Are coups good for democracy?

Introduction

On 18 February 2010, troops stormed the Nigerien presidential palace and arrested President Mamadou Tandja. ¹ Though Tandja was elected in free and fair elections in 1999, his government took an autocratic turn a decade later when (among other things) he dissolved the National Assembly and passed a constitutional referendum to extend his rule another three years. Upon ousting Tandja, the coup leaders formed the Supreme Council for the Restoration of Democracy (CSRD) to rule the country, and to ‘make Niger an example of democracy and good governance.’ Citizens celebrated, and the opposition proclaimed it an opportunity to restore democracy.

Observers around the globe wondered whether the coup was a positive development (Miller, 2011). News headlines included, ‘Niger coup: Can Africa use military power for good?’ (Armstrong, 2010) and ‘Niger: A coup for democracy?’ (BBC News, 2010). Their optimism was well founded. The CSRD oversaw free and fair elections in early 2011, and former opposition leader Mahamadou Issoufou assumed the presidency a few months later (Freedom House, 2012).

This was not the first time the Nigerien military staged a coup to impose democratization. The competitive elections that installed Tandja in 1999 came after a coup that killed Ibrahim Bare Mainassara, a strongman who had ruled Niger since 1996. Nor is the Nigerien experience an isolated one. So-called ‘good coups’ – or coups against dictatorships that pave the way for democracy – have occurred in places ranging from Portugal in 1974 to Mali in 1991 to Guinea-Bissau in 2003.

These events have generated arguments that coups have the potential to be good for democracy. By providing a ‘shock’ to the system, coups may create opportunities for political liberalization that would not exist otherwise (Thyne and Powell, 2014: 2–3). Summarizing these arguments, Paul Collier writes that ‘coups and the threat of coups can be a significant weapon in fostering democracy’ (Collier, 2009).

Can coups really foster democracy? We explore this question by looking at the political systems that follow coups in autocracies, as well as the ensuing levels of repression. Our aim is simply to establish what the empirical patterns look like. ² Because scholars have observed that ‘good coups’ have dramatically increased in frequency since the end of the Cold War (for reasons we summarize in the section that follows), we emphasize throughout how these patterns compare during and after the Cold War (Marinov and Goemans, 2014).

We begin by noting that coups against dictatorships replace the dictator and sometimes other leaders as well. The new leaders then choose one of three possible outcomes: no regime change (e.g. the 1975 Nigerian coup that replaced General Yakubu Gowon with Brigadier Murtala Muhammad, without changing either the group in power or the rules for governing); the ouster of the incumbent dictatorship and the establishment of a new one (as in the 1971 Ugandan coup that toppled the civilian government led by Milton Obote and brought General Idi Amin to power); and the ouster of the dictatorship followed by democratization (as in Niger in 1999 and 2010).

We find that since the Cold War’s end, regime change of some sort increasingly follows successful coups: 68% pre-1990, compared to 90% afterward. ³ (The rest are coups that only reshuffle the leadership.) ⁴ Though democratization coups have become more frequent, the most common outcome is still the ouster of the incumbent dictatorship by a different group of autocrats intent on replacing one arbitrary distribution of benefits and suffering with another. At least half of all coups – 56% during the Cold War and 50% from 1990 to 2015 – initiate new authoritarian regimes. For example, even though Nigerien coups in 1999 and 2010 brought democracy, coups in 1974 and 1996 established new dictatorships. Basic summary statistics indicate that both before the Cold War and afterward, coups are most often followed by new dictatorships.

Statistical tests that take into account a variety of potentially confounding factors tell a similar story: the association between coups and democratization is statistically insignificant, while that between coups and the establishment of new dictatorships is robust.

Looking at failed as well as successful coups fails to yield a rosier picture. Though some have argued that coup attempts, such as the Burundian military’s failed power grab in 2015, can set in motion changes leading to democratization (Noyes, 2015), our study indicates that this is unlikely. Regime change is less likely to follow a coup attempt than a successful coup, and when it does, the regime that emerges is more likely to be a new dictatorship than a democracy.

We also examine the association between different types of coups and repression. Using an annual latent measure of repression, we find that coups that launch new dictatorships are followed by an increase in repression in the calendar year after the coup compared to the year before it. In contrast, coups that lead to democracy are associated with a decrease in repression, but only for post-Cold War cases. We supplement this with an analysis of event data for 49 coup attempts that occurred after 1989, by comparing state-sanctioned civilian deaths in the 12 post- and 12 pre-coup months. We find only one case of a coup followed by a drop in deaths and numerous cases of increases in violence after different types of coup events.

The central message of this study is clear: though coups against autocrats have sometimes led to democratization, more often they install a new set of autocratic elites and expose citizens to higher levels of repression.

The study is organized as follows. It first provides some background: a definition of coups, and a description of how patterns associated with them have changed over time. Next, it evaluates the link between coups and regime change empirically. It then investigates the connection between coups and repression. The final section summarizes our findings.

Background

Coups are successful efforts ‘by the military or other elites within the state apparatus to unseat the sitting executive using unconstitutional means’ (Powell and Thyne, 2011: 252). Though civilians may support coups, at their core coups are a technology that can only be used by the military, police, and security forces.

The most basic goal of a coup is to bring about a change in leadership, but often coup plotters also seek more substantial political transformation. They may announce their intention to hold democratic elections in the near future and even offer a timeline for the transition. However, they do not always fulfill such promises, as in Egypt where the military that promised a transition later ousted elected President Morsi in 2013 and remains in control.

Recent research, however, suggests that the consequences of coups may have changed since the end of the Cold War. Not only have coups declined in frequency, but those that occur are increasingly followed by competitive elections (Marinov and Goemans, 2014). From 1950 to 1989, 14% of successful coups against dictatorships led to democracy within two years, while 40% did so from 1990 to 2015.

A variety of factors could explain this. For one, conspirators often state that the desire to save their countries from autocratic incumbents motivated them. They may opt for free and fair elections simply because they value them. Even when junta members are less altruistic, coups may lead to democratization because of the incentives created by international pressure in the post-Cold War era (Marinov and Goemans, 2014). Threats to suspend foreign aid can motivate the military to hold elections quickly.

In summary, there are a variety of mechanisms through which coups might pave the way for democratization (Thyne and Powell, 2014), as well as compelling arguments for why they might be more likely to do so since the Cold War’s end (Marinov and Goemans, 2014). Have coups become good for democracy, though? In the next section, we evaluate this question empirically.

Coups and regime change

An investigation of the consequences of coups should take into account the kind of leadership the coup aimed to replace. From 1950 to 1989, nearly half of all dictatorships (49%) suffered from at least one coup, while only a third of democracies did (35%). Since 1990, coups have been much less frequent but just as likely to occur in democracies as in dictatorships (12%). Democracies ended by coups, however, tend to rebound quickly (Marinov and Goemans, 2014). One reason competitive elections more often follow coups may therefore be that a larger proportion of them now occur in places that tend to redemocratize quickly.

Yet, few would argue that coups against democratic governments are ‘good’ for democracy, even in cases like 2009 Honduras, where the Supreme Court sided with the military and the autocratic interlude was short. Democracy, at a minimum, guarantees that transitions from one leader to the next occur through a free and fair process. Thus coups in democracies indicate democratic backsliding at a minimum and usually outright democratic breakdown.

To avoid treating cases of quick redemocratization after coups against democracies as ‘good for democracy’, we limit our investigation to coups against dictatorships. We examine how coups influence the likelihood of two distinct types of autocratic regime collapse: democratization and adverse regime change (Geddes et al., 2014). ⁵ The latter constitute events during which one group of autocratic elites replaces another – as for example, when coups replaced monarchies with military-led dictatorships in Egypt, Iraq, Libya and Yemen. Our main independent variable is an indicator of whether a coup occurred in the observation year or either of the two years prior (Thyne and Powell, 2014).

We estimate a linear model with regime-case fixed effects. ⁶ Doing so enables a within-regime comparison of what follows a coup, while conditioning-out all differences between autocratic regimes (e.g. level of development, colonial history, how the regime seized power). Crucially, this approach accounts for autocratic regime type, including whether the incumbent autocracy is ruled by the military. Accounting for differences among regimes, such as whether a military junta leads the dictatorship, is important, because military dictatorships often use coups as a method for reshuffling leaders, and they are also more likely to democratize, though not necessarily because of coups (Geddes, 1999). The specification also controls for regime duration, leader duration and year fixed effects. To examine the effect of the end of the Cold War, we estimate separate variables for Pre-1990 coups and Post-1989 coups. This is similar to estimating an interaction between time period and coups; the year fixed effects subsume the time period constituent term.

The sample covers all autocratic regime-years in 285 dictatorships from 1950–2015. A model with democratization as the dependent variable estimates the effect of coups on the probability of transitioning from autocracy to democracy, from one calendar year to the next: P r ( D e m o c r a c y t | A u t o c r a c y t − 1 ) . A model with adverse regime change as the dependent variable estimates the effect of coups on the probability of transitioning from autocracy to a new autocracy, from one calendar year to the next: P r ( N e w A u t o c r a c y t | A u t o c r a c y t − 1 ) .

The top panel of Figure 1 plots the estimates. The top coefficient in each pair (circle) is from a model with democratization as the dependent variable, while the bottom coefficient (triangle) is from a model in which adverse regime change is the dependent variable.

OPEN IN VIEWER

Figure 1.

Coups and autocratic regime transitions.

The figure shows that successful coups during the Cold War are not correlated with democratization; post-Cold War coups are, but the estimate is not statistically significant. A successful coup is, however, associated with an increase in the chance of transition from one autocracy to another; a 19% increase during the Cold War and 27% afterward. Both estimates are statistically significant. This suggests that coups destabilize dictatorships not by improving prospects for democratization, but by boosting the odds that a new dictatorship replaces the old one.

We next consider whether coup attempts – regardless of whether the coup effort succeeds or fails – are associated with higher chances of democratization. It is possible that when coups fail, they set in motion events conducive to democratization down the road. Thyne and Powell (2014) argue, for example, that failed coups increase the likelihood of democratization by sending a signal to incumbents that military support for the status quo has declined. Incumbents, fearing a future successful coup, then opt for political liberalization, which gives them a better position from which to negotiate their exit than would a coup.

The data, again from Powell and Thyne (2011), show that about half of all attempted coups fail. The likelihood of failure varies with time period: 48% of attempted coups failed during the Cold War and 63% after it. In any case, regime change rarely follows failed coups: 6% of failed coups were followed by democratization and 3% by the establishment of a new dictatorship. In summary, coups fail frequently, and when they do, regime change rarely follows.

The lower panel of Figure 1 is similar to the top panel, but looks at all coup attempts (successful or failed). It reveals a similar pattern: coup attempts are correlated with an increased probability of adverse regime change, but the effect on democratization is not statistically significant.

These findings contrast with recent research on the democratizing effects of coups, notably Thyne and Powell (2014). To explore why our results differ, we re-examine their statistical tests in the Online Appendix. We show that the ‘democratizing effect’ of coups is not robust to including either a control for military regime or regime-case fixed effects. These are not trivial changes to the specification because they address the most important omitted variable – namely, military regime leadership – which is correlated with both coups and democratization. Furthermore, we show that even using their specification, the association between coups and the probability of democratization is quite small in the post-Cold War period (1.2%).

Regime change is not the only way coups can affect citizen welfare, however. In the next section, we investigate the relationship between coups and repression.

Coups and repression

On 23 December 2008, Captain Dadis Camara staged a coup in Guinea, just a day after the death of longtime dictator Lansana Conté. Camara established the National Committee for Democracy and Development (CNDD) to control government (Walker, 2008). Though some citizens initially welcomed the coup as a respite from the repressiveness of Conté’s government, their hopes of greater freedom were soon dashed. Reports emerged in the months that followed of arbitrary arrests and detentions, restrictions on citizens’ political rights, and criminal activities organized by the military (Human Rights Watch, 2009). The violence peaked on 28 September 2009, when security forces killed more than 150 citizens participating in anti-government protests, sparking international condemnation of the Camara regime and triggering sanctions against it (Reuters, 2009).

The 2008 coup in Guinea led to the establishment of a new dictatorship. The events that followed it highlight the possibility that coups can have damaging consequences for citizens beyond regime change. At the extreme, coups precipitated deadly civil conflicts in Algeria (1991–1999) and Rwanda (1994), while a 1999 coup in Côte d’Ivoire unleashed a decade of political violence and repression, culminating in foreign intervention. We explore the coup–repression nexus in this section.

To measure repression, we use mean estimates of a latent measure of respect for human rights from Fariss (2014). This measure accounts for the changing standards of accountability over time, as human rights norms and reporting standards have become stricter. We invert it so that larger values indicate higher repression.

We disaggregate coups into four categories: failed coups (Failed coups), coups that launch new autocracies (Adverse change), democratization coups (Democratizing) and coups that merely reshuffle leaders, causing no regime change (No change). To estimate the effect of coups on repression, we compare levels of repression in the year before the coup/attempt to repression in the year after it, limiting the sample to dictatorships that experienced coups. The before-and-after comparison allows us to rule out the possibility that coups and repression are correlated because coups occur more frequently in more repressive countries.

Let C ∈ ( F , A , D , N ) denote coup type and define

E C = ∑ k = 1 K C ( R t + 1 k − R t − 1 k ) K C

where k indexes K C coup events of type C , and R t + 1 − R t − 1 is the difference in repression between the post- and pre-coup year. To recover an estimate of E C (for coups of type C), we fit the following linear model

R i , t C = α + β 1 P r e C o u p i , t C + β 2 P o s t C o u p i , t C + γ i + ξ t + ε i , t

which includes binary indicators of pre- and post-coup years (with coup years as the reference category) and country ( γ i ) and time-period fixed effects ( ξ t ). ⁷ The estimated within-country change in repression over the 3-year coup window ( t − 1, … , t + 1 ) is the difference between the coefficients of the post- and pre-coup indicators. For example, the estimated change in repression around adverse regime change coups ( C = A ) is E A = β ^ 2 − β ^ 1 . If this is positive, it indicates there is more repression in the post-coup than the pre-coup year.

Figure 2 illustrates the results, with each estimate coming from a separate model for each type of coup: Failed coups, Adverse change, Democratizing and No change (reshuffling). ⁸ The left panel includes data from 1950 to 2010, while the right panel restricts the sample to the post-Cold War years. During the full period, repression is lower after failed coups, democratizing coups and reshuffling coups, but the effects for Failed coups and No change are small and not statistically different from zero. Higher repression follows coups that initiate new dictatorships, however, consistent with the experience of Guinea after the 2008 coup. In other words, dictatorships that replace other dictatorships via a coup tend to use more violence against citizens than the dictatorships they replaced. The post-Cold War pattern looks similar, but the worsening of repression following coups that replace one dictatorship with another is even greater, as is the reduction in repression after democratizing coups.

OPEN IN VIEWER

Figure 2.

Coups and repression.

We next look at the effect of coups on repression using event data. We use two sources: the Worldwide Atrocities Dataset (WAD) (Ulfelder and Schrodt, 2009), and the Social Conflict in Analysis Database (SCAD) (Salehyan et al., 2012). The former codes deaths of noncombatants and includes the years 1995 to 2014. The latter captures broader manifestations of instability, such as protests and riots, in Africa and Latin America from 1990 to 2013. Given our focus on repression, we rely mostly on WAD, but also use lethal events included in SCAD from 1990 to 1994, in order to analyse as many coups as possible. Using these two sources, we create a variable measuring the change in number of deaths recorded in the 12 months following each coup event compared to the 12 months preceding it. These data allow us to focus on repression in the months immediately before and after coups, unlike the data used in Figure 2. In order to capture repression, we only aggregate deaths from government-sanctioned lethal atrocities coded in WAD and deaths from pro-government violence coded in SCAD. ⁹

Figure 3 displays this information for all coups that occurred in dictatorships in the years 1990-2014, by coup type. This includes 30 failed coups, seven democratizing coups, seven adverse regime change coups and five leader-reshuffling coups. Boxplots summarize the distribution of changes in deaths. They show that, for all except reshuffling coups, the median within-country change is zero – the median is the dark line in each box. Though we cannot be statistically confident that repression increases after coups – even for reshuffling coups – we nevertheless believe there are some patterns worth noting.

OPEN IN VIEWER

Figure 3.

Coups and state-sanctioned civilian deaths (boxplots).

First, only one (failed) coup event is followed by a decrease in civilian deaths. No lethal violence occurred either pre- or post-democratizing coups (D); hence the whole distribution falls on the zero line. Substantial increases in repression against citizens follow other kinds of coup, however. A considerable amount of violence follows adverse regime change coups (A), consistent with the annual data evidence in Figure 2. Leader-reshuffling coups (N) are also associated with an increase in violence. However, this is owed mostly to one large outlier (which has been removed from the graph to aid interpretability). Finally, increases in post-coup deaths follow a number of failed coups (F), the modal type in this period.

To sum up the evidence in this section, most coups, successful or failed, are followed by greater repression against citizens. Although yearly indices, which exclude the year of the coup itself, suggest a possible decrease in repression after some coups, particularly democratizing ones, event data, which track the months immediately after coups, document no short-term decreases. The annual data allow a confident claim that substantial increases in repression follow coups initiating new dictatorships, and the monthly data supplement this claim with suggestive evidence that more atrocities are committed against citizens.

Conclusion

Several scholars have recently challenged the standard interpretation of military coups as anti-democratic. Instead, they argue, a coup can help usher in democracy. Our results give pause to such optimism.

To assess whether coups are associated with democratization, we investigated what follows coups against dictators, excluding coups against democracies. We show that coups are not systematically correlated with democratization, either during the period from 1950 to 2014 or post-Cold War. On the contrary, the perpetrators of coups tend to oust dictators only to impose new ones.

Further, we show that many coups in dictatorships lead to increases in human rights abuses. Using a yearly repression measure, we show that coups that launch new dictatorships are followed by increases in state violence against citizens. With more fine-grained event data, we note that the year after all but one post-Cold War coup against dictatorship brings either an increase or no change in civilian deaths. In short, the months that follow coups can be bloody, even if the coups themselves were not.