralize control over the legislative agenda because she will usually share policy authority with a prime minister supported by the legislature. Rather, semi-presidential constitutions that grant substantial legislative powers to an elected president tend to increase the potential for conflict between the latter and a prime minister supported by a potentially hostile majority (Roper 2002; Sedelius and Ekman 2010: 524; Sedelius and Mashtaler 2013). Intra-executive conflict, in turn, fosters cabinet instability and submits parties to cross-cutting pressures, leading to lower levels of party unity. On the other hand, when presidents are weak and lack formal legislative powers to challenge undesired policies pursued by a competing prime minister, the odds of success for a confrontational strategy are much lower. In these settings, presidents will face stronger incentives to pass their agenda by coordinating with the prime minister and the legislative majority and, as a consequence, one should observe higher levels of party unity.

Based on the theoretical expectations outlined above, we test three hypotheses in the next sections: (1) as the legislative authority of the president increases, party unity will increase under pure presidentialism, but decrease under a semi-presidential constitution; (2) weak presidents with restricted legislative authority will be associated with significantly more cohesive legislative party behaviour in semi-presidential than in presidential democracies; and (3) levels of party unity should not differ substantially between presidentialism and semi-presidentialism when presidential legislative authority is sufficiently high.

To test these hypotheses, we rely on the comparative measure of party cohesion developed by the Varieties of Democracy (V-Dem) project and on our own extension of Matthew Shugart and John Carey's (1992) original data on presidential legislative powers.¹ Our data set covers a total of 72 democracies from the late 1970s to the mid-2010s, with a total of 1,586 observations.

Our results point to the importance of considering the combined effects of presidential legislative powers and regime type on legislative party unity. We find that weak presidents are associated with higher levels of party unity under semi-presidentialism, but not under presidentialism. Statistical results also show that presidents with wide legislative authority are associated with higher (lower) levels of party unity in presidentialism (semi-presidentialism), which challenges the assumption that presidential strength and party unity are linearly and inversely related. The findings also contribute to the growing literature on semi-presidential democracies. The semi-presidential regime type is indeed unique (Schleiter and Morgan-Jones 2009: 891), in that presidents' legislative authority generates opposite effects when compared to those observed under pure presidentialism.

Presidents, prime ministers and party unity

Voting unity within legislative parties is, most often, a result of two distinct types of political processes. First, legislators may vote together because party leaders have the resources and authority to reward loyalty and punish breaches in discipline. Second, voting unity may result from party cohesion – that is, the extent to which the party's delegation is composed by legislators with similar preferences (Carey 2007). In this article we focus on the more easily measured and directly observable phenomenon of party unity, defined as the degree to which party members act in unison (Sieberer 2006). We assume that variation in party unity across distinct political systems is likely to reflect variation in the willingness and capacity of party leaders (or other party principals) to enforce discipline, as well as variation in terms of shared preferences (cohesion). Discipline and cohesion are, therefore, intervening, often unobservable variables that account for the level of party unity.

As a general rule, presidential systems tend to create weak incentives for the construction of party organizations that can either enforce discipline or secure adherence of party members to the party's programme. The opposite is true in parliamentary countries, where political elites have strong incentives to create legislative party organizations that can form and sustain governments. Maximizing legislative seats is a major goal of political parties in parliamentary countries because obtaining a majority (or a plurality of seats, in a multiparty legislature) is often a necessary condition for a party to have the opportunity to form the government (Clark and Wittrock 2005; Samuels and Shugart 2010). Once a party or coalition forms the cabinet, the members of the prime minister's party (or coalition) need to unite behind the government's agenda (Diermeier and Feddersen 1998; Huber 1996; Owens 2003; Shugart 1998).

In presidential countries, party-building incentives are rather distinct. First, the president's party is always and necessarily the formateur party, regardless of the size of the president's legislative contingent. Second, government survival does not depend on the support of any party in the legislature. Under separation of powers, therefore, controlling the executive does not require building a strong parliamentary organization.² Rather the contrary: parties tend to concentrate their efforts and resources on the executive election as opposed to the legislative election (Samuels and Shugart 2010).

Parties ‘presidentialize’ by specializing in competition over the presidency. More often than not, parties will recruit candidates that can cultivate a personal vote above and beyond the party's electoral base to compete effectively in the presidential race. Therefore, presidentialization creates incentives for downplaying the party's platform and diluting ideological appeals, undermining party development (Samuels and Shugart 2010).

In semi-presidential democracies, a popularly elected president coexists with a prime minister dependent on the confidence of the legislature. In this situation, incentives for party presidentialization arising from executive elections might be counterbalanced by the legislature's power to make and break governments (Elgie 2011; Samuels and Shugart 2010). Regardless of variation in presidential powers, legislative parties in semi-presidential systems are more powerful vis-à-vis the president than they are in any pure presidential polity. Even in the situation of unified government, when both the prime minister and the president belong to the same party or coalition, interparty or intra-party conflicts are a common cause of prime ministerial turnover (Cheibub and Chernykh 2009; Samuels and Shugart 2010).

Although we do not dispute the view that semi-presidential constitutions create cross-cutting incentives which are significantly different from those associated with presidential systems (Sauger 2009), the level of institutional heterogeneity among democracies with elected executives is substantial, especially with regard to the legislative powers of the head of state (Cheibub et al. 2014; Siaroff 2003; Tavits 2009).

More often than not, strong presidencies have been associated with weak and uncohesive parties (Clark and Wittrock 2005; Fish 2006; Ishiyama and Kennedy 2001). If one assumes that politicians will seek to construct strong legislative organizations mainly in the expectation of gaining control of the legislative agenda, then there is less rationale to do so in the presence of an executive with the ability to shape policy through its unilateral powers (Clark and Wittrock 2005: 176). In a similar vein, Shugart and Carey (1992) argue that strong presidencies foster a division of labour between the executive and the legislature in pure presidentialism. Lacking the means to either control the legislative agenda or select the government, legislative parties would specialize in pork-barrelling and local constituency service.

Overall, if these theoretical accounts were correct, we would expect legislative parties in countries with strong presidencies to organize mostly as opportunistic coalitions of patronage-seeking politicians, lacking in internal cohesion and ideological consistency. Presidential legislative powers would reinforce the detrimental impact of elected presidencies on party unity under both presidential and semi-presidential constitutions.

In this article, we argue in favour of an alternative understanding of the linkages between presidential strength and party unity. Under pure presidentialism, an executive endowed with substantial legislative powers has the means to centralize agenda setting and coordinate with a legislative majority to approve government bills. These powers include the prerogative of issuing decrees, exclusive legislative initiative in selected policy areas, control over the budget and veto powers (Cheibub et al. 2014; Colomer and Negretto 2005; Cox and Morgenstern 2001). By giving a first-mover advantage to the executive in addition to the ability to veto undesired policies, broad legislative powers imply that legislators are much more likely to exert a significant influence over policymaking if they cooperate with the executive. On the other hand, because presidential powers are not absolute (at least not under a working democratic regime) and are most effective when effectively supported by legislative parties, presidents tend to be better off by avoiding a unilateral strategy of government.

Although previous research has argued that incentives for interbranch cooperation should be low when presidents are endowed with substantial legislative authority (Mainwaring and Shugart 1997; Shugart and Carey 1992), empirical evidence on coalition formation under presidentialism does not support these claims.³ Argelina Figueiredo et al. (2012) analysed over 200 cabinets formed by minority presidents in Latin America and demonstrated that high levels of presidential agenda power reduced the likelihood of a minority government being formed. In other words, when minority presidents have the ability to set the agenda of the legislature – for instance, due to the exclusive power of legislative introduction in some key policy areas – they are more likely to succeed in forming a multiparty, majority cabinet.

Under either coalition or single-party cabinets, presidents can use their agenda-setting powers in favour of the legislative majority, not against it (Cheibub 2007: 130). The president and the leader (or leaders) of the governing majority in the legislature will usually coordinate around a common agenda and take advantage of the legislative powers of the executive to centralize policymaking and reduce the opposition's ability to advance its initiatives. As legislative parties form two well-defined blocs – government and opposition – legislative behaviour becomes more predictable, fostering higher levels of party unity (Alemán and Navia 2009; Amorim Neto et al. 2003; Aninat 2006; Figueiredo and Limongi 2000).

This dynamic is similar, to some extent, to that observed in parliamentary countries. The confidence vote procedure allows prime ministers to control the legislative agenda, because legislators anticipate that the head of the government will link votes on specific bills to government survival if they fail to support the executive's agenda (Huber 1996). According to Cheibub (2007: 124), the key causal mechanism that explains party discipline under parliamentarism is agenda control afforded by the confidence vote. Agenda control secures future policy gains, which in turn explains why legislators may opt to vote against their own preferences in the short term. Cheibub (2007) further argues that strong presidential powers may create similar incentives under presidentialism, despite the independent survival of the cabinet and the legislative majority.

When presidents lack the ability to set the legislative agenda, they will have fewer resources to influence legislators' votes. At the same time, legislative parties will typically enjoy substantial authority over policymaking, regardless of their disposition to cooperate with the executive. In such scenarios, presidents will depend to a substantial extent on their partisan powers to obtain stable legislative support. Often, minority presidents will rely on other, more informal powers, such as the ability to distribute pork and patronage among legislators, to approve policy. In the latter case, legislative coalitions will be built mainly on the basis of individualistic rewards, leading to lower stability of voting behaviour (Amorim Neto and Santos 2001). All else being equal, therefore, coordination between the executive and legislative parties in pure presidentialism should be harder to achieve when presidents lack formal powers over the legislative agenda. As a result, levels of party unity should be lower.

In semi-presidential polities, executive authority is divided between an elected president and a prime minister dependent on the confidence of the legislature. Under this dual-authority design, presidents cannot rely on their formal powers to control the legislative agenda because the legislature has the power to bring down the cabinet. By the same reasoning, the prime minister cannot rely on the confidence vote procedure to control the agenda in the same way a prime minister operating under a parliamentary constitution can, due to the existence of an elected president whose survival in office is independent from legislative support. Intra-executive conflict is thereby endemic to semi-presidentialism.

When the president is endowed with substantial legislative powers, she may rely on her constitutional authority to challenge undesired policies pursued by the prime minister. This is the case even in premier-presidential systems, where the power to dismiss the cabinet rests only with the legislature and, therefore, the president cannot replace a prime minister enjoying majority support (Roper 2002). Actually, although the features of premier-presidentialism may favour ‘parliamentarized’ government, a president lacking the power to dismiss a cabinet supported by a hostile majority but yet counting on substantial legislative powers is likely to face rather strong incentives to try to block the prime minister's legislative initiatives. Under a president-parliamentary constitution, because the power to dismiss the cabinet is shared among the president and the legislature, there is an inbuilt tendency to conflict and stalemate. Still, as presidents increase their legislative agenda-setting powers, one would expect the potential for conflict to increase even further. In sum, broad presidential legislative powers likely foster cabinet instability and submit parties to cross-cutting pressures (Sedelius and Ekman 2010; Sedelius and Mashtaler 2013). To the extent that strong presidential powers are associated with divided control of the legislative agenda and tend to increase the potential for intra-executive conflict, they are also likely to undermine party unity in semi-presidential systems.

In contrast, when presidents are weak and lack formal legislative powers to block undesired policies pursued by a competing prime minister, the odds of success of a confrontational strategy are much lower. In these settings, presidents will face stronger incentives to pass their agenda by coordinating with the prime minister and the legislative majority. In sum, low presidential legislative powers lead to greater integration between the cabinet and the legislative majority, producing higher levels of party unity.

On the one hand, the detrimental effect of pure presidentialism on party unity is partially offset by the presence of substantial legislative powers, but strong presidents tend to undermine semi-presidential constitutions' greater incentives to party unity (relatively to presidentialism). On the other hand, when the legislative authority of the president is restricted, semi-presidential constitutions are more favourable to party unity than pure presidentialism.

Due to the contrasting effects of presidential legislative powers under presidential and mixed systems, the performance of each of these regime types regarding party unity should differ less and less as presidents become more powerful in terms of legislative authority. Therefore, for very high levels of presidential legislative powers, levels of party unity should differ only slightly (or not differ at all) between presidentialism and semi-presidentialism.

Based on the theoretical expectations outlined above, we propose three hypotheses:

Hypothesis 1: As the legislative authority of the president increases, party unity will increase under pure presidentialism, but decrease under a semi-presidential constitution.

Hypothesis 2: Weak presidents with restricted legislative authority will be associated with much more cohesive and disciplined legislative party behaviour in semi-presidential than in presidential democracies.

Hypothesis 3: Levels of party unity should not substantially differ between presidentialism and semi-presidentialism for sufficiently high levels of presidential legislative authority.

Cases and data

Previous research has relied mostly on party-level and aggregate measures of legislative behaviour, such as the Rice index, to operationalize party unity (Carey 2007; Hix 2004; Hix et al. 2005; Sieberer 2006). This approach has had important limitations in view of the paucity of cross-national data. In this article, we utilize an indirect measure of party unity obtained from the V-Dem data set (version 8, 2018).⁴ One important advantage of this measure is that it covers a very large number of countries and periods. The V-Dem project relies on expert information on 173 countries to code multiple regime characteristics, including those pertaining to political parties. The legislative party cohesion score (v2pscohesv) is based on the averages of country expert responses to a general question on parties' voting behaviour: ‘Is it normal for members of the legislature to vote with other members of their party on important bills?’ Responses range from 0 to 3. Null scores imply that many members of the legislature are elected as independents and party discipline is very weak. The maximum score is attributed to settings in which legislators vote in unison most of the time.

One potential limitation of reputational measures produced through aggregation of experts' responses is that raters may diverge in their coding due to either differences in opinion or mistakes. The V-Dem project deals with these issues by relying on Bayesian item response theory (IRT) modelling techniques to account for differences in rater reliability and rater thresholds and thus yield reasonable estimates of the latent concepts being measured (Pemstein et al. 2018: 4). The IRT measurement models take ordinal values as input to produce an interval-level, standardized estimate of the given latent trait, considering patterns of cross-rater disagreement, bridge and lateral coding ratings (Coppedge et al. 2019: 19).⁵ We rely on these standardized scores as a proximate measure of party unity.

Another limitation of reputational measures is that error and uncertainty in experts' ratings tend to increase as the period of coverage of a given measure increases. That is, an expert who participated in the 2018 V-Dem wave would be more likely to have first-hand knowledge of legislative behaviour in her country in, say, 2012, than in 1990. Given this potential shortcoming, we ran alternative models relying on a restricted sample that excludes all observations before the year 2005. Reassuringly, we found that results using this subset of the data were very similar to those obtained using the full sample. Thus, we opted to utilize all observations in the analyses shown in the remainder of the article. Alternative models using a smaller time frame are reported in Online Appendix C.

As a simple validity test, we compared the V-Dem standardized party unity scores with average Rice scores calculated by Coman (2015) for the 24 countries present in both our data set and in Emanuel Emil Coman's sample of 33 national chambers. The reputational V-Dem measure correlates strongly with average Rice scores (r = 0.7), which indicates that expert surveys' evaluations of legislators' behaviour are rather close to actually observed voting patterns in the floor.⁶

We coded regime types by relying on the classification proposed by David Samuels and Matthew Shugart (2010). In addition to the classification of countries and periods listed in Samuels and Shugart (2010), we relied on various other sources to code the cases. These include the V-Dem data set, the Database of Political Institutions (DPI), as well as countries' constitutions, obtained from the Comparative Constitutions Project.⁷

Presidential legislative powers were coded according to the classificatory scheme developed by Shugart and Carey (1992) and updated by Lee Metcalf (2000). Legislative powers include partial and package veto, decree powers, reserved policy areas, budgetary powers and proposal of referenda. Each of these dimensions received a score ranging from 0 to 4. The scores on each dimension were summed to arrive at an overall score of presidential legislative powers. We substantially expanded Shugart and Carey's (1992) original data set and revised it according to the coding rules proposed by Metcalf (2000). We included a significant number of additional countries and updated those cases for which we identified relevant changes to presidents' legislative authority. We relied mainly on the Comparative Constitutions Project to obtain information on presidential legislative powers. We also utilized Timothy Frye's (1997) database of political institutions in the post-communist world. We gathered data on 49 countries with elected presidents, in addition to a few parliamentary polities in which the head of state is a non-elected president endowed with more than only ceremonial powers. All other parliamentary countries received a score of 0 in our aggregate measure of legislative powers. Most of our data set covers the post-1970s period, but we gathered data for the older democracies from the 1950s onwards, whenever we could find adequate sources.

Because our hypotheses require comparing the effect of the legislative authority of the president under presidentialism and semi-presidentialism, we opted for setting the latter system of government as the reference category. Therefore, we included dummies for presidentialism and parliamentarism, and interacted each of these variables with our measure of legislative powers. With the inclusion of these two interaction terms, the coefficient estimated for legislative powers indicates the effect of presidents' legislative authority for the reference category (semi-presidentialism).

Electoral systems have been said to exert a very significant influence over legislators' behaviour. On the one hand, where party leaders control candidates' access to the ballot and party lists cannot be altered in any significant way by voters, it is to be expected that legislators will pursue individualistic rather than collective electoral and legislative strategies. On the other hand, where politicians have both incentive and opportunity to cultivate a personal vote, party leaders will have fewer instruments with which to enforce party discipline (Carey 2007; Hix et al. 2005; Sieberer 2006).

In addition to electoral rules, legislators' incentives to behave in a disciplined fashion depends on how centralized the process of candidate selection is. Where candidates are selected by the national leadership, with no input from local party leaders and electoral rules are party-centric, the incentives produced by the electoral and legislative arenas will be consistent with each other. Legislators will have very strong incentives to follow the national party leadership, leading to high levels of party unity. Consistency between the electoral and legislative arenas will also occur when candidates are selected by local party leaders or through primary elections and vote personalization is high, but in this latter scenario party unity tends to be substantially lower due to the weak incentives to follow the national party leadership. Finally, where candidate selection is highly centralized, but the electoral system allows for personalization of the vote, legislators will be forced to respond to the competing demands of voters and party leaders, who are in control of the distribution of resources within the assembly (Carey 2007). As legislators are submitted to competing pressures, party unity likely decreases.

We expect centralized candidate selection to increase party unity only in those instances where the incentives to cultivate a personal vote are low or non-existent. Where electoral rules favour personalization of the vote, centralized candidate selection should produce the opposite effect, undermining party cohesion and discipline.

We account for the incentives to cultivate a personal vote by relying on the ‘Particularism around the World’ database, developed by Joel Johnson and Jessica Wallack (2012). The database classifies 180 countries from 1978 to 2005, according to the 13 positions in Carey and Shugart's (1995) original ranking of electoral systems.

In those cases in which there were different tiers of legislators, elected according to distinct electoral rules, Johnson and Wallack (2012) attributed separate scores to each tier. We used the scores attributed to the largest or dominant tier – that is, the tier that accounts for the majority of the members in the assembly. The resulting personal vote score varies from 1 to 13. We have updated the original database by adding information on countries' electoral systems after 2005.⁸

We measure centralized candidate selection by relying on the V-Dem measure of candidate nomination. Because the V-Dem original score (v2pscnslnl) attributed higher (lower) values to more decentralized (centralized) nomination procedures, we inverted the scale. Also, we reduced the number of categories from six to three, putting together countries that utilize primary elections and selection by local party organs. Finally, we divided the index by two to make it range from 0 to 1.

Given our theoretical expectations regarding the conditional effect of candidate selection, we interact the personal vote index with the measure of centralized candidate nomination. We subtracted 1 from the personal vote score to make it range from 0 to 12 and thus facilitate interpretation of interaction effects.

Variation in legislative party unity across countries and over time may also reflect differences in the levels of maturity and stability of democratic regimes and party systems. In part, differences in party strength are likely related to the timing of democratization. In early democratizing countries that transitioned to democracy before the second half of the 20th century, parties were vehicles of social and political integration of masses of new citizens. In most Third Wave democracies, in contrast, parties have been less central in the struggle to expand citizenship (Mainwaring and Zoco 2007; Randall and Svåsand 2002). Controlling for these timing effects is crucial to estimate correctly the impact of government systems on party unity because most of the old democracies have opted for either parliamentarism or semi-presidentialism, whereas presidentialism is over-represented in the group of late democratizing countries. In view of these aspects, part of the differences in levels of party unity observed across government systems might be related to the fact that the overwhelming majority of presidential democracies emerged in settings less favourable to the construction of strong parties, whereas the opposite is true for most parliamentary and, to a lesser extent, semi-presidential countries. Thus, all models include a dummy for old democracies, defined as those polities that transitioned to a democratic regime before the 1950s and have remained democratic since then.

Party development over time and the associated patterns of behaviour in the legislature are also likely associated with social and economic modernization. In low-income societies characterized by low levels of geographic mobility and few channels of mass communication, and where most citizens are employed in low-skilled occupations, voters are less likely to demand collective goods provided by programmatic parties. Instead, parties will have both opportunity and incentive to develop clientelistic linkages with voters, by supplying localized benefits (Stokes 2007). Therefore, one should expect parties to be weaker and less cohesive in poor as compared to wealthy democracies. Our models control for these factors by including a measure of GDP per capita in thousands of dollars, obtained from the Maddison Project.⁹ GDP per capita was logged to deal with the asymmetry in the distribution.

Politicians' incentives to coordinate around a common party programme are also dependent on the nature and extent of social divisions. In particular, comparative research has shown that political parties in ethnically divided societies compete mainly by providing selective benefits to ethnic groups, especially where these divisions are politicized (Hale 2007; Van de Walle 2003). Thus, in the presence of salient ethnic divisions, parties tend to downplay ideological and programmatic differences, relying instead on ethnic-based electoral appeals. As a result, parties' behaviour in the legislature will be less cohesive. We control for these factors by including in the regressions a measure of ethnic fractionalization taken from Alberto Alesina et al. (2003).

Finally, we include controls for world regions to deal with the geographic clustering of government systems. Latin American democracies, with very few exceptions, are all presidential, whereas semi-presidentialism and parliamentarism predominate in European countries. Moreover, semi-presidential countries with strong presidencies are much more frequent in Eastern Europe and Africa than in Western Europe. Such geographic clustering implies that both the choice of specific regime types and the properties of party systems might be related to unmeasured characteristics of world regions. Thus, we include controls for regional effects to deal with this potential confounding factor.

The final data set covers 72 democracies from the mid-1970s to the mid-2010s, with a total of 1,586 observations.¹⁰ Of these 72 democracies, 49 had a directly elected president (25 presidential and 24 semi-presidential systems). Considering the three countries which experienced a change of the system of government and, thus, were included in more than one category, the data set includes a total of 26 parliamentary countries.¹¹

Results and discussion

Table 1 presents the distribution of presidential legislative powers by system of government for all countries included in our data set. As expected, the median legislative authority of presidents is substantially higher in presidential countries than in either parliamentary or semi-presidential countries. Still, there is substantial variation among semi-presidential democracies, as the extent of legislative powers range from none to a score of 10. The group of semi-presidential cases located in the third quarter of the distribution of legislative powers is similar to the presidential cases with broad legislative authority, whereas the first quarter of the distribution actually resembles the distribution of legislative powers for parliamentary cases.

Table 1. Distribution of Presidential Legislative Powers by System of Government

Sources: V-Dem (2018); Samuels and Shugart (2010); Database of Political Institutions; Comparative Constitutions Project.

Table 2 presents initial evidence on the distribution of our dependent variable across presidential and semi-presidential with low and high levels of presidential legislative powers. We defined the low and high categories with reference to the first and third quarters of the distribution of legislative authority for all cases with elected presidents. Consistent with theoretical expectations (H1), we find that average party unity substantially decreases within semi-presidential countries as we move from low to high presidential legislative authority. On the other hand, mean party unity is slightly higher in presidential democracies with strong presidents than in presidential democracies with weak presidents.

Table 2. Mean Legislative Cohesion by System of Government (Presidential × Semi-Presidential) and Legislative Authority of the Presidency

Sources: V-Dem (2018); Samuels and Shugart (2010); Database of Political Institutions; Comparative Constitutions Project.

To test whether the mean differences observed in Table 2 hold in the presence of adequate statistical controls, we ran a series of statistical models, whose main results are presented below. Issues of cluster heterogeneity and within-cluster dependence of observations are pervasive to time series cross-section data, which makes the utilization of standard OLS regressions unfeasible. Moreover, because some of our independent variables change rather slowly over time (e.g. electoral system properties and presidential legislative powers), or do not change at all (dummies for presidentialism and parliamentarism), a fixed-effects specification is also inappropriate to model the data.

To deal with these data structure issues, we rely on GEE (generalized estimating equations) models (Table 3). Whereas the coefficients estimated by random-effects models represent an average of within- and between-cluster effects that is hard to interpret, the GEE approach models the marginal or population-averaged expectation of the dependent variable as a function of the covariates. In this sense, GEE models are especially advisable when the research is interested in making comparisons across groups or subpopulations (as it is for the case of our empirical strategy). Different from both fixed and random effects models, the GEE approach deals with intra-cluster correlation and panel heterodasticity by estimating a working correlation structure instead of estimating cluster effects (Zorn 2001). We opted for the AR1 correlation structure to deal with high levels of serial correlation.

Table 3. GEE Models for the Determinants of Party Unity

Notes: Main explanatory variables in bold. Dependent variable: standardized V-Dem party cohesion score (v2pscohesv). *** p < 0.01; ** p < 0.05; * p < 0.10.

Model 1a presents the baseline specification, with no interaction terms. Model 1b includes all interactions and Model 1c controls for regional effects.

The effect of the main explanatory variables – presidentialism and presidential legislative powers – is consistent across Models 1b and 1c. The negative coefficient estimated for presidentialism indicates that presidential countries where the legislative powers index is equal to 0 and, thus, presidents are weak in terms of legislative authority (H2), have substantially less disciplined and cohesive parties, compared to the reference category (semi-presidentialism). However, the interaction of presidentialism and legislative powers is positive and highly statistically significant, which is consistent with our claim (H1) that party unity should increase as the legislative authority of the president increases under presidentialism. The negative coefficient estimated for presidential legislative powers indicates the effect of this variable for semi-presidential countries, which is also consistent with theoretical expectations. The main effect for parliamentarism dummy lacks statistical significance in all specifications. This result indicates that semi-presidentialism and parliamentarism do not differ in such concerns as party unity when presidential legislative powers are 0.

As expected, early democratizing countries have higher levels of party unity, on average, than late democratizing ones (positive coefficient for old democracy). However, this effect is only significant in Models 1a and 1b, and disappears when we include controls for world regions. Ethnic fractionalization has a negative effect on party unity, as expected, but all coefficients have large standard errors. Our control variable for economic and social modernization (GDP per capita) came with the wrong sign (negative) and did not achieve statistical significance in any of the models.

To ascertain the effect of centralized candidate selection conditional on the incentives to cultivate a personal vote, we estimate marginal effects and 95% confidence intervals (plotted in solid and dashed lines, respectively, in Figure 1). Figure 1 shows that centralized candidate selection increases party unity when electoral rules are mostly party-centric (personal vote score ranging from 0 to 2). This is consistent with our claim that party unity should be higher when the incentives arising from the legislative and electoral arenas are both conducive to party-oriented behaviour. However, for extreme levels of personalization of the vote, ranging from 8 to 12, centralized candidate selection has a negative effect on party unity. In other words, party unity is undermined when legislators are submitted to competing pressures from voters and party leaders.

Figure 1. Marginal Effect of Centralized Candidate Selection on Party Unity Conditional on Personal Vote (Model 1c, 95% Confidence Interval)

To test our first hypothesis, we relied on the R package Zelig to obtain expected values of party unity for different levels of presidential legislative powers and distinct government systems and calculate first-differences.¹² All simulations utilize the coefficients estimated in the full Model (1c).

To test H1, we calculated the effect of legislative powers while keeping regime type constant (presidentialism or semi-presidentialism). As we move from the first to the third quarter of the distribution of legislative powers for the subgroup of semi-presidential countries (scores of 1 and 4, respectively), party unity decreases from 1.29 to 0.84. We also simulated the difference in expected values for low and extreme legislative powers, defined as the median value plus two times the standard deviation of the distribution (score of 7).¹³ In this latter scenario we find an even larger decrease of 0.89 in party unity. All results are significant at the 99% confidence level.

As for the group of presidential countries, we find a positive and statistically significant effect, as expected. The difference between the expected values for a presidential country with high and low presidential legislative powers (first and third quarter of the distribution) is 0.63. This result indicates that moving from low to high legislative powers under presidentialism increases party unity by over half standard deviation of the distribution of party unity for all observations. Finally, as we move from low to extreme levels of presidential powers (score of 11), party unity increases by 1.26.

Consistent with our first hypothesis, we find that the legislative authority of presidents has a strong negative impact on party unity in semi-presidential countries. We also find evidence in favour of the claim that presidents with substantial legislative powers under presidentialism are associated with higher levels of party unity.

To test H2 and H3, we estimated the differences in expected values of party unity for presidential and semi-presidential countries, while keeping presidential legislative powers constant. We used the same simulation procedures adopted to test H1. Recall that our second hypothesis states that presidents with restricted legislative authority should be associated with higher levels of party unity in semi-presidentialism than in pure presidentialism. H3 claims that we should not observe a relevant difference in levels of party unity between semi-presidentialism and presidentialism when presidential legislative authority is sufficiently high.

We created three categories to simulate expected values and differences across systems of government. Since H2 and H3 require that we make comparisons across regime types, we define the low (first quarter), high (third quarter) and extreme legislative powers (third quarter plus two standard deviations) categories using the distribution of legislative powers for both presidential and semi-presidential countries. Table 4 reports the results of these simulations for Model 1c.

Table 4. Differences in Party Unity between Presidentialism and Semi-Presidentialism according to Levels of Presidential Legislative Authority (Model 1c)

Source: Table 3.

Note: ***p < 0.01.

For low levels of presidential legislative powers, we find that party unity is substantially lower in presidential than in semi-presidential countries, consistent with H2: the difference between expected values for presidentialism and semi-presidentialism is −1.47 (p < 0.01). As presidential legislative authority increases, the advantage of semi-presidentialism shrinks, up to the point it is reverted in favour of presidentialism, although these differences fail to achieve standard levels of statistical significance.

In addition to the models presented in Table 3, we ran an alternative specification using David Doyle and Robert Elgie's (2016) aggregate measure of presidential powers. These additional tests, which are mostly consistent with our theoretical claims, are reported in Online Appendix E.

Final remarks

Our findings are mostly supportive of the hypotheses presented earlier in this article. As the legislative authority of the president increases, the levels of party unity also increase under pure presidentialism, but decrease under a semi-presidential constitution (H1). We also found significant differences between presidential and semi-presidential countries. Weak presidents are associated with substantially higher levels of party cohesion in semi-presidential than in presidential democracies (H2). However, as we move from low to high levels of presidential legislative authority, these differences become increasingly smaller. Moreover, when presidential legislative powers are extreme, the difference is slightly favourable to presidentialism – although this result lacks statistical significance. Overall, model results are consistent with H3: levels of party unity should not differ substantially between presidentialism and semi-presidentialism for sufficiently high levels of presidential legislative authority.

These original findings have important implications for the debates on legislative behaviour and systems of government with popularly elected presidents. First, we show that standard categorical classifications of regime types can still be useful to assess the functioning of the legislature. Second, the results also suggest that more nuanced approaches may be important to explore differences both across and within the regime types.

Indeed, our empirical findings provide support for a more nuanced and complex view about the effect of presidential powers on the workings of democracies with elected executives than that which has prevailed in the comparative literature so far (a point also stressed by Margit Tavits (2009) and Cheibub et al. (2014). Previous theorizing on the matter has assumed that executive strength and party strength are linearly and inversely related and, therefore, presidential systems in which presidents are endowed with substantial legislative authority should almost of necessity develop weaker and less cohesive parties than any other regime type. By the same reasoning, a presidency with limited powers was viewed as the best alternative in terms of constitutional design, and especially so under pure presidentialism (Mainwaring and Shugart 1997; Shugart and Carey 1992). What we find, however, is that presidential powers vary a lot among popularly elected presidents, and that weak presidents are associated with higher levels of party unity under semi-presidentialism, but not under presidentialism.

The findings have important implications in terms of constitutional design. Presidential countries with weak presidents, such as Costa Rica or Nicaragua, can increase party unity in the legislature by opting for a semi-presidential constitution, while maintaining presidential legislative authority unaltered. As a second-best option, they can maintain the presidential system of government but increase the legislative authority of the president. For semi-presidential countries with strong presidents (e.g. Ukraine, Peru, Senegal), in contrast, the best way to increase party unity is by substantially reducing presidents' legislative powers. A presidential system is not an advisable option for these cases, since the simulations reported in Table 4 indicate that there are no statistically significant differences between presidential and semi-presidential systems with high or extreme levels of legislative authority. Constitutional engineering efforts need to take into consideration the fact that changes in regime type (e.g. from presidential to semi-presidential) and in the legislative authority of presidents should always be evaluated together, to avoid the risk of fostering inter-branch and/or intra-executive conflict, in addition to undermining party unity.

The robustness and relevance of our results notwithstanding, we are aware that our research design suffers from some limitations. First, due to the lack of available comparative data, we do not account for the internal organization of legislatures in our analyses. This is a relevant dimension that deserves to be explored in future studies, to the extent that levels of party unity are likely higher where party leaders exert centralized control over intra-parliamentary resources (Sieberer 2006). Second, because we rely on a reputational, country-level measure of party unity, we cannot account for variation due to party-level factors, such as party size (Owens 2003). Unfortunately, there is a trade-off between the spatial and longitudinal coverage of the data, on the one hand, and the number of explanatory dimensions that can be assessed, on the other. Thus, despite the above-mentioned limitations, our research makes an important contribution by developing one of the most comprehensive analyses of the institutional determinants of party unity to date in comparative politics.

The results of this article are related to previous work on the political consequences of institutional variation across regimes with directly elected executives (Tavits 2009). In particular, presidential-parliamentary countries where presidents have broader powers regarding the appointment and dismissal of the cabinet have been found to perform poorer than pure presidentialism regarding government performance and democratic stability (Sedelius and Linde 2018). Our research indicates that presidential legislative authority is also a relevant dimension on its own, as it exerts a strong and significant impact on levels of party unity under semi-presidentialism.