Sage Journals: Discover world-class research

Abstract

How do dictators successfully counterbalance (fragment their coercive apparatus) despite the significant risk of military retaliation? Drawing on recent insights that the timing of coup-proofing is essential to its success, I argue that dictators are more likely to increase counterbalancing efforts in the aftermath of failed coups. I test this proposition in a difference-in-differences framework, using novel data on coups and counterbalancing, and find a statistically significant effect of coup failure. I substantiate my analysis with two illustrative examples from Sierra Leone and Turkey that probe the plausibility of my theorized mechanism. My findings contribute to the growing literature on the effects of failed coups by opening up the discussion on their long-term structural consequences for the dictator’s security apparatus.

Keywords

civil-military relations coup D’Etat authoritarian power-sharing counterbalancing

Introduction

Perhaps once thought of as a relic of the Cold War, military coups, defined as “the removal or attempted removal of a state’s chief executive by the regular armed forces through the use or threat of use of force” (Thompson 1973, 6), are on the rise again. Since 2020, fourteen coups have been attempted against incumbent dictators in Sub-Saharan Africa, nine of which succeeded (Duzor and Williamson 2023). The most recent success saw the overthrow of President Bongo in Gabon on 30 August 2023 (Beaumont 2023). Three-quarters of all autocrats were removed via coup between 1946 and 2019 (Chin et al. 2021); forty percent of those removed were either executed, incarcerated or exiled within the first year of losing office (Geddes et al. 2018, 72).

To insulate themselves from being overthrown by their armed forces, dictators try to ‘coup-proof’ their regimes. Coup-proofing refers to a broad range of strategies the purpose of which is to deplete or eliminate the capacity of potential coup-plotters from staging a coup (Quinlivan 1999). One such strategy is to ‘institutionally’ coup-proof or ‘counterbalance’¹ the military (Böhmelt and Pilster 2015; De Bruin 2018). This involves fragmenting a country’s coercive manpower into rival security organizations that check and balance one another. This can entail creating separate branches within the military (e.g., creating a coastal guard in addition to the navy), or establishing parallel security organizations with a command structure outside of the regular armed forces (e.g., a presidential guard) (Belkin and Schofer 2003, 2005).

Efforts to counterbalance the military, however, entail significant risks (Sudduth 2017b). In a bid to preserve its corporate interests and autonomy, the army might intervene in a pre-emptive coup, displacing the autocrat prior to losing its coup-making capacity. For example, in Algeria, President Ben Bella was overthrown in 1965 by his officers after he tried to establish a system of popular militias that threatened the army’s monopoly on the use of violence (Zartman 1970). The 1966 coup in Ghana, the 1973 coup in Niger and the 1977 coup in Pakistan were similarly staged in response to rulers’ efforts to undermine their militaries (De Bruin 2020; Higgot and Fuglestad 1975).

Despite the often-hazardous consequences associated with the strategy, counterbalancing remains a central element in any coup-proofing effort (Pilster and Böhmelt 2012). Sixty-three percent of all autocratic regimes between 1946 and 2010 succeeded in counterbalancing their militaries (Geddes et al. 2018, 163), while, as shown in Figure 1, the average number of counterweights (parallel security forces) adopted by dictators worldwide has almost doubled since 1960.² How is it that dictators successfully counterbalance despite this significant risk of military retaliation?

Figure 1.

Average number of counterweights in autocracies over time (dashed line depicts sample mean).

Drawing on Sudduth’s (2017a) insight that the timing of coup-proofing is essential to its success, I argue that rulers are more likely to increase counterbalancing efforts in the aftermath of failed coups. A failed coup serves as an internal shock to the regime that temporarily shifts the balance of power between the autocrat and his military elite in the dictator’s favor. This provides the dictator with a window of opportunity to lock in long-term organizational reforms to the security apparatus, while the military is weakened and unable to resist such coup-proofing efforts. In contrast to arguments that emphasize the democratizing nature of failed coups (e.g., Thyne and Powell 2016), I argue that the rare opportunity presented by the post-coup environment is more conducive to gradual institutional consolidation in autocratic regimes instead.

To test this argument, I carry out a large-n panel analysis leveraging a difference-in-differences design. Using data on autocratic leaders from 1960 to 2010, I find a statistically significant effect of coup failure on counterbalancing. Given the difficulty of identifying causal effects with observational data, I draw on two illustrative cases to probe the plausibility of the theorized mechanism. These cases are: (1) the 1971 failed coup in Sierra Leone, which led to the creation of the Internal Security Unit, a parallel military organization akin to a presidential guard, and (2) the 2016 failed coup in Turkey, which resulted in the institutional re-design of the Turkish Armed Forces and the establishment multiple additional counterweights to the military.

My findings contribute to two adjacent literatures. First, literature on the outcomes of coups has undertheorized the effects of failed coups, despite coup failure accounting for half of all overt coup attempts (Powell and Thyne 2011). More recent studies show that coup failure leads to a greater number of cabinet purges (Bokobza et al., 2022) and a short-term increase in regime personalization (Timoneda et al. 2023). However, while purges occur in the immediate aftermath, or ‘the morning after’ failed attempts, the long-term structural consequences of failed coups are rarely considered, particularly as they pertain to the autocrat’s coercive apparatus.

Second, I contribute to the growing literature on the determinants of counterbalancing. Despite the existence of a vast literature on the effectiveness of counterbalancing in deterring coups (De Bruin 2018), little scholarly attention has been paid to the origins and development of a state’s coercive institutions. Outside of case studies (Frazer 1994; Greitens 2016), systematic examinations of the timing of the institutional (re-)design of the dictator’s security apparatus are lacking. Meanwhile, existing measures of ‘coup risk’ that try to capture the relationship between coups and counterbalancing are insufficient in gauging their direct causal relationship. The following section contextualizes my contribution within the broader literature. I then introduce and test my theorized mechanism, drawing on two examples to substantiate my findings.

Literature Review

From the perspective of coup-plotters, coups can be understood as an outcome of a decision-making process in which satisfaction with the incumbent leader is weighed against the perceived benefits and costs associated with the attempt (Feaver 1999). These costs include individual-level consequences, such as imprisonment or exile; group-level consequences, such as the abolition of the military; and the worst possible outcome of all – the escalation of a coup into a fratricidal conflict. Every conspirator wants to avoid their attempt from escalating into a violent confrontation and potentially a civil war (Powell 2012).

Contrary to common sense expectation that coups are intense battles akin to civil wars or foreign invasions, coups are better understood as ‘coordination games’ (Singh 2014). The goal of a coup is not to overpower the rest of the armed forces, but rather to quickly capture strategically important targets, such as symbolic sites and broadcasting stations, and convince other members of the military that the coup d’état is a ‘fait accompli’ – a process Singh (2014) calls ‘making a fact’. As the rest of the military would want to avoid fratricide, they will rally to whichever side they perceive will prevail, thus making coup success a self-fulfilling prophecy. Informational uncertainty is crucial for such endeavors, as coup-plotters can only shift expectation about the inevitability of their victory when other actors have incomplete information about their true strength (Geddes et al. 2018).

Counterbalancing changes this equation by increasing the associated costs of carrying out a coup (Quinlivan 1999). Within the literature, this challenge manifests itself in two arguments. First, it aggravates the collective action problem inherent in executing coups. Additional counterweights not only complicate coordination efforts by increasing the number of actors whom the coup-plotters have to either eliminate or seduce (Luttwak 1979), it also creates obstacles to cross-unit coordination among the coup-plotters themselves (Böhmelt and Pilster 2015). This leads to a greater risk of tactical errors, which impedes the seizure of coup targets, making it more difficult for coup-plotters to shape expectations about the coup’s likely outcome (De Bruin 2020).

Second, counterbalancing creates strong incentives for some soldiers to resist coup attempts (De Bruin 2018). Typically, parallel security forces, such as the presidential guard, are not only created by the dictator, but are also directly bound to him through personal or ethnic ties (take, for example, Muammar Gaddafi’s Revolutionary Guard). This creates a vested interest among the organization’s members to maintain the status quo and defend the regime in case of a coup (Escribà-Folch et al. 2020). Consequently, resistance during a coup attempt increases the likelihood it escalates into a violent confrontation, possibly a civil war, which should deter rational actors from staging one to begin with.

Empirical evidence on the effectiveness of counterbalancing, however, has been mixed. While some authors point to a partial deterrent effect (Albrecht and Eibl 2018; Powell 2012), others find a curvilinear one (Böhmelt and Pilster 2015), still others find no effect whatsoever (Albrecht 2015). Recent work by De Bruin (2020) seems to suggest that while counterbalancing has no effect on coup attempts, it reduces the likelihood a coup will succeed once underway.

Notwithstanding this mixed evidence, counterbalancing has been linked to a plethora of (unintended) consequences for autocratic regimes. One implication is that counterbalancing decreases the battlefield effectiveness of the dictator’s army. Not only does the creation of parallel military structures siphons away resources from the regular military and invests them into organizations whose primary responsibility is not to defend the state (Quinlivan 1999), but factionalizing the armed forces may also inhibit cross-unit coordination crucial for conducting complex military operations (Talmadge 2015). This translates into more battlefield deaths suffered in military engagements (Pilster and Böhmelt 2011), lower counter-insurgency capacity (Powell 2019), more indiscriminate use of violence against civilians (Greitens 2016) and the initiation of nuclear weapons programs to offset the loss of military deterrent capacity against foreign threats (Brown et al. 2016).

Despite these significant implications and the growing literature on the effectiveness of counterbalancing, little has been written on the determinants of counterbalancing itself. In light of this shortcoming, regime type (Escribà-Folch et al. 2020; Pilster and Böhmelt 2012) and colonial legacy (Mehrl and Choulis 2021) have emerged as key predictors of autocrats’ decision to fragment their militaries.

Others have looked at ‘coup risk’ as the main culprit behind the proliferation of counterbalancing efforts. The argument follows that as coup risk rises, the dictator is incentivized to protect himself from a potential coup by counterbalancing his military (Belkin and Schofer 2003, 2005). However, using an alternative measure of coup risk, Sudduth (2017b) challenges Belkin and Schofer’s (2003, 2005) findings and shows that counterbalancing actually decreases as coup risk rises given the potential that such efforts will be retaliated against by the military in a pre-emptive coup. Instead, dictators will fragment their security apparatus when coup risk is low: when the military is unable to resist the dictator’s opportunism (Sudduth 2017a).

That said, latent measures of coup risk, although highly nuanced, are insufficient at gauging the direct causal relationship between coups and institutional coup-proofing. Not only do they lead to inconsistent findings, but such aggregate measures are constructed from broad indices, such as strength of civil society, regime legitimacy and previous coups (Belkin and Schofer 2003), which are arguably selected or weighed arbitrarily, and are endogenous to counterbalancing itself.³ Newer measures are derived from estimating a model of coup success and then using the yearly predicted probabilities from that model as an estimate of coup risk (e.g., Boutton 2019; Sudduth 2017b). However, given that so few determinants of coups have been found to be statistically robust (Gassebner et al. 2016) and given that coup outcomes are argued to be quasi-random (Lachapelle 2020), models that try to predict either outcome tend to be underspecified. As such, the predicted probabilities derived from them tend to produce noisy measures with high levels of random error. Instead, a direct measure of coup failure might be more appropriate at testing the causal relationship between coups and counterbalancing.

When it comes to coup failure, a prominent debate that has emerged in recent years is whether coups can enhance the prospects for democracy in autocratic states. Proponents of the ‘democratic coup hypothesis’ argue that successful coups can provide an opportunity for democratization by forcefully extricating staunch autocrats from power (Kuehn 2017). Early empirical scholarship corroborates this proposition by finding that, at least in the post-Cold War era and in aid-dependent states, military seizure groups have sought to legitimize their intervention by opening up the political process and holding elections (Chacha and Powell 2018; Marinov and Goemans 2014; Thyne and Powell 2016).⁴ Less attention, however, has been given to the role of failed coups in this process, despite coup failure accounting for half of all overt coup attempts (Powell and Thyne 2011). The literature that does exist is sparse.

On the one hand, Thyne and Powell (2016) maintain that in the aftermath of failed coups, dictators, vulnerable to future attempts to unseat them (as surviving a coup in no way guarantees political stability), will adopt liberalizing reforms with the intention of legitimizing their rule. As advocated by the authors, “leaders will realize that the best path to continued survival includes strengthening the economy, establishing political legitimacy, and improving relations with the outside world. Each of these is best achieved through opening the political process to the population” (199). On the other hand, skeptics have argued that failed coups are conducive to lengthening autocratic tenure and further entrenching rulers whose position was not as secure prior. Bokobza et al. (2022) find that failed coups are associated with extensive cabinet reshuffling, while Easton and Siverson (2018) demonstrate that such post-coup purges tend to prolong a dictator’s tenure. More recently, using a latent measure of regime personalism, Timoneda et al. (2023) find that failed coups lead to a sharp increase in the dictator’s mean personalism score, which reflects the surviving autocrat’s proclivity to abolish institutional constraints, purge rivals and personally appoint cabinet positions.

Taking this debate further requires disaggregating latent concepts, such as personalism, and specifying the mechanisms underpinning regime transformations. Although purges have emerged as the go-to instrument of autocratic power consolidation (Bokobza et al. 2022), less attention has been paid to institutional coup-proofing in this process. Likewise, whereas purges happen in ‘the morning after’ failed attempts, the long-term structural consequences of failed coups, particularly when it comes to the dictator’s security apparatus, have been neglected. Timoneda et al. (2023) do disaggregate their analysis and look at the effect of failed coups on the personalization of security forces. However, security force personalization is conceptually and empirically distinct from its fragmentation, which warrants a separate analysis.

The literature that does exist on security force fragmentation is largely case-based and links the emergence of autocratic coercive institutions to either the ‘dominant perceived threat’ autocrats face at the time of their incumbency (Greitens 2016) or shared colonial struggle (Frazer 1994). This begets more systematic theorization and analysis.

A Theory of Failed Coups and Counterbalancing

The goal of all autocratic leaders is to survive in office (Bueno de Mesquita et al. 2003). To achieve this goal, they enter into power-sharing arrangements with members of the elite whose support is necessary for their survival. This group is known as the ruling coalition (Svolik 2009).

In any authoritarian regime, high-ranking military officials are an indispensable part of the ruling coalition. They are in control of the state’s coercive apparatus, so securing their acquiescence is necessary to prevent insider coups (Meng and Paine 2022). Power-sharing between the dictator and his military elite takes the form of a quid-pro-quo arrangement where in exchange for pivotal cabinet positions, autonomy from oversight, and greater say in political decision-making, officers promise not to oust the leader (Brooks and White 2022).

However, this brings attention to the fundamental problem of authoritarian power-sharing: the lack of an independent authority that could enforce power-sharing agreements between the key political actors (Svolik 2012). There are no guarantees that the military elite will not use its access to state resources to overthrow the dictator, which incentivizes the autocrat to accumulate power whenever the opportunity to do so arises. In an ideal world for autocrats, the intra-regime balance of power is such that the ruling elite can no longer credibly threaten the dictator with his removal, but rather finds itself subordinate to his whims.

In order to achieve this, dictators must make a series of risky power grabs, which entails institutionally coup-proofing their military to reduce its coup-making potential. Such efforts, however, entail significant risks. Literature on civil-military relations has long argued that civilian intervention in military affairs strongly encourages coup attempts by military officers seeking to preserve their autonomy and corporate interests (Thompson 1973).

Dividing the military into rivaling organizations or creating additional paramilitary units hurts these interests as it challenges the military’s monopoly over the use of violence and depletes its budget (Geddes et al. 2018). Furthermore, as removal via coup is the only means by which the ruling coalition can keep its leaders accountable and commit resources to their benefit (Svolik 2009, 2012), it is crucial that the military elite stop autocrats’ coup-proofing efforts by removing them beforehand. This motivates senior officers to intervene pre-emptively.

In light of this, Svolik (2012) argues that power aggrandizement is likely to occur under the guise of secrecy – when such efforts go unnoticed by other elites. However, it is highly unlikely that the creation of an additional security force would go unnoticed by the military. Instead, more recent evidence seems to suggest that dictators are more likely to coup-proof their regimes when such efforts go unretaliated (Bokobza et al. 2022; Sudduth 2017a, 2017b). This happens “when the probability that elites can successfully remove the dictator via coup is temporarily low” (Sudduth 2017a, 7). For Sudduth (2017a), this occurs in the immediate aftermath of a successful coup – when the number of elites who are willing to challenge the new dictator is small and their capacity to coordinate against him is low. However, this argument can also be extended to failed coups.

A failed coup serves as an internal shock to the regime, which temporarily shifts the balance of power between the dictator and his ruling coalition in favor of the incumbent. Momentarily finding himself in a position of power vis-à-vis his military elite, the dictator has both an incentive and an opportunity to lock in institutional changes to the security apparatus, while the military elite is unable or unwilling to effectively resist such coup-proofing efforts. The following sections unravel these two logics in more detail.

Incentive

First, a failed coup signals to the dictator that his position is not sufficiently entrenched and that existing arrangements with the military are not efficient in deterring rebellion (Timoneda et al. 2023). As such, the incumbent has an increased incentive to prevent a repetition of such attempts by further consolidating his power.

One way in which consolidation occurs is through post-coup purges (Bokobza et al. 2022; Easton and Siverson 2018). Purges, however, are limited for two reasons. First, they are short-term remedies that punish immediate belligerents without addressing the underlying institutional incentives for repeated challenges, such as the presence of a strong military deeply embedded into the politics of the state (Bratton and Van de Walle 1997). Second, failed coups might not effectively expose the full internal opposition to the regime. While those holding guns might be punished, undetected co-conspirators and sympathizers can simply fade into the crowd of regime officials and bide their time for another opportunity (Woldense 2022).

In light of these limitations, autocrats will seek to consolidate power in the longer-term by undermining established political and military institutions via a process Baturo and Elkink (2021) call ‘deinstitutionalization’. This entails two strategies: (1) increasing personal control over the institution in question (i.e., Timoneda et al. 2023) and (2) reducing the effectiveness of the institution in question by introducing an element of deliberate redundancy (80). The latter speaks to the logic of counterbalancing.

Some have argued that failed coups may lead to an extension of power-sharing instead. For example, Thyne and Powell (2016) argue that dictators, vulnerable to future efforts to unseat them, would seek to legitimize their rule by opening up the political process, which entails further redistributing power to the ruling coalition. Such tactics, however, instead of addressing the underlying commitment problem (the lack of a third-party enforcement mechanism) that led to the initial bargaining failure, simply enhance the elite’s ability to coordinate against the dictator and more credibly threaten him with his removal down the line.

Moreover, the discussion so far has assumed that the military is a unitary actor, reflecting the interests of its senior officers. Although this is the case in professionalized militaries, where individuals’ career success is inextricably bound to following orders from their commanders, a common occurrence of the past century has been the emergence of factionalized militaries. In such forces, personal, partisan or ethnic loyalties can crosscut the military’s strict hierarchy and result in ‘rogue’ coups organized by junior officers or faction leaders seeking a faster way to promotion. With factionalized militaries, senior officers included in the ruling coalition cannot commit their subordinates to abide by the power-sharing bargain as unconditional obedience to a strict hierarchy is absent (Geddes et al. 2018, 100–101). Consequently, the dictator has little incentive to share power with his senior officers if they cannot credibly guarantee the obedience of their lower ranks. Instead, the autocrat will seek to undermine the ability of rogue factions to launch coups by fracturing the military apparatus.

Opportunity

Arguably, rulers always have an incentive to consolidate power, however, they might not be able to act out on this incentive save for the presence of distinct opportunities to do so. Failed coups provide this opportunity by reducing the military’s ability and willingness to strike again.

Regarding ability, failed coups weaken the organizational capacity of remaining rivals to stage a repeat attempt. Informational incompleteness is paramount to a successful coup (Singh 2014). As opponents to the attempt are revealed and military factions choose sides, the same level of informational uncertainty necessary for coup success is absent in its aftermath. By exposing military allegiances and loyalties, failed coups undermine remaining rivals’ ability to shape the narrative surrounding their likely success in any repeat attempt. This deters dissatisfied elites from launching another attack. Furthermore, failed coups enable the dictator to rat out and eliminate rivals found to have joined or supported a coup through extensive purges (Bokobza et al. 2022; Easton and Siverson 2018), which further depletes their organizational capacity.

Regarding willingness, a new post-coup environment reshapes the incentive structure of those elites who did not join the conspiracy (Timoneda et al. 2023). A failed coup informs these elites about the possibility of another faction’s takeover and, in turn, the risk of them losing their privileged position inside the ruling coalition. This risk induces loyalty towards the incumbent, making them more likely to acquiesce to the dictator’s power grabs since remaining inside the inner circle leaves them better off than being excluded or risking failure in a new challenge (ibid., 7). Likewise, purges in the aftermath of failed coups lead to the narrowing of the dictator’s ruling coalition. This increases the loyalty norm among the remaining backers as it (1) increases the share of spoils allocated to them (as there are now fewer rivals to share patronage rents with) and (2) increases the chance of their exclusion from a subsequent winning coalition if a repeat coup were to succeed (Bueno de Mesquita et al. 2003)⁵.

By diminishing rivals’ organizational capacity and weakening the remaining elites’ willingness to challenge the autocrat, failed coups provide a window of opportunity for dictators to lock in long-term organizational reforms to the security apparatus that transcend immediate purges, while risking little retaliation from the military elite. Purges likely play a major role in ‘softening up’ the ruling coalition to such changes and are thus likely to temporally precede counterbalancing. This suggests a longer time horizon for the strategy to be implemented, but once counterweights to the military are conceived, they take a path-dependent nature towards their materialization. For example, while the creation of the Internal Security Unit in Sierra Leone was announced 11 days after the failed coup, on 4 April 1971, the new security unit became a full member of the regime’s coercive apparatus only after completing its training by late 1972 (Turay and Abraham 1987, 150).

To summarize, with the coup plotters’ ability to resist the ruler eliminated and the remaining elites’ willingness to do so diminished, dictators are offered a rare opportunity to aggrandize power by setting up parallel security organizations for their protection. In contrast to arguments emphasizing the extension of power-sharing in the aftermath of failed coups (Thyne and Powell 2016), I argue that the unique opportunity presented by the post-coup environment is more conducive to institutional coup-proofing instead, in which case failed coups should lead to greater counterbalancing efforts.

Failed military coups lead to an increase in counterbalancing.

Research Design, Data and Methodology

To test the hypothesis, I combine a large-n quantitative analysis with two illustrative examples of the theorized mechanism. Unlike other studies that have used the quasi-random property of coup outcomes (coup success vs. coup failure) for causal identification (e.g. Bennett et al. 2021; Lachapelle 2020), the conditional independence assumption necessary for identification in the case of counterbalancing might be violated due to simultaneity bias (the strategy affects coup outcomes). Instead, I assume selection on observables and treat years prior to failed coups as the control.

I use panel data with autocratic leader-years as the unit of analysis. I derive my list of authoritarian leaders from the Geddes et al. (2018) data on authoritarian regimes. This yields a total of 3,899 unique leader-year observations across the period of 1960–2010, featuring 243 autocratic regimes with 482 leader-spells. A leader-spell is defined as the number of sequential years a given autocrat stays in power. The average length of a leader-spell is 9 years. Finally, leader-spells inform the group-level at which the treatment is applied.

Counterbalancing

To date, the most comprehensive data collection effort on counterbalancing has been carried out by De Bruin (2020). As such, for the outcome variable, I rely on her State Security Forces dataset, which codes measures for 375 security forces in 110 countries between 1960 and 2010. The author defines a security force as a ground combat-capable force with a concrete leader and name that is under the administrative control of a recognized state beyond the army, navy and air force that comprise a traditional military (De Bruin 2021, 317).

A security force is coded as a ‘counterweight’ to the military if it is: (1) stationed within 60 miles (100 km) of the capital and (2) is placed outside of the normal (military) chain of command (De Bruin 2021, 318). Security forces that have capital deployment potential are relevant to coups as they are in the immediate vicinity of coup targets, while independence from the military is crucial for carrying out autonomous operations during an attempt (Luttwak 1979). Such security forces are thus explicitly created or have the ability to counterbalance the military.

To reduce skewness and heteroskedasticity in the count data, I use the log-transformed number of counterweights available in each autocratic leader-year (‘Log Counterweights’) as the main dependent variable. As an alternative measure, I use an operationalization of whether in a given leader-year at least one counterweight was present (coded as 1, otherwise 0). As argued by De Bruin (2018), “because coups can be staged (and prevented) by even a small number of men, we should expect the logic of counterbalancing to hold no matter the size of the counterweight” (1,441). This makes the dichotomous measure (‘One Counterweight’) sufficient for testing the hypothesis. For additional robustness, I also include a measure of whether in a given year, the dictator employs an above-average number of counterweights (‘Mean Counterbalancing’). This operationalization is used amply by other authors who look at counterbalancing (Escribà-Folch et al. 2020; Mehrl and Choulis 2021) and is coded a 1 if the number of counterweights in a given leader-year exceeds the sample average of 1.468 and zero if not.

The mean log number of counterweights available in each year is 0.77 with a standard deviation of 0.51, a minimum of zero and a maximum of 2.30. 76 percent of all observations are coded as having at least one counterweight, whereas 44 percent of all autocratic leader-years have an above-average number of counterweights.

Failed Coups

For a measure of failed coups, I rely on the Colpus Dataset, which categorizes coup attempts into military and non-military categories, with the former requiring the participation of “at least one current and active member of the regime’s military and/or security apparatus” (Chin et al. 2021, 3). Although a stretch from Thompson’s (1973) definition introduced prior, Colpus offers the most comprehensive data on military coups to date. The authors rely on an unprecedented amount of documentation to justify, provide transparency for and bolster confidence in their coding of coup events. Alternative datasets either fail to distinguish military coups from coups organized by other within-regime elites (Powell and Thyne 2011), do not distinguish between successful and failed coups (Belkin and Schofer 2005), or provide a definition of coups that does not target the sitting executive (Buddy et al., 2025). That said, I reproduce my findings using the widely cited Powell and Thyne (2011) (PT) coup dataset, as well.

A failed coup is broadly defined by Colpus as a coup event in which the regime leader does not lose power during the calendar dates of the event (Chin et al. 2021, 11). If multiple failed coups occur in a single leader-year, they are coded as a single event. For the period 1960 to 2010, 129 such events are identified. Figure 2 shows that despite declining in frequency since the 1970s, there have only been 7 years that did not experience a failed coup in the sample.

Figure 2.

Failed coups time trend.

Treatment

The main independent variable of interest is ‘Post-Fail Coup’, which is coded as zero for all leader-year observations that have not experienced a failed coup prior and 1 for all observations after a failed coup within a given leader-spell. If more than one failed coup occurs, these observations are also coded as 1 until the end of that leader-spell. The treatment is coded at the group-level, which allows one to distinguish between three groups. First, the ‘never-treated’, or leader-spells that have never experienced a failed coup and are thus coded as a zero throughout. Second, the ‘untreated’, or observations within leader-spells prior to the introduction of the treatment (also coded as 0) and third, the ‘treated’ – observations within leader-spells after the introduction of the treatment (coded as 1). The treatment itself is a binary indicator of whether a given leader-year experienced a failed coup.

Within the sample, 2,879 observations are coded as never-treated (73.8 percent), 296 observations are coded as untreated (7.6 percent) and 724 are coded as treated (18.6 percent). Figure 3 depicts the average number of counterweights by treatment status. While the never-treated and untreated units have an average of 1.381 and 1.422 counterweights respectively, treated units have an average of 1.778 counterweights. A basic two-tailed t-test reveals that this difference is statistically significant at the 1 percent significance level (t-score of 2.701). While this preliminary test provides some evidence in favor of the hypothesis, it is likely to suffer from confounding bias. A more robust empirical strategy is thus necessary to gauge the causal effect.

Figure 3.

Average number of counterweights by treatment status.

Empirical Strategy

To test the hypothesis empirically, I rely on a difference-in-differences (DID) design. Drawing on the potential outcomes framework, the estimand is the average treatment effect on the treated, or the difference in the potential outcome under treatment post-treatment, minus the potential outcome under control post-treatment. As one cannot observe the latter, DID imputes the missing potential outcome under control through parallel trends – the idea being that the never-treated group can act as a counterfactual to the unobserved control as long it behaves in a similar manner to the control in the pre-treatment period. For causal identification, DID requires the following assumptions to be satisfied.

First, the treatment must be conditionally independent of the outcome. This means that previous counterbalancing efforts should not determine when a failed attempt occurs (conditional on covariates). Although it is difficult to rule out sources of endogeneity, such as anticipation and reverse causality,⁶ it is possible to account for short-run anticipation effects by including a 1-year lead treatment indicator in the static DID model. Doing so partials out any confounding from treatment effects that manifest before actual treatment introduction and simultaneously tests for the presence of 1-year anticipation. Additionally, the absence of significant pre-trends in the event study models of the following section provides further reassurance that longer-run anticipation and reverse causality are unlikely to bias the estimated post-treatment effects.

The second assumption is parallel trends. Figure 4 provides descriptive evidence for its plausibility by plotting yearly averages of counterweights in the treatment group (untreated and treated observations) against a synthetic control group generated from the pool of never-treated observations for each failed coup stacked in event-time.⁷ The figure shows similar pre-trends that diverge sharply only in the post-treatment period. That said, pre-treatment similarity is neither necessary nor sufficient to guarantee the existence of a parallel counterfactual – a definitive test does not exist. To account for potential violations, I employ the following (DID) specifications:

1. Naïve OLS

{C B}_{i g t} = β {P o s t F a i l}_{i g t} + μ_{g} + ε_{i g t}

where

i

indexes each unit within group

g

in period

t

, while

μ

captures group-level fixed effects and

ε

is the error term. Given that the treatment is coded at the group level, all models include group-level fixed-effects. The purpose of this Naïve estimator is to make sure that the effect is not spuriously generated through the addition of extra parameters.

2. Two-Way Fixed-Effects (TWFE)

{C B}_{i g t} = β {P o s t F a i l}_{i g t} + {γ L e a d}_{1 i g t} + {α X}_{i g t - 1} + μ_{g} + ρ_{t} + ε_{i g t}

where

i

indexes each unit within group

g

in period

t

X

is the vector of all time-varying covariates,

μ

captures group-level fixed-effects;

ρ

captures period or time fixed-effects and

ε

is the error term.

{L e a d}_{1}

captures 1-year anticipation effects. The second model is the baseline TWFE model with group and period fixed-effects, as well as covariates lagged by a year.

3. Two-Stage Difference-in-Differences (2SDID)

First Stage : {C B}_{i g t} = {γ L e a d}_{1 i g t} + {α X}_{i g t - 1} + μ_{g} + {ρ_{t} + ε}_{i g t}

Mid - Stage : {\hat{C B}}_{i g t} = {C B}_{i g t} - \hat{μ}_{g} - {\hat{ρ}}_{t}

Second Stage : {\hat{C B}}_{i g t} = β {P o s t F a i l}_{i g t} + {γ L e a d}_{1 i g t} + {α X}_{i g t - 1} {+ ε}_{i g t}

where

i

indexes each unit within group

g

in year

t

X

is the vector of all time-varying covariates;

{L e a d}_{1}

captures 1-year anticipation effects;

μ

captures group-level fixed-effects;

ρ

captures period fixed-effects and

ε

is the error term. One of the assumptions of DID is that treatment introduction is homogenous across groups. If this is not the case, then some variation in the outcome that is due to the treatment will be absorbed by group and period fixed-effects in the TWFE model (Goodman-Bacon 2021). To adjust for this negative bias, Gardner (2021) proposes an intuitive two-stage DID estimator. It works by residualizing the outcome for group and period fixed-effects using the sub-sample of untreated and never-treated observations (

{P o s t F a i l}_{i g t} = 0

) in the first stage and then regressing the residualized outcome on the treatment for the full sample in the second stage. As is best practice, I add time-varying covariates to both the first and second stage of the estimator (Gardner 2021). Across all specifications,

β

is the parameter to be estimated.

Figure 4.

Descriptive parallel trends.

For the binary outcome variables, I rely on Linear Probability Models (LPMs) to estimate the main effect. LPMs are selected over logistic regression due to their ease of interpretation, the fact that they outperform logistic models in rare-events data (Timoneda 2021) and rarely yield out-of-bounds [0–1] predictions when the target quantity is the average causal effect of the treatment on a binary outcome (Gomila 2021).

To address concerns stemming from omitted variable bias, I add multiple time-varying covariates to all relevant models. This includes all potential confounders of coup attempts and counterbalancing found to be relevant in the literature, such as ‘GDP per Capita’, ‘Economic Growth’, ‘Oil Rents’, ‘Latent Personalism’, ‘Military Spending per Soldier’, ‘Interstate’ and ‘Intrastate’ wars, the proportion of ethnic population excluded from power (‘Excluded Pop.’), ethnic heterogeneity in the military (‘Ethnic Military’), and ‘Defense Pacts’ between states.

Finally, I include a measure of ‘Leader Tenure’, or the number of sequential years an autocrat stays in power, and ‘Time Since Last Successful Coup’ as additional time-varying controls. Table A1 in the online appendix provides explanations and descriptive statistics for all relevant variables. To deal with missingness, I impute missing values for all covariates using a random forest imputation algorithm in R. I cluster all standard errors by leader-spell and lag all time-varying covariates (except leader tenure and time since last coup) by a year to reduce the likelihood of post-treatment bias.

Results

Table 1 shows the results for the nine specifications in which I estimate the effect of failed coups on counterbalancing. The table suggests that the treatment has a positive effect on the coup-proofing strategy across all specifications. Looking at the main dependent variable (Models 1–3), failed coups are associated with a 0.13–0.23 log-point increase in the log number of counterweights available to autocrats in each year. Coup failure also leads to an 8.6–20.7 percent point increase in the likelihood the dictator establishes at least one counterweight (Models 4–6) and an 18–25.3 percent point increase in the likelihood he adopts an above-average number of counterweights (Models 7–9). Furthermore, the anticipation indicator fails to reach conventional significance across all relevant specifications, which suggests the absence of observable treatment effects in the year prior to treatment introduction.

Table 1.

Results for the effect of failed coups on counterbalancing (Colpus)

	Log counterweights			One counterweight			Mean counterbalancing
	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6	Model 7	Model 8	Model 9
	Naïve	TWFE	2SDID	Naïve	TWFE	2SDID	Naïve	TWFE	2SDID
Post-fail coup	0.134* (0.054)	0.136+ (0.081)	0.229* (0.099)	0.086* (0.037)	0.084+ (0.046)	0.207* (0.089)	0.180** (0.063)	0.188* (0.094)	0.253* (0.118)
Anticipation		0.016 (0.059)	−0.010 (0.011)		−0.036 (0.048)	−0.013 (0.010)		0.056 (0.069)	−0.005 (0.013)
Covariates	✗	✓	✓	✗	✓	✓	✗	✓	✓
Group FE	✓	✓	✓	✓	✓	✓	✓	✓	✓
Year FE	✗	✓	✓	✗	✓	✓	✗	✓	✓
N	2,731	2,410	2,142	2,731	2,410	2,142	2,731	2,410	2,142
Adjusted R²	0.910	0.920	0.167	0.880	0.893	0.143	0.862	0.872	0.161

Note. **p < 0.01; *p < 0.05; + p < 0.1. All models show OLS estimates. Standard errors, clustered by leader-spell, are in parentheses. For full results see Table A4.1 in the online appendix.

After exponentiating the log coefficient, the Naïve estimator suggests that failed coups are associated with a 14.3 percent increase in the number of counterweights employed by the dictator in a given year. This rises to 14.5 percent for the TWFE and 25.7 percent for the 2SDID models. The likelihood of having at least one counterweight increases by 8.6 percent points (Naïve), 8.5 percent points (TWFE) and 20.7 percent points (2SDID) in the aftermath of a failed coup, while the likelihood of having an above-average number of counterweights increases by 18 percent points (Naïve), 18.7 percent points (TWFE) and 25.3 percent points (2SDID). All estimates reach conventional statistical significance of at least 10 percent, with the majority being significant at the 5 percent level.

In addition to the baseline results, I estimate a series of event-study models to assess dynamic treatment effects under staggered treatment adoption. I use the ‘heterogeneity-robust’ 2SDID estimator with leads and lags of up to 9 years. Figures 5 –7 serve as further evidence for the existence of parallel trends as the coefficients in the pre-treatment period fail to reach conventional significance levels. This means that the difference-in-differences between the treatment and control groups followed a similar trend prior to treatment and that divergence in the post-treatment period is very likely attributed to treatment introduction (Cunningham 2021). Likewise, no significant pre-trends also implies that the outcome is not causing the treatment prior to treatment introduction, which is suggestive of the absence of observed anticipation.

Figure 5.

Event study: Staggered treatment – log counterweights.

Figure 6.

Event study: Staggered treatment – one counterweight.

Figure 7.

Event study: Staggered treatment – mean counterbalancing.

Departing from the event study models, although the estimates jump in the immediate aftermath of failed coups, they obtain significance only 2 years later. This likely reflects a delayed process of institutional creation – a pattern consistent with my theoretical expectations. Likewise, the effect seems to grow the longer the autocrat stays in power, which provides evidence for a gradual process of institutional consolidation with time since failed coup. Consequently, I fail to reject my main hypothesis. Instead, I find a significant effect of failed coups on counterbalancing in autocratic regimes that increases with leader tenure.

Robustness Checks and Limitations

Table A4.2 in the online appendix presents the results of my analysis using the PT coding of coups. While the PT data largely replicates my findings, it does not achieve conventional significance in certain specifications, likely due to differences in how coups are operationalized. Additionally, my findings are robust to logistic regression (Table A6.3), alternative dependent variables (Table A6.4), alternative DID estimators (Figures A3.1-9), and dynamic panel models that account for reverse causality, including lagged dependent variable and Arellano-Bond GMM estimators (Table A5). Finally, my results are robust to clustering standard errors at both the regime and country-levels, accounting for the implicitly nested structure of the data (Tables A6.1-2).

That said, limitations to my empirical strategy are apparent. First, the timing and location of failed coups are not randomly assigned, which means that causal identification relies on selection on observables. This opens up the estimation to the presence of time-varying unobserved heterogeneity, as the treatment may be correlated with unobserved confounders.

Second, the identifying assumption necessitates that the treatment is independent of the outcome in the pre-treatment period conditional on the inclusion of confounders (Cunningham 2021). Given the observational nature of the data, it is impossible to fully account for reverse causality and anticipation, especially if it is unobserved. Fears that the dictator might fragment the military in the future may cause the armed forces to step in and intervene pre-emptively (Sudduth 2017a), while reverse causality may happen contemporaneously if failed coups occur within the same leader-year of establishing a new counterweight. Given these limitations, I probe the plausibility of the theorized mechanism through two illustrative examples of failed coups.

Case Studies

The two examples I leverage are (1) the 1971 failed coup in Sierra Leone and (2) the 2016 failed coup in Turkey. Whereas the former is a within-sample historical illustration of how failed coups can lead to the creation of particular counterweights, the latter is an out-of-sample, timelier, case meant to highlight the broader implications of failed coups for the ruler’s coercive apparatus.

March 1971 Coup in Sierra Leone

Siaka Stevens and his All People’s Congress (APC) came to power in April 1968 following a mutiny of the lower ranks against the Southern ethnic Mende-dominated military government. Shortly after taking power, Stevens, a Northern Limba, began ethnically reshaping the military by pensioning off Mende officers and stacking the armed forces with officers of Northern ethnic decent. This included the promotion of Brigadier John Bangura to Force Commander in May 1969, who hailed from the largest Northern ethnic contingent – the Temne (Horowitz 1985, 475–476). Within 18 months of Stevens’ rule, the Mende share of officers was halved (Harris 2014, 64). More than a third of the officer corps came to consist of Temne and another 30 percent of Limba and Koranko Northerners, who had earlier formed an insignificant part of the total officer body (Horowitz 1985, 477).

Stevens’ policy of ethnic stacking, however, led to fears among the Temne that they were being slighted in favor of Creoles (putative descendants of slaves) and other ethnic minorities (Harkness 2018, 120). These fears manifested when Stevens began to bypass Bangura for key consultations, favoring instead his immediate subordinates, including Colonel Joseph Momoh (a Limba) and Lieutenant Colonel Samuel King (a Creole) (Vidler 1998, 209).

In response, Temne officers launched a coup on 23 March 1971. However, the plot, led by Bangura and Major Falawa Jawara, did not find support among the loyalist factions of the military and was put down in a successful counteroffensive within 24 hours of its onset. On 26 March, purges of those involved ensued. Bangura and Jawara were detained 2 days later, tried in May and executed on 29 June 1971 (Harkness 2018, 121).

In the immediate aftermath of the coup attempt, Stevens took decisive steps to consolidate his power. Having lost confidence in his army, he signed a defense pact with Guinean President Sékou Touré on 28 March, requesting for an additional 200 Guinean soldiers to be stationed in Sierra Leone. Meanwhile, Stevens sought out new, internal, ways of controlling the military and neutralizing army opposition. This culminated in the creation of the Internal Security Unit (ISU) by late 1972 (Turay and Abraham 1987, 150).

The proposal to create a new security unit came only 11 days after the abortive coup, on 4 April 1971 (Vidler 1998, 203). Trained by Cuban, British and Israeli special forces the ISU was meant to replace the Guinean military presence in the country following its departure in 1973 as the sole unit responsible for Stevens’ personal protection (Krogstad 2012). The loyalty of the forces was largely guaranteed as it was comprised of young APC supporters from the North, many of whom were Limba and had kinship ties to the party leadership. According to Horowitz (1985), Stevens “recruited heavily from Kambia, [his] home district” (552). The ISU operated under Stevens’ direct command and in addition to countering the military, it was used as a repressive agent to intimidate his opponents (Harris 2014).

The main incentive for creating the new security force was the failed coup. As discussed by Vidler (1998), “it was designed as a deterrent to future coup plots or, if failing that, as a means of defeating attempts at intervention [by] breaking the military’s monopoly on the instruments of coercion [and] putting the army ‘in its place’” (203). As noted by Stevens himself, he intended to “put an end to this nonsense [of attempted coups] by rapidly reducing the importance of the army (cited in Vidler 1998, 261). Almost immediately, the military resented its new rival, which culminated in low-level skirmishes between the two forces (Turay and Abraham 1987, 155). However, this did not result in another coup attempt, which can be explained by the strategic timing of the ISU’s creation.

The failed coup provided Stevens with a pretext to purge a large part of the Temne component from the army at minimal risk, which depleted the organizational capacity of any remaining rivals in the military establishment. It also allowed him to restructure the army so that all commanders were either Creole, Limba or Koranko (Horowitz 1985). Momoh was appointed Force Commander in September 1971 given his role in providing intelligence to Stevens during the coup attempt, while King became his deputy for his role in the successful counteroperation (Turay and Abraham 1987, 153). Consequently, none of the top brass challenged the creation of the ISU due to their personal loyalty to Stevens and their knowledge that other senior officers, as well as a large part of the rank-and-file supported him. The coup had revealed their allegiances.

Additionally, the presence of Guinean forces played a key role in dissuading the army from challenging Stevens’ decisions. The dictator also undertook several reconciliatory gestures towards the armed forces, which included making a public statement lauding its role in the abortive coup and ensuring its supremacy over the ISU. This was followed by huge increases in military expenditure, which convinced officers to turn a blind eye to the creation of its rival (Turay and Abraham 1987, 155).

Although Stevens had previously flirted with the idea of creating a youth militia as a counterweight to the army (Vidler 1998, 203), such efforts never materialized. It was only after the failed coup that efforts to fragment the security apparatus were undertaken. Intense purges of the Temne as well as the execution of Bangura and Jawara depleted the military’s mobilizational capacity on grounds of ethnic grievances, while the co-optation of loyalist officers combined with other appeasement strategies decreased the willingness of a repeat coup. The post-coup environment thus enabled the creation of the ISU with little resistance from the weakened and unwilling military.

July 2016 Coup in Turkey

Having secured victory in the November 2002 elections, leader of the Islamist Justice and Development Party (AKP), Recep Tayyip Erdoğan became Prime Minister of Turkey in March 2003. The AKP’s first order of priority was to reduce the military’s traditional influence over politics in order to consolidate its rule (Kaynar 2022b). Indeed, the Turkish Armed Forces (TSK), under the mandate of its dual guardianship role,⁸ had previously intervened via coup in 1980 and 1997 against openly Islamist parties with the intention of restoring secular government (Kutay 2016). Fortunately for the AKP, two developments coincided with its rise that allowed for a radical transformation of Turkish civil-military relations.

First was Turkey’s negotiations over accession to the European Union, which mandated the normalization of civil-military relations under the ‘democratic control of the armed forces’ initiative. Following the 2004, 2006 and 2010 EU reform packages, constitutional changes were introduced to the structure of the military, which drastically curbed its role as a veto player in politics (Erdag 2019). Second, Erdoğan found common cause with Fethullah Gulen, founder of the ‘Hizmet’ (Service), a transnational religious and social movement (Phillips 2017). During the early years of Erdoğan’s rule, both actors collaborated to erode the military’s popularity among the public. Gulen instructed his followers to infiltrate the security apparatus and falsify evidence against to the army, which led to a series of sham trials that irreversibly damaged the army’s reputation (Yavuz and Koç 2018, 40). The most famous of these trials were the Ergenekon and the Sledgehammer trials, which discredited army commanders as terrorists, sentencing 325 officers to imprisonment for 13–20 years (Kaynar 2022a, 386).

By the end of 2010, however, deep ideological divisions began fermenting between Erdoğan and Gulen, leading both to be at odds with each other over the control of key government positions and state resources (Yavuz and Koç 2018). In December 2013, Gulenist prosecutors launched a graft probe against Erdoğan’s close ministers and family members. Erdoğan retaliated with cabinet purges (Satana 2022). Believing it was their last chance to stop Erdoğan’s attempt to purge the military as well, the Gulenists wooed many dissatisfied TSK officers to join them in a bloody coup attempt that took the lives of over 250 and injured more than 2,700 (Yavuz and Koç 2018, 87).

What followed the failed coup was the largest purge of military and public officials in Turkish history. In the immediate aftermath, a state of emergency was declared in Turkey, which was supposed to last for 3 months, but was retained for 2 years (Kaynar 2022a). Under the state of emergency, the government detained 40,000 individuals and arrested 163 generals, which accounted for 45 percent of the military’s total (Kaynar 2022b). On 16 April 2017, Erdoğan held a referendum to change the political system from a parliamentary to a presidential republic and was sworn in as executive president in 2018.

Aside from purges, both the state of emergency and Erdoğan’s first years as president saw changes to the institutional design of the armed forces. Counterbalancing formed the core of this re-design. First, the general commands of the Gendarmerie and the Coastal Guard were removed from the armed forces and brought under the control of the Minister of the Interior (Erdag 2019, 61). Both forces were subsequently redefined as general law enforcement bodies, establishing them as counterweights to the TSK – given their distinct command structures (Kaynar 2022a).

Second, paramilitary forces, like the Village Guards and the Border Guards were re-established and granted extensive powers. In addition, a new police unit called the Auxiliary Task Force Police Unit was created and first deployed in the capital in July 2018 and later in Istanbul in 2020. Composed of 500 Rapid Force personnel, the purpose of the mobile unit was to respond to emergencies, ranging from natural disasters to military coups. Legislation passed in January 2021 solidified the police force, including Auxiliary Task Force, as a paramilitary organization (Kaynar 2022a, 394).

Third, new security forces with ties to the President were created. The Presidential Guard Regiment, which was previously in charge of directly protecting the president and his family was abolished following its participation in the July 2016 coup attempt. In its place, on 10 July 2018 Presidential Decree No. 1 founded the Department of Administrative Affairs of the Presidency and on 19 January 2019, the General Directorate of Security Services was established (Kaynar 2022a, 395). Both were tasked with directly protecting the president. Subsequently, Erdoğan had both an incentive and an opportunity to reshape the armed forces.

In the aftermath of the coup, Erdoğan publicly remarked that “this uprising is a gift from God to us because this will be a reason to cleanse our army” (cited in Dettmer 2016). Given that the coup was organized by a rogue faction within the military, extending power-sharing and democratizing civil-military relations made little sense for Erdoğan’s personal protection. Instead, fragmenting the military apparatus and deterring a repetition of such attempts proved to be the favored strategy. In doing so, Erdoğan met little opposition from other political parties and the military establishment at large. Given that the Gulenists had previously hurt many other factions within society, opposition parties acquiesced to Erdoğan’s extensive purges. By the time of his executive presidency in 2018, Erdoğan managed to consolidate enough power that even if such resistance was present, it would not have mattered (Satana 2022). Likewise, Erdoğan portrayed the coup-plotters not as military officers, or even proper members of the armed forces, but rather as members of the ‘Fethullahist Terrorist Organization’. This allowed many high-ranking officers to save face, making them more than likely to acquiesce to Erdoğan’s restructuring of the military in order to remain part of the ruling elite (Kaynar 2022b, 9).

Consequently, by reshaping the security apparatus in his image, Erdoğan removed the military’s historic role as a check and balance to executive power aggrandizement, allowing the president to consolidate immense power and transform the military tutelage of the previous century into that of a personalist strongman regime.

Conclusion

How do dictators successfully counterbalance their militaries despite the significant risk of military retaliation? I argue that dictators are more likely to counterbalance in the aftermath of failed coups, when the military is weakened and unwilling to challenge the dictator in a repeat attempt. This provides the autocrat with an increased incentive and opportunity to fragment his coercive apparatus without risking retaliation.

Empirical findings support this proposition: failed coups have a significant effect on counterbalancing, illustrating that coup failure not only prompts short-term purges and regime personalization, but also facilitates longer-term fragmentation of autocratic coercive institutions. Consequently, by expanding on the relationship between coup failure and counterbalancing, this study advances scholarly understanding of the determinants of autocratic survival strategies and their far-reaching implications for military effectiveness, civil war and nuclear proliferation.

In contrast to arguments emphasizing the democratizing effects of failed coups, my findings suggest that coup failure is associated with long-term deinstitutionalization – an effect that increases with time since coup. This effect holds despite the short-term nature of post-coup personalization (Timoneda et al. 2023).

That said, future work should further consider the scope conditions underlying these findings and the ways in which counterbalancing is combined with extensive purges and other appeasement strategies that ‘soften’ the risk of potential repercussions. Additionally, future research should consider looking at the effects of successful coups on counterbalancing, while also examining the conditions under which counterweights are disbanded, not only created.

Supplemental Material

Supplemental Material - Civil-Military Relations in the Aftermath of Coups: How Does Coup Failure Affect Counterbalancing in Autocratic Regimes?

Supplemental Material for Civil-Military Relations in the Aftermath of Coups: How Does Coup Failure Affect Counterbalancing in Autocratic Regimes? by Artem Kyzym in Journal of Conflict Resolution.

Supplemental Material

Supplemental Material - Civil-Military Relations in the Aftermath of Coups: How Does Coup Failure Affect Counterbalancing in Autocratic Regimes?

Supplemental Material for Civil-Military Relations in the Aftermath of Coups: How Does Coup Failure Affect Counterbalancing in Autocratic Regimes? by Artem Kyzym in Journal of Conflict Resolution.

Footnotes

Acknowledgements

I thank Lars-Erik Cederman, Andreas Juon, Anders Sundell and Nikolay Marinov for their guidance and supervision throughout the different stages of this project. I am also grateful to my colleagues at the Department of Political Science at the University of Gothenburg and to participants of the PhD peer presentation series for their thoughtful feedback on earlier drafts. Finally, I am indebted to the editor and two anonymous reviewers for their constructive comments, which elevated the quality of my work.

ORCID iD

Artem Kyzym

Funding

The author received no financial support for the research, authorship, and/or publication of this article.

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Data Availability Statement

All replication materials necessary to replicate the findings of this study are available in open access at Harvard Dataverse: https://doi.org/10.7910/DVN/KQCWG0 ().

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Albrecht

2015. “The Myth of Coup-Proofing: Risk and Instances of Military Coups D’État in the Middle East and North Africa.” Armed Forces & Society 41 (4): 659–87.

Albrecht

Eibl

. 2018. “How to Keep Officers in the Barracks: Causes, Agents, and Types of Military Coups.” International Studies Quarterly 62 (2): 315–28.

Baturo

Elkink

. 2021. The New Kremlinology: Understanding Regime Personalization in Russia. Oxford University Press.

Beaumont

2023. “Gabon Military Officers Declare Coup After Ali Bongo Wins Disputed Election.” The Guardian. https://www.theguardian.com/world/2023/aug/30/gabon-coup-military-takeover-gabonese-election-disputed

Belkin

Schofer

. 2003. “Toward a Structural Understanding of Coup Risk.” Journal of Conflict Resolution 47 (5): 594–60.

Belkin

Schofer

. 2005. “Coup Risk, Counterbalancing, and International Conflict.” Security Studies 14 (1): 140–77.

Bennett

D. L.

Bjørnskov

Gohmann

S. F.

. 2021. “Coups, Regime Transitions, and Institutional Consequences.” Journal of Comparative Economics 49 (2): 627–43.

Böhmelt

Pilster

. 2015. “The Impact of Institutional Coup-Proofing on Coup Attempts and Coup Outcomes.” International Interactions 41 (1): 158–82.

Bokobza

Krishnarajan

Jacob

Sakstrup

Aaskoven

. 2022. “The Morning After: Cabinet Instability and the Purging of Ministers After Failed Coup Attempts in Autocracies.” The Journal of Politics 84 (3): 1437–52.

10.

Boutton

2019. “Coup-Proofing in the Shadow of Intervention: Alliances, Moral Hazard, and Violence in Authoritarian Regimes.” International Studies Quarterly 63 (1): 43–57.

11.

Bratton

Van de Walle

. 1997. Democratic Experiments in Africa: Regime Transitions in a Comparative Perspective. Cambridge University Press.

12.

Brooks

White

P. B.

. 2022. “Oust the Leader, Keep the Regime? Autocratic Civil-Military Relations and Coup Behavior in the Tunisian and Egyptian Militaries During the 2011 Arab Spring.” Security Studies 31 (1): 118–51.

13.

Brown

C. S.

Fariss

C. J.

Blake McMahon

. 2016. “Recouping After Coup-Proofing: Compromised Military Effectiveness and Strategic Substitution.” International Interactions 42 (1): 1–30.

14.

Buddy

Bajjalieh

Shalmon

Martin

Soto

. 2025. Cline Center Coup d’État Project Dataset Codebook”. Cline Center Coup d’État Project Dataset. Cline Center for Advanced Social Research. V.2.1.3. January 30. University of Illinois Urbana-Champaign. doi:10.13012/B2IDB-9651987_V8

15.

Bueno de Mesquita

Smith

Siverson

R. M.

Morrow

J. D.

. 2003. The Logic of Political Survival. MIT Press.

16.

Chacha

Powell

. 2018. “Economic Interdependence and Post-coup Democratization.” Democratization 24 (5): 819–38.

17.

Chin

J. J.

Carter

D. B.

Wright

J. G.

. 2021. “The Varieties of Coups D’État: Introducing the Colpus Dataset.” International Studies Quarterly 65 (4): 1040–51.

18.

Cunningham

2021. Causal Inference: The Mixtape. Yale University Press.

19.

De Bruin

2018. “Preventing Coups D’État: How Counterbalancing Works.” Journal of Conflict Resolution 62 (7): 1433–58.

20.

De Bruin

2020. Preventing Coups D’État: How Counterbalancing Works. Cornell University Press.

21.

De Bruin

2021. “Mapping Coercive Institutions: the State Security Forces Dataset, 1960-2010.” Journal of Peace Research 58 (2): 315–25.

22.

Derpanopolous

Frantz

Geddes

Wright

. 2016. “Are coups good for democracy?” Research & Politics 3 (1): 1–7.

23.

Dettmer

2016. Erdogan Launches Crackdown After Failed Coup in Turkey. Voice of America. https://www.voanews.com/a/erdogan-launches-crackdown-after-failed-coup-in-turkey/3421297.html

24.

Duzor

Williamson

. 2023. Coups in Africa. Voice of America. https://projects.voanews.com/african-coups/

25.

Easton

M. R.

Siverson

R. M.

. 2018. “Leader Survival and Purges After a Failed Coup D’État.” Journal of Peace Research 55 (5): 596–68.

26.

Erdağ

2019. “After the Failed Military Coup: the Need for Organizational Reform in the Turkish Military.” All Azimuth: A Journal of Foreign Policy and Peace 8 (1): 55–6.

27.

Escribà-Folch

Böhmelt

Ulrich

. 2020. “Authoritarian Regimes and Civil–Military Relations: Explaining Counterbalancing in Autocracies.” Conflict Management and Peace Science 37 (5): 559–79.

28.

Feaver

P. D.

1999. “Civil-Military Relations.” Annual Review of Political Science 2: 211–41.

29.

Frazer

J. E.

1994. Sustaining Civilian Control: Armed Counterweights in Regime Stability in Africa. Doctoral dissertation, Stanford University. https://searchworks.stanford.edu/view/2861834

30.

Gardner

2021. “Two-Stage Differences in Differences.” Working Papers 2203: 05943. https://jrgcmu.github.io/2sdd_current.pdf

31.

Gassebner

Gutmann

Voigt

. 2016. “When to Expect a Coup D’État? An Extreme Bounds Analysis of Coup Determinants.” Public Choice 169 (1): 293–33.

32.

Geddes

Wright

Frantz

. 2018. How Dictatorships Work: Power, Personalization, and Collapse. Cambridge University Press.

33.

Gomila

2021. “Logistic or Linear? Estimating Causal Effects of Experimental Treatments on Binary Outcomes Using Regression Analysis.” Journal of Experimental Psychology: General 150 (4): 700–79.

34.

Goodman-Bacon

2021. “Difference-In-Differences with Variation in Treatment Timing.” Journal of Econometrics 225 (2): 254–77.

35.

Greitens

S. C.

2016. Dictators and their Secret Police: Coercive Institutions and State Violence. Cambridge University Press.

36.

Harkness

K. A.

2018. When Soldiers Rebel: Ethnic Armies and Political Instability in Africa. Cambridge University Press.

37.

Harris

D. J.

2014. Sierra Leone: A Political History. Oxford University Press.

38.

Higgot

Fuglestad

. 1975. “The 1974 Coup D’État in Niger: towards an Explanation.” The Journal of Modern African Studies 13 (3): 383–98.

39.

Horowitz

D. L.

1985. Ethnic Groups in Conflict. University of California Press.

40.

Kaynar

A. K.

2022a. “Post-2016 Military Restructuring in Turkey from the Perspective of Coup-Proofing.” Turkish Studies 23 (3): 383–46.

41.

Kaynar

A. K.

2022b. “The Last Two Decades of Civil-Military Relations in Turkey Under the Shadow of Courts.” European Journal of Turkish Studies 34 (1): 1–27.

42.

Krogstad

E. G.

2012. “Security, Development, and Force: Revisiting Police Reform in Sierra Leone.” African Affairs 111 (443): 261–80.

43.

Kuehn

2017. “Midwives or Gravediggers of Democracy? the Military’s Impact on Democratic Development.” Democratization 24 (5): 783–80.

44.

Kutay

2016. Civilian and Military Relations in Turkey: a Historical Survey. Chr. Michelsen Institute (CMI Working Paper WP 2016:11). https://www.cmi.no/publications/6028-civilian-and-military-relations-in-turkey

45.

Kyzym

2025. “Replication Data for: Civil-Military Relations in the Aftermath of Coups: How Does Coup Failure Affect Counterbalancing in Autocratic Regimes”. Doi:10.7910/DVN/KQCWG0

46.

Lachapelle

2020. “No Easy Way Out: the Effect of Military Coup on State Repression.” The Journal of Politics 82 (4): 1354–71.

47.

Luttwak

E. N.

1979. Coup D’État: A Practical Handbook, Revised Edition. Harvard University Press.

48.

Marinov

Goemans

. 2014. “Coups and Democracy.” British Journal of Political Science 44 (4): 799–825.

49.

Mehrl

Choulis

. 2021. “The Colonial Roots of Structural Coup-Proofing.” International Interactions 47 (4): 750–76.

50.

Meng

Paine

. 2022. “Power Sharing and Authoritarian Stability: How Rebel Regimes Solve the Guardianship Dilemma.” American Political Science Review 116 (4): 1208–25.

51.

Phillips

D. L.

2017. An Uncertain Ally: Turkey Under Erdogan’s Dictatorship. Routledge.

52.

Pilster

Böhmelt

. 2011. “Coup-Proofing and Military Effectiveness in Interstate Wars, 1967-99.” Conflict Management and Peace Science 28 (4): 331–50.

53.

Pilster

Böhmelt

. 2012. “Do Democracies Engage less in Coup-Proofing? on the Relationship Between Regime Type and Civil—Military Relations.” Foreign Policy Analysis 8 (4): 355–71.

54.

Powell

J. M.

2012. “Determinants of the Attempting and Outcome of Coups D’État.” Journal of Conflict Resolution 56 (6): 1017–40.

55.

Powell

J. M.

2019. “Leader Survival Strategies and the Onset of Civil Conflict: a Coup-Proofing Paradox.” Armed Forces & Society 45 (1): 27–4.

56.

Powell

J. M.

Thyne

C. L.

. 2011. “Global Instances of Coups from 1950 to 2010: a New Dataset.” Journal of Peace Research 48 (2): 249–59.

57.

Quinlivan

J. T.

1999. “Coup-Proofing: Its Practice and Consequences in the Middle East.” International Security 24 (2): 131–65.

58.

Satana

N. S.

2022. “The New Civil-Military Relations in Turkey”. Middle East Institute. https://www.mei.edu/publications/new-civil-military-relations-turkey

59.

Singh

2014. Seizing Power: The Strategic Logic of Military Coups. John Hopkins University Press.

60.

Sudduth

J. K.

2017a. “Strategic Logic of Elite Purges in Dictatorships.” Comparative Political Studies 50 (13): 1768–181.

61.

Sudduth

J. K.

2017b. “Coup Risk, Coup-Proofing and Leader Survival.” Journal of Peace Research 54 (1): 3–15.

62.

Svolik

2009. “Power Sharing and Leadership Dynamics in Authoritarian Regimes.” American Journal of Political Science 53 (2): 477–94.

63.

Svolik

2012. The Politics of Authoritarian Rule. Cambridge University Press.

64.

Talmadge

2015. The Dictator’s Army: Battlefield Effectiveness in Authoritarian Regimes. Cornell University Press.

65.

Thompson

W. R.

1973. The Grievances of Military Coup-Makers. Sage Publications.

66.

Thyne

C. L.

Powell

J. M.

. 2016. “Coup D’État or Coup D’Autocracy? How Coups Impact Democratization, 1950-2008.” Foreign Policy Analysis 12 (2): 192–23.

67.

Timoneda

J. C.

2021. “Estimating Group Fixed Effects in Panel Data with a Binary Dependent Variable: How the LPM Outperforms Logistic Regression in Rare Events Data.” Social Science Research 93 (1): 102486.

68.

Timoneda

J. C.

Escribà-Folch

Chin

. 2023. “The Rush to Personalize: Power Concentration After Failed Coups in Dictatorships.” British Journal of Political Science 53 (3): 878–91.

69.

Turay

E. D. A.

Abraham

. 1987. The Sierra Leone Army: A Century of History. Macmillan.

70.

Vidler

1998. Regime Survival in the Gambia and Sierra Leone: A Comparative Study of the People’s Progressive Party (1965-1994) and the all People’s Congress (1968-1992). PhD dissertation, University of Newcastle. https://theses.ncl.ac.uk/jspui/handle/10443/1024?mode=simple

71.

Woldense

2022. “What Happens when Coups Fail? the Problem of Identifying and Weakening the Enemy Within.” Comparative Political Studies 55 (7): 1236–65.

72.

Yavuz

M. H.

Koç

. 2018. “The Gulen Movement Vs. Erdogan: the Failed Coup.” In Turkey’s July 15th Coup: What Happened and Why, edited by Yavuz

M. H.

Balci

, 78–97. University of Utah Press.

73.

Zartman

I. W.

1970. “The Algerian Army in Politics.” In Soldier and State in Africa, edited by Welch

C. E.

Jr , 224–49. Northwestern University Press.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.83 MB

0.00 MB

2.03 MB