With a Little Help from the Opposition? Relaxing Term Limits in the Argentine Provinces,

Transcription

1 With a Little Help from the Opposition? Relaxing Term Limits in the Argentine Provinces, Adrián Lucardi Washington University in St. Louis María Gabriela Almaraz Universidad Católica Argentina - Universidad Torcuato Di Tella - CONICET adrianlucardi@wustl.edu galmaraz@utdt.edu July 18, 2015 Abstract How can incumbents relax term limits when they cannot impose their preferences unilaterally? Modeling constitutional reforms as a bargaining game between a term-limited executive and the opposition, we argue that reforms involving term limits should be more likely when (a) the incumbent party can change the constitution unilaterally or (b) the opposition is pessimistic about its future electoral prospects. Moreover, (c) this second effect should be stronger when a single opposition party has veto power over a reform, because this precludes the executive from playing a divide-and-rule strategy. We examine these claims with data from the Argentine provinces between 1983 and The results show that the probability of initiating a reform is highest when the executive's party controls a supermajority of seats, but falls sharply when a single opposition party has veto power over a reform, and this party expects to do well in the next executive election. Keywords: institutional change constitutional change executive term limits Argentina subnational politics 1

2 Although presidential powers have long captured the attention of political scientists, 2 the issue has gained prominence in recent years as several presidents in the developing world notably Venezuela s Hugo Chávez, Bolivia s Evo Morales and Ecuador s Rafael Correa have promoted constitutional reforms aimed at relaxing term limits and dismantling checks and balances. 3 Such attempts often led to hot-heated debates and even constitutional breakdown, as was the case in Honduras (2009), Burkina Faso (2014) and Burundi (2015), where the president s reluctance to respect term limits triggered a political crisis that ended in a military coup. 4 Given the obvious restrictions that term limits place on incumbents, the fact that executives want to get rid of them is unsurprising. What is puzzling is that so many executives manage to relax term limits despite lacking the legislative majority to impose their preferences unilaterally. For example, 12 of the 38 elected presidents who managed to relax term limits between 1960 and 2009 (32%) controlled less than a two-thirds majority of seats in the national legislature. 5 Similarly, in 22 of the 30 constitutional reforms that took place in the Argentine provinces since 1983 (73%), the governor s party did not control enough seats to change the constitution without opposition support. How do executives manage to relax term limits when they cannot impose their preferences unilaterally? According to the existing literature, such reforms often result from an agreement between the executive and the opposition: the later votes in favor of relaxing term limits in exchange for some valuable institutional concession(s). 6 But this raises the question of why some opposition parties are willing to negotiate with the executive while others remain adamantly opposed to such a move. The empirical evidence cannot help answer this puzzle because it either treats the reform process as a black box ignoring all interactions between players 2

3 altogether or limits its attention to the subset of cases in which a constitutional reform was already under way that is, after an agreement between the executive and the opposition had taken place. In this paper we seek to overcome these limitations through a study of the constitutional reforms that took place in the Argentine provinces between 1983 and Our approach improves upon the existing literature in two ways. Theoretically, we model the process of constitutional reform as a bargaining game between a term-limited executive and the opposition. We agree that incumbents will remove term limits unilaterally if they can. But we also show that when the option of unilateral imposition is off the table, the probability of relaxing term limits should depend on both the unity and the expectations of the opposition. Intuitively, the opposition faces a trade-off between maximizing the chances of winning the next executive election by keeping term limits in place, and accepting the compensation payment(s) offered by the incumbent. Thus, opposition leaders will be more open to a deal when they do not expect to win the next executive election. And this effect should be stronger when a single opposition party can veto a constitutional reform, because this precludes the executive from playing a divide-and-rule strategy against his adversaries. On the empirical side, we study both the initiation of the reform process that is, whether the provincial legislature approved a law mandating a constitutional reform and its final outcome whether term limits were effectively relaxed. This approach allows us to open the black box of constitutional reform without restricting the analysis to successful cases. In both cases, the empirical results provide substantial support for our claims. The rest of this paper is organized as follows. We begin with a brief discussion of the multiple steps involved in a constitutional reform. The following section introduces our 3

4 theoretical argument and develops the hypotheses that we will examine in the rest of the paper. We then describe our research design and variables, and present our empirical results. The final section concludes. The process of constitutional reform Most constitutional reform processes can be divided into two main stages, which are summarized in Figure 1a. In the initiation stage, the legislature decides whether to approve a reform proposal presented by some player the executive, a political party, or a certain number of voters and convoke a constituent assembly or pass a constitutional amendment that will be submitted to a referendum. In the ratification stage, some additional player(s) voters, a constituent assembly, or subnational legislatures must choose between ratifying the reform proposal and rejecting it. 7 The distinction underscores the fact that reform attempts can fail in many ways, and for different reasons: some are never proposed (presumably because they would be easily rejected); others are rejected by the legislature; and others fail at the ratification stage, for example if an anti-reform party gains control of the constituent assembly. As seen in Figure 1b, however, the existing literature has ignored these nuances: the work of Baturo examines only the final outcome of the process, treating all intermediate steps as a black box (see dotted line), 8 while Shugart and Negretto restrict their attention to those cases in which the reform process was already initiated (see broken line). 9 As the plot shows, this is problematic because it ignores those cases in which the legislature did not initiate a constitutional reform, either because no reform proposal was made or because the legislature rejected it outright. 4

5 Figure 1. The process of constitutional reform. Panel (a) summarizes the main steps typically involved in the reform process, distinguishing between the initiation stage, the ratification stage, and the final outcome. Panel (b) compares alternative research designs for studying constitutional reform. The shaded areas illustrate the approach pursued in this paper: the main analysis studies the initiation of the reform process, while an extension considers the ratification stage; compare with the approach followed by Shugart and Negretto (broken line) and Baturo (dotted line). (a) The steps of the reform process (b) Comparing research designs Given the constraints involved in studying complex processes with relatively few observations, these decisions are understandable. Nonetheless, they come at the price of a substantial loss of information. To overcome this problem, in this paper we adopt the two-step approach illustrated by the shaded areas of Figure 1b. Our main analysis focuses on the initiation stage: the goal is to determine when the legislature will pass a constitutional reform bill that allows for the possibility of relaxing term limits. We then investigate whether our argument can explain why, during the ratification stage, some constituent assemblies effectively relax term limits while others keep them in place. 5

6 Explaining constitutional reforms Existing literature When do incumbents relax term limits? In answering this question, the existing literature has underscored the role of two factors. The first is that not all executives place the same value on staying in office: incumbents for whom the spoils of office are particularly large or the probability of leading a comparable life outside of power is low should be especially wiling to introduce (consecutive) reelection. 10 This suggests that the removal of term limits should be more common in poor, corrupt and under-institutionalized countries where politics rather than the private sector constitutes the main avenue for personal enrichment and a former ruler s assets can be easily expropriated after he steps down. 11 However, term limits have been relaxed in countries with very different levels of development, and modernization arguments are not always successful at explaining why some incumbents accumulate more power than others. 12 Moreover, these considerations are only valid when the incumbent is relatively unconstrained; if the opposition can veto a constitutional reform, increasing the value of office should decrease the probability of reform, because opposition leaders will be more motivated to capture the executive office for themselves. And some measures of the executive s willingness to remain in power might be actually capturing his capacity to remove term limits: a large public sector or a high risk of expropriation certainly increase the stakes of office, but they also make it easier for the executive to bribe or threaten the opposition. Second, a large literature argues that the probability of observing power-concentrating institutional reforms rather than power-sharing ones depends on two factors: (a) whether a single player can modify the relevant legislation unilaterally, and (b) the probability that this player will remain in office in the future. When a single player can impose any institution it likes and 6

7 expects to remain in office in the future, we should observe power-concentrating reforms, like a more powerful executive or a majoritarian electoral system. But if an institutional reform requires the agreement of multiple players, or the most powerful player expects to lose the next election, institutional reforms should be power-sharing, like a proportional electoral system or an independent judiciary. 13 The problem with this argument is that it cannot explain why some incumbents manage to relax term limits even when their party does not control enough seats to change the constitution unilaterally. Acknowledging this fact, some authors have argued that these reforms are possible because of a grand bargain between the executive and the opposition. For example, opposition leaders might let the incumbent run for reelection in exchange for a more proportional electoral system or an independent judiciary. 14 But this begs the question of what conditions facilitate (or hinder) such agreements: Why some opposition parties are willing to negotiate a deal with the executive, while others remain adamantly opposed to such a move? The argument Our theoretical argument builds upon this literature but introduces two crucial differences: it pays more attention to who is fragmented and whose expectations matter. While a fragmented political system might lead to power-sharing institutions, a fragmented opposition can facilitate the introduction of power-concentrating reforms, which is the opposite of what the literature predicts. Similarly, when the opposition can veto a constitutional reform, it is the opposition s expectations that matter. Simply put, opposition parties that expect to win the next executive election will be more reluctant to let the sitting executive run for a new term than those that expect to lose anyway. 7

8 More specifically, we treat the reform process as a bargaining game between a term-limited executive whom we also call the incumbent and one or more opposition parties. Removing term limits requires the support of a supermajority of legislators. We assume that executives running for reelection enjoy an electoral advantage and that the executive office is the most valuable political position in the polity, a reasonable assumption in most developing countries. Thus, other things equal opposition parties will prefer to keep term limits in place, though their capacity to achieve such an outcome may vary. Specifically, the players might find themselves in one of three mutually exclusive scenarios, each with different implications for the probability of reform. In the first ( supermajority ), the incumbent party controls a supermajority of seats; in the second ( single party veto ), a single opposition party controls enough legislators to veto a constitutional reform; and in the third scenario ( fragmented opposition ), the incumbent party does not control a supermajority of seats, but no opposition party can veto a reform unilaterally. We consider each of them in turn. (1) Supermajority. If the party of the sitting executive controls a supermajority of seats in the legislature, it can approve a constitutional reform regardless of what the opposition does. In other words, the opposition is irrelevant, which makes a reform particularly likely. For example, in 1997 Namibian president Sam Nujoma took advantage of his party s legislative supermajority to run for a third consecutive term. 15 Similarly, when Venezuelan president Hugo Chávez sought to scrap term limits altogether, the fact that his party controlled almost all seats in the National Assembly the opposition had boycotted the previous election greatly facilitated legislative passage. 16 This suggests the following hypothesis: 8

9 H 1. (Supermajority). A constitutional reform should be more likely when the executive s party controls a supermajority of seats in the legislature. (2) Single party veto. When a single opposition party can block a constitutional reform, the outcome will depend on whether the executive and the opposition can reach a mutually satisfactory agreement. Since the incumbent needs opposition support to accomplish his reform project, he has obvious incentives to offer some valuable compensation in return. But the preferences of the opposition party are less clear: while its leaders value the compensation payments that the incumbent can offer, they also know that relaxing term limits will decrease their chances of capturing the executive in the upcoming election. 17 Thus, the opposition s choice will depend on the relative importance of three factors: (a) the generosity of the compensation payment(s) offered by the incumbent; (b) the value of controlling the executive; and (c) its probability of winning the next executive election. In general it makes sense to assume that (b) will be much larger than (a), not only because the executive is the most valuable office in the polity, but also because the incumbent will not be interested in offering concessions that trump the value of holding executive office. However, the extent to which the opposition can expect to win the next executive election can vary substantially. This implies that the electoral expectations of the opposition should play a key role in determining whether an agreement can be reached: opposition parties that do not expect to do well in the upcoming election will accept the incumbent s offer and agree to relax term limits, while those with good chances of winning office in the future will fight tooth and nail to block the incumbent s reelection drive. 9

10 The Argentine constitutional reform of 1994 offers a good example of this. Since his party was forty deputies short of the two-thirds majority required to enact a constitutional change, president Carlos Menem s reelection project dependent on the support of the main opposition party, the Unión Cívica Radical (UCR). 18 Thus, Menem adopted a two-pronged strategy. On the one hand, he called a (nonbinding) plebiscite to exploit the UCR s unpopularity with voters. 19 On the other, he showed willingness to introduce institutional reforms that the UCR valued, for example eliminating the electoral college, shortening the presidential term, or placing formal limits on the president s decree authority. 20 Eventually an agreement was reached, and Menem was able to seek reelection in This suggests the following hypothesis: H 2. (Expectations). If the incumbent party does not control a supermajority of seats, a reform should be less likely the more optimistic the electoral expectations of the opposition. (3) Fragmented opposition. This scenario is similar to the previous one, with the difference that no opposition party can veto a reform unilaterally. That is, the incumbent needs opposition support, but no single opposition party is indispensable: if A will vote against the reform no matter what, the executive may get what he wants by reaching an agreement with B. Thus, in addition to the previous considerations, opposition leaders must also consider what other opposition parties are likely to do. Intuitively, if the executive reaches an agreement with A but not with B, then B will pay the cost of running against an incumbent seeking reelection but without having received any compensation payment(s) in return. In other words, a fragmented opposition allows the executive to play a divide-and-rule strategy, taking advantage of each opposition party s fear that the other will try to negotiate a better deal. Of course, opposition 10

11 leaders understand this and may seek to negotiate jointly with the executive, but the credibility of such an agreement cannot be taken for granted. The implication is that when the opposition is fragmented, its electoral expectations should also matter for the probability of reform, but to a lesser extent than if a single opposition party has veto power. The (failed) Ukrainian constitutional reform of 2004 provides a real-life example of this scenario. Because of his alleged involvement in corruption scandals and politically motivated assassinations, president Leonid Kuchma placed a high value on remaining in power after his second term expired in International pressure meant that the cost of removing term limits was prohibitive, so Kuchma sought to weaken the presidency and strengthen the prime minister s office, hoping to swap roles at the end of his term. But the president s allies were sixty deputies short of the two-thirds majority required to approve a constitutional amendment, forcing Kuchma to seek an agreement with the opposition. The parties on the right expected to win the next executive election, and thus had little incentive to weaken the presidency. But the parties on the left had little chance of capturing the executive, so they agreed to support Kuchma s proposal in exchange for a more proportional electoral system that would increase their legislative representation. They delivered, though in the end the reform failed by a narrow margin due to some last-minute defections among the president s own supporters. 21 A strikingly similar process had unfolded in the Argentine province of La Pampa in Governor Rubén Marín was term limited and his party was a few deputies short of the two-thirds majority required to approve a constitutional reform. The two opposition parties with legislative representation the UCR and Convocatoria Independiente (CI) opposed the governor s reform project, but neither could veto it single-handedly. Thus, Marín threatened to replace the PR electoral system then in place with a majoritarian arrangement. This would have been disastrous 11

12 for CI, and since changing the electoral rules required a simple majority of seats, the governor s threat was credible. Therefore, CI agreed to support Marín s reelection bid if the PR system was enshrined in the constitution ensuring that it could not be changed by a simple majority in the future. After the agreement became known, the UCR dropped its opposition to the governor s reelection in exchange for a voice in the reform process. 22 Thus, the final hypothesis is the following: H 3. (Fragmented opposition). The effect of the expectations of the opposition should be weaker when no single opposition party can veto a constitutional reform. Research design We examine these hypotheses with data on the Argentine provinces between 1983 and Like the U.S. states, the Argentine provinces enjoy a substantial degree of autonomy for designing local institutions, including executive term limits. When the country returned to democracy in 1983, no provincial governor was allowed to stand for reelection at the end of his term, but by 2015 this restriction only remains in place in two districts. Moreover, the timing of these reforms differed substantially between provinces some introduced executive reelection as early as 1986, while others waited until 2007, and some provincial legislatures initiated multiple reforms. At the same time, other provincial institutions display relatively little variation: all provinces have a presidential system, all gubernatorial terms last four years, almost all provincial executives are directly elected by plurality rule, 23 and constitutional changes require the approval of a two-thirds majority of (both chambers of) the provincial legislature. 24 This contrasts with the cross-country literature on institutional reform, where the rules governing 12

13 institutional change often differ widely between units and cannot be summarized along a common metric. 25 The analysis is divided in two parts. We begin by examining the initiation of the reform process: the goal is to see when a provincial legislature will pass a constitutional amendment or approve a special law calling for a constituent assembly. In this case, the unit of observation is the province-two-year; that is, we divide each four-year gubernatorial term into two two-year periods, treating each as a separate observation. We do not use entire gubernatorial periods because several provinces hold midterm elections, which might alter the composition of the provincial legislature. 26 This yields a maximum of 14 observations per province, 27 though the actual number is generally lower, for two reasons. Since the universe of interest are those governors who may have wanted to change the provincial constitution in order to run for another term, we exclude all observations in which the sitting governors faced no term limits at the beginning of the period. 28 We also restrict the sample to elected governors who remained in office for at least half of the two-year period; 29 acting vice-governors and interim governors are weaker political players, which makes them unlikely to initiate a reform process. These factors explain why the main sample contains 177 observations instead of the approximately 330 that would be the case if all two-year periods were included. 30 The second analysis studies the ratification stage. As illustrated in Figure 1b, this means that we examine a constituent assembly s choice between introducing reelection and keeping term limits in place. To that end, we employ a sample of 22 constituent assemblies that followed from the initiation decisions examined in the first analysis

14 Variables In the main analysis the outcome is Initiation, a dummy that takes the value of one if the provincial legislature (a) approved a law calling for a constitutional reform that (b) would have allowed the sitting governor to stand for reelection at the end of his term. 32 A total of 29 such reforms were initiated between 1983 and 2013; this exceeds the number of provinces (24) because some reform attempts ended in failure, and some provinces relaxed term limits more than once. In the second analysis the dependent variable is Reelection, a dummy that takes the value of one if the constituent assembly introduced a clause allowing the sitting executive to run for reelection at the end of his term. 33 Almost four out of five assemblies (17 of 22) allowed the governor to run for a new term. 34 According to our hypotheses, the probability of reform depends on two factors: the scenario in which players find themselves and the electoral expectations of the main opposition party. We capture the first with two dummies. (Super)majority takes the value of one if the governor s party controlled enough seats to initiate (or approve) a constitutional reform unilaterally. This corresponds to a two-thirds majority in (both chambers of) the provincial legislature or to an absolute majority in the constituent assembly. Single party veto takes the value of one when a single opposition party could veto a reform by itself. This corresponds to more than one third of the seats in (at least one chamber of) the provincial legislature, or to half of the seats in a constituent assembly. The fragmented opposition scenario obtains when both (Super)majority and Single party veto equal zero. 35 Operationalizing the expectations of the opposition is trickier. Survey data is either unavailable or nonexistent for most provinces. We thus employ electoral data, under the assumption that past electoral results provide opposition leaders with information about their 14

15 future electoral prospects. Thus, Expectations is defined as the vote share of the main opposition party in the last gubernatorial election that took place in the province. 36 This variable is certainly correlated with Single party veto, 37 but the two are not equivalent: provincial electoral systems are often biased in favor of the incumbent party, 38 and the fact that several provinces hold midterm elections means that the composition of the provincial legislature is not entirely determined in years with gubernatorial elections. Indeed, Figure 2c shows that Expectations can vary substantially even when Single party veto is equal to one. We also include the following controls. Reelected governor is a dummy that takes the value of one if the sitting executive was not serving his first consecutive term in office. To the extent that a third consecutive term is more difficult to defend in public than a second one, this variable should have a negative effect on the probability of reform. National reform is a dummy that takes the value of one for the period; it accounts for potential contagion effects from the national constitutional reform of National transfers indicates the amount of revenues per capita that the province received from the national government, averaged over four years. Controlling for this variable makes sense because more revenues make the governorship more attractive, while at the same time increasing the governor s capacity to compensate the opposition and making provincial politics less competitive. 39 Partisan cultures and the structure of party organizations can also affect the probability of reform: decentralized parties might be easier to co-opt while in opposition, but more difficult to discipline while in government. 40 Thus, we control for the identity of the Incumbent party and the main Opposition party, both of which are factors with three categories: PJ (Partido Justicialista), UCR (Unión Cívica Radical), and Other

16 Specification For the main analysis we fit random effects probit models of the form Pr (Y it = 1) = Φ (α j + β S S jt + β V V jt + β E E jt + β VE V jt E jt + γ C jt ) α j ~ N (µ α, σ α 2 ), where Pr (Y jt = 1) is the probability that a constitutional reform law will be approved in province j in period t, Φ is the normal CDF, α j is a random intercept that varies by province, S jt and V jt stand for Supermajority and Single party veto respectively, E jt indicates the Expectations of the opposition, and C jt is a vector of controls. The second analysis keeps the underlying linear specification, but with two important differences. Since most provinces held a single constituent assembly, we get rid of the random intercepts. Furthermore, perfect separation between Majority and Reelection precludes the use of generalized linear models, so we fit linear probability models, which are immune to this problem. The interpretation of the results follows directly from the hypotheses. The supermajority hypothesis predicts β S > 0: the probability of reform should increase when the governor s party controls a (super)majority of seats. According to the expectations hypothesis, a more optimistic opposition should be less likely to acquiesce to a reform, implying β E < 0. And the fragmented opposition hypothesis predicts β VE < 0: when a single opposition party can veto a reform, the negative effect of Expectations should be larger in magnitude. Results Initiating constitutional reforms Since we are trying to estimate a complex relationship with a relatively small sample, some readers might wonder whether our findings constitute a statistical artifact rather than a feature of 16

17 Figure 2. Distribution of Initiation, conditional on the Expectations of the main opposition party, under different scenarios. Filled dots denote governors in their first term in office, while open dots indicate governors who had already been reelected. the data. Thus, in Figure 2 we plot the distribution of Initiation, conditional on Expectations, for each of the three scenarios considered by our argument. Prima facie, the data seems consistent with our hypotheses. Figure 2a shows that whenever the governor s party controls a supermajority of seats, a reform is very likely. Figure 2b indicates that when the opposition is fragmented, the probability of reform does not seem to depend on Expectations. This runs counter to the expectations hypothesis, but notice that many of these observations correspond to reelected governors, who might be less likely to pass a reform for other reasons. Finally, Figure 2c shows that when a single opposition party can veto a reform, Expectations has a negative effect on the outcome, and the relationship is not being driven by observations with abnormally low values of this variable: most observations are located along the [0.20:0.50] range, but instances of initiation are overwhelmingly concentrated in the [0.20:0.40] interval. Table 1a explores whether these patterns persist after accounting for province-specific effects and multiple variables. Model 1 only includes Supermajority, Single party veto, Expectations and the interaction between the last two as explanatory variables. Model 2, which is our preferred specification, adds a dummy indicating whether the governor had already been reelected. Models 17

18 3 to 5 add controls for National reform, National transfers, and the identity of the incumbent and opposition parties, respectively. Model 6 includes all control variables simultaneously. Table 1. Constitutional reforms in the Argentine provinces, Panel (a) reports the results for the main analysis, with Initiation as the outcome. Panel (b) reports the results for the constituent assembly sample, with Reelection as the outcome. Standard errors in parentheses. (a) Outcome is Initiation (random effects probit) (b) Outcome is Reelection (linear probability model) (1) (2) (3) (4) (5) (6) (7) (8) (9) (Super)majority (β S ) (0.36) (0.38) (0.38) (0.38) (0.38) (0.39) (0.17) (0.17) (0.18) Single party veto (β V ) (0.95) (0.99) (1.00) (1.00) (1.03) (1.04) (0.21) (0.77) (0.83) Expectations (β E ) (1.87) (2.00) (2.02) (2.02) (2.28) (2.32) (0.58) (0.61) Veto x Expectations (β VE ) (2.71) (2.82) (2.86) (2.84) (2.95) (2.99) (2.85) (3.14) Reelected governor (0.32) (0.32) (0.33) (0.34) (0.36) (0.21) National reform (0.32) (0.33) National transfers (logged) (0.27) (0.29) Incumbent party: PJ (0.52) (0.53) Incumbent party: UCR (0.46) (0.48) Opposition party: PJ (0.55) (0.57) Opposition party: UCR (0.35) (0.36) Intercept (0.60) (0.67) (0.66) (1.99) (0.79) (2.09) (0.14) (0.23) (0.24) AIC BIC log-likelihood Deviance num. observations num. provinces Provincial variance (σ j ) Residual variance

19 In line with Figure 2, we find support for the supermajority and fragmentation hypotheses. First, the coefficients for Supermajority are always positive and reliable. Second, and contrary to the expectations hypothesis, the point estimates for Expectations are positive, though the large standard errors mean that we cannot reject the claim that the actual effect might be zero. Finally, the interaction between Single party veto and Expectations has a negative sign and is large in magnitude, in line with the claim that the expectations of the opposition are more relevant when a single party can veto a constitutional reform. The point estimates are not entirely reliable due to the small sample size, but most are statistically significant at the 0.10 level. Moreover, the effect of the interaction term is much larger in magnitude than that of Expectations, indicating that when the opposition is unified the net effect of this variable is negative. Including control variables weakens this effect, but the negative estimate for the interaction term remains in place. The controls often have the expected sign (positive for National reform and National transfers; negative for Reelected governor), but only the point estimates for Reelected governor are substantial in magnitude and reliably estimated. To get a sense of the magnitude of these findings, Figure 3 displays the predicted probability of initiating a reform as Expectations increases across its range, conditional on the opposition s capacity to veto a reform. 42 Figure 3a supports the claim that the effect of Supermajority is large in magnitude and independent of the expectations of the opposition. Although the slope of the curve is positive, the substantive effect is small: increasing Expectations along its interquartile range (from 0.28 to 0.42) raises the probability of initiating a reform from 0.54 to This is not a large effect, especially if we consider that the probability of initiating a reform was already high at the lower quartile of Expectations. Figure 3b indicates that when the opposition is divided, the probability of initiating a reform increases with Expectations, but again the 19

20 Figure 3. Predicted probability of initiating a constitutional reform, conditional on Expectations, under different scenarios. Broken lines indicate 80% confidence intervals. All results based on model 2, Table 1, assuming Reelected governor = 0. substantive effect is modest: as Expectations moves along its interquartile range, the probability of initiation goes from 0.20 to Finally, Figure 3c shows that when a single opposition party can veto a reform, a similar increase in Expectations cuts the probability of initiating a reform by half, from 0.24 to That is, we go from expecting one reform every two gubernatorial terms (four two-year periods) to one every four terms. 43 Constituent assemblies and the re-election clause Table 1b reports the results for the sample of constituent assemblies. Model 7 includes only Majority and Single party veto as predictors. Consistent with the first hypothesis, the point estimate for the first variable indicates that if the governor s party controls an absolute majority of seats in the assembly, the probability of relaxing term limits increases by 33 percentage points indeed, all assemblies in which the incumbent party controlled an absolute majority of seats introduced executive reelection. Conversely, the negative point estimate for Single party veto indicates that if a single opposition party can veto the assembly s decisions, the probability that the governor will be allowed to run for a new term decreases by 42 percentage points. The next two models examine whether the probability of introducing reelection is also driven by the 20

21 Expectations of the opposition. This seems to be the case: as in the previous section, the point estimates for Expectations are positive but small in magnitude and very unreliable, while the interaction with Single party veto is large and negative. Certainly, the estimates are quite unreliable, but this is to be expected given that we are including an interaction in a very small sample. Model 9 also shows that including a dummy for Reelected governor does not change the results. This is consistent with the interpretation suggested before: reelected governors might find it harder to convince the public (and the legislature) of the necessity of reforming the constitution in order to run for a third term, but once this obstacle is surmounted there is no reason why they should be less successful in a constituent assembly. Discussion and conclusion This paper began with the question of why some incumbent executives are able to introduce power-concentrating institutions such as relaxing term limits when the option of unilateral imposition is off the table. To answer this puzzle, we modeled the process of constitutional reform as a bargaining game between a term-limited executive and one or more opposition parties. This generates two implications that had been overlooked by the existing literature. First, opposition parties that expect to do well in the upcoming election have more to lose if the executive can stand for reelection, and thus will be more inclined to keep term limits in place. Second, this effect will be stronger when a single opposition party can veto a constitutional reform by itself, because this prevents the executive from playing a divide-and-rule strategy. In line with these claims, the empirical findings show that Argentine governors were most likely to initiate a reform involving term limits when their party controlled a supermajority of seats in the provincial legislature, and least likely when a single opposition party could veto a reform and 21

22 expected to do well in the next executive election. The small sample size reduces the reliability of the estimates, but nonetheless the magnitude of the effects is substantial: a governor whose party controls a supermajority of seats has a more-than-even chance of initiating a reform; but if an opposition party can veto a reform, increasing Expectations across its interquartile range decreases the probability of reform from one in four to one in eight. A similar logic explains the behavior of constituent assemblies. Although we chose to focus on the Argentine provinces for methodological reasons, it is worth noting that our argument can shed light on processes of constitutional reform taking place elsewhere. As mentioned above, our argument can extend Negretto s account of constitutionmaking in Latin America 44 by making more precise predictions about the conditions that should facilitate agreements between the executive and the opposition. Moreover, while Negretto s sample is limited to constituent assemblies, the research design proposed in this paper can lead to a better understanding of how these assemblies originate in the first place. Our argument can also help explain the origins of competitive authoritarian regimes, that is political regimes that combine formal democratic institutions with an electoral playing field that is heavily skewed in favor of the ruling party due to electoral fraud, say, or because the incumbent monopolizes media access. Although these regimes have received a lot of attention recently, 45 few authors have noticed that their origin often lies in the restriction of democratic competition through a self-coup or the gradual erosion of political liberties rather than in the (incomplete) liberalization of a non-electoral regime. 46 This suggests a similar puzzle to the one that motivates this paper, namely: How can democratically elected executives manipulate elections to such an extent that future alternation in power becomes unlikely? How can incumbents get away with large-scale fraud and other forms of undemocratic behavior (like 22

23 closing the legislature) in a context in which formal institutions are explicitly designed to prevent such kind of actions? The fact that most Argentine provinces, though democratic, 47 share important similarities with competitive authoritarian regimes including pervasive patronage, weak legislatures, and a fragmented opposition opens the possibility that similar mechanisms might be at work in both cases. Indeed, the examples of Peru and Russia in the 1990s and Venezuela after 1998 suggest that a divided and discredited opposition can play a key role in allowing incumbents to get away with overtly undemocratic behavior. 48 It remains to be seen whether a similar logic holds in a broader array of countries. 1 We thank Germán Lodola, who suggested that we should work in this paper together, as well as Margit Tavits, Frank Thames, Guillermo Rosas, Marcelo Nazareno, Brian F. Crisp and eight anonymous reviewers for their helpful comments. We are grateful to Carlos Gervasoni, Agustina Giraudy, Germán Lodola, Mark P. Jones and Sebastián Saiegh, who kindly shared their data on Argentine provincial politics. Earlier versions of this paper were presented at the 2013 MPSA Annual Meeting, the 2013 Symposium on Scholarship and a Free Society and the XI Congress of the Argentine Association of Political Analysis, where it was awarded the Guillermo O'Donnell Prize on Comparative Politics. We thank participants at those venues for their helpful comments and encouragements. All remaining mistakes are our entire responsibility. 2 Matthew Soberg Shugart and John M. Carey, Presidents and Assemblies. Constitutional Design and Electoral Dynamics (Cambridge: Cambridge University Press, 1992); John M. Carey and Matthew Soberg Shugart, eds., Executive Decree Authority (Cambridge: Cambridge University Press, 1998); Eduardo Alemán and George Tsebelis, The Origins of Presidential Conditional Agenda-Setting Power in Latin America, Latin American Research Review 40, no. 2 (2005): 3 26; George Tsebelis and Eduardo Alemán, Presidential Conditional Agenda Setting in Latin 23

24 America, World Politics 57, no. 3 (2005): ; Brian F. Crisp, Scott W. Desposato, and Kristin Kanthak, Legislative Pivots, Presidential Powers, and Policy Stability, Journal of Law, Economics, and Organization 27, no. 2 (2011): See Gideon Maltz, The Case for Presidential Term Limits, Journal of Democracy 18, no. 1 (2007): ; Alexander Baturo, The Stakes of Losing Office, Term Limits and Democracy, British Journal of Political Science 40, no. 3 (2010): ; Alexander Baturo, Democracy, Dictatorship, and Term Limits (Ann Arbor: University of Michigan Press, 2014); Tom Ginsburg, James Melton, and Zachary Elkins, On the Evasion of Executive Term Limits, William and Mary Law Review 52, no. 6 (2011): ; Daron Acemoglu, James A. Robinson, and Ragnar Torvik, Why Do Voters Dismantle Checks and Balances?, Review of Economic Studies 80, no. 3 (2013): ; Gabriel L. Negretto, Making Constitutions: Presidents, Parties, and Institutional Choice in Latin America (Cambridge: Cambridge University Press, 2013). We prefer to speak of relaxing term limits rather than removing them because constitutional reforms often let the incumbent run for an additional term but fall short of eliminating term limits altogether. 4 Defying the outside world, The Economist, July 2, 2009; Not so pretty now, The Economist, November 8, 2014; and Good coup, bad coup, The Economist, May 23, The list of presidents who relaxed term limits comes from Table 3.5 of Baturo, Democracy, Dictatorship, and Term Limits. The numerator is restricted to those executives who controlled less than a two-thirds majority and passed a constitutional reform through the legislature. 6 Negretto, Making Constitutions; María Gabriela Almaraz, Ambición Política por la Reelección en las Provincias Argentinas, Revista SAAP 4, no. 2 (2010):

25 7 Between 1960 and 2009, 80% of reforms involving term limits were promulgated by either a legislative body or a referendum (most of which were preceded by a legislative decision; see Baturo, Democracy, Dictatorship, and Term Limits, Table 3.1). 8 Baturo, The Stakes of Losing Office, Term Limits and Democracy ; Baturo, Democracy, Dictatorship, and Term Limits. 9 Matthew Soberg Shugart, The Inverse Relationship between Party Strength and Executive Strength: A Theory of Politicians Constitutional Choices, British Journal of Political Science 28, no. 1 (1998): 1 29; Gabriel L. Negretto, Political Parties and Institutional Design: Explaining Constitutional Choice in Latin America, British Journal of Political Science 39, no. 1 (2009): ; Negretto, Making Constitutions. 10 Term limits can take many forms: some executives are barred from running again forever; others must spend some period(s) out of office before running again; and yet others can remain in power for only two consecutive terms (Ginsburg, Melton, and Elkins, On the Evasion of Executive Term Limits ). Since our focus is on consecutive reelection, we classify an incumbent as term-limited if he is legally barred from running in the next executive election. 11 Baturo, The Stakes of Losing Office, Term Limits and Democracy ; Baturo, Democracy, Dictatorship, and Term Limits. 12 Mario D. Serrafero, Reelección y Sucesión Presidencial. Poder y Continuidad en Argentina, América Latina y EE.UU. (Buenos Aires: Editorial de Belgrano, 1997); Carlos Gervasoni, A Rentier Theory of Subnational Regimes: Fiscal Federalism, Democracy, and Authoritarianism in the Argentine Provinces, World Politics 62, no. 2 (2010): Jon Elster, Forces and Mechanisms in the Constitution-Making Process, Duke Law Journal 45, no. 2 (1995): ; Carles Boix, Setting the Rules of the Game: The Choice of Electoral 25

26 Systems in Advanced Democracies, The American Political Science Review 93, no. 3 (1999): ; Kenneth Benoit, Models of Electoral System Change, Electoral Studies 23, no. 3 (2004): ; Jodi Finkel, Judicial Reform as Insurance Policy: Mexico in the 1990s, Latin American Politics and Society 47, no. 1 (2005): ; Gabriel L. Negretto, Choosing How to Choose Presidents: Parties, Military Rulers, and Presidential Elections in Latin America, The Journal of Politics 68, no. 2 (2006): ; Negretto, Political Parties and Institutional Design ; Negretto, Making Constitutions; John A. Ferejohn, Frances Rosenbluth, and Charles Shipan, Comparative Judicial Politics, in The Oxford Handbook of Comparative Politics, ed. Carles Boix and Susan C. Stokes (Oxford: Oxford University Press, 2007), ; Marcelo Leiras, Agustina Giraudy, and Guadalupe Tuñón, Who Wants an Independent Court? Political Competition and Supreme Court Instability in the Argentine Provinces ( ), The Journal of Politics 77, no. 1 (2015). 14 Negretto, Making Constitutions. See also Almaraz, Ambición Política por la Reelección en las Provincias Argentinas. 15 Peter VonDoepp, Party Cohesion and Fractionalization in New African Democracies: Lessons from Struggles over Third-Term Amendments, Studies in Comparative International Development 40, no. 3 (2005): Voters initially rejected the proposal in the 2007 referendum, but they were more supportive two years later, when Chávez insisted with a similar project (Javier Corrales and Michael. Penfold-Becerra, Dragon in the Tropics: Hugo Chavez and the Political Economy of Revolution in Venezuela (Washington, DC: Brookings Institution Press, 2011). 17 The same reasoning holds if they believe that voters will interpret an agreement as a sellout and punish them at the polls. We thank Marcelo Nazareno for suggesting this interpretation. 26

27 18 Technically, the UCR was two deputies short of the eighty-six required to veto a constitutional reform. But if the UCR voted against the reform, Menem would have needed the support of almost all other legislators, some of whom belonged to small parties that strongly opposed his government. That is, the transaction cost of negotiating with all other opposition deputies was so high that for practical purposes the UCR can be considered a veto player. 19 The UCR was discredited for its mismanagement of the economy during the 1980s. In the 1993 legislative election it received 30% of the vote, and in 1995 its presidential candidate would hit a record low of 17%. 20 Carlos H. Acuña, Algunas Notas sobre los Juegos, las Gallinas y la Lógica Política de los Pactos Constitucionales (Reflexiones a Partir del Pacto Constitucional en la Argentina), in La Nueva Matriz Política Argentina, ed. Carlos H. Acuña (Buenos Aires: Nueva Visión, 1995), ; Negretto, Making Constitutions, ch Baturo, Democracy, Dictatorship, and Term Limits, pp Juan Pablo Micozzi, Procesos de Reforma Institucional en las Provincias de Chaco, Chubut y La Pampa (Working Paper, Proyecto Arg 007, PNUD-Jefatura de Gabinete de Ministros, 2001). 23 Three provinces had an electoral college until 1993, and four use some kind of runoff system. The rest employ simple plurality rule. 24 Provincial constitutions can be changed in two ways: (a) through a constituent assembly specifically called for that purpose, or (b) via a legislative amendment that voters must ratify in a referendum. In both cases, the reform must be initiated by a supermajority of at least two-thirds of provincial legislators. 25 Richard S. Katz, Why There Are so Many (or so Few) Electoral Reforms?, in The Politics of Electoral Systems, ed. Michael Gallagher and Paul Mitchell (Oxford: Oxford University Press, 27