Regime and Leader Instability under Two Forms of Military Rule Nam Kyu Kim∗ Alex M. Kroeger† Abstract The finding that military regimes are more fragile than other authoritarian regimes represents one of the few stylized facts in comparative politics. However, the existing literature contains substantial differences in the theoretical explanations for military regime instability and operationalizations of military rule. To assess competing explanations, we examine regime and leader instability after distinguishing between collegial and personalist military rule. We show that regime and leader insecurity characterize only collegial military regimes. Particularly, the fragility of collegial military regimes comes from a heightened likelihood of democratization, not more frequent transitions to alternative autocratic regimes. Additionally, leaders of collegial military regimes face higher risks of both regular and irregular turnovers than other autocrats. Also, irregular exits of collegial military leaders tend to occur through reshuffling, rather than regime-changing, coups. The results strongly support theories focusing on military officers’ preference for unity over other explanations. Forthcoming in Comparative Political Studies ∗ Corresponding Author: Assistant Professor, Department of Political Science, University of NebraskaLincoln. Email: [email protected]. † Graduate Student, Department of Political Science, University of Nebraska-Lincoln. Email: [email protected]. Introduction The finding that military regimes are more unstable than other authoritarian regimes represents one of the few stylized facts in comparative politics. Since Geddes (2003) established this finding, several follow-up studies confirm that military regimes have shorter tenures than other autocratic regimes and are more likely to end in democratization (Debs, 2016; Gandhi, 2008; Geddes, Wright, & Frantz, 2014; Lai & Slater, 2006; Magaloni, 2008). Thus, the fragility of military regimes appears to be a foregone conclusion. However, a closer look at the existing literature reveals substantial differences in the theoretical explanations for military regime instability and the operationalization of military rule. The fragility of military regimes is explained either by military officers’ preferences for preserving the unity and corporate interests of the military (Geddes, 2003), the lack of political parties and infrastructural power to maintain control over society without relying on sheer repression (Gandhi & Przeworski, 2007; Lai & Slater, 2006; Magaloni, 2008), or the ability of military leaders to secure improved post-tenure fates by initiating democratic transitions (Debs, 2016; Geddes, 2003). Scholars also define military rule differently. While some focus on military-led dictatorships, emphasizing the importance of a leader’s current or former military service in explaining regime fragility (Debs, 2016; Gandhi, 2008), others discuss collegial military regimes where a group of officers governs and controls access to political offices (Geddes, 2003). Geddes, Frantz, and Wright (2014), extending the work of Geddes (2003), emphasize both the military-civilian and the personalist dimensions of military rule by distinguishing between two types of military regimes: collegial military regimes and personalist military regimes. As a result of these theoretical and definitional differences, our knowledge of military regimes is much less coherent than often thought, with several important puzzles remaining to be answered. For example, it is unclear whether regime and leadership insecurity apply to both collegial and personalist military rule. It may be that the differences between military-led and civilian-led dictatorships are driven primarily by the difference between collegial military 1 regimes and other autocracies. In addition, given the disagreement about what explains the fragility of military regimes, we still do not know which mechanism is responsible for the observed empirical pattern. Are the differences in regime stability and leader turnover between military and civilian dictatorships simply a product of whether the leader wears, or once wore, a military uniform or are other factors driving this relationship? It is also unclear whether common assumptions used to explain the fragility of military rule, such as the lack of political parties or the improvement of post-tenure fates following democratization, hold for both collegial and personalist military regimes and leaders. We systematically examine the relationship between military regimes and regime/leadership instability by identifying and evaluating as many observable implications of existing theories as possible while also distinguishing between collegial and personalist military rule. Our empirical analyses reveal several important qualifications of the existing literature. First, we find that regime and leader insecurity characterize only collegial military regimes. As such, not all military-led dictatorships are more unstable than civilian dictatorships. Second, collegial military regimes democratize more quickly than others, but do not experience more frequent transitions to alternative autocratic regimes. Third, leaders of collegial military regimes experience more regular and irregular turnovers than those of other autocracies. Fourth, collegial military leaders are also more prone to coups than other autocratic leaders, but these coups are primarily aimed at reshuffling leaders rather than changing regimes. These findings demonstrate the importance of distinguishing between collegial and personalist military regimes in both theories and empirical analyses of regime breakdown, leader survival, and democratization. They also provide empirical support for all of the observable implications of Geddes’s (2003) theory. While we acknowledge the difficulties of directly testing Geddes’s argument about military officers’ preferences for unity, we find only qualified support for the predictions of alternative theories stressing the importance of political institutions and post-tenure fates. The predictions of these alternative theories are only consistent with our findings when applied to collegial military regimes, contrary to their 2 original application to all military regimes. Our results also emphasize the need for scholars to carefully match measures of military rule to the theories they are testing. Furthermore, scholars should carefully consider whether alternative measures of military regimes are truly substitutes when conducting robustness checks. This study contributes to the literature on military regimes by providing the first empirical analysis that systematically analyzes the three main theories of military regime fragility. It also extends the work of Geddes, Frantz, and Wright (2014) and Geddes, Wright, and Frantz (2014) by differentiating between two types of military rule and linking them to different forms of regime and leader turnover. Our findings confirm the importance of distinguishing between collegial and personalist military regimes when studying the relationship between military regimes and regime/leadership insecurity. Two Types of Military Rule Many studies of authoritarian regimes identify military dictatorships as a distinct sub-type of authoritarianism. Existing scholarship has found systematic differences between military and civilian dictatorships across a range of important political outcomes. The distinctive features of military dictatorship can be summarized as follows. First, military dictatorships are led by individuals who specialize in the use of force. A higher capacity for violence makes military dictatorships better equipped to use repression in response to popular dissent (Escribà-Folch & Wright, 2010). However, the higher coercive capacity of military regimes does not translate into more durable regimes and longer leader tenure. Both military regimes and their leaders have the shortest life span of all authoritarian regime types and leaders (Gandhi, 2008). Although the tenure of military regimes and leaders is more likely to end in negotiated democratic transitions than other dictatorships (Geddes, 2003), military dictators also face a high risk of being violently ousted by other officers, and suffer severe post-tenure punishment such as imprisonment and death (Debs, 2016; Debs & Goemans, 2010). While important differences have been found between military and civilian dictatorships, 3 recent research by Geddes, Frantz, and Wright (2014) emphasizes that there are two distinct forms of military rule: collegial military rule and personalist military rule. These regimes differ in both the decision-making procedures and the make-up and influence of the leadership group, even though they are both ruled by current or former members of the military. Rule by the military institution implies that political power resides within a governing junta composed of key members of the armed forces. The junta institutes mechanisms to prevent the leader from concentrating power individually, and the leader consults with a group of officers representing the junta when making important decisions (Geddes, Frantz, & Wright, 2014). Examples include the regime in Brazil after the military coup in 1964, the three military regimes in Pakistan since 1958, the 50 year military rule in Myanmar, and the regime in Greece after the military coup in 1967. By contrast, personalist military rule takes the form of “one-man rule,” characterized by uncontested control of executive powers. A single military officer controls the decision-making structure and personalizes the security apparatus, while other officers have little influence on important political decisions. Mobutu Sese Seko’s regime (1960-1997) in Zaire, Rafael Trujillo’s regime (1930-1961) in the Dominican Republic, and the Gaddafi regime (1969-2011) in Libya are notable examples of personalist military rule. As Geddes, Frantz, and Wright (2014) emphasize and we show below, the distinction between collegial and personalist military rule has important theoretical and empirical implications for research on the survival of military regimes and leaders. However, the definitions and operationalizations of military rule in existing studies are not always compatible. This is the case, for instance, when scholars theoretically discuss collegial military rule but operationalize military regimes broadly to include collegial and personalist military regimes.1 In other cases, definitions and theories of military rule are ambiguous as to whether military rule should be operationalized broadly as military-led dictatorship or as two distinct types, 1 Following Geddes, Frantz, and Wright (2014), we describe broad operationalizations of military rule that focus on the current or former identity of the leader without considering the regime structure as military-led dictatorships. 4 making operationalization and theory testing difficult. Thus, while military regimes and leaders appear to experience distinct patterns of survival, theoretical ambiguity on the types of military rule and the use of incompatible measures of military rule make it difficult to distinguish among competing explanations. Gandhi (2008) provides an example of the theoretical and empirical ambiguity described above. Empirically, Gandhi codes regimes as military-led dictatorships when “[t]he effective head of the government...is or was a member of the institutionalized military prior to taking power” (p. 32). However, as Geddes, Frantz, and Wright (2014) point out, her discussion of military led-dictatorships appears to suggest a particular type, the collegial military regime. Gandhi states that “[w]hat constitutes military rule is the fact that the armed forces are the institution through which rulers govern” (p. 25). This assumes that military governance is carried out through a junta where power is shared among a group of high ranking military officers, an assumption that does not apply to personalist military regimes (Geddes, Frantz, & Wright, 2014). Gandhi also justifies the inclusion of a dummy variable for military leaders in analyses of leader survival based on the work of Geddes (2003) who claims that military leaders voluntarily leave office to preserve military unity. Yet, Geddes’s argument and Gandhi’s measure of military dictatorship are not compatible since Geddes’s argument only applies to collegial military regimes. This ambiguity has been transferred to research investigating the effects of military regimes on other outcomes. For example, Steinberg and Malhotra (2014) examine the effect of authoritarian regime type on exchange rate policy. According to their explanation, tenure security promotes undervalued exchange rates, something that can be economically beneficial in the long run but painful in the short run. They claim that military-led regimes are less likely to maintain undervalued exchange rates than civilian autocracies or monarchies because military rulers are less secure in power, particularly during hard economic times. However, this assumes that both collegial and personalist military leaders are plagued by job insecurity, something that has not yet been verified empirically. Steinberg and Malhotra’s robustness 5 checks are also problematic. Their main empirical specifications rely on the measure of military-led dictatorships from Cheibub, Gandhi, and Vreeland (2010). As a robustness check, they use the alternative measure of collegial military regimes from Geddes, Wright, and Frantz (2014). Yet, as Geddes, Frantz, and Wright (2014) note, the Cheibub et al. (2010) and Geddes, Wright, and Frantz (2014) measures capture different conceptions of military rule that often cannot be used to test the same theoretical mechanisms.2 Of course, the distinction between collegial and personalist military regimes may not be important when studying the relationship between military rule and regime/leader survival. Both collegial and personalist military regimes may be more fragile than their civilian counterparts. Furthermore, several scholars disagree with the coding of personalist dictatorships, such as personalist military regimes, as distinct types of authoritarianism (Magaloni, 2008; Wahman, Teorell, & Hadenius, 2013). This disagreement leads some to treat personalism “as a continuous trait that may be more or less present in a regime” (Hadenius & Teorell, 2007, p. 145). Thus, it may be unnecessary to distinguish between collegial and personalist military regimes. Therefore, comprehensive empirical analyses are needed to determine whether the differences between the two types of military rule are, indeed, relevant to the study of regime and leader survival. If both collegial and personalist military rule are more fragile than civilian rule, differentiating between them is not important when studying regime/leader survival or democratization. Yet, if Geddes, Frantz, and Wright (2014) are correct, this distinction is very important. Determining whether both types of military rule are fragile will also help further our theoretical understanding of the precise characteristics that produce leader and regime fragility. 2 Steinberg and Malhotra (2014) also argue that military autocracies are more likely to maintain fixed exchange rates than civilian autocracies because “a military junta typically depends on a handful of colonels or generals to form the selectorate” (p. 408). Using the Cheibub et al. (2010) measure of military-led dictatorship is not well-suited to testing their argument. 6 Three Explanations of Leader and Regime Instability We reexamine the survival of military regimes and leaders by focusing on three specific theories of fragility found in the existing literature. These theories explain survival through either a preference held by military officers for unity within the military institution, the presence of political parties and infrastructural power, the regime’s institutional control over society and political opponents, or the post-tenure fates of leaders. We systematically review each argument and attempt to generate testable hypotheses, with a particular focus on whether each theory speaks to both types, or only one type, of military rule. The first theory explaining the fragility of military rule focuses strictly on collegial military regimes, defined as those where “a group of officers decides who will rule and exercises some influence on policy” (Geddes, 2003, p. 51). Building on the previous studies (Decalo, 1976; Finer, 2002; Janowitz, 1977; Nordlinger, 1977; Stepan, 1971), Geddes (2003) argues that military officers value the survival and efficacy of the military over anything else and thus emphasize the preservation of unity and capacity within the military institution. When political crises threaten to fracture the officer corps, one group is likely to prefer a return to the barracks. When this group of officers makes a credible first move returning to the barracks, most other officers acquiesce to their decision.3 As Geddes (2003) succinctly states, “military regimes thus contain the seeds of their own destruction” (p. 131). However, Geddes, Frantz, and Wright (2014) note that returning to the barracks is only possible in collegial military regimes where junta members are able to constrain leader behavior. Personalist military regimes are different in that leaders consolidate power individually. Compared to collegial military regimes, Geddes expects personalist regimes to be less vulnerable to elite splits as their leaders have a greater ability to monitor and undermine threatening individuals. Ruling elites in a personalist regime tend to cling to the regime since they are likely to lose all benefits when the regime collapses. Accordingly, personalist regimes are relatively safe 3 Geddes (2003) use a coordination game to illustrate this argument: both majority and minority factions prefer to either stay in power together or return to the barracks together. 7 from internal splits except when economic crisis disrupts patronage networks. Thus, Geddes expects collegial military regimes to be more fragile than personalist military regimes. Additionally, the military unity argument can also be expected to produce different modes of regime transition. As Geddes (2003) notes, transitions from collegial military regimes are much more likely to result in democratization than transitions from personalist military regimes. This is attributed to the ability of junta members to negotiate transitions back to the barracks that protect their military careers and associated perks (see also Huntington (1991) and Nordlinger (1977)). When they decide to extricate themselves from power, junta members tend to negotiate their exit to arrange an orderly transition that protects their interests in the post-transition period. Members of outgoing collegial military regimes are also unlikely to compete in transitional elections, reducing opposition harassment and encouraging democratic transitions (Geddes, Frantz, & Wright, 2014; Ofosu, 2013). Conversely, transitions from personalist rule tend to be initiated using violence and coercion and result in authoritarian transitions where those ousting the regime consolidate power for themselves (Geddes, 2003). Therefore, it is more likely that the breakdown of collegial military regimes will be initiated without coercion and more likely to result in democratization when compared to the breakdown of personalist military regimes. The expectation of short regime tenure following the theory of Geddes (2003) also leads to the hypothesis that leaders of collegial military regimes have shorter tenures than personalist military leaders. However, the end of a leader’s tenure does not always cause the breakdown of the existing regime (Geddes, Wright, & Frantz, 2014). As Geddes (2003) finds, coups against leaders of collegial military regimes are common and are primarily used as a mechanism to maintain power-sharing by reshuffling leaders without changing the regime. In this way, coups in collegial military regimes are “the analogue of votes of no confidence in parliamentary systems” (Geddes, 2003, p. 66). When a faction of officers removes the incumbent leader, other officers tend to go along with this decision. This provides another reason to expect leaders of collegial military regimes to have shorter tenures than personalist 8 military leaders and also suggests different modes of exit for each type of military leader. Additionally, military officers in collegial military regimes develop power-sharing agreements of rotating leadership in order to deter the personalization of power and preserve military unity. Leaders of collegial military regimes should be vulnerable to both regular and irregular exits, with irregular exits taking the form of reshuffling, rather than regime changing, coups (see Aksoy, Carter, and Wright (2015)). Another prominent theory explaining the fragility of military rule focuses on the presence of political parties and infrastructural power. Party-based explanations claim that military regimes are fragile because they lack the institutional capacity to neutralize threats from regime opponents without resorting to outright repression. Political parties provide a mechanism to co-opt potential opponents by offering the opportunity for career advancement and access to state resources in exchange for party membership (Gandhi, 2008; Gandhi & Przeworski, 2007).4 Political parties with strong institutions at the local, regional, and national levels allow incumbents to mobilize party members and leverage the party’s ideological resources in support of the regime during difficult political and economic times (Lai & Slater, 2006; Slater, 2003). From this perspective, military-led regimes, lacking strong party institutions and ideological resources, are less able to counter dissent through non-violent means and often resort to overt repression when facing political and economic challenges. The lack of a politically and ideologically dominant party in military-led regimes is also argued to increase the likelihood of splits among ruling elites (Magaloni, 2008; Svolik, 2012), making leaders more vulnerable to internal challenges. Magaloni (2008) claims that authoritarian political parties allow leaders to make credible power-sharing commitments with elites by delegating the ability to fill political offices to the party. Party-based regimes thus encourage “sunk political investments,” whereby junior party members provide greater party service in exchange for fewer rewards in the hopes that they will be promoted to 4 Gandhi and Przeworski (2007) and Gandhi (2008) also argue that authoritarian legislatures as well as political parties broaden support for the regime and lengthen its rule. According to their explanation, however, political institutions are a best response to the dictator’s environment. Thus, we may not observe a positive correlation between institutions and tenure. 9 senior positions which offer greater rewards for less party service (Svolik, 2012). This is not to say that authoritarian leaders cannot renege on their power-sharing commitments. However, party-backed leaders have incentives to maintain power-sharing commitments to avoid “conspiracies, military coups, and violent rebellions” (Magaloni, 2008, p. 716). Magaloni claims that military dictators are unable to offer the same level of power-sharing because their dominant position within the military gives them substantial powers to select individuals into positions of political power. Furthermore, the reliance of military dictators on repression as a method of authoritarian control creates problems of moral hazard by empowering the state’s repressive agents (Svolik, 2012). Hence, military dictatorships are less capable of solving commitment problems, making military leaders more vulnerable to elite splits and coups. Taken together, these theories predict that military-led dictatorships are more fragile than civilian dictatorships because they lack strong party institutions.5 Military-led dictatorships are less able to quell threats from inside and outside the regime. Accordingly, they are expected to be more vulnerable to regime breakdown than party-based regimes. Military dictators should also be short-lived in office, facing greater threats from revolts and coups. In contrast to Geddes (2003), party-based arguments do not specify whether the breakdown of military-led dictatorships is more likely to produce democratization or whether coups in military-led dictatorships will be regime changing or reshuffling. They only expect a higher likelihood of regime breakdown through coerced means. Scholars using party-based theories to explain fragility define military rule in several ways. While Geddes classifies military rule based on the locus of decision making authority, Slater’s (2003) view of military rule shifts the focus to infrastructural power. Slater classifies regimes as either party-backed or military-backed based on the organization that executes the leader’s orders. Magaloni (2008) also stresses the importance of political parties for autocratic survival, but defines military regimes as those where “the head of the executive is an 5 However, Wright and Escribà-Folch (2012) propose a dissenting view about the effect of authoritarian parties on regime survival and democratization. They argue that authoritarian parties reduce collective action costs for regime opponents and protect the interests of authoritarian elites in a subsequent democracy. Accordingly, they destabilize authoritarian regimes and increase the likelihood of democratization. 10 active member of the military” (p. 732). According to Magaloni, what distinguishes military dictatorship from party dictatorship is the organization that controls access to political office and the main power positions. As a result, military regimes that create political parties are not classified as party autocracies because military dictators share power with their ruling coalition through the military institution (Magaloni, 2008). Last, as explained above, Gandhi (2008) adopts the most inclusive definition by classifying regimes as military dictatorships when the effective head of government is or ever was a member of the military. Despite disagreements over the definition of military regimes, party-based theories do not emphasize the difference between collegial and personalist military rule. Magaloni (2008, p. 740) does consider this distinction in a footnote, but, contrary to Geddes (2003), predicts that collegial military regimes will be more stable than personalist military regimes. Thus, party-based theories provide few reasons to expect differences in regime transitions, leader survival, or coups between the two types of military rule. Moreover, where two types of military regimes are discussed by scholars taking this approach, collegial military regimes and leaders are predicted to be more stable. We assess assumptions in party-based theories that military dictatorships tend to lack political parties when compared to civilian dictatorships. The left panel of Figure 1 explores the percentage of country-years having political parties associated with an authoritarian leader, using data from Svolik (2012). It is clear that collegial military (Junta) regimes are less likely to have political parties associated with the leader than other types of authoritarian regimes, including personalist military (Strongman) regimes.6 However, there is nearly an equal probability that military personalist (Strongman) and civilian personalist (Boss) dictatorships have a political party associated with the leader.7 We also examine the relationship between 6 We follow the previous studies’ labels (Lai & Slater, 2006; Weeks, 2012). Collegial civilian regimes are called Machine. 7 This result is consistent with the finding that after the seizure of power, military dictators from less professionalized militarizes created a civilian support party or allied themselves with a pre-existing party about 60% of the time (Geddes, Frantz, & Wright, 2014, 151). According to Geddes (2011), this high percentage of strongmen with a civilian support party may be the result of these military leaders’ strategic choices. Leaders from less professionalized militaries are more insecure in tenure, because credible power-sharing agreements are more difficult to enforce. Thus, these leaders are more likely to create a mass-based civilian party or 11 Political parties associated with leader 1.00 ● ● ● 0.75 ● 0.75 ● Percent Percent Elected legislatures 0.50 ● 0.50 ● ● 0.25 0.25 0.00 ● Monarchy Junta Strongman Boss ● 0.00 Machine Monarchy Junta Strongman Boss Machine Figure 1. Political parties and legislatures by autocratic regime type, 1946 – 2008. Dots show the mean values, and vertical line segments associated with dots show the 95% confidence intervals. the existence of an elected legislature and authoritarian regime type, given that Gandhi and Przeworski (2007) and Gandhi (2008) emphasizes the role of legislatures in coopting potential opposition members. The right panel of Figure 1 shows that Junta regimes are less likely to establish elected legislatures than other autocracies, with the exception of monarchies. Also, more civilian personalist dictatorships rule with elected legislatures than military personalist dictatorships. Taken together, Figure 1 demonstrates the importance of differentiating between collegial and personalist military regimes. If ruling parties enhance regime stability, only collegial military regimes should be less stable than other autocratic regimes, with the exception of monarchies. Additionally, insofar as elected legislatures contribute to regime stability, the greater propensity of personalist civilian regimes to rule with elected legislatures relative to personalist military regimes should result in greater regime stability. Finally, the fragility of military rule has also been explained by the post-tenure fates of ally themselves with a pre-existing party to counterbalance military forces. However, we cannot reject the possibility that the role or influence of political parties may be different across regimes: political parties may be more influential and effective under civilian personalist rule than under military strongman rule. Given the lack of data on details of authoritarian regime parties, we are not able to verify this point. 12 regime elites. Geddes (2003, p. 63) states that all but the highest ranking regime officers in collegial military regimes have incentives to voluntarily leave office because doing so allows them to continue their military careers with few consequences. However, she states that personalist leaders are less likely to initiate transitions because they are likely to lose their privileged positions in society, and possibly lose their lives. Accordingly, personalist military leaders are more likely to resist negotiating transitions to democracy with domestic opponents than are leaders of collegial military regimes (Geddes, Wright, & Frantz, 2014).8 More recent scholarship has extended Geddes’s argument by claiming that even military regime leaders, not just a large portion of the officer corps, can improve their post-tenure fates by voluntarily leaving office. However, these explanations focus on military-led dictatorships broadly. Using the Cheibub et al. (2010) measure of military-led dictatorship, Debs and Goemans (2010) find that, when compared with other dictators, military dictators face a higher likelihood of being exiled, imprisoned, or killed when exiting office. However, military dictators face substantially better post-tenure fates when their exit occurs through a negotiated democratic transition (Debs, 2016). Debs (2016) explains that democratic transitions solve a commitment problem between former leaders and their successors. Military leaders, as specialists in violence, pose a threat to their successors. This is particularly the case with autocratic successors who can improve their survival prospects by eliminating other specialists in violence. Democratic successors, however, are expected to have weaker incentives to harshly punish military leaders for two reasons. First, democratic transitions favor politicians who can win elections, not those with a high capacity for violence. This means that democratic leaders often lack the capacity to punish outgoing military leaders. Second, the protection of political rights in democracies means that democratic leaders expect substantial opposition to emerge following any non-constitutional military intervention in politics. This reduces the threat of future military intervention and decreases the incentive to exile, imprison, or kill outgoing military leaders during democratic transitions. Military leaders are thus expected 8 Wright and Escribà-Folch (2012) also argue that authoritarian elites are less reluctant to accept democratization when they believe they can protect their interests in a subsequent democracy. 13 to democratize quickly to secure the best possible post-tenure fate. By extension, democratic transitions from military rule can also be expected to take place without coercion. Although Debs (2016) does not distinguish between collegial and personalist military leaders, his argument implies that both types of leaders improve their post-tenure fates through democratization. If this assumption is correct for both collegial and personalist military leaders, we should not expect the distinction between the two types of military rule to be consequential for regime or leader survival. However, if the two types of leaders and regimes face different prospects following democratization, as Geddes (2003) suggests, the type facing the most improved status following democratization should be the most short-lived. Unlike the argument of Geddes, no clear hypotheses can be deduced from this argument on the likelihood or type of coups against the two types of leaders. We empirically analyze the post-tenure fates of authoritarian leaders after making the additional distinction between collegial personalist military leaders. Figure A1 in the Supporting Appendix shows that the distinction between collegial and personalist leaders produces a different picture. As the top panel of the figure illustrates, Strongman leaders of personalist military dictatorships tend to have worse post-tenure fates that leaders of other authoritarian regimes. Conversely, more than the majority of Junta leaders avoid harsh post-tenure punishments. We further differentiate post-tenure fates following transitions to another autocratic regime from post-tenure fates following democratic transitions since Debs (2016) emphasizes post-tenure fates after democratization. Following autocratic transitions (middle panel), there is little difference in the post-tenure fates of Junta, Strongman, and Boss leaders. However, when we focus on post-tenure fates following democratic transitions (bottom panel), the percentage of OK fates for Junta leaders jumps substantially. By contrast, both military and civilian personalist leaders see little improvement in their post-tenure fate after democratization, which is consistent with the argument of Geddes, Wright, and Frantz (2014). The personalist dimension seems to play a greater role in determining post-ouster fates than does the military-civilian dimension. 14 Additionally, the number of dictators facing democratization, presented in the bottom panel of Figure A1 challenges Debs’s argument. Of 121 Junta leaders, 79 (65%) of them experienced democratization, while only 9 of 38 Strongman leaders (20%) experienced democratization. This shows that not all types of military dictators democratize quickly. If, as Debs argues, the consequences of losing power loom large in autocrats’ minds, only Junta leaders should democratize quickly. This is not to say that the scholars discussed above have not considered the distinction between collegial and personalist military rule. For example, Debs (2016) discusses Geddes’s coding of collegial military regimes, arguing that this coding is problematic for predictive purposes because it relies on the observed behavior of leaders (see Pepinsky (2014)). Instead, he uses the CGV classification of military-led dictatorships. This measure does have advantages for isolating certain causal processes as it only contains information available prior to the leader’s ascension to power. At the same time, this measure is problematic because it does not distinguish between leaders and political regimes. Leaders and regimes are not conceptually equivalent. When we use Gandhi’s approach, we can only connect leaders’ personal background to regime instability or coup risk. Unlike Gandhi’s argument, a dictators’ military background does not necessarily imply that they rule through the armed forces, as emphasized in Geddes, Wright, and Frantz (2014). When we are interested in how the structure of control over policy, leadership selection, and the security apparatus influences regime/leadership instability, we must use other measures. Furthermore, while the Geddes coding of regimes may reflect the strategic behavior of leaders, this does not preclude using these measures to analyze regime and leadership survival. Any measure of political regimes that considers rules, procedures, and institutions of political power reflects the strategic behavior of political actors, since political regimes are equilibrium outcomes. For example, scholars argue that for democracy to work, it must be a self-enforcing equilibrium in the sense that “incumbents choose to hold regular, competitive elections and comply with the results” (Fearon, 2011, p. 1661; Przeworski, 1991; Przeworski, 2005). The 15 same can be said of the measure of single-party or multiparty autocracies in the Wahman et al. (2013) dataset and Magaloni (2008) dataset in that they reflect key political actors’ strategic behavior. Nevertheless, we revisit these concerns after presenting our findings by providing an assessment of how they might influence our results. For the ease of comparison, Table 1 summarizes the existing theory’s observable implications for military dictatorships’ regime and leadership instability. Regime failure Autocratic breakdown Democratic transitions - Coerced democratization - Non-coerced democratization Autocratic transitions Leadership failure Regular leadership turnover Irregular leadership turnover Coups - Regime-changing coups - Reshuffling coups Scholars Officers’ preferences for unity Party-based Post-tenure fates 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 Slater (2003) Magaloni (2008) Geddes (2003) Debs (2016) 3 Geddes (2003) Table 1. Summary of Existing Arguments. The check mark (3) implies that a theory associates military dictatorships with a greater probability of the variable in the first column. Data We examine the relationship between military regimes and regime and leader failure rates to test competing arguments on the fragility of military regimes. Our unit of analysis is the country-year. Our main sample covers around 3,200 observations on 213 authoritarian regimes between 1952 and 2009, subject to data availability. 16 Measuring Collegial and Personalist Military Regimes To measure authoritarian regimes, we use the regime-type data constructed by Geddes, Wright, and Frantz (2014) (hereafter GWF). The GWF dataset classifies autocracies as military regimes, dominant-party dictatorships, personalist regimes, hybrids of these three pure types, and monarchies. To distinguish between dictatorships, GWF focus on “whether control over policy, leadership selection, and the security apparatus is in the hands of a ruling party (dominant-party dictatorships), a royal family (monarchies), the military (rule by the military institution), or a narrower group centered around an individual dictator (personalist dictatorships)” (p. 318).9 Using this dataset, we differentiate between authoritarian regimes using both personalist and military-civilian dimensions, resulting in the following regime categories: personalist military (Strongman), collegial military (Junta), civilian personalist (Boss), and collegial civilian (Machine) regimes. Emphasizing rule by military institutions, GWF define military regimes as those in which “the dictator consults with other high-ranking officers and can be constrained by them” (p. 152). Military dictatorships in Argentina 1955–1983, Brazil 1964–1985, and Uruguay 1973–1984 belong to this category. We code military regimes identified by GWF as Junta regimes. To measure Strongman, we follow GWF’s recommendation: a country-year qualifies as military strongman rule when it is coded both as personalist by GWF and as military by Cheibub et al. (2010).10 The Cheibub et al. classification of dictatorships depends solely on the identity of the regime leader and does not consider the institutional configuration and composition of the political leadership. It codes a dictatorship as a military dictatorship if the effective leader is a military officer or was a military officer prior to seizing power. Thus, military dictatorships coded by Cheibub, Gandhi, and Vreeland identify military-led autocracies, and the strategy of combining the GWF dataset with the Cheibub et al. dataset 9 For example, they code a regime as a military regime if the proportion of questions regarding military rule answered by “yes” is high and the proportion of questions regarding personalist and party rule answered by “yes” is low. A regime with high scores in multiple categories is coded as a hybrid regime. 10 The Supporting Appendix presents the list of collegial military rule and strongman military rule. 17 allows the identification of personalist military (Strongman) rule, a subset of military-led autocracies. Idi Amin in Uganda (1971–1979), Yahya Jammeh in Gambia (1994–), Ali Abdullah Saleh in Yemen (1978–2012) are notable examples of Strongman rule. Additionally, we include the civilian counterparts of personalist and collegial military regimes, Boss and Machine regimes. We code the remaining civilian personalist regimes in the GWF data as Boss. Dominant-party dictatorships are coded as Machine. In addition to these four autocratic regimes, we include Monarchy as a separate regime category. Dependent Variables To test our hypotheses regarding autocratic regime survival, we probe three dependent variables: autocratic regime collapse, leadership removal, and coups. Following Geddes, Wright, and Frantz (2014), a regime is defined as a set of “rules that identify the group from which leaders can come and determine who influences leadership choice and policy” (p. 314). Regime change then occurs when there are significant changes in leadership and the set of formal and informal rules for choosing leaders and policies. As GWF emphasize, this definition has two important implications: 1) the removal of a dictator does not necessarily coincide with regime change, and 2) autocratic breakdown includes autocracy-to-autocracy transitions as well as democratization. To identify the two forms of autocratic breakdown, we again use the GWF’s dataset that provides information on when and how the autocratic regime starts and ends and what the subsequent regime is. We further distinguish between coerced transitions and non-coerced transitions. Coerced transitions refer to regime changes via foreign invasions, coups, uprisings, or ouster by insurgents. Non-coerced transitions imply regime changes through elections or rule changes made by insiders. A second dependent variable is leadership turnover. To obtain the information on leadership turnovers, we rely on the Archigos database (version 4.1) (Goemans, Gleditsch, & Chiozza, 2009), covering the period of 1875–2015. Since we expect that the leaders of Junta regimes are short-lived and reshuffled by coups, we examine both regular and irregular 18 turnover, as coded in the Archigos dataset. A leader turnover is coded as regular when a leader is “removed in accordance with explicit rules or established conventions of his or her particular country” and is coded as irregular when a leader removal is “in contravention of explicit rules and established conventions.” Regular turnovers include leader changes through term limits and defeat in elections, while irregular turnovers include turnovers through coups, revolts, and assassinations. Leadership turnovers do not include deaths from natural causes, suicides, retirement due to illness, or leaders deposed by other states, all of which are right-censored. The last dependent variable is coups, the modal type of irregular leader removal in dictatorships. We examine both all coup attempts and successful coup attempts. The information on coups is taken from the dataset constructed by Powell and Thyne (2011), covering the period of 1950 to the present year. By carefully differentiating coup attempts from other types of anti-regime activities, such as riots, protests, or civil wars, they define coup attempts as “illegal and overt attempts by the military or other elites within the state apparatus to unseat the sitting executive” (p. 252). We further differentiate between reshuffling coups and regime changing coups, relying on the dataset developed by Aksoy et al. (2015). They define reshuffling coups as those “in which the leader atop the regime changes but the group of elite in power remains the same” and regime changing coups as those that change “the group of elite who hold power and the way decisions are made” (p. 428). To identify regime changing coups, they combine the GWF regime data with the coup data from Powell and Thyne (2011). They not only distinguish successful regime changing coup attempts from reshuffling coup attempts but also do so for failed coup attempts. To this end, they assess whether a failed coup attempt would have led to a reshuffling of elite within the regime, or to the establishment of a new regime if it had been successful.11 In the sample of autocratic countries from 1971 to 2006, their dataset identifies 38 successful regime change coups, 77 failed regime changing attempts, 40 successful reshuffling coups, and 73 failed reshuffling attempts. 11 See Aksoy et al. (2015) for details of their codings. 19 Control Variables To control for potential confounding variables, we include a number of control variables based on the existing literature on political liberalization and democratization. All variables are lagged by one year. First, we control for several measures of political instability. We include two dichotomous variables Interstate war and Civil war, which equal 1 if a country has experienced the corresponding war in the previous year and zero otherwise. These variables are taken from the Correlates of War data (Sarkees & Schafer, 2000). We also include a measure of mass political mobilizations using data drawn from Banks and Wilson (2013). We sum the incidents of antigovernment demonstrations, general strikes, and riots that occur in the previous year and log-transform the sum. Second, we control for domestic political conditions that might affect regime and leader survival. More liberalized regimes may be more likely to break down and suffer from coups (e.g., Powell, 2012). We include the electoral democracy index (v2x_polyarchy), ranging from 0 to 1, constructed by the Varieties of Democracy (V-Dem) project (Version 6.2) project (Coppedge et al., 2016). This index is formed by averaging the following indices: clean elections, elections for the chief executive, extensive suffrage, freedom of association, and freedom of expression. These indices capture the political characteristics of the regime that the GWF measures of authoritarian regimes do not address. A higher electoral democracy index is expected to be associated with a greater likelihood of regime change. Next, the level of economic development and economic growth affect autocratic regime and leader survival. If the modernization theory holds true, economic development may generate greater domestic pressure for democracy. On the other hand, short-term economic growth may have the opposite effect on regime changes. We thus include a natural log of real GDP per capita, and the annual percentage change of real GDP per capita. Data on GDP per capita are taken from Penn World Table 7.0 (Heston, Summers, & Aten, 2011). Fourth, global and regional environments may influence autocratic regime instability. 20 For example, international factors such as the Cold War struggle between the United States and the Soviet Union or regional preponderance of democracies may independently affect both military intervention and rivalries. Magaloni and Kricheli (2010) contend that “Cold War conditions were particularly conducive to the emergence of closed dictatorships, and in particular to military dictatorships, which strategically used the communist threat to get funds from international financial institutions and from the United States to repress their opponents” (p. 135). Conversely, regional preponderance of democracies may provide the context for autocratic breakdown. We thus include a dummy variable for the post-Cold War period and the proportion of democratic neighbors.12 Last, we include the natural log of regime duration or leader tenure, ln(t), in order to control for negative duration dependence. Empirical Analysis Autocratic Regime Breakdown We first examine the relationship between military regimes and regime survival. Table 2 presents the results of multinomial logit models in which the dependent variable can take three possible values: autocratic regime survival, transition to another autocratic regime, and democratization. This test allows us to compare the effect of military regimes on two different types of autocratic breakdowns. The survival of the current autocratic regime is set as the baseline category. Model 1 includes two dummy variables, Military dictators and Monarch, and compares them with Civilian dictators, the baseline category. Consistent with existing studies on military dictatorships, Military dictators are more likely to democratize than other types of dictators. However, when we examine the likelihood of transitions to an alternative autocratic regime, the difference between Military dictators and Civilian dictators 12 We define a country’s neighbors as all countries within 1001 km minimum distance, as reported in the cshapes R library. 21 (1) (2) Transition to Transition to Military dictators Monarchy Democracy Autocracy 0.995∗∗ (0.295) -0.893 (1.002) 0.024 (0.292) -0.583 (0.744) 79 78 <0.001 79 Monarchy Protest ln(GDP per capita) GDP growth Neighbor democratic transition Post-Cold War ln(t) Constant H0: Junta=Strongman (p-val.) Transitions N Log-Likelihood 0.45 78 2.710∗ (1.057) 0.513∗ (0.248) 0.215 (0.560) 0.248 (0.183) -0.328∗∗ (0.117) -0.035∗∗ (0.009) 0.504∗ (0.224) -1.519∗∗ (0.357) -0.205 (0.137) -1.610∗ (0.816) Boss Interstate War 0.774∗ (0.334) 0.536+ (0.324) 1.006∗∗ (0.345) -0.327 (0.717) 3.074∗∗ (0.922) 0.477∗ (0.239) 0.228 (0.591) 0.190 (0.192) -0.312∗∗ (0.111) -0.040∗∗ (0.010) 0.507∗ (0.225) -1.717∗∗ (0.371) -0.050 (0.138) -2.471∗∗ (0.724) 3.894∗∗ (1.024) 0.117 (0.293) 0.123 (0.600) 0.756∗∗ (0.127) 0.120 (0.134) -0.032∗ (0.013) 0.441+ (0.233) -0.032 (0.338) -0.025 (0.111) -6.471∗∗ (0.941) Strongman Civil conflict Autocracy 3.050∗∗ (0.379) 0.349 (0.578) 1.559∗∗ (0.385) 0.007 (1.063) 4.092∗∗ (1.012) -0.137 (0.350) 0.197 (0.649) 0.598∗∗ (0.155) -0.068 (0.132) -0.053∗∗ (0.014) 0.437+ (0.247) 0.081 (0.385) 0.566∗∗ (0.164) -7.322∗∗ (1.082) Junta Electoral democracy index Democracy 3111 -661.21 3111 -626.63 Table 2. Multinomial logit model of autocratic regime change , by transition outcomes. All covariates are lagged by one year. Standard errors are clustered by country (reported in parentheses): +p < 0.1, *p < 0.05, **p < 0.01. 22 Democractic Transition Autocractic Transition ● Junta ● ● Strongman ● ● Machine ● ● Boss ● ● Monarchy 0% ● 5% 10% 15% 20% 0% 5% 10% 15% 20% Figure 2. Predicted probabilities of autocratic breakdown across different autocratic regime types. Dots show the point estimates, and vertical line segments associated with dots show the 95% confidence intervals. is close to zero and not significant. This suggests that, contrary to what many existing studies argue, military-led dictatorships are not simply more fragile than civilian-led dictatorships. They are, however, more likely to experience autocratic collapses that result in democratization. Model 2, which sets collegial civilian Machine regimes as the baseline category, shows that not all types of military-led dictatorships are prone to democratization by displaying a significant difference between Junta and Strongman. For ease of interpretation, Figure 2 displays the predicted probabilities of democratic transitions and autocratic transitions across different autocratic regime types.13 Junta regimes are most likely to democratize, and the difference between Junta and other regimes is substantial. For instance, the estimated annual probability of democratization in Junta regimes is 15 percent. Junta regimes are 15 times more likely than Machine regimes (1 percent) and 10 times more likely than Strongman regimes (1.5 percent) to democratize. In contrast, Strongman regimes are less likely to democratize than their civilian personalist counterpart Boss regimes (1.5 percent versus 3 13 We adopt the observed value approach, setting all the other covariates to the values observed for each observation, and obtain average effects. 23 percent; p-value < 0.05) and are not significantly different from collegial civilian Machine regimes (p-value ≈ 0.38) . This suggests that the difference between military-led and civilianled dictatorships, found in Model 1, is mainly driven by the difference between Junta regimes and other autocratic regimes. When we examine autocracy-to-autocracy transitions, little evidence shows that the fragility of military rule applies to autocratic transitions. Boss regimes are most prone to autocratic collapses that result in transitions to subsequent dictatorship (about 4 percent), and both types of military dictatorships (Junta: 3.2 percent and Strongman: 3 percent) are more likely to experience autocratic transitions than Machine regimes (2.7 percent). Thus, we cannot conclude that military dictatorships are more likely to be replaced by another autocratic regime than civilian dictatorships. Additionally, no significant difference between Junta and Strongman regimes is found. These results are more consistent with Geddes’s argument than with other arguments that aggregate two types of military autocracies. No significant evidence suggests that personalist military dictatorships are less durable than personalist civilian dictatorships. Indeed, we find evidence that suggests the opposite, supporting the claim of Geddes, Frantz, and Wright (2014) that simply focusing on military dictators conceals substantial variation between the two forms of military rule. Furthermore, the fragility of collegial military regimes in comparison to other autocracies is only the result of a higher likelihood of democratic, not autocratic, transitions. This finding supports the predictions of Geddes and challenges the predictions of party-based theories that military dictatorships are generally fragile. The propensity for democratic transitions under collegial military rule is also more consistent with Geddes’s argument that military officers return to the barracks when faced with political crises that threaten to undermine unity within the military institution. We further differentiate between coerced and not-coerced breakdowns in Table 3. Differentiating between transition outcomes and transition modes generates four kinds of autocratic regime breakdowns: coerced democratization, non-coerced democratization, coerced auto- 24 Transition to Transition mode Junta Strongman Boss Monarchy Electoral democracy index Civil conflict Interstate War Protest ln(GDP per capita) GDP growth Neighbor democratic transition Post-Cold War ln(t) Constant H0: Junta=Strongman (p-val.) Transitions N Log-Likelihood Democracy Non-coerced Democracy Coerced Autocracy Non-coerced Autocracy Coerced 3.389∗∗ (0.483) -0.241 (0.835) 1.040+ (0.577) -0.193 (1.105) 5.066∗∗ (1.269) -0.531 (0.438) -0.527 (1.122) 0.519∗∗ (0.201) -0.029 (0.170) -0.050∗∗ (0.015) 0.505+ (0.289) 0.342 (0.465) 0.551∗∗ (0.204) -8.337∗∗ (1.316) 2.522∗∗ (0.680) 1.092 (0.824) 2.312∗∗ (0.607) 0.332 (1.148) 1.616 (1.560) 0.325 (0.447) 0.870 (0.753) 0.760∗∗ (0.214) -0.179 (0.182) -0.055∗ (0.028) 0.299 (0.367) -0.213 (0.589) 0.566+ (0.290) -7.117∗∗ (1.509) 1.032 (0.834) -0.198 (1.314) -13.356∗∗ (0.831) -12.965∗∗ (0.756) 0.016 (2.828) 1.082 (0.926) -13.604∗∗ (0.682) -0.422 (0.471) 0.288 (0.318) -0.004 (0.038) -0.061 (0.545) -0.360 (0.956) -0.390 (0.344) -6.901∗∗ (2.285) 0.642 (0.415) 0.636+ (0.336) 1.212∗∗ (0.341) -0.145 (0.730) 3.515∗∗ (0.962) 0.366 (0.295) 0.470 (0.576) 0.278 (0.199) -0.409∗∗ (0.122) -0.044∗∗ (0.010) 0.575∗ (0.248) -1.908∗∗ (0.416) -0.004 (0.161) -2.250∗∗ (0.814) <0.001 53 0.03 26 0.21 10 0.99 68 3111 -687.59 Table 3. Multinomial logit model of autocratic regime change, by transition outcomes and modes. All covariates are lagged by one year. Standard errors are clustered by country (reported in parentheses): +p < 0.1, *p < 0.05, **p < 0.01. cratic transition, and non-coerced autocratic transition. Table 3 estimates multinomial logit regressions using these four types of transitions as possible outcomes. Again supporting Geddes’s argument, Junta regimes dominate other autocracies in the probability of non-coerced democratization, and its differences from other autocracies are statistically significant at the 25 25% 15% ● ● 10% ● ● Failure rate Failure rate 20% 15% ● 10% ● ● 5% ● 5% 0% 0% Military Civilian Monarch Junta Strongman Machine Boss Monarchy Figure 3. Autocrat’s failure rate by autocratic regime type, 1946 – 2010 1% level. Meanwhile, Junta regimes are more likely than collegial civilian Machine regimes to democratize via coerced means such as coups, uprisings, and ouster by insurgents but are nearly equally likely to engage in coerced democratization as civilian personalist Boss regimes. Next, the examination of autocratic transitions reveals a different pattern. Junta regimes are not statistically different from Machine and Strongman regimes. Last, little evidence suggests that Strongman regimes are fragile, regardless of how we measure autocratic transitions.14 Leader removal Given the difference between regime and leadership survival, we examine the relationship between autocratic regime types and leadership failure. We first take a descriptive look at data. The left panel of Figure 3 plots leadership failure rates per year by autocratic regime type. Military dictators tend to lose power at a higher annual rate than their civilian counterparts. Yet a different pattern emerges when we differentiate authoritarian regimes using the personalist dimension. As shown in the right panel of Figure 3, leaders of Junta regimes are most insecure in office and military strongmen stay in power slightly longer on average than civilian bosses. This suggests that the difference in tenure between military and civilian leaders is largely attributed to Junta leaders. 14 There are only 10 non-coerced autocratic transitions, and Boss and Monarchy experienced no non-coerced autocratic transitions. This explains the large, negative coefficients on Boss and Monarchy. 26 Table 4, presenting logit estimates of leadership turnover, shows that this finding holds when we control for other confounding variables.15 Model 1 shows that leadership failure rate is higher for military dictators than for other types of dictators. However, when Model 2 distinguishes between Junta and Strongman leaders, leaders of Junta regimes have the greatest risk of removal, and failure rates of Strongman leaders are not statistically different from those of civilian leaders. Simply focusing on leadership failure does not directly test Geddes’s argument that military leaders in Junta regimes are vulnerable to both regular and irregular exits and military personalist leaders are not. We thus differentiate between regular and irregular exits: Models 3 and 4 examine only regular leadership turnovers including those through term limits and electoral defeats, and Models 5 and 6 consider only irregular leadership turnovers including those through coups, revolts, and assassinations. Figure 4 illustrates the predicted probabilities of each type of leader failure across different regime types. The estimated annual probability that a Junta regime experiences a regular leadership turnover is 8.4 percent, which is substantially greater than those for other authoritarian regimes (the differences being statistically significant at 5% level). Nonpersonalist civilian Machine regimes also tend to have more frequent regular leadership changes than both personalist regimes (4.2 vs. 2.2 percent; p-values < 0.001). When we examine irregular exits, Junta leaders are most prone (8.5 percent), although their difference from civilian personalist Boss leaders is not statistically significant at any conventional level. Machine leaders are least prone to irregular exits (2.4 percent), showing the importance of political parties in establishing credible power-sharing deals between a ruler and his ruling coalition. Last, Strongman leaders face a lower risk of irregular leader turnover than Boss leaders, their personalist civilian counterparts, but the difference between the two personalist regimes is not statistically different from zero (4.3 percent vs. 7.2 percent; p-value of 0.22). Taken together, the results show that leadership instability characterizes only collegial 15 We also reestimate all the models of Table 4 by using the full sample including democracies. Results remain similar. 27 Leader exit (1) Military dictators Monarchy 0.677∗∗ (0.165) 0.123 (0.485) 3.192∗∗ (0.680) 0.025∗∗ (0.007) -0.509 (0.388) 0.733∗∗ (0.188) 0.294∗∗ (0.086) -0.027 (0.072) -0.006 (0.008) -0.797∗∗ (0.198) 0.284∗ (0.122) -0.406∗∗ (0.088) -4.070∗∗ (0.588) 294 3171 -899.01 0.00 294 3171 -891.91 Strongman Boss Monarchy Age Interstate war Civil war Protest ln(GDP per capita) GDP growth Post-Cold War Neighbor democratic transition ln(Tenure) Constant H0: Junta=Strongman (p-val.) Turnovers N Log-Likelihood Irregular exit (3) (5) (4) 0.629∗∗ (0.230) 0.214 (0.847) 1.015∗∗ (0.209) 0.270 (0.258) 0.333+ (0.191) 0.124 (0.518) 2.784∗∗ (0.601) 0.022∗∗ (0.008) -0.467 (0.399) 0.554∗∗ (0.173) 0.265∗∗ (0.084) -0.066 (0.071) -0.007 (0.008) -0.742∗∗ (0.194) 0.305∗ (0.124) -0.327∗∗ (0.095) -3.659∗∗ (0.644) Junta Electoral democracy index (2) Regular exit (6) 0.654∗∗ (0.215) -0.011 (0.374) 3.814∗∗ (0.738) 0.037∗∗ (0.010) -0.270 (0.559) 0.684∗∗ (0.243) 0.391∗∗ (0.118) 0.190 (0.116) 0.024∗ (0.011) -0.218 (0.226) 0.260 (0.161) -0.490∗∗ (0.115) -7.470∗∗ (0.823) 0.673∗∗ (0.261) -0.708+ (0.392) -0.676+ (0.356) 0.061 (0.844) 3.277∗∗ (0.676) 0.031∗∗ (0.010) -0.234 (0.540) 0.538∗∗ (0.206) 0.387∗∗ (0.109) 0.109 (0.113) 0.021+ (0.012) -0.023 (0.227) 0.263 (0.165) -0.401∗∗ (0.124) -6.301∗∗ (0.825) 144 3171 -521.53 0.00 144 3171 -515.85 1.868∗ (0.859) 0.014 (0.010) -1.306∗ (0.606) 0.608∗∗ (0.229) 0.201 (0.139) -0.275∗∗ (0.095) -0.022∗ (0.010) -1.493∗∗ (0.278) 0.378∗ (0.155) -0.297∗∗ (0.093) -2.140∗∗ (0.736) 1.360∗∗ (0.276) 0.827∗ (0.325) 1.232∗∗ (0.246) 0.203 (0.401) 1.585∗ (0.793) 0.012 (0.011) -1.284∗ (0.612) 0.426+ (0.226) 0.145 (0.142) -0.264∗∗ (0.086) -0.024∗ (0.011) -1.609∗∗ (0.279) 0.425∗∗ (0.158) -0.202∗ (0.098) -2.477∗∗ (0.808) 145 3171 -542.63 0.09 145 3171 -527.61 Table 4. Logit model of leader exit in autocratic regimes. All covariates are lagged by one year. Standard errors are clustered by country (reported in parentheses): +p < 0.1, *p < 0.05, **p < 0.01. 28 military regimes, not personalist military regimes. On the other hand, no significant evidence shows that personalist military rulers are more likely to be removed than civilian rulers, regardless of whether we examine regular or irregular removal. This supports Geddes’s argument and challenges alternative explanations regarding the tenure of military rulers. Regular exit Irregular exit ● Junta ● ● Strongman ● ● Machine ● ● Boss ● ● Monarchy 0% 4% ● 8% 12% 0% 4% 8% 12% Figure 4. Predicted probabilities of leader exits across different autocratic regime types. Dots show the point estimates, and vertical line segments associated with dots show the 95% confidence intervals. Coups Last, we probe a specific mode of leadership turnover–coups d’etat. While the modal dictator exits office through a coup d’etat (Svolik, 2012), determining whether collegial and personalist military regimes face different risks of coups is an important step toward broadening our understanding of the fragility of military regimes. Focusing specifically on coups also allows us to assess Geddes’s argument that coups in collegial military regimes are used primarily as a mechanism of leader, and not regime, change. Table 5 reports logit estimates, and Figure 5 presents the substantive impact of autocratic regime types on different types of coup attempts. Models 1 and 2 examine successful coups. Model 1 shows that Military dictators are more likely to experience successful coups than Civilian dictators (Model 29 1), and Model 2 demonstrates that Junta leaders are more likely to be ousted via coups than other autocrats, including Strongman leaders (Model 2). Congruent with the results in Table 4, coup risk for Junta leaders greatly outweighs the coup risk for other autocrats. The same cannot be said of Strongman leaders. However, Models 3 and 4, examining failed coup attempts, display a different pattern. Military dictators are more likely than civilian dictators to experience failed coup attempts, while Junta and Strongman leaders are not significantly different from each other and are more coup-prone than other civilian dictatorships. Remaining models of Table 5 differentiate reshuffling coup attempts from regime-change coup attempts.16 Military-led dictatorships are more likely to experience reshuffling coup attempts than civilian-led dictatorships (Model 5), but this does not hold when we examine regime-change coup attempts (Model 7). When accounting for the personalist dimension, Model 6 demonstrates that Junta leaders are most prone to reshuffling coup attempts, manifesting a substantial difference from other regime types. Conversely, Junta leaders are less prone to regime changing coups than civilian personalist Boss and Monarchy leaders. The results suggest that most coups in Junta regimes just reshuffle leaders without altering the rules of the regime or the leadership group. This supports the argument of Geddes (2003) that collegial military regimes often use coups as a method for reshuffling leaders. This also fits well with the findings reported in Tables 2 and 3 that Junta regimes are less durable but their insecurity manifests only in the likelihood of democratization, but not in the likelihood of transition to another autocracy. Robustness Checks As a robustness check, we use an alternative measure of Junta regimes. Following GWF’s suggestion, our main regime measures include all party-hybrids and oligarchies in the category of dominant-party dictatorships by prioritizing the party dimension, grouping only “personalmilitary” with military regimes and classifying pure “personal” as personalist autocracies. To explore how this coding rule affects estimation results, we use an alternative, broad 16 We also examine only successful reshuffling and regime-change coup attempts. Results remain similar. 30 Successful coups (1) Military dictators Monarchy 2.131∗ (0.939) 0.255 (0.261) -1.421 (1.046) 0.099 (0.174) -0.193+ (0.112) -0.021+ (0.011) 0.333 (0.218) -1.250∗∗ (0.340) -0.182∗∗ (0.062) 0.010∗ (0.004) -0.000∗ (0.000) -1.809∗ (0.780) 100 3114 -403.88 0.03 100 3114 -398.98 Boss Monarchy Interstate War Protest ln(GDP per capita) GDP growth Neighbor democratic transition Post-Cold War t t2 t3 Constant H0: Junta=Strongman (p-val.) Coups N Log-Likelihood (4) 0.602∗ (0.245) 0.322 (0.349) 1.040∗∗ (0.280) 0.406 (0.305) 0.745∗ (0.311) -0.522 (0.657) 1.838∗ (0.897) 0.108 (0.258) -1.407 (1.041) 0.026 (0.167) -0.243∗ (0.104) -0.023∗ (0.011) 0.362+ (0.217) -1.346∗∗ (0.357) -0.170∗∗ (0.064) 0.010∗ (0.004) -0.000∗ (0.000) -1.663∗ (0.757) Strongman Civil war (3) 0.504∗ (0.232) -0.585 (0.636) Junta Electoral democracy index (2) Failed coups Reshuffling (5) (6) 1.705∗∗ (0.462) -0.227 (1.016) 0.697 (1.308) 0.403 (0.325) 0.104 (0.493) 0.264 (0.170) -0.074 (0.097) 0.008 (0.017) 0.370 (0.256) -0.735∗ (0.302) -0.225∗∗ (0.059) 0.010∗∗ (0.004) -0.000∗ (0.000) -2.349∗∗ (0.791) 1.200∗∗ (0.331) 1.075∗∗ (0.398) 0.851∗ (0.376) 0.598 (0.423) 0.456 (1.228) 0.326 (0.307) 0.125 (0.497) 0.190 (0.173) -0.094 (0.094) 0.007 (0.018) 0.382 (0.255) -0.904∗∗ (0.303) -0.216∗∗ (0.060) 0.010∗∗ (0.004) -0.000∗ (0.000) -2.533∗∗ (0.806) 102 3114 -412.93 0.64 102 3114 -405.50 Regime-changing (7) -0.076 (0.255) 0.128 (0.694) 1.034 (1.358) 0.147 (0.400) 0.148 (0.584) 0.305 (0.193) -0.177 (0.126) -0.011 (0.014) -0.067 (0.297) -1.154∗ (0.540) -0.338∗∗ (0.100) 0.018∗∗ (0.007) -0.000+ (0.000) -2.345∗ (1.171) 2.578∗∗ (0.403) 1.234∗ (0.505) 0.381 (0.773) -0.190 (1.010) 0.029 (1.391) -0.287 (0.285) 0.286 (0.618) 0.135 (0.213) -0.305∗ (0.146) -0.016 (0.018) -0.051 (0.299) -0.847+ (0.485) -0.281∗∗ (0.087) 0.015∗∗ (0.006) -0.000∗ (0.000) -1.586 (1.240) 2.399∗∗ (0.916) 0.519+ (0.311) -0.869 (0.994) -0.205 (0.278) -0.441∗∗ (0.138) -0.030∗ (0.012) 0.280 (0.303) -1.018∗ (0.438) -0.345∗∗ (0.122) 0.018+ (0.010) -0.000 (0.000) 0.304 (0.865) 0.094 (0.386) 0.404 (0.352) 1.103∗∗ (0.386) 0.414 (0.703) 2.671∗∗ (0.906) 0.433 (0.354) -0.981 (0.995) -0.230 (0.288) -0.379∗∗ (0.146) -0.029∗ (0.013) 0.292 (0.305) -1.382∗∗ (0.452) -0.352∗∗ (0.124) 0.019+ (0.010) -0.000 (0.000) -0.461 (0.949) 66 2412 -261.61 0.00 66 2412 -245.87 63 2412 -260.75 0.46 63 2412 -256.59 Table 5. Logit model of coup attempts and successful coups in autocratic regimes. All models include a cubic polynomial of years since the last coup. All covariates are lagged by one year. Standard errors are clustered by country (reported in parentheses): +p < 0.1, *p < 0.05, **p < 0.01. measure of Junta regimes by coding all military hybrids, including “party-military” and “party-personal-military” hybrids as Junta. 31 (8) Successful coup ● Junta 0% 10% 0% ● ● ● 5% ● ● ● ● ● ● ● ● Boss Regime−changing ● ● ● Machine Reshuffling ● ● Strongman Monarchy Failed coup ● ● 5% 10% 0% ● 5% 10% 0% 5% 10% Figure 5. Predicted probabilities of coups across different autocratic regime types. Dots show the point estimates, and vertical line segments associated with dots show the 95% confidence intervals. We code residual party-based autocracies as Machine. Using these alternative measures of Junta and Machine regimes, we re-estimate all models of Tables 2 to 5 and report the results in the Supporting Appendix (Tables A2 to A4). Our central findings are not altered by this alternative coding of regimes. We also employ alternative estimators. For each of our dependent variables, we include country random effects to control for unobserved heterogeneity among countries and cluster standard errors at the regime level (reported in Supporting Appendix). Our main findings are robust to these alternative estimators. Discussion and Conclusion Our empirical analyses demonstrate the importance of distinguishing between collegial and personalist military rule when studying regime and leadership survival. We show that the stylized facts of regime and leader instability under military rule are only applicable to collegial military rule. Regime and leadership instability are not pronounced in personalist military regimes. Also, the breakdown of collegial military regimes is more likely to result in 32 democratization, but does not increase the likelihood of transitions to alternative authoritarian regimes. The leaders of collegial military regimes face a higher risk of both regular and irregular leader turnover than other autocrats, and are particularly vulnerable to coups. However, coups against collegial military leaders tend to be reshuffling rather than regime changing coups. Previous findings showing that military-led dictatorships are more unstable than civilian dictatorships are thus driven by collegial, not personalist, military rule. These results provide strong support for Geddes’s predictions about the effect of military regimes on regime survival, democratization, and coups. They are consistent with her argument that preferences for military unity in the officer corps make collegial military regimes more likely to break down and exit office voluntarily when facing popular threats or economic crises than other authoritarian regimes. The leaders of collegial military regimes face higher risks of both regular and irregular exits from office, with irregular exits often coming in the form of coups that reshuffle leaders without toppling the regime. Our findings call into question party-based explanations of the fragility of military rule. Contrary to these explanations, personalist military dictatorships are not more fragile than civilian dictatorships. Additionally, the fact that collegial military dictatorships are not particularly prone to autocratic transitions and that democratic transitions from collegial military rule tend to be voluntary and orderly are not congruent with the party-based arguments. Our results provide mixed support for Debs’s argument which connects military regime and leadership instability to the post-tenure fates of leaders. Distinguishing between collegial and personalist military rulers again proves to be theoretically and empirically important. Although Debs applies his theory to all military rulers, we find that only collegial military rulers see improvements in their post-tenure fates following transitions to democracy. This finding, while questioning Debs’s classification of military rulers, is actually supportive of his theoretical reasoning as more frequent transitions to democracy should only be expected when such transitions improve the leader’s post-tenure fate. 33 One potential concern with our findings is that the GWF coding of collegial military regimes contains information about the leaders’ observed post-seizure behavior, which is a strategic response to the leader’s expectation about regime and leader instability.17 For instance, our empirical results are significantly undermined if the GWF measures of collegial military regimes simply reflect information about regime and leadership instability of collegial military regimes. However, we find few reasons to believe this is the case. The existence of power-sharing and leader constraints in collegial military regimes does not predict that these regimes are short-lived. Even if the existence of power-sharing and leader constraints simply reflect the leader’s political weakness, this can only explain leader instability, not regime instability. In addition, as Magaloni (2008) argues, we can make the opposite argument that power-sharing and leader constraints should result in collegial military regimes having greater leadership and regime stability than personalist military regimes. Last, power-sharing and leader constraints do not explain why collegial military regimes are more likely to experience democratic, but not autocratic, transitions than other dictatorships. Importantly, however, this study does not represent a final judgment of the efficacy of Geddes’s theory over party-based and post-tenure fate theories.18 The empirical analyses only provide support for the empirical implications of Geddes’s theory; they do not directly test officer preferences for military unity. Additional work is needed to better understand the motivations behind voluntary exits from office by military officers. Also, scholars should pay particular attention to how to interpret frequent coups in collegial military regimes. Are they motivated by a desire to maintain unity within the military government and prevent a single leader from concentrating power in his hands? Or are they merely a proxy of internal power struggles? If the latter is true, frequent coups contradict the assumption that military officers emphasize corporate unity within the military institution. Finally, this study has not investigated the internal consistency of the theories tested. We have only tested the empirical implications of existing arguments. Geddes, for example, 17 18 We thank an anonymous reviewer for raising this question. We thank an anonymous reviewer for calling this point to our attention. 34 uses a coordination (battle-of-sexes) game to explain why officers voluntarily return to the barracks when political divisions among officers threaten military unity. Pepinsky (2014) argues that “more theoretical work is required to produce the prediction that military regimes are more likely to break down after internal splits because they play a co-ordination game” (p. 638). This is because the mere existence of the two equilibria, {Intervene, Intervene} and {Barracks, Barracks}, says nothing about the stability of a particular equilibrium, and it is not clear whether internal splits imply a change in the relative preference orderings of the military factions or the absolute difference in the preferences (see Pepinsky (2014) for details). In addition to questions of internal consistency, there are also questions about the development of post-tenure fate theories. For example, Debs’s (2016) theory links the post-tenure fates of military leaders to regime breakdown and democratization. Geddes (2003) provides a similar argument, but only claims that the majority of officers, not the most senior leaders in military regimes, have incentives to return to the barracks and initiate democratic transitions to improve their post-tenure fates. These two arguments, while both discussing post-tenure fates, suggest distinct processes of regime breakdown and democratic transition that have yet to be investigated. 35 References Aksoy, D., Carter, D. B., & Wright, J. (2015). Terrorism and the fate of dictators. World Politics, 67 (03), 423–468. Banks, A. S., & Wilson, K. A. (2013). Cross-national time series data archive. Jerusalem, Israel: Databanks International. Cheibub, J. A., Gandhi, J., & Vreeland, J. R. (2010). Democracy and dictatorship revisited. Public Choice, 143 (1-2), 67–101. Coppedge, M., Gerring, J., Lindberg, S. I., Skaaning, S.-E., Teorell, J., Altman, D., et al. (2016). V-Dem Codebook v6. Varieties of Democracy (V-Dem) Project. Debs, A. (2016). Living by the Sword and Dying by the Sword? Leadership Transitions in and out of Dictatorships. International Studies Quarterly, 60 (1), 73–84. Debs, A., & Goemans, H. (2010). Regime type, the fate of leaders, and war. American Political Science Review , 104 (03), 430–445. Decalo, S. (1976). Coups and army rule in Africa: Studies in military style. New Haven: Yale University Press. Escribà-Folch, A., & Wright, J. (2010). Dealing with tyranny: International sanctions and the survival of authoritarian rulers. International Studies Quarterly, 54 (2), 335–359. Fearon, J. D. (2011). Self-enforcing democracy. Quarterly Journal of Economics, 126 (4), 1661–1708. Finer, S. (2002). The man on horseback: The role of the military in politics. Transaction Publishers. Gandhi, J. (2008). Political institutions under dictatorship. New York: Cambridge University Press. Gandhi, J., & Przeworski, A. (2007). Authoritarian institutions and the survival of autocrats. Comparative Political Studies, 40 (11), 1279–1301. Geddes, B. (2003). Paradigms and sand castles: Theory building and research design in comparative politics. Ann Arbor: University of Michigan Press. 36 Geddes, B. (2011). How the military shapes “democratic” institutions in dictatorships. In How autocracies work conference, University of Michigan. Geddes, B., Frantz, E., & Wright, J. G. (2014). Military rule. Annual Review of Political Science, 17 , 147 -162. Geddes, B., Wright, J., & Frantz, E. (2014). Autocratic breakdown and regime transitions: A new data set. Perspectives on Politics, 12 (02), 313–331. Goemans, H., Gleditsch, K., & Chiozza, G. (2009). Introducing Archigos: A dataset of political leaders. Journal of Peace Research, 46 (2), 269 –283. Hadenius, A., & Teorell, J. (2007). Pathways from authoritarianism. Journal of Democracy, 18 (1), 143-157. Heston, A., Summers, R., & Aten, B. (2011). Penn World Table Version 7.0. Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania. Huntington, S. P. (1991). Democracy’s third wave. Journal of Democracy, 2 (2), 12–34. Janowitz, M. (1977). Military institutions and coercion in the developing nations. Chicago: University of Chicago Press. Lai, B., & Slater, D. (2006). Institutions of the offensive: Domestic sources of dispute initiation in authoritarian regimes, 1950–1992. American Journal of Political Science, 50 (1), 113–126. Magaloni, B. (2008). Credible power-sharing and the longevity of authoritarian rule. Comparative Political Studies, 41 (4/5), 715–741. Magaloni, B., & Kricheli, R. (2010). Political order and one-party rule. Annual Review of Political Science, 13 (1), 123–143. Nordlinger, E. A. (1977). Soldiers in politics: Military coups and governments. Prentice Hall College Div. Ofosu, G. (2013). Explaining variation in the fairness of transitional elections across authoritarian regime types. In Unpublished manuscript, University of California, Los 37 Angeles. Pepinsky, T. (2014). The institutional turn in comparative authoritarianism. British Journal of Political Science, 44 (03), 631–653. Powell, J. (2012). Determinants of the attempting and outcome of coups d’état. Journal of Conflict Resolution, 56 (6), 1017–1040. Powell, J., & Thyne, C. L. (2011). Global instances of coups from 1950 to 2010. Journal of Peace Research, 48 (2), 249 –259. Przeworski, A. (1991). Democracy and the market: Political and economic reforms in Eastern Europe and Latin America. New York: Cambridge University Press. Przeworski, A. (2005). Democracy as an equilibrium. Public Choice, 123 (3-4), 253–273. Sarkees, M. R., & Schafer, P. (2000). The correlates of war data on war: An update to 1997. Conflict Management and Peace Science, 18 (1), 123–144. Slater, D. (2003). Iron cage in an iron fist: Authoritarian institutions and the personalization of power in Malaysia. Comparative Politics, 36 (1), 81–101. Steinberg, D. A., & Malhotra, K. (2014). The effect of authoritarian regime type on exchange rate policy. World Politics, 66 (3), 491–529. Stepan, A. (1971). The military in politics: Changing patterns in Brazil. Princeton: Princeton University Press. Svolik, M. W. (2012). The politics of authoritarian rule. New York: Cambridge University Press. Wahman, M., Teorell, J., & Hadenius, A. (2013). Authoritarian regime types revisited: updated data in comparative perspective. Contemporary Politics, 19 (1), 19–34. Weeks, J. L. (2012). Strongmen and straw men: Authoritarian regimes and the initiation of international conflict. American Political Science Review , 106 (02), 326–347. Wright, J., & Escribà-Folch, A. (2012). Authoritarian institutions and regime survival: Transitions to democracy and subsequent autocracy. British Journal of Political Science, 42 (02), 283–309. 38
© Copyright 2025 Paperzz