‘Congruent We Govern’: Cost of Governance in Multiparty Presidentialism

Abstract

The received wisdom on executive–legislative relations in multiparty presidential systems is that the size of the president's majority in Congress is the key factor explaining governance patterns, particularly the president's legislative success. However, in many cases presidents enjoying a nominal majority have been unable to pass legislation and have faced institutional instability. The article departs from a conventional definition of divided government and focuses on the preference incongruence between the governing coalition and the floor of Congress. It argues that the ideological distance between the floor and the coalition is a key factor explaining the president's cost of governing (which includes factors such as the distribution of cabinet portfolios and budgetary transfers to coalition partners). The article provides an empirical test with data from Brazil that find that the greater the ideological incongruence, the higher the cost of governing for the president.

Much of our theorizing about presidential success and survival in office, explicitly or implicitly, draws on the two-party US case. In this case, the president's party either controls both the executive and legislative branches – majority government – or it does not – minority government. However, multiparty presidential regimes are also the modal form of democratic governance in Latin America, Africa, Asia and post-communist Europe (Chaisty et al. 2014, 2018). Recent contributions have explored the dynamics of coalitional presidentialism by showing how presidential agenda-setting powers (Alemán and Tsebelis 2011, 2016), coalition management strategies (Raile et al. 2011), coalition durability and defection (Martinez-Gallardo 2011), legislative support (Amorim Neto 2006) and executive/legislative preference alignment (Cheibub and Limongi 2011) matter for successful governance.

In this article we show that, when we control for these factors, presidential success hinges on the ideological congruence between the president's coalition (parties holding cabinet portfolios) and that of the chamber's floor. This finding helps explain why some constitutionally powerful presidents enjoying great support from proportional coalitions have failed to pass their agendas in Congress (chamber of deputies) and vice versa: the paradox of minority presidents' success in approving their political agenda.

Parliamentary regimes require legislative majorities to form governments and are consequently more likely to have congruence between the two branches of government. By contrast, the delegation chain in presidential systems is broken into two separate bodies, elected by different constituencies, which enhances the likelihood of incongruence between president and assemblies. Hence congruence plays a crucial role in governance outcomes under presidentialism.

Our point of departure is the scholarship on coalition management under presidentialism (Cheibub et al. 2004), which shows that government coalitions are less frequent, although not exceptional, under presidentialism. When the president's party does not control a majority of seats, coalitions occur more than half of the time and they are more likely to occur when the legislature is more fractionalized.

Minority presidents have therefore to build and sustain post-electoral coalitions. Drawing on data from Latin America, André Borges et al. (2020) show that pre-electoral coalitions and existing partisan agreement are crucial variables in this process. The received wisdom on multiparty presidential administrations (Figueiredo et al. 2009), however, is that proactive constitutional powers held by presidents and their agenda-setting powers are the key determinants or sufficient conditions for a successful government. Eduardo Alemán and George Tsebelis (2016) claim that there are three dimensions of agenda-setting: the partisan (a stable and cohesive majority), the institutional (legislative rules regulating who makes the initial proposal and who can amend it under what conditions and by how much) and the positional (ideological positions of all the actors involved) for the success of elected presidents.

Other contributions (Bertholini and Pereira 2017; Pereira et al. 2016) have argued that in addition to the constitutional and agenda-setting powers of the president, other factors contribute to presidential success, including the size of the governing coalition measured by the number of partners, its ideological heterogeneity, and the proportionality of resources shared with coalition allies. That is, in a multiparty presidential setting, the president works as a kind of CEO managing the preferences and interests of several stakeholders, who do not necessarily share the same political agenda.

In ‘Divided We Govern’, a seminal contribution to the debate, David Mayhew finds that in the US the legislative success of the government is not related to the divided/unified status of the government (Mayhew 1991), an argument that generated extensive theoretical and empirical debates (Binder 2003; Conley 2002; Fiorina 1996; Krehbiel 1992). Marcelo Camerlo and Cecilia Martínez-Gallardo (2018), analysing portfolio allocation in Latin America, argue that the internal alignment of the government is maximized through the appointment of agents whose individual preferences are close to the president.

We focus on a related but distinct issue and argue that the size of the presidential party and/or coalition and its ideological incongruence with the floor of the Congress are not necessarily correlated, a fact not fully acknowledged in the extant literature. If an ideological mismatch occurs, the president will have a higher cost of governing, despite the powers that the executive possesses (constitutional, budgetary and agenda-setting), the size of the majority of the governing coalition and its cohesiveness or the level of partisan fragmentation.

This article aims to contribute to this conceptual debate by proposing an alternative approach to thinking about the governing of multiparty presidential systems. We suggest an objective criterion capable of measuring preference incongruence between and within governments and claim that the degree of congruence between the preference of the presidential coalition and the preference of the floor of the Congress is the crucial factor. This article tests the following hypothesis:

Hypothesis 1: All else constant, the higher the preference incongruence between the president's coalition and the floor is, the more expensive the coalition management will be.

In the next section we engage the literature on congruence and coalition management under multiparty presidentialism and build an alternative to the minority–majority dichotomy that is applicable to multiparty presidential regimes. Our empirical focus is the ideological incongruence in 18 Latin American countries in the last three decades. We show that there is a great deal of variation in the occurrence of incongruent government across countries but also within each country.

The evidence we present here suggests that the majority status of presidential coalitions does not usually translate into congruent governments, and vice versa. By incongruent government we mean the distance between the coalition's and floor's preferences, and not the level of cohesiveness among coalition partisan allies as suggested by Taeko Hiroi and Lucio Rennó (2016: 75), who refer to the ‘the ability of the members of the coalition to act in unity in legislative matters’. That is, because a majoritarian coalition may not necessarily be a representative subset of Congress, there might be a mismatch between the preference of the floor and the governing coalition. In other words, the president could be able to build a majority coalition, but if the median preference of this coalition is away from the median preference of the floor of the Congress, that majority could not be translated in the capacity of governing. Therefore, a focus on divided governments in multiparty settings is incomplete at best.

This article argues that the congruence between the median ideological preference of the presidential coalition and the median ideological preference of the floor of the Congress is the fundamental aspect for understanding the functioning of presidential systems in fragmented institutional settings.

We have selected Brazil for a case study to test our hypothesis, as it has been the paramount example of hyper-fragmented presidentialism in the world (Kingstone and Power 2017). By focusing on a single country, we can control for institutional features that have remained constant since the enactment of the Constitution of 1988, such as agenda-setting and other constitutional powers, which was the focus of previous research (Alemán and Tsebelis 2016). In addition, the wealth of data and the observed variation of how presidents have managed their governing coalitions make Brazil the best fit to test our hypothesis. We expect that our theoretical framework and empirical findings on an extreme case (Gerring 2008) travel well to other settings.

Coalition incongruence in multiparty presidential regimes

Presidents are the central players in a multiparty presidential system. They are responsible for choosing the parties with whom they will govern and to whom they allocate power. Depending on the particular institutional context, the president holds a large ‘toolbox’ (Raile et al. 2011) for managing coalitions. Understanding the game between partners in multiparty coalitions is important because it generates gains from exchange and cooperation over time between political actors (Pereira and Melo 2012).

Such choices engender different trade-offs and costs for the executive (Pereira et al. 2016). That is, each of these presidential choices engenders consequences with regard to the capacity of presidents to govern, their legislative successes and the cost of governing.

It is important to note, however, that these presidential choices are not free from restrictions. Legislative branch elections actually set the initial parameters for these choices, since they define the size (number of seats) of the president's party, as well as the size of potential allies and parties that will oppose the government. However, even in the face of such restrictions, it is the president, the formateur, who ultimately defines how many, which, how much power and resources will be shared with allied parties, and how congruent this assembled coalition will be with regard to Congress's preferences.

The president has to build a governing coalition as a result of its minority status (Martinez-Gallardo 2005). Yet, as Alemán and Tsebelis (2016) argue, presidents have been able to navigate the lack of majority without facing the perils of deadlock. They claim that this is partly the result of the interaction of the agenda-setting prerogatives in the hands of the president: ‘in the case of cohesive partisan majorities, the institutional dimension will not be very influential. But in their absence (or if such majorities are weak, undisciplined, or not cohesive), one must focus on the institutions regulating the prerogatives of each actor. These rules, along with the relative positions of the actors, will shape policy-making outcomes’ (Alemán and Tsebelis 2016: 2).

Nevertheless, it is not a rare event that the preferences of the president dramatically differ from Congress's preferences. This potential incongruence arises because candidates for the presidency can gather votes in the entire country with a nationally oriented policy agenda. Legislators, on the other hand, are elected in smaller constituencies with a very narrow and local policy agenda. Based on these separate origins and survival strategies, the chief of the executive and legislators may not share the same preferences.

The composition of Congress and its preferences, therefore, can be very distinct from the president's preferences and his/her coalition. A liberal president, for instance, may be concomitantly elected with a very conservative Congress, and vice versa. Thus, in addition to dealing with the minority status, the elected president has to face a Congress that does not necessarily share similar preferences with her/him.

As the formateur of the coalition, therefore, the elected president frequently faces a difficult dilemma. He/she may either decide to build a coalition ideologically consistent with the preferences based on the campaign platform or may prefer to build a coalition congruent with Congress' preferences.

This decision has very important implications both for the policymaking and for the cost of governing. If the president decides to ignore voters' preferences and build a majority coalition congruent with the median preference of the Congress, she/he runs the risk of quickly becoming an unpopular president; however, by aligning his/her coalition with what Congress wants, the president may achieve legislative success at relatively low costs. On the other hand, should the president's coalition be consistent with voters' preferences and incongruent with legislators', she/he could be very popular but, most likely, at the expense of many governability problems at relatively higher costs. This is the key trade-off elected presidents face in multiparty settings.

The unexplored frequency of minority presidents in multiparty presidential regimes and the consequent outcome of post-electoral coalition governments has led to a research agenda exploring the puzzle of minority coalitions as an equilibrium (Chaisty et al. 2014, 2018; Figueiredo et al. 2009). The lack of a partisan majority backing the president in one or both legislative houses does not necessarily mean that a single opposition party must automatically control the legislature in a multiparty presidential regime. Instead, the most frequent outcome is a post-electoral majority coalition supporting the president. Presidents in multiparty presidential institutional contexts tend to be constitutionally strong and capable of attracting partisan support. What is more, many parties are motivated to work as satellites of the president as an opportunity to extract political and financial benefits under the discretion of the executive in exchange for political support in the legislature.

The extant empirical literature on divided governments has focused on a range of issues, from the executive's strategies in contexts of minority status (Amorim Neto 2006; Cheibub and Limongi 2011; Cox and McCubbins 2007) to the reasons why majority coalition governments fail to emerge (Chaisty et al. 2014; De Figueiredo et al. 2013; Elgie 2001).

José Antonio Cheibub et al. (2004) and Paul Chaisty et al. (2014) have investigated to what extent ideological congruence affects the incentives for the existence of both single-party minority governments and their majority variant. Our approach considers a related question: namely, what the consequences of ideological incongruence between the president's coalition and Congress are, and – more importantly – how ideological incongruence can be used analytically as a criterion for identifying governments and their capacity to govern.

The focus on preference incongruence proposed in this article sheds light on several puzzles and even unexplored issues in the literature regarding presidential success, survival, legislative oversight and cost of governing, as suggested in the empirical section of this article. Keeping this interpretation of congruent/incongruent government in mind, the next section examines the distribution of this event in Latin America and differentiates the occurrence of majority and minority status of the president's party or coalition from incongruence in comparative perspective.

Congruent government: an empirical comparative assessment

By incongruence we mean the ideological distance of the governing coalition from the floor of the Congress. We expect that the greater that distance, the higher the cost of governing. On the other hand, the closer the coalition is from Congress's preferences, the smaller the cost of governing. In this section we present data on coalition–floor congruence in Latin American governments over the last three decades or so. This comparative exercise is very useful because it allows us to observe that the nominal majority status of the government in Congress does not necessarily translate into higher government congruence.

Our definition of incongruence, therefore, implies a continuous measure, which we call ‘Ideological Distance Coalition–Congress’.¹ It represents the absolute value of the difference between the average Congress ideology and the average coalition ideology. This measure goes from 0 – perfect congruence – to 9 – highest incongruence. This measure is not affected by the direction of ideological incongruence – that is, it does not consider whether a governing coalition leans towards the left or the right of Congress's average position.

For visualization and simplicity, we also use a variable called ‘Ideological Difference Coalition–Congress’, which can vary from −9, when the coalition is on the extreme left of the political spectrum and the floor is on the extreme right of this theoretical continuum, to 9, when the very opposite takes place. That is, the same as the ideological distance, but without using absolute values.

We use data from the Database of Political Institutions (DPI) 2015 data set (Cruz et al. 2016) and ideology scores from several different sources, such as Andy Baker and Kenneth Greene (2011), Sebastián Saiegh (2015) and Cesar Zucco and Timothy Power (2019). When more than one ideology scale was available for a given party of a given country in a given year, we used the average between the indexes. Ideology scores were re-scaled to a 1–10 scale, where 1 is extreme left and 10 is extreme right.

The left–right ideological positions of political parties have been used to analyse a great variety of topics, including macroeconomic, policymaking, legislative institutional choice, electoral competition, voting behaviour, political representation, cabinet stability and so forth. However, data that are comparable over time and across countries are not available. Some contributors – see Matthew Gabel and John Huber (2000) – argue that it is possible to overcome these limitations by extracting information from party election manifestos. The Manifestos Research Group (MRG) has carefully coded the content of party election manifestos for a large number of parliamentary democracies. However, the availability of data from presidential regimes, especially in Latin America, is very restricted and it is not party specific. Notwithstanding, in recent years there have been several empirical efforts to place Latin American political parties on an ideological continuum. Most of these data are based on experts' opinions, collected through surveys. The DPI 2015 data set, for instance (Cruz et al. 2016), uses a 1–5 scale to score political parties between extreme left and extreme right. The advantage of DPI over any other data source is its vast availability of observations (country–party–year). Baker and Greene (2011) provide data on Latin American parties from 1993 to 2016 on a five-point scale, similarly to the DPI. Finally, we rely on Saiegh (2015), who uses joint scaling methods from three large-scale surveys to place parties from different Latin American countries on a common ideological space.

In the particular case of Brazil, we rely on data produced by eight rounds of the Brazilian Legislative Survey by Zucco and Power (2019). Their ideology scale positions parties across a left–right continuum using a one-shot Bayesian estimation procedure over surveys responded by legislators. They assume each party in each survey has a ‘true’ unobservable ideological position. Legislators, then, ‘observe these true positions with error, and also shift and stretch the scale according to their idiosyncratic interpretations of the left–right continuum’ (Power and Zucco 2012: 9). The authors assume that legislators apply the same shift and stretch factors to their own positions when reporting them over time, and ‘this means that legislators who answer more than one survey have one true position that is stable over time, and that differences in self-reporting are due to changes in the meaning of the scale from year to year’ (Power and Zucco 2012: 9).² This scale will be our benchmark. In order to take advantage of all the information in the data we first re-scale the scales to fit our benchmark – that is, to a 1–10 scale. Then we combine data for country–party–year and use the average value for each observation.³

Figure 1 shows the distribution of ideological differences between Congress and presidential coalitions for several Latin American countries, from 1990 to 2015. We considered the first coalition built by the elected government in each country only. The x-axis represents a point estimation of the ideological difference between coalition and Congress, and the y-axis a combination of country and year (cccyy). We used the variable allhouses from the DPI to identify if a specific country in a given year was formally considered a majority or a minority government. The first column of Figure 1 contains only countries that had majority governments in a given year. The second and third columns contain country-years that were minority governments from 1990 to 2001 and from 2002 to 2015, respectively. We considered countries with four or more parties with representation in Congress only.

Figure 1. Majority vs Minority Government in Latin America by Ideological Differences

Countries close to the zero line are considered to have a coalition more congruent to Congress, and those countries far from it are considered more incongruent. Strictly speaking, a congruent government should occur when the coalition ideology exactly mirrors the Congress ideology only. However, in multiparty settings, some of them highly fragmented, it would be extremely difficult for this congruent condition to take place.

Figure 1 shows great variation of countries with both majority and minority status, either under congruent and/or incongruent governments. The Brazilian case, which we will analyse in detail in the next section, is represented with the black dots.

This descriptive assessment defines as congruent some governments that the classic literature would identify as minority governments. This is precisely the case with countries that are close to the zero line in the centre and right columns of Figure 1: minority governments in Argentina in 2001, Peru in 2006, Panama in 1999 and Colombia in 1994, for instance.

Figure 1 also shows that several countries in which the government enjoys the majority status tend to have more congruent governments. Although less common, it is possible to identify cases that combine majority governments with greater incongruence, such as Honduras in 2001, Brazil in 2006 and Venezuela in 2010. In other words, the simple numeric majority status, measured by the number of seats held by the party of the president or his/her coalition, does not necessarily mean that the government is congruent with the floor. The opposite can also be observed. There are cases in which the government coalition does not enjoy majority status but, nevertheless, is ideologically congruent with Congress.

It is important to bear in mind that a congruent government does not necessarily generate a welfare-enhancing government. This measure also does not suggest that parties are stronger or that legislators should be more reliable. Rather, our measure of coalition incongruence with Congress's preferences mostly suggests that the executive will face less pressure to expend greater resources with coalition allies.

The Brazilian case

The Brazilian case is a good environment to test our hypothesis, given that we control for institutional variations and the availability of ample data with considerable granularity and significant variation in our main independent variable. Data concerning ideology in particular are far more reliable than those available for other Latin American countries. The ideology data for Brazilian parties come from Zucco and Power (2019), who organized an extensive data set with eight waves spanning 28 years (1990–2019). We interpolate data for missing years, when there was no wave.

Figure 2 shows that Brazil has experienced a great deal of variation with regard to ideological incongruence between governing coalition and Congress after its re-democratization. The two extreme values in the data coincide with the impeachment of former president Fernando Collor de Melo in 1992 and a major corruption scandal during Lula's first term in office in 2005.

Figure 2. Ideological Differences Coalition-Congress in Brazil

President Sarney's coalition (1986–9) started out ideologically aligned with Congress. As he was the first president of Brazil after the military regime, elected indirectly by the Congress, this result might be expected, given that his government was a consequence of a grand coalition between the two largest parties in Congress, the Party of the Brazilian Democratic Movement (PMDB) and the Party of the Liberal Front (PFL). After the first direct presidential election in 1989, an ideological misalignment between the presidential coalition and Congress took place: Collor de Melo built an incongruent coalition to the right of the Congress's ideology, equivalent to one point on our scale. It was indeed a right-wing coalition away from the average ideology of the floor of Congress. Thus, it might not be a coincidence that Collor de Melo was the first president to be impeached in Brazil. The other extreme value in the Brazilian data is found in 2005, which coincided with one of the most important political crises of the decade: the mensalão corruption scandal during Lula's first administration (2003–9) (Raile et al. 2011).

After Collor's impeachment his vice president, Itamar Franco, took office. He managed to build a very large coalition that was pretty much congruent with the ideological preference of Congress for most of his term in office. As a consequence of the great success of the Real plan, which finally controlled hyperinflation and provided macroeconomic stability, Franco's former minister of finance, Fernando Henrique Cardoso, was elected president in 1994. He built a majority coalition that matched the ideological average of Congress for his two terms in office. His successor, Luiz Ignacio Lula da Silva, built a large, ideologically heterogeneous and disproportional coalition, whose ideological profile, however, did not match Congress's ideology. In fact, the Workers' Party (PT) governments embodied a left-wing shift away from the conservative average of legislators within Congress.

Figure 2 also indicates that this ideological mismatch under PT administrations only diminished in Dilma Rousseff's second term. This seems to be a consequence of her unsuccessful survival strategy during the process leading to her impeachment. The data suggest that Rousseff indeed tried to build a coalition more congruent with Congress's preferences. But it seems that it was too late. Her administration paid the price for her hostility towards coalition partners during previous Workers' Party administrations. Bear in mind that Lula's coalitions monopolized political and financial resources to the PT and under-rewarded coalition allies (Pereira et al. 2011). Therefore, presidents may sustain a coalition at a high cost in the short run, but impeachment might emerge latter.

Under Michel Temer's administration, the new president decided to aim for the median preference of the Congress, building and managing a governing coalition that mirrored Congress's ideological preferences.

Estimating the effect of ideological incongruence on the cost of governing in Brazil

The discussion about strategic presidential decisions converges on the idea that presidents in multiparty settings have under their control and discretion a pool of political and monetary resources – a toolbox – that executives use to build and sustain post-electoral majorities (Raile et al. 2011).

We empirically estimate the effect of the level of incongruence between the government coalition and the floor of Congress in Brazil, having as the dependent variable the costs of managing a coalition, which is measured by the Governability Cost Index (GCI). The expectation is that ideological incongruence between the coalition and the floor makes the executive–legislative relation more difficult to coordinate and, as a consequence, costlier for presidents to govern. The variables and data on coalition management in Brazil come from Frederico Bertholini and Carlos Pereira (2017), and the ideology scores were taken from Zucco and Power (2019). The data are organized monthly and span from October 1988 to December 2016.

To estimate the political and monetary resources the executive has to spend on managing the coalition partners, we built the Governability Cost Index. We want to be consistent with the literature, which assumes that political and monetary resources – seen as tradable goods – are imperfect substitutes (Raile et al. 2011). Policy can be another source of costs; when a coalition partner gets too expensive to buy, the president has to concede.

The GCI is composed of three complementary measures of presidential costs associated with coalition management: (1) the number of cabinet positions (and state secretaries with the status of a minister) that a president decides to include in the administration on a monthly basis;⁴ (2) the natural log of the total amount of budgetary resources allocated to each cabinet position on a monthly basis; and (3) the natural log of the total amount of legislators' individual amendments to the annual budget (i.e. pork) appropriated by the government on a monthly basis. The pork variable attempts to capture legislators' intent to allocate resources to their main electoral constituency. Although legislators have the prerogative to amend the budget bill, the executive has the power to pursue the budget under his/her discretion. To account for economic cycle effects, we adjust the GCI using a GDP correction. Ample evidence in the literature (Raile et al. 2011 and Alston and Mueller 2005) demonstrates that the president makes strategic use of the appropriation of legislators' amendments in the budget as a governing tool to achieve legislative success.

The GCI is a formative indicator, meaning that causality flows from the measurable variables to the latent variable, and the latter is the result of a linear composition from the first (Diamantopoulos and Siguaw 2006). Formative indexes are a plausible approach for combining a number of indicators to form a construct without any previous assumptions as to the patterns of inter-correlations among these items (Bagozzi 2007; Edwards 2011). The construction of formative indicators implies a decision on the weights to be assigned to each variable in its composition. One possible solution is to assign weights based on consolidated theory or expert opinion. Another option is to let the variance matrix set these weights. In the absence of consolidated theory, we chose to assign weights for each variable in the index based on its contribution to explaining the total variance, which is the factor loading of this variable in the composition of factors.

The technical procedure used to calculate the GCI was a principal component analysis (PCA). The PCA analytical solution is produced through decomposition of the correlation matrix by identifying the first principal component or the linear combination of the variables that explains the larger part of the total variance (Coltman et al. 2008). According to this method, better solutions involve a greater proportion of the total variance being contained in the first component (Fisk 1977).

In our case, the principal component analysis produced a composite indicator that explained 70.8% of the total data variance while generating only one factor with an eigenvalue higher than 1. Weights automatically assigned to the variables that make up the index were: 0.619 for the number of cabinet posts, 0.608 for allocations to the ministries and 0.497 for pork spending. The decision on the variables that compose the GCI meets the objective criteria of data availability.

We were able to gather data for each of the three components of the GCI for the period from January 1995 to December 2016. A month becomes the unit of analysis in our data. This period corresponds to four different administrations: Fernando Henrique Cardoso (PSDB), 96 months; Luis Inacio Lula da Silva (PT), 96 months; Dilma Rousseff (PT), 36 months; and Michel Temer (MDB), 8 months. Figure 3 shows that the costs Brazilian presidents have incurred in the management of their coalitions have varied widely. This variance applies both to the total cost and to the extent to which the costs are distributed to the president's party versus other coalition parties.⁵

Figure 3. Governing Costs Index in Brazil – 1995 to 2016

As shown in Figure 3, the management costs of the presidential coalition varied considerably in Brazil during the period analysed. It varied not only with regard to the total cost, but also in regard to the composition of this cost. For instance, the first term of President Cardoso was characterized by a relatively low total cost (average of 36 points). Cardoso's costs with his own party, the PSDB, were lower than the combined costs of other parties in his coalition (an average of 14.3 and 21.7 points, respectively). In his second term in office, however, there was a notable increase in the total governing costs (average of 59.5 points) and a reversal in the composition of costs, with the allocation of a greater portion of resources to the PSDB (average 32.4 points) vis-à-vis the cost of the other parties of his coalition (average of 27.1 points).

When the PT took office in 2003, not only did the total cost of government increase (average 90.6 points), but the president's party also became by far the main recipient of expenditure in relation to the other parties of the president's coalition (average of 62.6 and 28.1 points, respectively). A similar pattern was reproduced in President Lula's second term in office. However, there was a further increase in total costs (average of 95.2 points). This growth was mainly due to an increase in costs of the other parties of his coalition in relation to the PT (average of 38.1 and 57.1 points, respectively). The first term of office of President Rousseff displayed a small decrease in total governing costs (average of 88.1 points). However, the costs of the president's party were still slightly greater than the costs of the sum of the expenditures related to other coalition parties (average of 44.9 and 43.1, respectively).

In Rousseff's second term in office, the total governing cost decreased considerably (an average of 58 points), reaching the level of Cardoso's second term in office. Again, the PT was over-rewarded relative to the other parties in the governing coalition (average of 31.9 and 26.1 points, respectively).

Finally, there was a sharp decline in total coalition management costs under Temer's administration (average of 15.4 points). At first sight, this does not make sense. It goes against any expectations that legislators might inflate the price of their support given the political vulnerability of the new president. Part of the answer to this puzzle has to do with the composition of the management costs. Temer preferred to allocate 70% of the governing resources to coalition allies and less than 30% to his own party. Temer combined a very high level (average of 80.4 points) of coalition coalescence (Amorim Neto 2006) – a measure of the proportionality of the distribution of resources given the political weight of coalition partners in Congress – with an extremely low level of ideological heterogeneity (average of 27.1 points) between coalition allies. This strategy of over-rewarding partners, instead of his own party, pleased ex ante partisan allies helping the president to be successful in Congress at relatively low cost, despite the great political vulnerability resulting from his abysmal popularity.

Even controlling for partisan fragmentation, presidential vulnerability and agenda size, the effect of presidential strategies of how to manage their governing coalition were very robust. That is, the larger the number of parties in the governing coalition, the higher the degree of ideological heterogeneity, and the smaller the power-sharing among coalition partners, the more expensive it is to run a multiparty government in a presidential system (Bertholini and Pereira 2017).

It is important to bear in mind, however, that partisan fragmentation does not necessarily imply higher ideological heterogeneity. It is not uncommon in highly fragmented party systems for some parties to have similar ideological preferences. A very good example was Temer's coalition, which was very fragmented but at the same time very ideologically homogeneous.

The model of cost of governing

In our preferred specification, we included as independent variables the measure of incongruent government, agenda size, average Congress ideology, effective number of parties, presidential popularity and nominal support.⁶

We expect that the greater the ideological distance between governing coalition and Congress, the higher the GCI score. We also expect that the greater the size of the legislative agenda of the president and the partisan fragmentation of Congress, the higher the governing costs. Concerning presidential popularity, we expect that it decreases the cost of governing. A popular president tends to transfer the costs of opposing executive legislative initiatives to members of Congress. With regard to the average of Congress ideology, we expect that left-wing coalitions would face high costs of governing, given that the median preference of Congress leans towards the right.⁷ In other words, a right-wing government has seldom invited left-wing parties to take part in the governing coalition in the last 30 years. But, it has been quite common for left-wing governments to invite right-wing parties into their coalitions. Furthermore, the average Congress ideology has consistently been on the centre-right in Brazil. However, it should be noted that the correlation between Congress ideology and ideological distance is small and not significant (−0.11, p > 0.99), as shown in Figure 6 in the Online Appendix.

Finally, concerning the size of the numerical support backing the president in Congress, we do not have a clear-cut expectation, given the controversial findings in the literature. William Riker (1962), for instance, established the ‘size principle’, according to which parties that form coalitions (formateur) will build coalitions that are large enough, or minimal-winning coalitions. Daniel Diermeier et al. (2003) considered the payoffs that would be shared between coalition partners. For him, interest in joining a coalition is directly associated with the ministries' budget, with possible opportunities for policymaking, and with expected electoral gains. Lawrence Dodd (1976) claims that, as the asymmetry of information among players increases, the more likely they are to increase the size of their coalitions. Cli Ord Carrubba and Craig Volden (2000) predict that, in order to create a more stable exchange environment that is less subject to high defection costs, parties create a minimal necessary coalition greater than the minimal-winning coalition. Tim Groseclose and James Snyder (1996) argued that supermajority coalitions may be cheaper than minimal-winning coalitions. Supermajorities occur more frequently because they prevent small parties from acting as the ‘casting vote’.

To address the possible correlation of errors across observations in our sample we employ the Newey–West Standard Error Model procedure (Newey and West 1987). This is a heteroskedasticity and autocorrelation consistent method of estimation that allows for improving the robustness of the model when it comes to hypothesis testing and the precision of the estimates when there is heteroskedasticity and serial correlation over time in the underlying data.

We use three different model specifications. Our first specification takes into consideration only nominal support and ideological distance between coalition and Congress. The idea is to compare a nominal support approach of majority government with our approach of coalition congruence. In our second specification, we include agenda size and Congress ideology, two control variables related to specific legislatures. The purpose of this model is to understand if our main hypothesis still holds for different Congress configurations, in terms of the size of the agenda proposed by the executive to legislators and the Congress's mean ideology. In our third specification we include control variables related to context: namely, presidential popularity and partisan fragmentation (using the effective number of political parties (ENPP)). All three specifications give support to our main theoretical expectation; that is, the higher the incongruence between the governing coalition and Congress, the higher the cost of governing will be.

Results in Table 1 indicate support for our main theoretical expectation, showing a statistically significant coefficient for the variable ideological distance between coalition and Congress. It suggests that the ideological distance between the presidential coalition and Congress plays an important role explaining the level of difficulty the president will face in coordinating his or her coalition. If a president is able to decrease incongruence by one point he would be able to ‘save’ approximately three points off his/her cost of governing. In other words, building and sustaining a governing coalition that mirrors the floor of Congress makes the life of the president easier and cheaper.

Table 1. Regression Models with Newey–West Correction for Autocorrelation – Governing Costs

Note: *p < 0.05; **p < 0.01. ENPP = effective number of political parties.

A very interesting result that challenges the importance of the distinction between divided/unified government in multiparty presidentialism is the lack of statistical significance of the variable ‘nominal support’ and the positive and significant influence of the ideological distance between the coalition and the floor. That is, regardless of the actual size of the governing coalition, ideological congruence seems to be the key element for understanding presidential support in Congress and the president's corresponding governing costs. The nominal size of the coalition seems to matter when we added other explanatory variables; however, its coefficient is significantly smaller than the coefficient of ideological distance.

The average of Congress ideology also behaved according to our expectations, suggesting that centre-right coalitions will face smaller costs of governing. Presidential popularity, on the other hand, presented a positive and statistically significant coefficient contradicting our initial expectation. The coefficient of the variable ‘nominal support of the president’ in Congress (the arithmetical definition of divided government) was significant and positive, suggesting that the greater the number of legislators backing the president in the legislature, the higher the cost of governing. This result crucially contradicts the extant consensus in the literature that suggests that the majority status of the president in Congress would facilitate governability and decrease the cost of governing. What really matters is the ideological congruence of the coalition with Congress, and not the nominal size of support for the president. This result suggests that having a large but incongruent coalition with the Congress's median preferences increases the cost of governing. Although these model specifications generated results supporting our main hypothesis, we are aware that a more robust model is necessary in order to deal with the time series characteristics of the GCI.⁸

Conclusions

This article has made two contributions. The first is that it shows how the analysis of governance in multiparty presidential systems – the modal regime type in new democracies – can benefit from a consideration of issues related to preference congruence between the president's coalition and the legislative floor. This helps shed light on puzzling empirical findings such as the cases of presidents with large multiparty coalitions but with poor ability to pass legislation at higher costs and, conversely, presidents counting on small coalitions that achieve strong legislative approval rates at lower costs.

The second contribution of the article is empirical. Despite the limitations of a case study, it illustrates how the congruence between the ideological preferences of governing coalitions and Congress matters for governance in multiparty presidential systems. The evidence shows that the effect of preference congruence between the governing coalition and the floor of the chamber of deputies looks robust, even controlling for the seat share of the governing coalition (the traditional arithmetical understanding of divided government) and the level of partisan fragmentation.

One of the key implications of our findings, especially for the comparative literature, is that we may need to consider downplaying the role of the constitutional and agenda-setting powers of the executive in overcoming the centrifugal governance problems and costs of managing in multiparty presidential regimes. Presidents cannot ignore the ideological preferences of Congress if they want to manage their coalitions successfully at low costs. This is particularly relevant when the composition of Congress and its median preference are not similar to the president's preference and its coalition. By managing a governing coalition that does not mirror Congress' preferences, increasing the cost of governing may not necessarily generate more legislative support.

The emphasis on preference congruence we implemented in this manuscript helps to elucidate its role related to the cost of governing in multiparty presidential settings, a key aspect that has been scantily explored in the scholarship so far.