Minority Presidents and Types of Government in Latin America. Draft: March 2003

Minority Presidents and Types of Government in Latin America Gabriel L. Negretto Centro de Investigación y Docencia Económicas (C.I.D.E) Draft: March 2003 Prepared for delivery at the 2003 meeting of the Latin American Studies Association, Dallas, Texas, March 27-29,2003

2 Abstract A widely accepted argument among students of presidential regimes is that interbranch cooperation is impaired when the president s party does not control a majority of seats in congress. This argument, however, fails to take into account three variables that should affect the performance of minority presidential governments: the location of the president s party in the policy space, the strength of the executive veto and the formation of executive coalitions. Based on a new typology of presidential regimes, I propose the hypothesis that the most conflictive forms of minority government are those in which the president s party does not control the median legislator, the president lacks effective veto power, and a majority or median minority executive coalition is not formed. Preliminary evidence to sustain this hypothesis is provided with data on interrupted Latin American presidencies in the period 1980-2002

3 Introduction A rich literature on presidential regimes has repeatedly associated party fragmentation, and its likely effect, minority presidents, with ineffective government, high levels of executivelegislative conflict, and even democratic instability. So far, however, neither single-case studies provide an accepted causal explanation of this association nor large-n statistical studies prove beyond doubt that the relevant variables are significantly correlated. This paper argues that the absence of conclusive theories or results is due to the lack of attention to variables other than the share of seats of the president s party in congress to explain the different performance of minority presidential governments. This paper proposes that there are several possible categories of minority government, not all them in principle related to poor government performance. Using an analytic model of executive-legislative relations, I argue that the most conflictive forms of minority presidential government are those in which the president s party does not control the median legislator in congress, the president lacks effective veto power, and majority or median minority executive coalitions are not formed. Preliminary evidence to sustain this hypothesis is provided with data on interrupted Latin American presidencies in the period 1980-2002. The argument is presented as follows. In section I, I critically review the literature on minority presidential governments. Section II proposes a new typology of presidential regimes and a series of hypotheses about inter-branch cooperation based on whether or not the party of the president controls the pivotal legislators necessary to pass a bill or sustain a veto. Section III analyzes the frequency at which majority coalitions or coalitions including

4 the median party in congress are formed in each type of government. In section IV, the level of inter-branch conflict in new Latin American democracies is tested using data on the early termination of a presidential term. I conclude by indicating the lines of research that should follow from this study. Presidentialism and minority government The separate origin and survival of presidents and assemblies has been the source of several hypotheses about the comparative performance of presidential democracies. The most radical of them was initially proposed by Linz (1990, 1994) and other critics of presidentialism (Valenzuela 1994; Skach and Stepan 1994; Linz and Stepan 1996), who argued that separate elections and fix terms induce inter-institutional conflict and make difficult if not impossible the resolution of political crises. According to this view, the fusion of powers induced by the logic of parliamentary constitutions is, under all political conditions, a much better formula to foster effective and stable democracies. In response to Linz and his colleagues, a new literature originated in the work of Shugart and Carey (1992) and followed by a series of influential studies (Mainwaring 1993; Jones 1995: Mainwaring and Shugart 1997) proposed that not all presidential regimes are equally problematic from the point of view of good governance. In particular, the so-called defenders of presidentialism proposed that presidents should be able to govern effectively when the right electoral system, by containing party fragmentation, provides them with partisan majorities in congress. 1 For this literature, then, the problem is not presidentialism per se but the difficult combination of presidentialism and multipartysm that breeds weak 1 From this perspective, the right combination of electoral variables is a president elected by plurality rule, legislators elected by party-list PR with moderate district magnitudes and concurrent electoral cycles for the election of presidents and assemblies. See Shugart and Carey (1992), Jones (1995), and Mainwaring and Shugart (1997).

5 and ineffective minority presidents. From this perspective, parliamentarism is not necessarily a better option than presidentialism, at least not than the majority type. A new turn in the study of the comparative performance of constitutional regimes took place after the large statistical study developed by Przewroski et al. (2000), which showed that presidential regimes are indeed more vulnerable than parliamentary ones to democratic breakdown. This means that there might be something wrong either with separation of powers as such or with some (non-structural) institutional characteristic that presidential regimes frequently present. However, neither this study nor those that followed from it (Cheibub 2002; Cheibub and Limongi 2002, Cheibub, Przeworski, and Saiegh 2002 ) provide much more than a list of negative findings to explain the relatively low survival of presidential regimes. On the one hand, Cheibub (2002) challenges the conventional wisdom that presidentialism and multipartym is the worst combination for democratic stability by showing that minority presidencies, even when they present conditions for deadlock, do not affect the survival of democracy. 2 On the other hand, Cheibub, Przeworski, and Saiegh (2002) show that singleparty minority governments are not less effective than majority or minority coalition governments and that the survival of presidential democracies is not affected by whether or not presidents fail to generate coalitions. So far, the only explanations advanced for the instability of presidential democracy are three, rather disparate hypotheses: the existence of three relatively equal political parties, the frequency of presidential term limits, and decentralized decision-making processes (Cheibub 2002; Cheibub and Limongi 2002). 2 He also indicates, however, that the survival of presidential democracies may be correlated with moderate rather than with extreme levels of party fractionalization.

6 In spite of the significant progress made during the last decade, the main problem with the current research on presidential regimes is that it still uses too crude a measure, namely the share of seats of the president s party in congress, to conceptualize different types of presidential government and make predictions about their performance. The most glaring example of this problem is the notion of minority or divided government. According to almost all definitions, presidential government is unified when the president s party holds more and divided when it holds less than 50 percent of seats in congress (or in one chamber, for bicameral assemblies). 3 But while the concept of unified presidential government is unambiguous, a number of substantially different situations can take place once presidents fail to obtain majority support in the legislature. One could expect, for instance, that if the president s party is centrally located in the policy space and a single dimension prevails, presidential proposals would be supported by a legislative majority even if that party controls only a minority of congressional seats. In other words, although opposition parties formally control the legislative process they may be unable to agree on a policy different from the one proposed by the executive. A second possible situation is that of a non-median minority president, but with an effective veto. This reactive power could be skillfully used to forge tacit legislative coalitions on different pieces of legislation that neither the president s nor opposition parties have the power to approve without mutual support. Finally, a non-median minority president, even without a veto, may still be able to govern with relative effectiveness through an executive coalition holding a majority of seats or including the median party in congress. Obviously enough, one cannot derive the same consequences from all these forms of government. 3 Perhaps only Shugart (1995) departs from this conventional definition to restrict the term divided government to situations in which a party different from the president s control more than 50 percent of the seats in congress. As we will see later, this corresponds to the notion of unified congressional government that I propose below.

7 Since the size of the president s party in congress is related to the number of parties holding seats in congress, different authors also use the effective number of congressional parties to explain the dynamics of minority governments. The typical statement here is that as fractionalization increases, the size of the president s party decreases thus making more difficult to induce legislative support either by ad hoc negotiations with congress or by making and maintaining majority executive coalitions (Mainwaring 1993, Jones 1995). This association, however, does not amount to a real explanation of the relation between fractionalization and executive-legislative conflict and/or regime stability. To see this, compare two situations, one where a few large parties hold most of legislative seats and another where many small parties hold a relatively equal share of seats. Suppose that in the first situation the party of the president is located at the extreme left, holding 40 percent of the seats, and that the main opposition party is a center-right party holding 50 percent of the seats, followed by a small rightist party, with 10 percent of the seats. The effective number of legislative parties here is 2.4. In the second situation, while the party of the president is a centrist party holding 30 percent of the seats the 3 opposition parties are distributed as follows: one in the center-left, holding 25 percent, one in the center-right also with 25 percent, and one in the right with 20 percent. The effective number of parties here is 3.9. Clearly enough, inter-branch cooperation should be extremely more difficult in the first situation, even though the level of fractionalization is low. This analysis indicates that any hypothesis or explanation about the consequences of minority presidential government should start by taking into account not only the share of seats of the president s party in congress or the effective number of legislative parties but also the location of the former in the policy space. I turn to this task in the next section.

8 Pivotal politics and types of presidential government Using a modified version of Krehbiel s model of pivotal politics (1996, 1998), it is possible to create a typology of presidential governments according to the particular location of three decisive actors: the president s party (PP), the veto party (VP) and the median party (MP) in congress (Colomer and Negretto 2003). PP is the party that controls the executive office, VP is the party that controls the legislator whose support is crucial to override a possible presidential veto, and MP is the party that controls the median legislator, that is, the legislator whose proposals gather the support of a legislative majority. The central assumptions of this model are 1) a unidimensional policy space, 2) single-peaked preferences, 3) disciplined parties, and 4) absence of agenda-setters. The central or single dimension of policy is assumed to be here the position of parties regarding state intervention in the economy and the adoption of redistributive social policies, where parties on the left support more and parties on the right support less state intervention and redistributive policies, with parties on the center advocating a moderate position. That preferences are single-peaked means that each player has an ideal point in the policy space so that utility never increases as policies move away (in any direction) from the player s ideal point. The assumption of disciplined parties, in turn, means that the median and veto legislators are expected to vote according to the policy position of the party they represent in the legislature. The absence of an agenda-setter means, in turn, that no special procedures give any player an advantage to get his or her proposals approved. I will later show the different predictions one can obtain by changing this assumption. Following this model, one can distinguish three basic forms of government in separationof-powers systems: presidential, divided, and congressional. As shown in Figure 1, presidential government occurs whenever the president's party has the support of both the

9 veto and the median legislator. This form of government logically includes two possible situations. One, which can properly be defined as unified presidential government, derives from an election in which the president s party obtained more than 50 percent of the seats in the assembly. The other, which can be labeled median presidential government, takes place whenever the president's party, even with less than 50 percent of legislative seats, is appropriately located around the 'center' of the policy space. While unified presidential government is usually the result of plurality formulas to elect presidents and concurrent electoral cycles (Mainwaring and Shugart 1997), there is in principle no institutional variable that guarantee that the president s party, short of a legislative majority, will be the median party in congress. 4 The effective number of legislative parties (ENP), which may be affected by electoral rules, is obviously related to the location of presidential governments in one or the other category. While unified presidential government is more likely when the effective number of legislative parties approach a two-party setting (e.g. ENP 1.8-2.4), median presidential government is more likely at a moderate level of multipartism (e. g. ENP 2.5-3.9). [Figure 1 here] Turning to hypotheses of government effectiveness and executive-legislative conflict, there is no doubt that, other things being equal, a president whose party controls by itself a majority of legislators should be more able than minority presidents to implement most of 4 According to Colomer (2001) and Colomer and Negretto (2003), majority runoff elections for president result in the selection of median candidates in a higher proportion than in plurality elections. This does not mean, however, that the party of the president would be in those cases the median party in congress. The reason is that there may be more parties competing in the congressional election than parties presenting candidates in the presidential election so that the medians in both election do not necessarily coincide.

10 his agenda with very little cost in terms of inter-branch negotiation. Thus the allure of unified presidential government among many scholars. Nevertheless, if the analytic model here presented holds, one should expect a minority president whose party is centrally located in the policy space to be also more effective and face less executive-legislative conflict than any other minority president. A median minority president may be in this sense equivalent to the phenomenon of minority governments in parliamentary regimes (Strom 1990; Laver and Schofield 1990). As Laver and Schofield point out (1990: 80), even in a regime that by definition requires majority legislative support to remain in power, a single-party minority government can be a viable government if there is no majority coalition of opposition parties which can defeat its proposals. If we apply this logic to presidential regimes, it is possible to argue that a median presidential government should have a record of legislative success and democratic stability not so different from unified presidential government. Figure 2 shows what in strict sense we can call divided presidential government, that is, a situation where the president s party controls the veto legislator but a party other than the president s has the support of the median legislator in congress. Whenever elections are not decisive in the sense of providing the same party with control over the presidency and the legislature, the occurrence of divided government is determined by the effective number of parties in the legislature and by the majorities required to override a presidential veto. In particular, the probability of having a divided government should increase as the effective number of legislative parties is equal or above 3 and the president can sustain a veto with the support of relatively few legislators, such as one-third in a unicameral assembly or in any of the two chambers in a bicameral assembly.

11 [Figure 2 here] This definition of divided government is closely associated to the notion of deadlock that many authors use (Przeworski et al; Cheibub 2002). This is so because it is precisely in this type of government that a party (or parties) other than the president s may gather a majority to pass legislation and still be unable to change the status quo if the president prefers the latter to the new policy and the legislative majority lacks the necessary votes to override his veto. In other words, neither the president (who lacks a majority) nor the opposition parties (who lack the votes to override a veto) are able to rule. The risk of a stalemate, of course, is the reason why this type of presidential regime is often supposed to constitute the most undecisive, costly, and conflictive form of minority government. But this association lacks theoretical foundations. The occurrence of gridlock under divided government depends on the location of the legislative status quo and the policy preferences of the pivotal actors, in this case, the president s (or veto) party and the median party (Krehbiel 1998). If the initial status-quo policy is located in between the preferences of the median party and the president with effective veto, policy changes may be, in effect, impossible. The president will veto any change approaching the outcome to the median party s preference and moving it away from the presidential one. It is for this reason that the policy space between the two decisive actors defines the 'gridlock interval', that is, the set of policy decisions that will be stable in spite of the existence of a legislative majority favoring policy change as illustrated in Figure 2. But whenever the legislative status quo is outside the gridlock interval, policy change is still possible. If the legislative status quo is located at a rather extreme position with respect

12 to the ideal points of the decisive actors, policy change may occur at exactly the ideal point of the median party. This is so because the president will not veto proposals made by the median party if they are closer to his preferences than the status quo is. In addition, if the legislative status quo is outside the gridlock interval but closer to the preferences of the president than the ideal point of the median party is, policy change is also possible if the median party moderates its proposal to make the president at least indifferent between using the veto or signing the bill. As Cameron (2000) point out, whenever the president has an incentive to veto policy changes and congress knows it, the latter will anticipate the veto and modify the content of the new policy to head it off. In other words, policy immobilism is not a structural trait of divided government. Presidents will support policy change when the new policy brings them more utility than the status quo does and also when the legislative majority, anticipating a veto, modifies legislation to induce the president to accept a new policy. This does not mean, of course, that we should expect from divided government the same level of government effectiveness and inter-branch cooperation as we can expect from median presidential government. To the contrary, divided government is likely to produce less legislative change and at higher cost than a minority but median presidential government. In addition, we should expect divided government to produce gridlock from time to time and, in the absence of perfect information among the players, inter-branch conflict as well. The central point is simply that looking at divided government as the worst possible situation for minority presidents may be looking at the wrong place. Following the logic of this analysis, a third possible category of government in a separation-of-powers system is that where a party different from the president s controls both the median and the veto legislator. To keep symmetry with the previous concepts we

13 can call this form of government congressional, which just like the presidential one, may include two possible situations. One, which we can call unified congressional, takes place when a party different from the president s controls not only 50 percent of the seats in congress but also the number of seats necessary to override a presidential veto. This form of government is likely to occur in two-party systems with a low veto override, usually an absolute majority. 5 A second situation, which we can call median congressional, is that where the policy position of an opposition party allows it to control the veto and the median legislator, even without having a majority of its own. This form of government is more likely in multiparty presidential systems, also with a low veto override. [Figure 3 here] Congressional government is in principle the most complex form of government in a separation-of-powers system. Not only the president has no control over the medial legislator but he also lacks an effective veto. For this reason, everything else being equal, one should expect congressional government to be less decisive and more conflictive than both median presidential and divided government. In Table I, I have listed the types of government that according to this classification have existed in 18 Latin American countries since the last inauguration of a democratic regime. Within each country, presidencies are classified according to the legislative share of the president s party and its policy location vis-à-vis the location of the median and the veto party on a left-right scale. Whenever a nonconcurrent congressional election changed any 5 As we will see, the only two cases of unified congressional government in Latin America since the beginning of the transition to democracy were the presidencies of Betancourt from 1982-86 and Pastrana from 1998-2002 in Colombia, a country that gathers the two characteristics just mentioned.

14 of these relative positions, the correspondent years of the presidency are counted as a different case. According to this methodology, the table gathers 97 cases, of which 33 correspond to unified presidential government, 22 to median presidential government, 22 to divided government and 21 to congressional government. The total number of minority presidencies located in the median, divided, and congressional category represents a clear majority of 64 cases, or 66 percent of the sample. [Table I here] It is important to note how this classification of presidential regimes, and the predictions one can obtain from it, differs from the conventional analysis based on the levels of party fragmentation. In the first place, as I noted before, it is not simply the ENP but the veto override rule what matters to classify certain kinds of minority governments. Both Chile under Alwyn (1989-94) and Perú under Fujimori I (1990-92) had an ENP in the Lower House above 5. However, while the Chilean constitution requires 2/3 of the vote in each chamber to override a presidential veto, in Perú congress could override a veto with only a majority of votes in each chamber. Thus the classification of Chile s Alwyn as divided and Perú s Fujimori I as congressional government. On the other hand, Colombia under Betancourt (1982-86), and Pastrana (1998-2002) had an ENP in the Lower Chamber of 1.97 and 2.09, respectively. However, since the president s party did not reach a majority of seats and the constitution required a majority of votes to override a veto, a party different from the president s was able to control the law-making process thus creating a unified congressional government. Table II shows the relation between ENP, veto strength and presidential types of government.

15 [Table II here] Perhaps more important is the fact that even if we hold constant the ENP and veto override rules within one country, presidencies may still differ according to the policy location of the decisive players. Bolivia is the case in point. Both during the presidencies of Paz Estenssoro and Siles Suazo the ENP in the Lower Chamber was above 4 and the constitution required 2/3 of a joint session of congress to override a presidential veto. Nevertheless, while Paz Estenssoro s MNR occupied the position of the median party in congress, Siles Suazo s UDP had a rather extreme location in the policy space and possessed only a veto to negotiate with the opposition. Logically, one cannot expect the same consequences from these two very different cases of presidential governance. Minority presidents and coalition-making In the previous section I have argued that, other things being equal, congressional government should be expected to be the least decisive and most conflictive form of minority government. Yet, congressional government opens up different possibilities, not all them implying the same negative consequences for democratic governance. A non-median minority president who also lacks an effective veto has two basic options: either to accept or to challenge congressional rule. Acquiescence to congressional rule would usually take the form of a coalitional government in which non-median presidents may try to overcome their disadvantage in congress by creating a majority executive coalition or a minority coalition including the median party in congress. Defiance to congressional rule, instead, would usually take the form of unilateral presidential

16 government, either by using powers granted by the constitution, like decrees, or by extraconstitutional actions, like the usurpation of legislative powers or the irregular dissolution of congress. Clearly enough, it is only when presidents prefer to challenge congressional rule that congressional government should be the most problematic form of government in a separation-of-powers system. In order to explain this choice, we need to take a step back and look more closely at the more general problem of coalition-making under presidential regimes. 6 The separate origin and survival of presidents and assemblies creates several structural differences between the process of coalition-making in multiparty presidential and multiparty parliamentary regimes. In the first place, whereas in a presidential regime the constitution establishes that the president is the formateur regardless of the representation of his party in congress and its policy location, in a parliamentary regime the formateur is usually the largest and/or the median legislative party (Laver and Schofield 1990). In addition, while the autonomy of presidents in terms of government formation is subject to variations, in a parliamentary regime cabinets are strictly dependent on legislative support to obtain investiture and win confidence votes. These differences, should and do make coalition governments a more frequent event under parliamentarism than under presidentialism (Samuels and Eaton 2002; Cheibub, Przeworski, and Saiegh 2002). Within presidential regimes, however, one should expect variations in the frequency and nature of coalitions depending on the type of government. Table III shows these variations according to whether an executive coalition is formed, whether the party members of the coalition hold together a majority of seats in congress, and whether a minority coalition 6 On cabinet coalitions in presidential regimes, see Deheza (1997), Amorim Neto (1998), Altman (2001), Chasquetti (2001), and Amorim Neto (2001).

17 includes the median party in congress. 7 An executive (or government) coalition is defined here as the set of legislators belonging to parties that hold cabinet positions. This coalition is considered to be a majority coalition only if the parties holding cabinet posts gather a majority of seats in one chamber for unicameral assemblies or in both chambers for bicameral assemblies. [Table III here] As one would expect, Table III shows that the frequency of executive coalition formation is the lowest in unified presidential government (.22). The reason is that in this category, obviously, presidents do not need outside support to accomplish their agendas. One of the frequent reasons why presidents form coalitions in spite of the fact that their parties have majorities in congress is the intention to integrate a National Unity government in the context of power-sharing agreements, as did Betancourt (1958-63) in Venezuela and the Colombian presidents who continued the practice initiated with the National Front (1958-74) until president Barco abandoned it (1986-90). Among the different categories of minority presidents, the lowest percentage of coalitions in general (.28) and of majority coalitions in particular (.00) corresponds to divided governments. In 2 out of 18 cases (.11), the presidencies of Alwyn (1989-94) and Frei (1994-200) in Chile minority coalitions included the median party, but only for the Chamber of Deputies. 8 This case is not immediately intuitive but it may fit the logic of divided government explained above. Since in this situation minority presidents have a 7 The presidencies of Pérez Balladares and Moscoso in Panamá were excluded due to the absence of information. 8 The median position in the senate has been consistently occupied since 1989 by the RN, which for obvious ideological reasons can never be part of a center-left executive coalition.

18 weapon (the veto) to force concessions from legislative majorities, they might not need an executive coalition to influence policy. At the same time, it may be the case that in divided government opposition parties also lack the incentives to give explicit support to the government by joining an executive coalition. Since party fragmentation is on average moderate in divided government, the median party (by definition the party whose support is necessary to provide the president with the support of a legislative majority) may be one of the main opposition parties which needs to differentiate its position from the government in order to compete in coming presidential elections. In median presidential governments coalitions are formed with a relatively high frequency of 52 percent for coalitions in general and 38 percent for majority coalitions. A minority president whose party controls the median legislator in congress may form a single-party government and still be able to obtain majority support in congress. For the same reason, however, a median president has nothing to lose in terms of policy and may actually obtain some electoral benefit (in terms of public image, for instance) by incorporating outside parties to the executive coalition. Opposition parties, in turn, may derive some utility from office by supporting policies that they would approve anyway. This is precisely the argument used in parliamentary regimes to explain why a median minority party, knowing it will be dominant in policy terms, may either chose to form a single-party government, a minimal winning or a surplus majority coalition (Laver and Schofield 1990). Congressional government, in turn, is the minority government with the highest number of coalitions. In 20 out 23 cases (.87) presidents in this category ruled through executive coalitions. Given that congressional government on average occurs at relatively high levels of party fragmentation, this result clearly shows that, against a previously held belief in the literature, neither coalitions are rare in minority governments nor party fragmentation

19 makes coalition-making more difficult. 9 The incentives of presidents to form executive coalitions seem clear in the case of congressional government: neither their parties have a pivotal role in policy-making nor they have an instrument, like the veto, to shape legislation. At the same time, given the presence of many small parties in this type of government it may be that parties in the opposition also have an interest in joining the government either because they have no chance of winning coming presidential elections or because even if they win, they will not be able to govern alone. Yet, although most minority presidents in this category managed to incorporate outside parties in the cabinet, not all them forged coalitions that would make congressional government viable in policy terms. Of the 23 cases, only 9 (.39) presidents managed to form a majority coalition. In addition, in 1 case (.04), the presidency of Lagos (2000-06) in Chile, a minority coalition included the median party, but only in the Chamber of Deputies. The obvious question, then, is why some presidents in congressional government did while others did not form viable coalitions. One possible answer is that not all presidents in this situation (regardless of the incentives of opposition parties) may have the same incentives to rely on outside support to govern. One of the assumptions of the pivotal model of decision-making outlined above was that no player has special agenda-setting powers. In reality, however, many presidents do have these powers and they can used them to change legislation without sufficient legislative support. This is the case, in particular, with the ability of presidents to issue decrees of legislative content. These powers may allow non-median minority presidents to change the legislative status quo and impose a new policy if the median party in congress prefers the latter to the reversionary outcome, that is, the outcome that would result from rejecting the 9 Cheibub, Przewroski and Saiegh (2002) show a similar finding.

decree after it produced effects (Figueiredo and Limongi 1998; Negretto forthcoming). If this is the case, then, a president invested with these powers may not feel the need to rely on a majority coalition as much as a president without them. At the same time, the incentives of presidents to make viable coalitions may also be determined by their degree of autonomy in terms of the formation and maintenance of cabinets. In spite of the separate origin of presidents and assemblies that characterizes presidential regimes, there is wide range of variation in terms of congressional control over cabinets (Shugart and Carey 1992; Colomer and Negretto 2003). A president who has the formal power to appoint and remove cabinet members may be in fact restricted in his degree of autonomy if legislators are able to vote a binding censure or impeach ministers by majority vote. The same could happen if in the absence of majority winners in a presidential election, congress regularly intervenes in the final selection of presidents, somewhat resembling the investiture vote in a parliamentary system. With these elements in mind, we can make sense of the choice that minority presidents in a congressional form of government have. The incentives to accept congressional rule by building a portfolio coalition large enough to muster majority support in congress should be greater when presidents are subject to congressional control and have no decree powers than when they face no congressional control and/or have decree powers. 10 Table IV lists the information about coalition-making in congressional government along with the relevant institutional variables for each case. A president was considered to have decree powers if the constitution explicitly grants him the authority to initiate policy by decree (constitutional decree authority or CDA) or when general emergency provisions include the capacity of presidents to make law and regulate rights by decree. A congress was 10 Amorim Neto and Tafner (2002) and Amorim Neto (2002) show that there is a relation between the use and existence of decree powers and coalition-making strategies. 20

21 considered to have control over cabinets when presidents are selected by congress or when legislators from opposition parties had the necessary votes established by the constitution to propose and decide a binding censure and/or impeach ministers. 11 [Table IV here] According to this table, a large majority of 18, out of 23 presidents in the category of congressional government had the capacity to issue decrees of legislative content. Only 6 (.26) of these presidents formed policy viable coalitions while 13 of them (.56) either did not form coalitions or formed a non-median minority coalition. In addition, while 13 presidents were subject to congressional control, only in 3 cases ( 2 in Bolivia and 1 in Uruguay) presidents were both subject to congressional control and had no decree powers. In these 3 cases presidents formed majority executive coalitions. This analysis cannot conclusive since too many presidents in this category hold explicit or residual decree powers. In fact, there may be a causal relation between minority presidents in congressional governments and strong proactive powers. 12 Whatever the reason, however, it may be the frequent presence of decree powers what explains why if the number of coalitions is so high in congressional government so few of them hold a majority of seats in congress or have the support of the median party in congress. To sum up, not all minority presidents inevitably need a majority coalition to influence policy and secure a minimum of inter-branch cooperation. Some minority presidents can 11 Opposition parties usually reach the necessary votes to control cabinets when the constitution requires a vote by absolute majority in congress but congressional control could also exist with a requirement of twothirds if the president lacks the support of more than one-third of legislators. 12 The existence of a correlation between CDA and separation of purpose in presidential regimes has been indicated by Shugart (1998) and Shugart and Haggard (2001).

22 compensate their lack of majority support with their control over the median legislator in congress or with their veto power. Coalition-making, however, may be crucial for minority presidents in congressional government and those who fail to do so in this category are expected to be the most problematic of all minority presidencies. Interrupted presidencies in Latin America There are several possible indicators to test the hypothesis that minority presidencies are different in nature and present different potential for government effectiveness and executive-legislative conflict. For government effectiveness, the percentage of bills initiated by presidents and approved by congress is one possible indicator (Saiegh 2002). For executive-legislative conflict, one could construct an index to measure the number of conflicts that each presidency experienced in terms of legislation and inter-institutional conflict (Jones 1995). However, in the absence of these data for all the presidencies within the period under consideration I will provisionally rely on a more crude but nonetheless significant indicator of presidential performance: the regular termination of the constitutionally defined presidential term. Presidents may end their constitutional terms prematurely either because they are ousted from office after a successful impeachment process, because they are forced to resign after the emergence of widespread social mobilizations against the government or a coup, or because they decide to dissolve congress as a way to govern effectively in the face of a particular crisis. Within the third wave of democracy in Latin America, there are 11 cases of interrupted presidencies. Table V provides a list of these presidencies indicating the president, year and reason for termination.

23 [Table V here] In 4 cases (Collor, Pérez, Bucaram, and Cubas), presidencies were terminated in the context of a process of impeachment either initiated or concluded by congress, and in 1 case (Suazo) in the context of a series of executive-legislative conflicts that included the threat of impeachment. In 3 cases presidents resigned or abandoned the office, 2 times (Alfonsín and De la Rúa) due to social mobilizations against the government and 1 time (Mahuad) due to an attempted coup. Finally, in 3 cases (Fujimori, Serrano and Chávez) presidents terminated their terms by their own decision to rule without the existing congress, although in 1 case (Serrano) the president failed in his attempt. In spite of these variations, the premature termination of a presidential term signals the breakdown of cooperation between presidents and congresses. Whatever the nature of the crisis that presidents and legislators had to face, it seems clear that either congress or the president saw the potential solution in the elimination of the other. From this perspective, the forced resignation or impeachment of presidents, on the one hand, or the dissolution of congress by presidents, on the other, are two possible non-cooperative outcomes of a bargaining game between executives and legislators. According to the hypothesis presented in this paper presidential regimes can be located in a continuum that goes from a maximum to a minimum of expected cooperation between separate branches of government. The maximum corresponds to unified presidential governments and the minimum to congressional governments without majority (or median minority) executive coalitions. The intermediate categories are median presidential, divided, and coalitional congressional government. In order to support this hypothesis, one should see some correlation between expected levels of cooperation and the number of

interrupted presidencies in each category. Table VI shows the existence of such a correlation. 24 [Table VI here] As expected, unified presidential government has the lowest (.03) and minority congressional government the highest (.38) percentage of interrupted presidencies. Also as expected, median presidential governments indicate a lower percentage of inter-branch conflicts than divided governments. Coalitional congressional governments, in turn, have no cases of interruption of presidential terms. This does not necessarily mean that coalitional congressional government is free from conflicts but, rather, that the difference between having or not the support of a majority or median coalition is crucial when the president does not have a veto or control the median legislator in congress. So crucial, indeed, that it may turn the most problematic form of minority presidential government into a successful one. It is suggestive to note that most minority congressional governments in which presidents did finish their terms in office also presented high levels of executive-legislative conflict that could have ended in the same way as those who did not. The presidencies of Febres Cordero, Borja (for the period 1990-92), and Durán-Ballén in Ecuador are all well-known cases of government infectiveness and executive-legislative conflict. All of these presidents intended to alleviate Ecuador s fiscal crisis and debt problems by means of austerity policies that were paralyzed or watered-down in congress. In addition, and as a demonstration of congressional reaction to massive social protests generated by those policies, opposition legislators continuously harassed the president through the

25 impeachment of his ministers (Conaghan 1995: 452-53; Isaacs 1996). In fact, in the last year of Borja s presidency, congress formally requested the Supreme Court to initiate impeachment procedures against the president himself and in 1995 (after the censure of several of his ministers) congress impeached Durán-Ballén s vice-president. Durán-Ballén responded with a failed attempt to call a constitutional reform to strengthen the powers of the president vis-á-vis the congress. Other presidencies in this category also present similar indicators of conflict. President Pastrana, á la Chávez, intended but failed to call a referendum on constitutional reform that included cutting the number of congressional seats by half and calling new congressional elections. During Wasmosy s presidency in Paraguay, executive and legislators were engaged in a bitter conflict that included a threat by congress to impeach the president. Finally, both terms of Caldera became a symbol of problematic presidencies in Venezuela. During his first presidency, Caldera faced permanent obstruction from congress and institutional attacks that eventually led to a constitutional crisis when the opposition approved a law depriving the president from his participation in the nomination of judges (Coppedge 1994: 339). 13 In his second presidency, Caldera decided to face the deep economic crisis affecting the country by suspending economic rights and implementing economic measures by decree (Crisp 2000). Once congress attempted to restore those rights, it was forced to back down by Caldera s threat to call a referendum on a proposal of constitutional reform which would have enabled him to dissolve congress. 14 In other words, if we were to measure the performance of presidential democracies under less dramatic measures of conflict than the early termination of a constitutional term, we 13 Accodring to Coppedge (1994: 339), those institutional attacks included at one point AD s attempt to change the presidential regime into a parliamentary one. 14 See Latin America Weekly Report, 11/08/94

26 would probably find that the number of critical cases within minority congressional government is much larger than the one presented here. This calls, of course, for a continuation of this research with more exhaustive tests. Conclusions Since the early 1990 s, the idea that presidential regimes are inherently unstable has been replaced by the hypothesis that it is not presidentialism but the combination of presidentialism, multipartysm, and low levels of party discipline what affects government s effectiveness and democratic survival in separation-of-powers systems. As a consequence, most research on presidentialism has focused since then on the electoral variables that are supposed to contain party fragmentation, grant the president substantive congressional support and secure at least moderate levels of party discipline. The difficult combination hypothesis is today under attack by a series of studies that question the correlation between multipartysm, legislative effectiveness and democratic survival and point out to other variables, such as presidential term limits or the degree of centralization of decision-making to explain the performance of presidential regimes. To date, however, no study has provided a plausible and theoretically grounded explanation of the causal mechanisms that foster or hinder cooperation between executives and legislators in minority presidential governments. This paper is a first attempt to elaborate such an explanation. I have argued that different forms of presidential government should be distinguished taking into account not only the size of the president s party in congress but also its location in the policy space and its relation with two pivotal actors: the median and the veto legislator. Following this logic, I proposed that there is a continuum from more to less expected cooperation between

27 presidents and assemblies that goes from unified presidential to congressional government, with minority median and divided government in between. I have also suggested that congressional government is the least effective and most conflictive form of minority government only when presidents fail or do not attempt to build executive coalitions holding a majority of seats or including the median party in congress. In order to test this hypothesis, I used the constitutional (impeachment, forced resignation) or unconstitutional (coups) interruption of a presidential term as an indicator of terminal institutional conflict between presidents and assemblies and found that the hypothesis, although not probed, is at least plausible. Of the 11 presidents that left office before the term specified in the constitution, the largest percentage corresponded to the predicted most conflictive category. Although the number of cases was small, no constitutional term was interrupted when presidents in congressional governments formed majority coalitions suggesting that coalition-making may indeed be a key variable to explain government survival under difficult circumstances. The analysis provided in this paper suggests that research on presidential democracies should shift from its exclusive focus on electoral variables affecting the effective number of parties to the electoral and institutional variables that make cooperation between branches possible in the absence of a legislative majority supporting the president in congress. In this respect, a new research agenda should pay attention to the electoral formulas that make more likely than others the selection of presidents whose parties occupy a median position in congress, the influence of presidential vetoes on policy output, and the interaction between the proactive legislative powers of presidents and congressional control over cabinets over the likelihood of majority coalition formation.