In February 1989, a group of military officers presented Sudanese Prime Minister Sadiq Al-Mahdi with an ultimatum: either give the military the means to achieve a battlefield victory against the secessionist regional rebellion in Sudan's south or end the conflict by political settlement. He was unable to do either. Four months later, he was ousted in a bloodless coup led by Brigadier Omar Al-Bashir (NYT 1989). The coup ushered in a military regime which has lasted 30 years and continues, despite Bashir's ouster in April 2019, in the face of popular mass mobilization.
The Sudanese case conforms to the central argument we make in this article. Regional rebellions – where rebels seek separation or autonomy from the political center, not its conquest – are associated with transitions to militarized regimes. Building on Slater (Reference Slater2010) and Slater (Reference Slater2018), this article furnishes a deductively grounded theory for this association between a distinctive type of war – regional rebellions – and one particular type of authoritarian regime: militarized regimes. To explain the link, we emphasize two causal mechanisms: (1) the disalignment of preferences between civilian and military leaders and (2) the military's autonomous operational experience in fighting regional wars.
When subjecting this macro-level association to empirical tests on a global dataset, we find strong empirical support for our argument that regional rebellions stand out among all forms of war in making militarization more likely and more enduring. Regional rebellions double the likelihood of the subsequent emergence of a military regime, and military regimes that experience a regional rebellion during their spell are 45 per cent less likely to democratize. We also provide illustrative qualitative examples of our mechanisms in Asian and African cases of regime militarization. Taken together, this study establishes regional wars as a key causal variable for explaining the emergence and durability of a specific type of authoritarian regime.
We make several theoretical contributions. While the authoritarianism literature has made great strides in explaining the effects of different regime types on various outcomes (for example, Chandra and Rudra Reference Chandra and Rudra2015; Weeks Reference Weeks2012), prior studies have not systematically considered why different types of authoritarian regimes emerge in the first place. The literature on war, conversely, has predominantly sought to explain war occurrence, while war as a producer of political outcomes other than state formation (for example, Thies Reference Thies2004; Tilly Reference Tilly, Evans, Rueschemeyer and Skocpol1985) has been neglected.
This article brings together these disjoint literatures and establishes war as an independent variable for regime type outcomes. The study also refines the literature on civil–military relations and coups, which has tended to lump together domestic conflicts of all kinds as determinants of military interventions, invoking concepts such as ‘mass, organized, and politically violent opposition’ (Svolik Reference Svolik2013, 767), ‘civil conflict’ (Piplani and Talmadge Reference Piplani and Talmadge2016) or ‘instability’ (Leon Reference Leon2014, 382). Our argument conceptually separates armed from unarmed domestic conflicts and, among armed conflicts, argues that only one specific type – regional rebellions – increases the likelihood of militarized regimes. More generally, our conceptual emphasis on militarization – a term we derive from seminal texts on civil–military relations – highlights the possibility of military rule through means other than coups.
In the following sections, we outline how this article relates to the coup literature, define the concept of militarization, and remark upon the surprising dearth of literature connecting types of war to types of political regimes. We then present our theoretical argument, which builds and expands on Slater (Reference Slater2010, Reference Slater2018), using illustrative case evidence to elaborate on the main causal mechanisms. Our quantitative analysis then tests the argument's generalizability at the global level. We conclude with a reflection on the broader relevance of our findings.
Coups and Militarization
Recent studies on coups have made significant strides in explaining militaries’ motivations to seize power, highlighting structural factors such as slow growth or lagging human development, and window-of-opportunity events (for example, weak electoral performance) as determinants (for example, Hiroi and Omori Reference Hiroi and Omori2013; Wig and Rod Reference Wig and Rod2016). Military involvement in politics in this line of inquiry boils down to a singular event – the coup – which is commonly defined as an ‘illegal attempt […] to unseat the sitting executive’ (Powell and Thyne Reference Powell and Thyne2011, 252).
While such a dichotomous conceptualization offers clear measurement advantages (Feaver Reference Feaver1999, 219), ‘coup-ism’ problematically defines the most extreme event as the conceptual benchmark and neglects other ways in which militaries exert political power (Croissant et al. Reference Croissant2010). The absence of a coup is not necessarily a sign of successful civilian control; rather, it may indicate that a politically powerful military has been granted influence and prerogatives ex ante, rendering a coup unnecessary (Feaver Reference Feaver1999, 218). Coups also frequently involve one military faction overthrowing another, rather than the displacement of civilian by military power (Geddes, Frantz and Wright Reference Geddes, Frantz and Wright2014).
In contrast, foundational texts on civil–military relations called attention to the different degrees to which regimes can be militarized, that is, the extent to which the military exerts political power in an ongoing fashion. Finer's (Reference Finer1962) seminal study distinguishes four levels of military involvement, ranging from the mere ‘displacement’ of critical with pliant civilian leaders to the full-blown ‘supplantment’ of civilian leadership through direct military takeover. Similarly, Stepan (Reference Stepan1988) highlights a range of institutionalized prerogatives of Latin American militaries. As Brooks (Reference Brooks2019) reminds us, civilian authority or, in Huntington's (Reference Huntington1957) terms, ‘objective control’ of the military by civilian leaders can be compromised without coups. We recover this concept of militarization here.
Following this literature, militarization means that the military usurps key decision-making power over policies that are unrelated to national security. As a result, militarization necessarily entails an autocratic form of government as unelected military officials decide major non-military policies. How militaries exert power has taken different institutional shapes (for example high-powered national security councils, military juntas, key ministerial portfolios occupied by military officers), and the degree to which militarized autocracies have used political institutions (such as parties and parliaments) and allowed political participation has been equally variable.Footnote 1
To make this empirical variation analytically manageable, we build on Finer's (Reference Finer1962, 4) concepts of ‘displacement’ and ‘supplantment’ and propose a corresponding distinction between two broad variants of militarization: direct and indirect. Militarization is direct when the chief executive is an active-duty officer,Footnote 2 who is often but not necessarily backed up by an institutionalized military decision-making body (for example, a junta); it is indirect when a civilian leads the executive and the military wields political power from the background, for instance through a national security council ‘advising’ the government. Both types can be described as militarized regimes.Footnote 3
Alternative conceptualizations of ‘military-led autocracies’ (for example, Geddes, Frantz, and Wright Reference Geddes, Frantz and Wright2014) do not encompass indirect military rule behind a civilian executive, although such indirect militarization is frequently observed.Footnote 4 For instance, the 1957 constitution of Honduras granted the military expansive powers to influence state spending and policies (Ruhl Reference Ruhl1996). Similarly, the National Security Council in Turkey historically allowed the military to influence civilian governments, and if necessary force them to abdicate (Narli Reference Narli2000). The Pakistani military undermined the power of post-Zia civilian governments to exercise authority over key state personnel and policies (Shah Reference Shah2014, 168). By using the concept of militarization, and distinguishing between direct and indirect variants, we capture the fact that militaries can wield political power without supplanting civilian governments.
While regime militarization often begins with a coup, a coup does not always commence regime militarization. Coups can initiate democratic transitions (for example, Marinov and Goemans Reference Marinov and Goemans2014) or represent a military effort to replace one set of civilian leaders with another. Conversely, regime militarization may involve a gradual military effort to acquire power and exert it indirectly, or situations in which overburdened civilian rulers’ hand over control to militaries. Different from coups, regime militarization is (to varying degrees) an enduring feature of political systems. It shifts our focus to the question of how (and why) militaries secure and sustain political power over long periods of time.
The Paucity of Theory on Wars and Regimes
Most empirical research on wars has sought to explain their occurrence. But what about the political effects of war? As Kasza (Reference Kasza1996, 370) noted nearly a quarter century ago: ‘every aspect of domestic politics has felt war's influence. It is time […] to give this pivotal phenomenon the attention it deserves’. Yet our relative inattention to war as a determinant of politics persists. This is particularly true for the most studied macro-level outcome of all besides war itself: democratic and authoritarian regimes.
Instead, scholars have mostly analyzed the effects of interstate and, more recently, intrastate war on state formation (for example, Thies Reference Thies2004). The nexus between war and regime formation has received much less interest. Ertman's (Reference Ertman1997) study is a telling example. Unlike most bellicist works, Ertman strives to explain both states and regimes in European political development. Yet war only helps explain states in Ertman's theory. To explain the rise of constitutionalist regimes, Ertman relies on a non-bellicist variable: the territorial vs. status basis of medieval representative assemblies.
When bellicist theory does discuss regime effects, the findings have been inconsistent. Some literature points to the democratizing force of external war as it leads to elite-level negotiations over taxes and state building (North and Weingast Reference North and Weingast1989; Tilly Reference Tilly1992). By contrast, Downing (Reference Downing1993) argues that the military revolution in early modern Europe pressed monarchs to disregard and disband parliaments. Beyond elite bargaining models, recent research suggests that high levels of popular mobilization during civil war can foster democratic inclusion after an armistice (Huang Reference Huang2017). Contrarily, invocations of ‘garrison states’ (Lasswell Reference Lasswell1941) imply that wars have authoritarian effects. Some authors argue that participation or military defeat in interstate war make regime breakdown through a military coup more likely (Bueno de Mesquita et al. Reference Bueno de Mesquita1992; Finer Reference Finer1962). Others have shown that interstate war diminishes the likelihood of military overthrows (for example, Piplani and Talmadge Reference Piplani and Talmadge2016). Desch (Reference Desch2001) and Staniland (Reference Staniland2008) report that intrastate conflict makes coups more likely. Bell and Sudduth (Reference Bell and Sudduth2017) argue that the risk of coups increases as a result of the costs imposed on combatants. Our analysis builds on their welcome shift in focus towards internal warfare, while more precisely specifying the type of civil war that is most likely to trigger this dynamic.
In summary, the war literature has devoted little sustained interest to regime formation. Even where it has done so, it has characteristically approached regimes as a dichotomy between democracy and authoritarianism. This is insufficient in light of the institutional diversity across authoritarian regimes that comparative research during the last two decades has established. The key distinction to emerge from this new wave of research is the contrast between authoritarian regimes in which power rests in a ruling party and those in which it rests in the military – a recurrent variation that remains to be systematically explained. Building on Slater's (Reference Slater2018) recent work, this article argues that a productive way to tackle this puzzle is to focus on the gap in the literature on the macro-level political consequences of war.
War and Regime Militarization: Towards Micro-Foundations
Our argument is rooted in Cold War Southeast Asia (Slater Reference Slater2010; Slater Reference Slater2018). Although virtually all Southeast Asian countries experienced severe internal warfare after World War II, they differed sharply in the predominant type of conflict. Some newly born countries confronted armed challenges from groups seeking to escape the newly formed state, while others faced rebels aiming to seize the new capital. This difference was significant because if violent conflict is to spark political development, it must unify a set of actors with sufficient resources to craft a new political order: either through revolutionary takeover or counter-revolutionary self-defense. Southeast Asia's experience with divergent forms of rebellion suggests that regional rebellions are the least likely type to inspire a cohesive counter-revolutionary coalition encompassing both military and civilian leaders.Footnote 5
Our core purpose in this section is to decontextualize Slater's original theory and parlay the existing, inductively generated findings into a more deductively grounded theory inspired by rational-actor approaches. This results in refined and generalizable micro-foundations for the causal link between regional rebellions and militarization. We thereby embed the theory in literatures on coups and civil–military relations, specifically their emphasis on information asymmetries and strategic assessment, coordination problems and corporate grievances.
We first elaborate a theory of two general mechanisms to explain the link between regional rebellions and regime militarization. We then present illustrative evidence from a variety of cases (starting in Southeast Asia) to elucidate how the stipulated mechanisms might work in practice. Our case material builds on a convenience sample and is only meant to serve as an empirical illustration, not a systematic test, of the mechanisms. As the causal mechanisms go beyond Slater (Reference Slater2018), the empirical material used to illustrate the mechanisms is new, although drawn from some of the same cases. The main point of this section is to theoretically motivate our general expectation that regional wars are more likely to lead to militarization – the core hypothesis and empirical focus of this article. A dedicated test of the causal mechanisms and their observable implications is beyond the scope of this current work.
We highlight two mechanisms – what we call the disalignment and operations mechanisms – through which the causal connection between regional rebellions and political militarization can unfold. These mechanisms are conceptually separate and can occur independently of each other, although it is conceivable that they influence each other in specific cases. Although these mechanisms are not necessarily exhaustive, we consider them to be the most systematic ways in which regional rebellions shape regimes.
Disalignment
When war reshapes states and regimes, it does so by reshaping political coalitions. Of particular interest for our argument here, political militarization requires a political disalignment in which militaries become both divorced from and more distrustful of civilian leadership. The civil–military literature posits that militaries can reject the legitimacy of specific missions or develop divergent views about how and when to use force (Betts Reference Betts1991; Feaver and Gelpi Reference Feaver and Gelpi2011).
We argue that regional rebellions are an especially likely candidate to disalign civilian and military elites in this fashion. Specifically, regional rebellions entail especially severe information asymmetries and hence prompt civilian and military leaders to develop divergent beliefs about the threat environment the regime is facing and how this threat should be countered.Footnote 6 This can heighten the military leadership's corporate grievances, which are key motivations for military interventions (Nordlinger Reference Nordlinger1977; Thompson Reference Thompson1980).
Mostly huddled in major cities, civilian elites are predominantly focused on political activities and threats originating in and around the capital and other major population centers (Bates Reference Bates1981; Waldner, Peterson and Shoup Reference Waldner, Peterson and Shoup2017). Given that taking rebellions into urban political centers is costly (Kalyvas Reference Kalyvas2006, 133) and that a signal's credibility increases with its costs, civilian elites should perceive rebel activity close to these centers, rather than far away from them, as highly threatening (Johnson and Thyne Reference Johnson and Thyne2018). What is more, civilian elites can often directly observe the effects of urban rebellion – for example if certain quarters or roads become inaccessible – conveying battlefield information directly to politicians.Footnote 7 Better information on urban armed conflict also stems from the nature of urban counter-insurgency, which typically involves civilian policing, surveillance and social welfare measures (Ron Reference Ron2003). This means civilian elites can rely on numerous civilian and military sources to form their beliefs about the threat environment.
By contrast, regional rebellions typically occur on the fringes of national territory, far from the inner perimeter of civilian elites' threat perception.Footnote 8 Fighting is delegated to military elites who are expected to sacrifice themselves and their soldiers to hold the country together. In addition, crushing regional rebellions typically involves the use of crude military force while the involvement of civilian actors is much reduced (Schutte Reference Schutte2017). As a result of its specialization in dealing with such threats, coupled with the conflict's geographical remoteness, the military has a significant informational advantage over its civilian principals (Brooks Reference Brooks2008; Egorov and Sonin Reference Egorov and Sonin2011; McMahon and Slantchev Reference McMahon and Slantchev2015).Footnote 9
This information asymmetry can disalign civilian and military elites in two distinct ways: a ‘hawkish’ scenario in which military leaders decide that civilian leaders provide insufficient support for fighting regional rebels, and a ‘dovish’ scenario in which military elites want to settle for peace while civilian leaders prefer to continue fighting. In both scenarios, the key factor is disagreement over the conduct of the conflict, and the outcome in terms of regime type – increased motivation of military elites to take over – is the same. The outcome as regards the approach to regional rebellion is different, depending on the type of scenario (hawkish or dovish). This, however, is not the main dependent variable of this article. The hawkish and dovish scenarios both entail a civil–military disalignment in which the regional rebellion creates extremely high stakes for the military – literally a matter of life and death – making them resent the civilian leadership for conducting the war too carelessly and cluelessly.
Hawkish scenario
In the hawkish scenario, the military signals that threat levels are high and requests more resources to crush the regional rebellion. Civilian elites receive this as a noisy signal. At this point the ‘civil–military problematique’ (Feaver Reference Feaver1996) kicks in: as civilian elites know that military elites have an incentive to exaggerate the threat to amplify their resources and, ultimately, political power, they will qualify or, worse, disregard the signal. As Feaver (Reference Feaver1996, 154) explains, ‘the military can describe in some detail the nature of the threat posed by a particular enemy, but only the civilian can decide whether to feel threatened’. The military, as a result, will feel increasingly reluctant to engage in and share information with joint civil–military assessment institutions, further exacerbating information asymmetries in a vicious circle (Brooks Reference Brooks2008, 45).
To the extent that civilian and military elites disalign over both how and how committedly to combat the regional rebellion, we can expect the military's frustration to rise and tolerance for civilian oversight to decay. Diverging threat perceptions can also lead to disagreements over strategy, doubts over civilian competence and concerns over the military's status, which have all been shown to trigger military intervention (Brooks Reference Brooks2008; Desch Reference Desch2001, 11–12; Finer Reference Finer1962, 27). Failing a move by the rebels closer to urban centers, military elites cannot make their assessment of rebel strength more credible.Footnote 10
The observable implications of this ‘hawkish’ variant of disalignment are that civilian rulers have limited access to independent information about the regional conflict; they discount the military's threat description, motivated by suspicions that the military exaggerates the threat strategically (which can indeed be the case), in turn frustrating military leaders and motivating political intervention. All this should be reflected in military and civilian strategic discussions. In practice these can be difficult to observe, though we descriptively illustrate much of the process in the Burmese case below.
Dovish scenario
In the alternative dovish scenario, the military concludes that the losses from fighting the regional rebellion are too high and an honorable negotiated settlement is needed to avoid pointless bloodshed. Meanwhile civilian elites, who have no personal skin in the game, believe a decisive military victory is attainable. Military elites could in principle simply tell their civilian principals that the rebels cannot be beaten. Yet in practice military elites will have incentives not to do so. As Goemans (Reference Goemans2008, 777) points out, acceptance of defeat by the military frequently entails a reorganization of the military, ‘with significant implications for the careers and prospects of the officer corps’. What is more, military elites themselves compete with other officers and thus are aware of the real risk of rival brass coming forward and declaring that s/he will be capable of defeating the rebels (Brooks Reference Brooks2008, 50). Save for putting their career and position in the regime coalition at stake, military elites may therefore decide that the only way of ending the conflict is to take control themselves.Footnote 11 Even in cases where military elites initially share their concerns with civilian rulers, extant research on strategic assessment suggests that, fearing that continued dissent will entail professional suicide, they will begin to withhold information, grow wary of participating in joint forums, and act more independently in the face of continuing disagreement with civilians (Brooks Reference Brooks2008, 5, 250). Ultimately, they may conclude that a fundamental change of strategy can only be achieved by forcefully sidelining civilian rulers.
The observable implications of this ‘dovish’ variant of disalignment are that military leaders become convinced that a military solution is impossible, cease communicating this to civilian elites (or never communicate it in the first place) due to fears of punishment and demotion, and eventually take this as a reason to intervene. This process is illustrated in the Thai example below.
Empirical illustrations
Burma's 1962 coup is a useful example of the hawkish scenario unfolding under conditions of information asymmetry. In April 1960, Prime Minister U Nu was re-elected after handing over power to a military caretaker government under General Ne Win in 1958. While rebel activity throughout Burma had markedly declined between 1958 and 1960, Nu's return to office in 1960 coincided with a rapid resurgence of rebel militancy in Shan state. Simultaneously, the Karen National Defence Organization intensified its military campaign in apparent co-ordination with Shan rebels. Additionally, with weapons allegedly supplied by the United States, remnants of the Chinese Kuomintang fighters present on Burmese territory joined forces with Karen and Shan rebels in early 1961 (Trager Reference Trager1963, 316–317). The army viewed the resurgence of rebel activity extremely anxiously (Smith Reference Smith1991, 185). In declassified Central Intelligence Agency (CIA) reports from January 1961, Chief of Staff Ne Win is quoted as saying that the situation has reached ‘serious proportions’ (CIA 1961a).
In stark contrast, Nu seemed to be in denial, rejecting the risk assessment put forward by the army. The split of the Anti-Fascist People's Freedom League (AFPFL) ruling party in April 1958 decimated local party infrastructure and left Nu with very limited sources of independent civilian information on developments in the countryside (Butwell and von der Mehden Reference Butwell and von der Mehden1960). In August 1960, he stated that ‘there's calm, peace, and lack of tension in the country, no one can deny it’ (New York Times 1960). In fact, he blamed rebel ‘luck’ and, paradoxically, the army's success in forcing rebels out of their hideouts for the worsening security situation and denied that the situation was ‘as bad as 1949–50’ – the worst year of the insurrection (Smith Reference Smith1991, 189–190; Trager Reference Trager1963, 317–318). This caused an outcry in the Burmese press, which demanded a vigorous increase in the size of the armed forces and more military spending (Trager Reference Trager1963, 317–318).
Yet this would have gone against Nu's goal of pushing back the influence of the army (CIA 1960a; CIA 1960b). Nu had previously accused the army of favoring the rival Stable AFPFL party before the election (Bigelow Reference Bigelow1960, 71). Accepting the military's risk assessment and boosting the army's resources and standing was thus not an option. Holding out vigorously against army budget requests (Callahan Reference Callahan2005, 6, fn. 42), Nu diverted troops to the Chinese border for survey and demarcation missions, reorganized the army and gave the civilian Ministry of Defense tighter control over all military activities (Callahan Reference Callahan2005, 202; Chang Reference Chang1969, 826). The CIA noted in May 1961 that the war effort against ethnic insurgencies had visibly declined (CIA 1961c). With Nu's refusal to acknowledge the heightened security threat, patience within the military evaporated (CIA 1961b), culminating in a military coup led by Ne Win in March 1962. In one of its first post-coup statements, the army promised ‘to stamp out’ the resurgent insurgencies (cited in Trager Reference Trager1963, 322).
Such hawkish scenarios differ meaningfully from the dovish scenario. The latter is well illustrated by the coup against Thai Prime Minister Thaksin Shinawatra in September 2006. Here, the military's frustration with Thaksin's heavy-handed handling of the insurgency in the country's south was a key driver (Funston Reference Funston and Funston2009; McCargo Reference McCargo2007; Pongsudhirak Reference Pongsudhirak2006; Sheridan Reference Sheridan2006). Thaksin responded to early incidents of violence in 2001 by berating the army in the south, threatening that ‘transfers will be ordered for those who are inefficient […] They won't be around here’ (McCargo Reference McCargo2006, 47). At the same time, and despite a worsening security situation from mid-2001, Thaksin disbanded the military-led Southern Border Provinces Administration Center and the joint civilian–police–military task force, CPM 43, and diverted control from the military to the police, with which he enjoyed closer political ties (Croissant Reference Croissant2007, 11; ICG 2005a, 33; McCargo Reference McCargo2006, 48).
As the conflict worsened from 2002 onwards, Thaksin ordered the police and army to respond with full force (Pathmanand Reference Pathmanand2006). Martial law was introduced to three Southern provinces in January 2004 and troop size increased radically, reaching 20,000 by October 2005 (ICG 2005a, 19). This repressive strategy resulted in a rapidly rising death toll – over 1,000 between 2004 and 2006 (Pathmanand Reference Pathmanand2006, 74) – and increasing government abuses, such as the suffocation of eighty-five Muslim men in detention in Bak Tai (ICG 2005a, ii). High-ranking military officers were deeply unhappy about this incident, which they saw as tarnishing their reputation (Pathmanand Reference Pathmanand2006, 82). The killing and public display of mutilated Thai marines in September 2005 had a profound psychological impact on troops in the south (ICG 2005b, 19). While military casualties were as high as police casualties (Chalk Reference Chalk2008, 10), decision making had been placed in the hands of the police, which had profound repercussions for strategic assessments and information sharing with civilian authorities.
While the police kept feeding Thaksin intelligence that blamed organized criminals for the resurgent violence and engaged in extrajudicial killings of former army informants (ICG 2005a, 34), the military command in the south gradually stopped sharing information with civilian leaders in Bangkok. According to General Vinai Pattiyakul, secretary-general of the National Security Council, ‘[military] intelligence agencies had discovered the establishment of separatist insurgency networks since 2002 but no one dared tell the prime minister’ (cited in Pathmanand Reference Pathmanand2006, 79). Thaksin's actions gradually established a belief among the army high command that, as a former police lieutenant colonel, Thaksin would rather listen to the police than the military and that speaking out would only add one's name to the list of inefficient officers to be rotated out (McCargo Reference McCargo2006, 49). Even Thaksin's cousin, General Chaiyasit Shinawatra, who briefly assumed the role of commander-in-chief between 2003 and 2004, quickly came to believe that Thaksin favored the police (McCargo Reference McCargo2006, 54): he was only one of five commanders-in-chief in an increasingly arbitrary pattern of punishment transfers (ICG 2007, 1; McCargo Reference McCargo2006, 53–54).
This breakdown of information sharing had a devastating impact on the government's ability to manage the rebellion. ‘A source familiar with Thaksin's intelligence-gathering capabilities argued that the prime minister lacked accurate information and analysis of developments in the South. By early 2004, the government had virtually no reliable human intelligence from the ground’ and ‘[i]t seemed abundantly clear that no one in power really understood what was happening around the southern border’ (McCargo Reference McCargo2006, 51; 54). US embassy cables repeatedly mention a phenomenon of stove-piping (often erroneous) information from local agencies to Bangkok (Wikileaks 2005; Wikileaks 2006a). At the same time, the army began to increasingly act independently of Thaksin, with soldiers sharing information only with senior army officials (ICG 2007, 1; 16; Wikileaks 2006b, 3). Police in effect boycotted the re-established joint command structures in 2004 and did not repair the informational breakdown (Pathmanand Reference Pathmanand2006, 82). Additionally, the southern army command – and indeed officers in central command – increasingly turned to the Privy Council and the king to voice their concerns. Harsh criticism of Thaksin's strategy by the council's president and vice president in 2004 can thus be seen as an expression of military frustration (ICG 2005b, 2; McCargo Reference McCargo2006, 62–63). Following a meeting between Malaysian Prime Minister Mahathir and Thai King Bhumibol, the army displayed its keenness to end the conflict by engaging in secret negotiations with the rebels under Mahathir's auspices, an approach rejected by Thaksin (Harish and Liow Reference Harish and Liow2007, 172; ICG 2007, 2).
Emboldened by his election landslide in 2005, Thaksin ‘supplemented martial law in the South with an emergency decree in July 2005[, which] transferred the responsibility for dealing with the insurgency […] to the prime minister's office’ (Storey Reference Storey2008, 42). Meanwhile, Thaksin paid lip service to a more conciliatory approach by establishing a national reconciliation committee while ignoring the dovish request of his commander-in-chief, Sonthi, for a negotiated solution (Harish and Liow Reference Harish and Liow2007, 170). Instead, he publicly chided Sonthi in June 2006 for failing to prevent a wave of bombings (ICG 2007, 1). Thus prior to the coup in September 2006, it was clear that ‘[t]he Army were deeply unhappy that they were being blamed for a problem they believed Thaksin and the police […] had largely created’ (McCargo Reference McCargo2007, 14). Days before the coup, General Sonthi declared: ‘Soldiers all of us, privates and generals, are risking our lives every day. I don't want to see […] politicians putting all the blame on field officers' (Human Rights Watch 2007).
Operations
The second mechanism linking regional rebellions to militarization relies on the particular operational experiences that militaries acquire in fighting regional rebellions. They tend to lead to (1) a stronger esprit de corps and social cohesion of the military; (2) improved co-ordination capacity across military branches; and (3) increases in the military's non-coercive organizational and bureaucratic capacity. We derive these effects from prior studies in the civil–military literature and illustrate them briefly using the cases of the Mauritanian and Indonesian military takeovers.
As mentioned above, militaries are not the only organizations that regimes have at their disposal to battle rebels. Institutions for civilian policing and bureaucratic governance can also be enlisted ‘to win hearts and minds’. Yet the remoteness and specificities of the terrain rebels hide in can greatly complicate the involvement of civilian actors. This means that militaries are often given a free hand, as much by default as by design, to suppress regional rebellions. Such campaigns tend to rely on military force alone, in contrast to joint civilian-military interventions prevalent in urban counterinsurgency (Ron Reference Ron2003). This is evidenced by Schutte's (Reference Schutte2017) finding that the use of crude military force without civilian involvement increases with distance from the capital city. This shared operational experience devoid of civilian support can significantly enhance the military's esprit de corps and increase its cohesion as a status group. Both esprit de corps and the symbolic status of militaries have been associated with a higher inclination to engage in politics (Finer Reference Finer1962, xv; 10; Nordlinger Reference Nordlinger1977, 65). Note that this process operates, in principle, independently of disalignment: the higher cohesion and social status that result from fighting regional wars do not automatically lead to disagreement over the conduct of operations, and vice versa; regional wars tend to trigger both in parallel.
The observable implications of this process are straightforward. Separate, autonomous operations during regional rebellions boost the military's social cohesion and elevate its social status over time. This, in turn, can facilitate political involvement. This is what we observe in the example of Mauritania, where the military's social status and collective identity as a leading national institution strengthened considerably throughout the conflict with Western Saharan rebels from 1975. While the military was embryonic when President Ould Daddah started the conflict, it gained significant combat experience and esprit de corps over the coming years, preparing the top brass for the 1978 takeover of power from a civilian leader perceived as detached and strategically inept (Pazzanita Reference Pazzanita1996, 47).Footnote 12
The coup literature further points to the importance of coordination capacity in making military takeovers possible (Pilster and Böhmelt Reference Pilster and Böhmelt2011; Singh Reference Singh2014). The conduct of regional military operations typically leads to the improvement of exactly such capacity between different military units as it often relies on different branches of the military working closely together. Again, this process is in principle clearly observable in terms of the increased co-operation of military branches over time, the more frequent interaction of senior officers it brings about, and the military elite's subsequent use of such a capacity to co-ordinate collective political action. For example, while Indonesia's army had been rife with competition between the army, navy and air force (Kingsbury Reference Kingsbury2003, 141), fighting rebels in Indonesia (1957–61) involved close co-operation between the country's air force, a sea blockade carried out by Indonesia's navy, and the deployment of infantry units, especially paratroopers (Doeppers Reference Doeppers1972; Feith and Lev Reference Feith and Lev1963). Institutionally, this led to the establishment of the Central War Administration, which coordinated military efforts during the state of emergency. Such co-operation leads to a better exchange of preferences within the military and better information on the likely support for military involvement in politics. While we would expect junior officers and rank-and-file soldiers to be busy fighting the rebellion, elite officers, having improved their co-ordination capacity, can potentially rely on troops not deployed in the restive region to overthrow the civilian government. This is supported in our data, as all military takeovers following regional rebellions but one (the Comoros in 1998) were carried out by the military leadership, often led by chiefs of staff.
Finally, waging war against regional rebels heightens the army's organizational and bureaucratic capacity. War efforts are typically accompanied by the professionalization and expansion of the military organization. Importantly, this expansion often means that the army branches out into civilian realms of government and production. This is directly observable in the widening of non-core operations and attendant organizational changes of the military, which in turn motivate and facilitate its political involvement and takeover. As Eck (Reference Eck2015, 232) remarks on the case of Burma, ‘[i]n 1952, Ne Win reorganized and expanded the resources dedicated to officer training [and] the armed forces ventured into business by setting up the Defense Services Institute (DSI) which, by 1960, was running banks, shipping lines, and the largest import–export operation in the country […] this organization was established to ensure that armed units were supplied’. Similarly, Lev (Reference Lev1964, 351) shows that martial law declared on the Indonesian islands of Sulawesi and Sumatra made the military the de facto rulers of the islands, subordinating the civilian administration to military orders. These processes of organizational and bureaucratic upgrading bestow precious managerial capacities upon military elites, putting them on par with civilian elites. As the Indonesian chief of staff, Nasution, declared in front of military cadets in 1958, the army ‘must be given an opportunity to make use of their skills in the Cabinet, the National Council, the National Planning Board, diplomatic posts, and elsewhere in government’ (Lev Reference Lev1964, 359). As a result of their ‘unconventional’ activities, the army would ‘feel they have as much to contribute as civilian leaders to the nation's development, perhaps more' (Lev Reference Lev1964, 364).
In sum, fighting regional rebels tends to augment the military's esprit de corps, coordination and organizational capacity, giving military brass more confidence in their abilities, including in the civilian realm. All three factors are progressively strengthened as fighting becomes more intense and protracted. As Maung (Reference Maung1969, 231) points out in the case of Burma, with the civilian government confined to major towns, the long duration of the rebellion forced officers to acquire the leadership skills of civilian administrators. Ultimately, their improved governance skills can lead the military to believe it is not only better placed to decide how to fight the war, but also – when disagreeing with civilian leaders – better able to run the country, either directly or indirectly.
The disalignment and operations mechanisms provide conceptually separate pathways to militarization, although they may influence each other. Heightened self-esteem and capacity can lead to more coherent interests that are more clearly disaligned with civilian leadership. Disalignment itself might also trigger greater efforts to operate autonomously. While such mutual causation is conceivable, the mechanisms can operate in parallel (as in our illustrative case material). They can also operate without each other. For example, operational capacities might remain low for political or social reasons (such as the class or communal cleavages of many militaries), yet military brass can disagree with civilian leaders about strategy in a regional conflict.Footnote 13
Testing External Validity: Quantitative Analysis
We now subject our theory to quantitative tests to establish whether the association between regional rebellion and militarization holds in a global context. Our theoretical discussion thus far implies the following hypotheses:
Hypothesis 1 Countries that experience regional rebellions are more likely to become militarized authoritarian regimes.
Hypothesis 2 More intense regional conflicts should lead to greater militarization of the regime.
We test these predictions by conducting two separate sets of regressions for the emergence of military regimes (Hypothesis 1) and for their extent of militarization (Hypothesis 2). In the online appendix, we also test a third hypothesis, namely that military rulers emerging from regional rebellions cling to power longer because their motivation and capacity to do so are stronger due to preceding preference disalignment and operational experience, resulting in a strong preference against a return to civilian rule. Our findings are supportive of this claim, albeit less unequivocally than for Hypotheses 1 and 2.
Data
Our dataset (Eibl, Hertog and Slater Reference Eibl, Hertog and Slater2019) relies on Svolik's (Reference Svolik2012) data on authoritarian regimes to measure the variable of primary interest – regime militarization. We prefer Svolik's data over alternative datasets for three reasons. First, it covers the most countries, yielding over 8,500 observations compared to 7,700 in the alternative Geddes, Wright and Frantz (Reference Geddes, Wright and Frantz2014) data. Secondly, the dataset allows military regimes to be unambiguously coded, which avoids the problem of manifold hybrids as in the Geddes et al. data. Thirdly, and critically, the data distinguishes between different degrees of military involvement in politics, enabling us to capture militarization below the level of a fully fledged military takeover.
We code the variable Military Regime for all autocracies as 1 if military involvement in politics is either direct or indirect, that is, ‘the head of the executive is a civilian executive but the military intervenes in government policies unrelated to national security’ (Svolik Reference Svolik2012, codebook). Direct military rule can either be institutionalized in a corporate fashion (for example, a classical junta) or be personal, with the military ruling through a personal leader without institutionalization (Slater Reference Slater2003). Non-military regimes are coded 0 and comprise both civilian autocracies and democracies. We also construct a categorical measure of military regimes (Military Regime Categorical) to test our militarization hypothesis (Hypothesis 2). The variable takes a value of 0 for non-military regimes, 1 for indirect military rule and 2 for direct military rule.Footnote 14
Data for our main independent variables – regional rebellions and, as contrast cases, center-seeking rebellions – are taken from the UCDP/PRIO dataset on armed conflicts (Allansson, Melander and Themnér Reference Allansson, Melander and Themnér2017; Gleditsch et al. Reference Gleditsch2002). We define a rebellion as regional if the incompatibility underlying the conflict is over territory. Center-seeking rebellions are characterized by incompatibility over government. As the total effect of rebellion is unlikely to unfold within a single year, we model the effect of rebellion with an exponential decay function. Specifically, the variable Regional Rebellion (Decay) takes a value of 1 whenever a regional rebellion is ongoing, after which it decays exponentially with a half-life of one year. Center-seeking Rebellion (Decay) is defined equivalently for rebellions with incompatibility over government.Footnote 15 To measure the intensity of rebellions, we also create an additional count variable Duration of Regional/Center-seeking Rebellion that measures the duration of an ongoing rebellion in years. We prefer this measure over alternative indicators, such as casualty numbers or the economic cost of destruction, which are frequently unreliable.Footnote 16
Although Piplani and Talmadge (Reference Piplani and Talmadge2016) have argued that interstate conflict makes regime military overthrows less likely, we err on the side of caution by including a control variable for the incidence of interstate wars in a given year (Allansson, Melander and Themnér Reference Allansson, Melander and Themnér2017; Gleditsch et al. Reference Gleditsch2002).Footnote 17 In addition, we include a battery of socioeconomic and political controls in our baseline models. As the democratization literature suggests a strong link between a country's wealth and regime type (Boix and Stokes Reference Boix and Stokes2003; Przeworski and Limongi Reference Przeworski and Limongi1997), Per Capita GDP (log) measures the logged per capita GDP in constant 2005 USD (taken from Bolt and van Zanden Reference Bolt and van Zanden2014).Footnote 18 We also include Population Size (log) using data from the World Development Indicators (World Bank 2017).
Establishing military rule in politically open systems might be harder; conversely, transitioning to democracy might be easier. We therefore include the variable Polity to approximate a country's degree of political openness using the standard Polity 2 indicator (Marshall, Gurr and Jaggers Reference Marshall, Gurr and Jaggers2010). The variable Military Capabilities (cube root) is based on the widely used composite index of national capability and captures, among others, material benefits directed towards the military (Singer et al. Reference Singer and Russet1972, v5.0). Transformation using a cube root function yields the closest approximation to a normal distribution.
The coup literature highlights that civil unrest increases the risk of a military takeover (for example, Johnson and Thyne 2016). Based on the Banks (Reference Banks2011) dataset, we therefore include the variable Unrest (log) in the regressions, which sums up and log-transforms the number of riots, demonstrations and general strikes occurring in each country-year. Conflicts leading to military rule might also be more likely in religiously fractionalized societies. We therefore include a measure of Religious Fractionalization (Alesina et al. Reference Alesina2003). Finally, we control for regional fixed effects by including region dummies.Footnote 19 Taking into account the effects of listwise deletion, our main models generally comprise the time period from 1961 until 2008.Footnote 20
Modelling Military Regime Emergence
Our first hypothesis posits that regional rebellions make the subsequent emergence of a military regime more likely. This calls for a model of military regime Onset, which we model as a Markov transition model. The model allows us to separate the effect of rebellions on regime onset from the effect on regime durability by conditioning the model on whether or not an event occurred in the previous year (Beck Reference Beck, Box-Steffensmeier, Brady and Collier2008, 490). In our case, this means that we condition the model on the absence of a military regime in the previous period. Modelling event onset in this way has been shown to be superior to alternative strategies, such as setting ongoing years to zero or missing (McGrath Reference McGrath2015). We opt for the widely used binary time-series cross-sectional regression model as our statistical workhorse. The model takes the following specification:
$$\Pr (Y_{i,t} = 1 \vert Y_{i,t-1} = 0) = {\rm Logit}(\alpha _0 + \beta _1R_{i,t-1} + \beta _2X_{i,t-1} + \delta _3 + \beta _4\vartheta _{i,t} + \mu _{i,t}),$$where α designates the constant, R our measure of regional rebellions, X a vector of covariates, and δ regional fixed effects. To deal with the issue of repeated events, we add the event counter 
$\vartheta $ measuring the number of previous transitions to a military regime (Beck, Katz and Tucker Reference Beck, Katz and Tucker1998). We add μ, a vector of cubic time polynomials to account for duration dependence (Carter and Signorino Reference Carter and Signorino2010).Footnote 21 All variables except for regional fixed effects and the event counter are lagged by one year to ensure the correct order of temporal dependence.
One characteristic of the above model is that, by pooling the data, it does not clearly separate between-country and within-country effects. This is potentially problematic as our findings could be driven by unit-level (country) effects, which the proposed specification fails to account for. We therefore also estimate a more conservative conditional logit model that controls for country-specific effects by conditioning on the number of successes in a group (Chamberlain Reference Chamberlain1980).Footnote 22 Conditioning on successes, in turn, means that all countries that did not experience a military regime are dropped from the analysis, and time-invariant variables cannot be estimated.Footnote 23
The results of our pooled and fixed effects transition models are summarized in Table 1. Columns 1–4 report the pooled logit model while successively adding control variables; Column 5 reports the results of the conditional logit model. Taken together, the results provide strong support for our claim (Hypothesis 1) that regional rebellions are associated with an increased likelihood of military regime emergence. Looking at the pooled logit model (Columns 1–4), the coefficients of Regional Rebellion are positive and highly statistically significant throughout. This contrasts with Center-seeking Rebellion, for which the model does not find a statistically significant effect. War, another potential confounder of the effect of regional rebellion, is also insignificant, suggesting that the cohesion or political ambition of military elites primarily, if not exclusively, stems from regional rebellions. Adding further controls diminishes the effect of regional rebellion only slightly. Our findings on control variables are mostly in line with theoretical expectations: military takeovers are less likely in politically more open countries (Polity) and whenever greater resources are directed toward the military (Military Capabilities), while Unrest facilitates military power grabs. Reassuringly, Regional Rebellion retains its strong positive effect in the conditional logit model.Footnote 24
Regional rebellions and military regime emergence
Note: pooled logit model in Columns 1–4. Conditional logit model in Column 5. Cluster-robust standard errors in parentheses. *p < 0.1; **p < 0.05; ***p < 0.01.
To illustrate the substantive effect of regional rebellions, we look at the percentage increase in the likelihood of a military regime emergence relative to the average baseline probability. In the pooled model (Column 1), the average annual probability of a transition to a military regime is 1.78 per cent. A regional rebellion increases the likelihood on average by 1.66 percentage points, representing a remarkable 93 per cent increase. In the case of the conditional logit model, the average increase is even bigger. Expressed in terms of odds ratios, the coefficient of Regional Rebellion in the conditional logit is 7.8, meaning that the odds of military regime emergence are nearly eight times higher when there is a preceding regional conflict.Footnote 25 Given our measurement of Regional Rebellion as decaying over time, we also illustrate the temporal decline in the probability of military regime emergence following a regional rebellion in Figure 1. Following a significant increase in the probability of a transition toward military rule in period t, the probability gradually declines in periods t 1 to t 4 until returning to levels around the baseline probability of about 1.8 per cent in t 5.
Effect of regional rebellion on military regime emergence
Note: marginal effects derived from Table 1, Column 1. Regional rebellion set at 1, 0.5, 0.25, 0.125 and 0.0625 according to decay function. All other covariates set at their observed values (Hanmer and Ozan Kalkan Reference Hanmer and Ozan Kalkan2013). All figure layouts are based on Bischof (Reference Bischof2017).
As emphasized in the theory section, the information asymmetries between military and civilian political elites result from the remoteness of regional rebellions from urban centers. An observable implication of this is that regional rebellions should make military regime emergence more likely the farther they are away from the capital. This prediction receives strong support in the data, as shown in Appendix Table A10 and Figure A6. This finding contrasts with Bell and Sudduth (Reference Bell and Sudduth2017), who find that coups during civil wars become increasingly likely the closer the war moves to the capital. While some differences in the models might be expected due to the different dependent variables (coups vs. militarized regimes), we suspect that their sample restriction to only cases with ongoing civil wars (1,441) accounts for most of the difference.
Modelling Regime Militarization
We use multinomial logistic regression to test our second hypothesis on regime militarization.Footnote 26 The dependent variable in this model is Military Regime Categorical, which comprises three different categories: 0 for non-military regimes, 1 for indirect and 2 for direct military rule. The intensity of regional rebellions is captured by our duration indicator Duration of Regional Rebellion. The multinomial logit takes the following functional form:
$$\Pr (Y_{i,t} = {\rm Mil}{\rm.} \,{\rm regime}\;{\rm categorical}) = {\rm Logit}(\alpha _0 + \beta _1D_{i,t-1} + \beta _2X_{i,t-1} + \beta _4\vartheta _{i,t} + \mu _{i,t}),$$where α designates the constant, D our measure of rebellion duration, X a vector of covariates, the event counter 
$\vartheta $,Footnote 27 and μ a vector of cubic time polynomials.Footnote 28 As before, all variables save for the event counter are lagged by one year. Note that we do not condition the model on the absence of a military regime in the previous year as we did in the case of the Markov transition model. This choice reflects the fact that further militarization can occur even after the regime has transitioned to a military regime. That said, we also run the model on the subset of military regimes only to test if the intensity of rebellion leads to further militarization after the transition to military rule has occurred, effectively making the multinomial logit a choice model between indirect and direct military rule.
Table 2 displays the results predicting Military Regime Categorical. Columns 1–4 report the regression coefficients for the unrestricted sample of military and non-military regimes combined. Column 5 presents the results for military regimes only. The base category in the first four models is non-military regimes. In Model 5, the base category is indirect military rule. As before, we add controls successively in Columns 1–4. The findings provide substantive evidence for our claim that regional rebellions militarize regimes by pushing regimes more towards direct military rule.
Regional rebellions and regime militarization
Note: multinomial logit regression with cluster-robust standard errors. Columns 1–4 unrestricted sample, base category: non-military regimes. Column 5 restricted sample on military regimes only, base category: indirect military rule. *p < 0.1; **p < 0.05; ***p < 0.01.
We illustrate the size of the effect for all regimes combined in Figure 2 and for the subset of military regimes only in Figure 3. Both figures plot the effect of our categorical variable for different durations of regional conflict and illustrate the increasing probability of direct military rule at the expense of both indirect and no military rule.Footnote 29 In the two cases, direct military rule becomes significantly more likely than indirect rule the longer regional conflicts endure.
Effect of regional rebellion duration on militarization (all regimes)
Note: marginal effects derived from Table 2, Column 4. All other covariates set at their observed values (Hanmer and Ozan Kalkan Reference Hanmer and Ozan Kalkan2013).
Effect of regional rebellion duration on militarization (military regimes only)
Note: marginal effects derived from Table 2, Column 5. All other covariates set at their observed values (Hanmer and Ozan Kalkan Reference Hanmer and Ozan Kalkan2013).
Robustness Tests
We subject our results to a panoply of robustness tests, which are all detailed in the online appendix. Specifically, we:
(1) use the alternative Mundlak-Chamberlain approach to run our fixed effects logit regression (Tables A5–A6),
(2) use a dummy version of regional and center-seeking rebellions instead of the decaying variable (Tables A7–8),
(3) test a binary indicator of military coups instead of military regimes, and obtain very similar findings (Table A9),
(4) test whether the distance of a regional rebellion to the capital increases the likelihood of military regime emergence, which is an observable implication of our theory (Table A10),
(5) use instrumental variables to take into account the potential endogeneity between military regime and rebellions (Table A11),
(6) add further or alternative controls, such as ethnic fractionalization, resource rents, population density, trade, a binary control for the post-Cold War period (which by and large has no substantive effect) and a dummy for prior regime type (Tables A12–14),
(7) control for potential diffusion effects of military regimes and democratization by controlling for the share of military regimes and democracies in the neighborhood, variously defined (Tables A15–17),
(8) test alternative time specification, such as temporal splines (Tables A18–20),
(9) rerun all our models using a coding of military regimes based on the alternative Geddes, Wright and Frantz (Reference Geddes, Wright and Frantz2014) (Tables A21–22),
(10) test alternative half-lives for our decaying variables (2, 3 and 5 years) (Tables A23–A28),
(11) exclude countries with the greatest leverage on our regression coefficient Regional Rebellion (Tables A29–30).
As expected, some of these tests render our findings weaker or less precise. That said, most of the robustness tests yield substantively identical or even stronger findings for military regime emergence and militarization. As for potential endogeneity concerns, strong findings from our instrumental variable regression strengthen our claim that it is regional rebellions that cause militarization, not the other way around.
Conclusion
If war makes states, and if states and regimes are empirically intertwined, it stands to reason that different types of wars make different types of regimes. We have found strong quantitative evidence in support of our claim that regional rebellions are systematically more likely than other types of wars to give rise to militarized authoritarian regimes.
This calls new attention to war as a causal variable in the ever-expanding literature on authoritarianism. While the focus in this piece has been on the specific case of militarized authoritarian regimes, our findings contribute to the study of authoritarianism more broadly by systematically addressing the question of why different types of authoritarian regimes emerge and endure. In equal measure, our analysis advances the bellicist literature by pivoting the debate from war as an outcome to be explained towards war as a producer of politics. Finally, we refine the literature on civil–military relations and coups by arguing and demonstrating that only one type of armed conflict – regional rebellions – increases the likelihood of military takeovers. We also re-emphasize the critical point that ‘military rule’ does not always originate from a coup. Military takeover of power can be gradually negotiated, such as in many Latin American countries in the 1950s–1970s (for example, in Uruguay), or be relinquished to the military as occurred in Burma in 1958, which is not coded as a coup by any of the prominent global coup datasets (Marshall and Marshall Reference Marshall and Marshall2014; Powell and Thyne Reference Powell and Thyne2011).
Much remains to be done to investigate the ways in which rebellions impact regimes. Although the correlation between regional rebellions and political militarization is robust and systematic, it is far more difficult to determine with confidence that the causal mechanisms we posit here are the most important ones driving this result. Future research should collect both further case study evidence and, as far as sources allow, standardized data on the causal mechanisms we have posited to assess their generalizability.
It will also be important to assess why regional rebellions sometimes do not give rise to regime militarization. In some cases, fear of a coup might pre-emptively align civilian leaderships with military preferences. Conversely, some categories of cases might be outside of the scope conditions of our theory, as institutional traditions of militaries suppress any political ambitions (as in richer democracies, but probably also some poor democracies like India). Considering how little we know about the effects of different types of war on different types of political regimes, when compared to how much we know about how war makes states, our contribution here is but a modest start towards what will hopefully become a much wider and sustained collective scholarly endeavor.
Supplementary material
Data replication sets are available in Harvard Dataverse at: https://doi.org/10.7910/DVN/DQ1W55 and online appendices at: https://doi.org/10.1017/S0007123419000528.
Acknowledgements
Ferdinand Eibl was the lead author of the article. Steffen Hertog and Dan Slater contributed in equal parts. We thank the editor, Tobias Böhmelt, and three anonymous reviewers for their insightful comments on the manuscript.
