Abstract
Current explanations of coup activity focus on the regime factors that makes them more likely to be successful rather than what them more likely to occur. While 60% of all coups succeed in autocracies, there has been less attention to the 40% that do not. Current empirical approaches testing the effect of legislatures on coup success implicitly treat the decision to attempt a coup as random and relegates it into the model’s residuals. This article discusses why this empirical approach is theoretically unrealistic and shows that the selection effect of coup attempt and success should be empirically accounted for. The analysis replicates and extends Bove and Rivera’s (2015) study on autocrats’ strategies to prevent coups, and it demonstrates that (a) institutionalisation reduces the risk of coup onset by 50%, (b) after accounting for the factors influencing coup attempt, an elected legislature does not by itself, help autocrats survive coups, and (c) purges increase the likelihood of coup attempts, but reduce their likelihood of success.
Introduction
In this article, I revisit the theoretical groundworks underlying the negative effect of civilian institutions on autocratic removal from power via a coup. To date, empirical studies find that civilian institutions – such as a legislature and political parties – extend autocrats time in office (Bove and Rivera, 2015; Gandhi, 2008; Svolik, 2012). By focusing exclusively on successful coups, a large amount of potentially useful information such as 40% of all attempted coups are unsuccessful (Powell and Thyne 2011) is discarded. I argue that it is theoretically important to disentangle the effect of civilian institutions on coup attempt and success. These institutions should influence the former, but not the latter. Legislatures and political parties are more a forum for political negotiations and compromise that alleviate the power-sharing problem autocrats face, rather than specialised units that can tilt the scales in an armed struggle.
Against this background, this article replicates and extends the study of Bove and Rivera (2015) on autocrats’ strategies to insulate themselves from coups in a time-series cross-sectional research design recording all coup incidences between 1950 and 2007. To empirically test the expectation that civilian institutions only deter coup attempts, but do not affect their success, I use a two-stage selection model for coup attempt and success (Böhmelt and Pilster, 2015; Powell, 2012). This allows me to relax the stringent theoretical assumption previous research makes about the factors that affect coup attempts. The coup determinants literature rests on the theoretical assumption that coups are attempted only if the payoff is greater than the benefits of the status quo (Böhmelt and Pilster, 2015) and the probability of victory offsets the consequences of failure (Powell, 2012). Thus, it follows that the decision to attempt a coup cannot be exogenous to the outcome, but it is implicitly treated as such in empirical analyses (Bove and Rivera, 2015). By not modelling the factors that affect the decision to attempt a coup in the first stage, the findings regarding the negative effect of civilian institutions on coup success suffers from potential bias in estimates. The bias creeps in because the factors determining coup attempts are treated as exogenous in relation to the outcome and relegated to the residuals (Wooldridge, 2002). To anticipate, when modelling the factors affecting the decision to attempt a coup and their effect on coup success, the residuals of the two stages are correlated, which warrants the use of a selection model estimating the effect of civilian institutions on coup attempt and success. However, the challenge to unbiased estimates requires the identification of a variable that explains coup attempts, but not their success. This article proposes that accounting for the status of political parties in the selection equation can plausibly satisfy this identification criterion.
The empirical analysis indicates that, once we account for the effect of institutions in the first stage (i.e. coup attempt), having an elected legislature reduces the likelihood of coup attempts, but has no effect in the second stage (i.e. coup success). Simply, it shows the negative effect of legislatures on coup success reported in previous studies is primarily driven by its effect on preventing coup attempts (Bove and Rivera, 2015; Svolik, 2012). Further, the results contrast with previous findings by showing that purging elites increase the likelihood of coup attempts, but reduces their likelihood of success (Bove and Rivera, 2015). More generally, it indicates that most factors commonly associated with coup success are found to be significant in deterring coup onset, but not affecting their success. However, these results hold to the extent that the status of political satisfies the exclusion criterion. 1
These results have three implications for the study of authoritarian survival. First, the fact that most coups are successful (Powel, 2012) suggests this is not due to chance alone. Yet, when empirically disaggregating coup determinants’ effect on the two theoretically distinct processes, it turns out we know little about what makes coups successful in autocracies. The literature on autocratic coup success cries out for novel theoretical propositions and rigorous empirical testing on its determinants. Second, the difference in the effect of the covariates points to a more complex dynamic of challenges in autocracies. There is a distinction between authoritarian vulnerability, the likelihood of an autocrat being challenged and authoritarian durability, the ability of the autocrat to survive challenges (Way, 2011). Future research investigating the effect of challenges on autocratic survival should incorporate this important theoretical and empirical distinction at the core of their study. Future research should also implement a theoretical and empirical framework that separates between explanations for vulnerability and durability; yet, the two should be studied jointly as expectations of the outcome influences decisions to challenge the regime in the first place. Finally, scholars unpacking this crucial distinction should be cautious in identifying variables that convincingly satisfy the identification criterion of a two-stage selection model.
Civilian institutions and coup activity in autocracies
Several scholars proposed that structural factors such as type of government, political culture, and state–society relations act as push and pull factors that motivate and create the conditions for military officers to launch coups (Belkin and Schofer, 2003; Luttwak, 1968; Powell, 2012). More recently, Svolik (2012) found that having a legislature or a single party reduces the hazard of exit from office via a coup in autocracies. Similarly, Bove and Rivera (B&R) (2015) found that autocracies with an elected legislature have a lower risk of coup attempts as they reduce information asymmetry between the leader and regime elites. B&R’s findings may suffer from selection bias because they use the Archigos data on leadership survival to capture coup attempts (Goemans et al., 2009). More precisely, Archigos captures only successful coups in removing the leader while excluding any unsuccessful coups from their sample. For example, Powell and Thyne’s data on coups (2011) records 285 coup attempts in authoritarian regimes, out of which 173 were successful between 1950 and 2007. By comparison, B&R only captured 165 successful coups attempts due to the coding criteria used by the Archigos data.
This is problematic for findings on the effect of legislatures on coup attempts because it induces selection bias in the coefficients. More precisely, the factors influencing the decision to attempt a coup are treated as exogenous to the outcome and relegated into the residuals. This affects our ability to draw valid inference about the effect of the covariates (Wooldridge, 2002). This empirical approach is quite surprising, considering a large body of literature indicates that the decision to attempt a coup is anything but random and very much premised on the expected outcome (Powell, 2012). The rationalist approach of the civil-military literature assumes that coups are attempted only if the expected payoff is greater than the benefits of the status quo (Böhmelt and Pilster 2015) and the probability of victory offsets the consequences of failure (Powell, 2012). Yet, empirical approaches modelling the effect of legislature on coup attempt and success do not follow this theoretical guiding as they fail altogether to account for the two-stage sequence of coup processes (Bove and Rivera, 2015; Svolik, 2012). Against this background, this article relaxes this theoretical assumption and revisits the hypotheses that legislatures reduce the likelihood of coup attempt and success.
Methodology
The article revisits models of coup attempt and success by estimating a two-stage selection model that accounts for the interdependent two-stage dynamic of coups. It uses a time-series cross-sectional research design of 114 authoritarian regimes, 2 with country-year as units of analysis between 1950 and 2007.
Outcome variable
Two dependent variables are used to model the effect of legislatures on coup attempt and success in autocracies. The first variable is binary and takes a value of 1 if there were ‘illegal and overt attempts by the military or other elites within the state apparatus to unseat the sitting executive’ (Powell and Thyne, 2011: 252), and 0 otherwise. The second dependent variable takes a value of 1 if the coup resulted in deposing the leader from office and 0 if it was unsuccessful. The data on coup success was culled from Archigos (Goemans et al., 2009) and updated with data from Powell and Thyne (2011).
Explanatory variables
The data capturing the status of the legislature in autocracies is obtained from Cheibub et al.’s (2010) dataset on political institutions. Similar to B&R, this variable takes a value of 1 if the legislature is elected, and 0 otherwise.
Model specification and identification strategy
The effect of institutions on coup attempt and success is estimated with a Heckman selection probit model because factors that lead to coups attempts could also influence their success (Powell, 2012; Van de Ven and Van Praag, 1981). Ignoring one of the two stages can potentially create bias through either over-estimation or under-estimation of results (Böhmelt and Pilster, 2015). The Heckman model returns unbiased estimators (or is identified) by including a variable that is theoretically related to coup attempts, but not to coup success. In other words, we need to identify a variable that influences the decision to attempt a coup, yet has no impact on the development of the coup situation once it has been attempted. This is quite challenging because decisions to attempt a coup are premised on their expected outcome. While I cannot establish that it is theoretically impossible for political parties to affect the outcome of coups, I will present several reasons as to why this is plausible.
Political parties are a bargaining tool between elites and the leader (Boix and Svolik, 2013; Magaloni, 2006), and are used as a co-optation tool to reward supporters (Gandhi, 2008; Svolik, 2012). As a bargaining tool, political parties increase the dictator’s credibility to share power by delegating elite selection and promotion to party cadres (Magaloni and Kricheli, 2010; Svolik, 2012). These party elites then have incentives to support the regime rather than conspire against it in exchange of promotion into rent-paying position. As a co-optation tool, parties ensure stability by controlling the masses and thwarting any challenges through collective mobilisation. By preventing dissent through co-optation, leaders reduce opportunities for would-be plotters to coordinate their actions and attempt a coup (Casper and Tyson, 2014). Once elites organise and attempt a coup, this coordination mechanism is no longer valid for strategic reasons: joining the losing side outweighs any preference over who wins (Svolik, 2012) and potential plotters, capable of tilting the balance, prefer to keep their heads down and wait to see who wins (Geddes, 1999). Since coups are fast-pace events, mobilisation through party ranks can be rather slow and it may come too late to matter for the outcome of the coup. The variable capturing political parties’ status takes a value of 0 if all political parties are banned, 1 if there a single regime party or united regime front, and 2 if multiple parties are allowed to exist outside the regime front. The supplemental materials contains a longer discussion of this issue and presents alternative model specifications.
Control variables
I include several control variables to account for potentially confounding factors regarding the effect of institutionalisation on coup attempts and outcomes (Böhmelt and Pilster, 2015; Bove and Rivera, 2015; Powell, 2012). These are grouped in economic (GDP/capita and GDP year-to-year change), political (instability, purges and previous coup history) and military factors (expenditure per soldier, year-to-year change in military expenditure and size of military), plus temporal controls. 3
Empirical analysis
Table 1 summarises the results of three Heckman probit selection models (Van de Ven and Van Praag, 1981). Models 1, 2 and 3 from Table 1 replicate and extend the study of Bove and Rivera (2015). The models consists of two equations that estimate the effect of institutions on coup attempts and success with splines included to account for time dependency as indicated by Beck et al. (1998). Model 1 is a sparse model with the main explanatory variables and GDP/capita, 4 and Model 2 replicates their exact model specification plus temporal control. Model 3 uses a categorical measure for the status of legislature to further differentiate the effect of each legislature status on coup attempt. The identification criterion of the Heckman probit selection models is ensured by the inclusion of the status of political parties in the attempt equation, but not in the success one. The test of independence of the two equations from Model 2 shows that we can reject the null hypothesis of no correlation of the errors between the two equations (Prob > χ² =0.0404). 5 Also, the athrho coefficient is statistically significant and substantively important because its corresponding rho coefficient is –.81. 6 This indicates that the selection model fits the data better than simple probit models estimating independently the effect of institutionalisation on coup attempt and success.
Heckman-probit selection models for coup attempts and success, 1950–2007.
Robust standard errors in parentheses: *** p<0.01, ** p<0.05. Splines for temporal correction are included in both equations of the model, but omitted from presentation.
Model 2 from Table 1 shows that when accounting for the selection effect of institutions on coups, having an elected legislature does not exert a statistically significant effect on the likelihood that coups are successful. An autocrat with an elected legislature is not more likely to survive a coup compared to his less institutionalised counter-part. This result contrasts with previous findings of the literature (Bove and Rivera, 2015; Svolik, 2012), but it makes intuitive sense since legislatures are a forum for political negotiation and compromise rather than specialised military units that can defend armed challenges to the leadership (Böhmelt and Pilster, 2015). This indicates that the negative effect of an elected legislature on coup success is actually driven by its effect on coup attempt when the selection effect is unaccounted for. Moreover, the results contrast again with the findings of B&R, and indicate that purging elites from the government reduces the chances of a successful coup. An explanation for this finding is that the leader has already purged elites that were strong enough to win a challenge and the remaining elites are incapable of defending the incumbent should it come to a confrontation between plotters and the leader. The rest of the control variables do not exert a statistically significant effect on the likelihood of a successful coup.
The coup attempt equations from Models 1 and 2 shows that autocrats with an elected legislature have a lower likelihood of coup attempts compared to leaders that closed the legislature. When accounting for the individual status of legislature (closed/appointed/elected) results indicate that an appointed or elected legislature reduces the likelihood of coup attempts 7 as illustrated in Model 3. A test for the difference between coefficients reveals that the coefficient for appointed legislature is not statistically different from the one for elected legislature. Substantively, this means that an open legislature, even unelected, reduces the likelihood of coup attempts in authoritarian regimes. This finding supports the hypothesis about the coup suppressing effect of autocratic legislatures, yet it also reveals that having an elected legislature does not have an effect on preventing coups. It is sufficient for dictators to have a legislature, even with appointed members, to reduce their coup attempt risk. If the leader is strong enough to appoint members of the legislature, then he/she can insulate him/herself from coups without giving up too much control (Gandhi, 2008). Furthermore, in line with previous research, leaders of single party regimes are less likely to be challenged by regime elites (Svolik, 2012). However, regimes that allow multiple parties to exist do not seem to have a lower likelihood of coup attempts compared to regimes that ban parties altogether.
The results also indicate that purging governmental elites increases the likelihood of coup attempts. This finding supports B&R’s claim that elites no longer feel safe and seek to overthrow the incumbent to prevent being victims of inter-elite violence. Similarly, increased levels of dissent increase the likelihood of coups because these events offer elites the window of opportunity to legitimise their coup (Casper and Tyson, 2014). The ‘coup-trap’ argument receives support as well since a higher number of previous coup attempts increases the likelihood of coup attempts (Londregan and Poole, 1990). Economic growth decreases the risk of coups, while higher levels of GDP/capita do not exert a significant effect on the risk of coup attempts. The effect of military factors on coup attempts is negative (except for increases in military expenditures which is not statistically significant) and mostly in line with findings of previous research (Powell, 2012). The results from Table 1 are robust to alternative specifications of the models and operationalisation of the key independent variables. The following robustness checks were run: a replication of the main model of B&R with time controls, a sparse model that includes only the main explanatory variables and temporal dependence variables, an alternative measures of coup success from Powell and Thyne (2011), and a logit model testing the effect of legislature status on coup attempts while controlling for ideal regime types (Geddes et al., 2014) as well as coup-proofing strategies (Böhmelt and Pilster, 2015). The justifications for these robustness checks has been clarified in the supplemental materials.
Figure 1 shows that a regime with an appointed or elected legislature has a 4% probability of a coup attempt while a regime with a closed legislature has a probability of about 13%. Closing the legislature more than triples an autocrat’s risk of being challenged by members of its ruling coalition. Furthermore, Figure 2 shows that having a regime with a single party have a 5% predicted probability of a coup attempt compared to 8% for no-party regimes and almost 8% for multi-party regimes. The likelihood of coups in single party regimes is nearly 70% lower compared to regimes that ban political parties or allow multiple parties to exist outside the regime front.
Effect of legislature on coup attempts.
Effect of political parties’ status on coup attempts.
Conclusion
This article showed that current models estimating the effect of civilian institutions on coup attempt and success in autocracies are biased in relation to the effect of legislature on coup success. The results indicate that having an open legislature, either appointed or elected, reduces threefold the risk of coup attempt. More importantly, the results show that, once we account for the two-stage dynamic of coup attempts and success, having an elected legislature no longer protects leaders from surviving a coup. This contrasts with previous findings (Bove and Rivera, 2015; Svolik, 2012) and shows that the suppressing effect of legislatures is mainly driven by its effect on coup attempts, not success. Similarly, a regime with a single political party or a united regime front reduces the coups risk by about 70% compared to regimes that chose to regulate political parties differently. Further, the results contrast with previous research and show that purges increase the likelihood of coup attempt, but decrease the likelihood of their success. The main implications of these results are that we still do not know what factors explain autocrats’ survival once a coup is attempted. Second, the literature on authoritarian survival and coup determinants should engage in thorough theorisation and rigorous empirical tests about the appropriate strategies for autocratic leaders to survive coup attempts. Lastly, future research should spend considerable effort in identifying variables that allow the estimation of unbiased estimates by satisfying the identification criterion of selection models.
Supplemental Material
Appendix_Coups_Research_Article_RAP – Supplemental material for Fait accompli, or live to fight another day? Deterring and surviving coups in authoritarian regimes
Supplemental material, Appendix_Coups_Research_Article_RAP for Fait accompli, or live to fight another day? Deterring and surviving coups in authoritarian regimes by Roman-Gabriel Olar in Research & Politics
Supplemental Material
Supplementary_materials_(RAp) – Supplemental material for Fait accompli, or live to fight another day? Deterring and surviving coups in authoritarian regimes
Supplemental material, Supplementary_materials_(RAp) for Fait accompli, or live to fight another day? Deterring and surviving coups in authoritarian regimes by Roman-Gabriel Olar in Research & Politics
Footnotes
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Supplemental materials
Notes
Carnegie Corporation of New York Grant
This publication was made possible (in part) by a grant from the Carnegie Corporation of New York. The statements made and views expressed are solely the responsibility of the author.
References
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.
