Counterrevolution

Counterrevolutions represent one of the greatest threats to new revolutionary regimes and are among the most obvious ways by which revolutionary transformations can be reversed or undermined. Yet despite their importance, there is limited research dedicated to explaining counterrevolution, as a number of recent works have pointed out (Mayer Reference Mayer2000; Slater and Smith Reference Slater and Smith2016; Weyland Reference Weyland2016). The scholarship on revolution does touch on the topic of counterrevolution, especially works that consider postrevolutionary trajectories or aftermaths. For example, some scholars note the ubiquity of counterrevolutionary threats following successful revolution (Gurr Reference Gurr1988; Levitsky and Way Reference Levitsky and Way2013; Mayer Reference Mayer2000; Skocpol Reference Skocpol1979).Footnote ¹ Goldstone (Reference Goldstone2014) argues that when revolutionary regimes survive fierce counterrevolutions they may become more ruthless and authoritarian. And Lawson (Reference Lawson2019) notes that counterrevolutionary challenges can breed protracted and destabilizing civil conflicts. But because they are concerned with theorizing revolution, these works focus mainly on how counterrevolutions shape postrevolutionary regime development. They therefore do not explicitly theorize or analyze counterrevolution as an outcome on its own terms, including the factors that make it more likely to emerge and succeed.

In part because there has been so little direct research on the phenomenon, there is little consensus on how best to conceptualize and operationalize counterrevolution. Indeed, though most scholars agree that counterrevolution involves opposition or resistance to revolutionary change, there are a variety of forms that such opposition can take. For example, Brinton (Reference Brinton1938) equated counterrevolution with “Thermidor,” or a period in which moderate revolutionaries take over and reimpose many of the rules, norms, and practices of the old regime. In his study of the Vendée counterrevolution in France, Tilly (Reference Tilly1964) defined counterrevolution as bottom-up resistance to the imposition of revolutionary rule. Others have conceptualized counterrevolution as a relatively unified, right-wing project directed at thwarting revolutionary movements worldwide (Halliday Reference Halliday1999; Mayer Reference Mayer1971).

Another prominent understanding of counterrevolution is an effort to defeat revolutionary movements before they come to power (Allinson Reference Allinson2019; Bisley Reference Bisley2004; Jones Reference Jones2013). For example, Slater and Smith define counterrevolution as “collective and reactive efforts to defend the status quo and its varied range of dominant elites against a credible threat to overturn them from below” (Slater and Smith Reference Slater and Smith2016, 1475), and argue that these counterrevolutions generate cohesion and strength among ruling coalitions. In another prominent recent study using this definition, Weyland (Reference Weyland2016) analyzes the diffusion of counterrevolutionary practices in Europe that led to the crushing of revolution during and after 1848.

In this paper, I choose to analyze a different manifestation of counterrevolution, which I call “restorative counterrevolution” because it seeks to restore the former regime to power following a successful revolution.Footnote ² A restorative counterrevolution is an irregular effort in the aftermath of a successful revolution to restore a version of the prerevolutionary political regime. Throughout the paper when I use the term “counterrevolution,” this is the version to which I refer (unless otherwise noted). I adopt this conceptualization, which is somewhat narrower than those used in previous studies, primarily in an effort to generate and test a more parsimonious and tractable theory. For example, a theory that makes sense of efforts to defeat a revolution before it comes to power would likely emphasize mechanisms like violent repression, concessions, or cooptation, which would mostly be ill-suited to explaining the conditions under which fallen incumbents restore themselves in office. In other words, by focusing only on restorative counterrevolutions I allow for the development of a more focused and coherent argument, while also recognizing that this leaves other manifestations of counterrevolution for future scholars to explore.

The definition of counterrevolution above immediately begs clarification of a second concept: successful revolution. Here, I mean that a regime has been overthrown through the mass mobilization of large numbers of everyday citizens. This understanding of revolution follows a long line of scholarship that has characterized “political revolution” as an exercise in popular regime change (Beissinger Reference Beissinger2022; Goldstone Reference Goldstone2001; Goodwin Reference Goodwin2001; Karl Reference Karl1990; Lawson Reference Lawson2019; Tilly Reference Tilly1978), distinguishing it from the far rarer phenomenon of “social revolution,” which strives for profound transformation in state institutions and societal relations (Huntington Reference Huntington1968; Skocpol Reference Skocpol1988). Moreover, as will become important for the arguments and analyses that follow, this more capacious understanding of revolution leaves open the question of revolutionary tactics; its only criteria are that the campaign be directed at regime change and that a sufficient number of everyday citizens be involved. These citizens can be organized in a nonviolent revolutionary movement staging protests and demonstrations, or they can be recruited into a revolutionary militia devoted to fighting the incumbent through guerrilla insurgency (Beissinger Reference Beissinger2022; Chenoweth and Stephan Reference Chenoweth and Stephan2012; Goldstone Reference Goldstone2001; Reference Goldstone2014). Indeed, in some cases these forms of resistance hybridize, with some revolutionaries opting for armed resistance while others remain committed to nonviolence. In other cases, a campaign that begins with nonviolent resistance may devolve into violent armed combat. The definition of revolution adopted here therefore follows existing works in treating revolutionary tactics, including the use of armed versus unarmed resistance, as a characteristic that varies—not as constitutive of the phenomenon itself.

A number of other terms in my definition are worth highlighting, particularly as they inform the coding of data and the empirical analyses that follow. First, a counterrevolution is defined as an effort rather than an outcome. Like revolutions, coups, rebellions, and other forms of regime change, counterrevolutionary efforts may either succeed or fail. This means that explanations of counterrevolution should be broken into two parts: first we must explain why counterrevolutions emerge and then why they succeed. Second, I note that a counterrevolution must take place through irregular action, like a coup, invasion, rebellion, or uprising. It therefore excludes efforts by former incumbents to capture power through the institutional channels established by the new regime, as, for example, when authoritarian successor parties return through elections in new democracies (Grzymala-Busse Reference Grzymala-Busse2002; Loxton and Mainwaring Reference Loxton and Mainwaring2018). Third, the goal of the effort must be restoration of a version of the prerevolutionary regime. In other words, it is insufficient to simply reimpose authoritarianism. Instead, a counterrevolution is an effort to resurrect a version of the governing rules that characterized the former regime and to restore members of that ruling elite to power.

Finally, it is important to distinguish my definition of counterrevolution from a related concept in an adjacent literature: democratic breakdown. The overthrow of a nascent democracy by antidemocratic forces has obvious affinities with my understanding of counterrevolution. Indeed, though not all democratic transitions are effected through revolution and not all breakdowns involve the return of the former regime, it is no doubt true that some of these breakdowns would qualify as counterrevolutions. When a democratic breakdown meets my criteria for counterrevolution—that is, it follows a political revolution, involves irregular action, and seeks to restore a version of the former regime—it is included in my dataset and analyses.

Yet despite some degree of conceptual overlap, there are good reasons to treat these phenomena as distinct for the purposes of developing theory. The most obvious of these reasons is that, as I explain in the next section, postrevolutionary transitions tend to be marked by a fairly well-defined distribution of power and interests between the old and new regimes, which directly shape counterrevolutionary processes. In contrast, democratization research tends to lump together transitions effected through revolutionary mobilization and those involving elite pacts. Moreover, the dynamics of these latter transition are somewhat distinct, in part because democratizers generally have far less initial leverage over former incumbents than they do following revolutions (Haggard and Kaufman Reference Haggard and Kaufman2016; Kadivar Reference Kadivar2018; Karl Reference Karl1990). As a result, the tactics that have been found to be important for avoiding democratic breakdown, like rapid demobilization, moderation in policies, and conciliation and compromise toward former officials, are likely to be ineffective in thwarting counterrevolutions (Diamond Reference Diamond2008; Higley and Burton Reference Higley and Burton1989; Huntington Reference Huntington1984; O’Donnell and Schmitter Reference O’Donnell and Schmitter1986). In fact, as I discuss in the Conclusion, this paper’s findings suggest that democratic revolutionaries’ best hopes for guarding against counterrevolution may be to remain as mobilized as possible so that they retain their ability to confront counterrevolutionary challenges with a return to mass protest. To summarize, although the scholarship on democratic breakdown may inform our understanding of counterrevolution, its arguments cannot alone explain this outcome.

Revolutionary Violence

Having now defined and conceptualized counterrevolution, we can begin developing some theoretical intuitions about when we might expect it to occur. As noted above, because counterrevolution is conceptualized as an effort that can either succeed or fail, we must consider counterrevolutionary outcomes in two parts: (1) when are counterrevolutions most likely to emerge and (2) when are they most likely to succeed?

One place to begin our theorizing is with the distribution of interests and coercive capacities between revolutionaries and old regime forces in the postrevolutionary critical juncture. On the one hand, we might expect patterns in counterrevolution to be shaped by the interests of the old regime. Of course, in all successful revolutions incumbents are forced from power against their will, and they are, as a result, likely to maintain a strong desire to return to office.Footnote ³ But the extent of this desire may vary considerably across revolutions, depending on the nature of the new regime. Revolutionary governments that undertake purges, pursue punitive justice, or threaten old regime interests through radical policies are likely to generate particularly acute desperation among these former elites, making it more likely that they will launch a risky gambit to grab back power.

In addition to interests, we might expect coercive capacities—of both the old regime and the new one—to pattern counterrevolutionary outcomes. Revolutions, by definition, entail that challengers have managed to overpower the old regime through some combination of armed insurgency and mass mobilization. But, like old regime interests, this revolutionary power advantage varies considerably in nature and extent across different revolutions. In some cases, the old regime is left with its coercive apparatus mostly or entirely intact, with revolutionaries only able to exert influence through their potential to return to revolutionary mass mobilization (e.g., the Egyptian and Tunisian revolutions of 2011). In other cases, revolutionaries may have an army of their own, and they can use this coercive apparatus to defend their revolutionary gains (e.g., the Chinese and Vietnamese revolutions). The relative coercive resources available to revolutionaries and former incumbents in the postrevolutionary critical juncture should shape the likelihood of both counterrevolutionary emergence and success. When old regime forces maintain some or all of their coercive might, they will be better equipped for a counterrevolutionary campaign. And when revolutionaries themselves have their own army or militia to draw on, they will be better positioned to defeat these counterrevolutionary challenges.

What are the primary determinants of these relative interests and coercive capacities in the postrevolutionary period? Here I argue that the nature of the revolutionary process—and, specifically, whether challengers embrace strategies of violent or nonviolent resistance—should determine the extent of old regime desperation and the distribution of coercive capacities after the revolution ends. In developing this argument, I draw on two scholarships. The first is the literature on nonviolent resistance, which has argued that nonviolence is a more effective strategy for toppling dictators than violent insurgency (Chenoweth and Lawrence Reference Chenoweth and Lawrence2010; Chenoweth and Stephan Reference Chenoweth and Stephan2012; Nepstad Reference Nepstad2011; Schock Reference Schock2005). The second is an older literature on social revolution, which traces the origins of political order and durable state institutions to violent conflict (Huntington Reference Huntington1968; Skocpol Reference Skocpol1979; Tilly Reference Tilly1993). Though these two literatures are often contrasted, interestingly they together point to a fairly consistent set of expectations regarding counterrevolutionary emergence and success.

On the one hand, we might expect counterrevolutions to be more likely following violent revolutions, because they induce heightened levels of desperation in former incumbents. Violent revolutions produce regimes that are far better able to purge and punish their enemies. For example, Mulholland (Reference Mulholland2017) argues that, throughout European history, revolutions achieved through military occupation were better able to subordinate resistant state bureaucracies. Old regime elites living under the rule of such regimes are therefore especially likely to fear for their lives, safety, and interests, and these fears can be a powerful spur to action. Slater and Smith (Reference Slater and Smith2016) find, for example, that the fears provoked by violent social revolutions that fail can bind the surviving ruling coalitions together for years or even decades afterwards. A similar mechanism of fear and desperation may make old regime elites living under violent revolutionary rule especially likely to attempt counterrevolutions.

However, if we think about how violence affects the coercive resources available to old regime elites, we reach a somewhat different set of conclusions. Nonviolent campaigns succeed by eliciting defections from the incumbent’s supporters, including, most crucially, members of the military (Chenoweth and Stephan Reference Chenoweth and Stephan2012; Goodwin Reference Goodwin2001; Schock Reference Schock2005; Tilly Reference Tilly1978). In contrast, violent campaigns achieve victory only when they are able to defeat an incumbent’s armies through guerrilla war or urban insurgency (Lachapelle et al. Reference Lachapelle, Levitsky, Way and Casey2020; Levitsky and Way Reference Levitsky and Way2013). The latter is a more difficult path to victory than the former—which is one reason that Chenoweth and Stephan (Reference Chenoweth and Stephan2012) argue nonviolent campaigns succeed more often than violent ones. But the flip side of this easy success is that nonviolent revolutions barely touch the old regime’s coercive apparatus. If these old regime agents later sour on the new government, they therefore have considerable means at their disposal with which to launch a counterrevolution.

Violence and nonviolence also crucially shape the capacities and strength of the new revolutionary government, including its ability to defeat counterrevolutionary challenges. As scholars of social revolution have long argued, the loyal military organizations built during violent revolutionary struggles become a crucial tool by which new regimes subsequently consolidated their rule (Huntington Reference Huntington1968; Skocpol Reference Skocpol1979; Tilly Reference Tilly1993). Waging violent revolution for any sustained period requires the development of at least some degree of armed organization, and these organizational resources can subsequently be channeled into the construction of a permanent revolutionary military. As Ted Gurr put it, “the revolutionary fighters and zealots become the cadre of new or transformed agencies of state security” (Reference Gurr1988, 53). New regimes rely on these loyal cadres in their initial years to put down bottom-up threats, which, according to Levitsky and Way (Reference Levitsky and Way2013), is one important reason that revolutionary regimes tend to be so strong and durable (also, Lachapelle et al. Reference Lachapelle, Levitsky, Way and Casey2020). Similarly, Meng and Paine (Reference Meng and Paine2022) argue that the process of waging violent rebellion generates loyalty and experience between rebel leaders and their military officers, which, after coming to power, helps these leaders solve the “guardianship dilemma”—that is, granting officers the power and authority to put down bottom-up threats (like counterrevolutions) without risking an internal coup.

In contrast, nonviolent revolutionary regimes may be more susceptible to counterrevolutionary threats. In fact, the same mechanisms that grant an advantage to nonviolent revolutions in toppling dictators may later generate opportunities for those same dictators to return. A crucial feature of nonviolent revolutions is the breadth and diversity of their revolutionary coalitions (Beissinger Reference Beissinger2022; Dix Reference Dix1984). Nonviolent campaigns are better able to attract large numbers of participants, and the more everyday citizens that pour out into the streets the more likely elites are to defect from the incumbent regime (Chenoweth and Stephan Reference Chenoweth and Stephan2012). But although broad coalitions are helpful in forcing dictators from power, they represent a liability when it comes to governing. As Rule and Tilly put it in their study of the 1830 revolution in France, “the initial seizure of control [by revolutionaries] requires a larger coalition than does the maintenance of control” (Reference Rule and Tilly1972, 56). Because of their broad coalitions, scholars like Goldstone (Reference Goldstone2014) and Nepstad (Reference Nepstad2011) have found that nonviolent revolutionary regimes are often weak and unstable, with a tendency to backslide from democracy back into authoritarianism. Beissinger (Reference Beissinger2022) similarly finds that the rapidly convened and coalitional nature of urban civic revolutions (which also tend to be nonviolent) make them less durable and less transformative than violent social revolutions. These revolutions lack the institutional and coercive resources to manage the squabbling and infighting that define postrevolutionary transitions (Beck Reference Beck2015; Foran and Goodwin Reference Foran and Goodwin1993; Goldstone Reference Goldstone2014). And because their campaigns were waged through nonviolent struggle, they lack their own loyal coercive force to help them put down threats, forcing them to rely instead on the military organization of the regime they just felled. For all these reasons, we might expect nonviolent revolutionary regimes to present potential counterrevolutionaries with far more opportunities to return to power.

This theoretical discussion points to a clear set of hypotheses about the relationship between revolutionary violence and counterrevolution. On the question of counterrevolutionary emergence we have two competing hypotheses. On the one hand, violent revolutions might induce more fear and desperation in the remnants of the old regime, pushing them to attempt a counterrevolution. On the other hand, violent revolutions destroy the coercive capacities of these former incumbents, making it harder for them to get a counterrevolution off the ground. On the question of counterrevolutionary success we have a more consistent prediction. Violent revolutions generate coercive resources that the new regime can then use to consolidate its rule and put down counterrevolutionary threats. In contrast, nonviolent revolutions produce fractious regimes based on broad coalitions that lack the coercive tools to defeat bottom-up challenges.

These hypotheses are summarized in Figure 1. We have two competing and opposite predictions regarding violence and counterrevolutionary challenges (H2a and H2b). Then we have a single and more consistent prediction that violence lowers the likelihood of counterrevolutionary success (H3). How these causal dynamics combine to shape counterrevolutionary outcomes in aggregate—that is, whether a revolution is reversed by counterrevolution or not—are somewhat hard to predict. However, given that there is a causal logic connecting violence both to lower rates of counterrevolutionary emergence and to lower rates of success, we might expect that, on net, the more violence deployed during a revolution the less likely that revolution is to be subsequently overturned (H1).

Figure 1. Diagram of Hypotheses and Mechanisms

Finally, it is important to note that revolutionary violence is unlikely to be the only variable that patterns counterrevolutionary emergence and success. For example, previous work on counterrevolution has emphasized the importance of international sponsorship (Allinson Reference Allinson2019; Bisley Reference Bisley2004; Halliday Reference Halliday1999; Jones Reference Jones2013; Walt Reference Walt1992). Domestic counterrevolutions are often funded and armed by sympathetic foreign powers, and foreign sponsorship can also be decisive in propping up new revolutionary regimes. Another potentially important variable is the existence of a strong revolutionary party, as a number of studies have argued that strong parties are associated with regime strength and durability (Anria and Cyr Reference Anria and Cyr2016; Huntington Reference Huntington1968; Levitsky and Way Reference Levitsky and Way2012; Slater and Smith Reference Slater and Smith2016). In the models and case studies below, I do control for and consider these alternative explanations. However, I leave fuller exploration of these potentially important causal pathways for future research.

Data on Counterrevolution

Because no one has ever collected data on counterrevolutions, we do not have even basic statistics about how often they occur. Therefore, I constructed an original dataset of all counterrevolutionary challenges globally from 1900 to 2015. To assemble these data, I began with a dataset built by Beissinger (Reference Beissinger2022) of successful political revolutions in the twentieth and twenty-first centuries (n = 123). In line with the discussion above, Beissinger adopts an understanding of political revolution as regime change involving the mass mobilization of everyday citizens (both armed and unarmed), and his dataset includes every campaign that meets these criteria. It therefore provided me with the full universe of cases in which counterrevolution was conceptually possible. Using a number of secondary sources (enumerated in Appendix A1), I conducted research on each of these cases to identify discrete instances of counterrevolution.

I was guided in my research by the three inclusion criteria in the definition above. First, the definition specifies that the mode of action must be irregular, so I looked for challenges to the revolutionary regime that occurred outside of institutional channels, including rebellions, mass protests, foreign invasions, and coups. The second criterion specifies that the goals of the challenge must be restoration of some version of the old regime. In evaluating whether an effort met this standard I looked at whether individuals who held official positions in the former regime led or joined the challenge, inferring from their participation that the goal of the effort included old regime restoration in some form. In other words, I did not include efforts that were simply led by those who supported the old regime (e.g., members of the former incumbent’s ethnic group or classes that benefited under the former regime). Finally, for the timing criterion—that a counterrevolution must occur in the aftermath of revolution, before the new regime has become institutionalized—I used a cutoff point of 10 years. In choosing a temporal cutoff as my method of operationalization, I followed the guidelines of Gasiorowski and Power (Reference Gasiorowski and Power1998), who operationalize democratic consolidation by identifying the temporal moment when democratic breakdown becomes far less likely. Similarly, I used the survival curve of revolutionary regimes in Beissinger’s (Reference Beissinger2022) dataset to identify an inflection point at years 7 to 10 when the probability of revolutionary regime collapse declines significantly. In identifying counterrevolutionary challenges I therefore focused on the decade following the end of the revolution (unless the revolutionary regime ended during this period, in which case the moment of termination marked the end point for my analysis).Footnote ⁴

Through this research, I identified 98 discrete counterrevolutionary challenges, which occurred across 65 cases of revolution (as some revolutions experienced multiple counterrevolutions). Of these counterrevolutions, 22 ended in success and 76 ended in failure. In other words, of the 123 revolutions that have occurred since 1900, 53% of them witnessed at least one counterrevolutionary challenge and 18% of them were actually overthrown by counterrevolutions. I use these data to construct the dependent variables in the statistical and set-theoretic analyses that follow. But before turning to these analyses let us briefly consider the list of revolutions that were actually overturned. In Figure 2, I enumerate these revolutions, along with the name of the counterrevolution that overturned it, the location, the end year of the revolution and the start year of the counterrevolution, and the number of people that were killed during the revolution (as a proxy for violence). These data on deaths were taken from Beissinger’s dataset.

Figure 2. Successful Counterrevolutions since 1900

In this table we see some compelling initial evidence that nonviolent revolutions are more likely to succumb to counterrevolutions. Many of the revolutions on the list involved very few deaths. They mostly occurred through unarmed mass mobilization and brought to power governments comprising diverse coalitions that lacked coercive organizations of their own. Counterrevolution then unfolded in a fairly similar way across cases: old regime generals soured on the revolutionary project and took advantage of opportunities created by coalition infighting to stage counterrevolutionary coups. This was the pattern in the handful of early twentieth-century Latin American revolutions on the list. It was also the pattern for the democratic revolutions during the Cold War—in South Korea, Thailand, Sudan, Ecuador, and Haiti. And most recently this pattern has played out following Egypt’s 2011 revolution, when divisions between Islamists and secularists in the revolutionary coalition paved the way for a return of military rule.

Furthermore, we see that the list contains none of the famous twentieth-century revolutions that appear in much of the literature—for example, the violent social revolutions in Russia, China, Vietnam, Cambodia, Algeria, Angola, Cuba, or Nicaragua. We do see a small number of other violent revolutions on the list—the Chechen civil war following Russia’s first counterrevolutionary invasion, the Taliban Revolution, and the Xinhai Revolution in China. But these cases may be the exceptions that prove the rule, as two of them were only overturned because powerful foreign actors decided to intervene militarily (i.e., Russia’s second invasion of Chechnya and the US invasion of Afghanistan). Of course, we can only infer so much from looking at the characteristics of “positive” cases. Therefore, in the next section I use statistical analyses to model the emergence and success of counterrevolution across the full universe of modern revolutions.

Statistical Analyses

Results

Results for the main analysis are depicted in Table 1, which includes three models, one corresponding to each measure of revolutionary violence: the number of deaths in the revolution (logged), the “civil war” binary variable, and the “revolutionary militia” binary variable. The results show that, regardless of how violence is operationalized, it is robustly and negatively associated with counterrevolution. In other words, in line with Hypothesis 1, revolutions in which the opposition resorted to violence are significantly less likely to be overturned by counterrevolutions. There are also some noteworthy results from certain control variables. For example, foreign sponsorship appears to complement revolutionary violence as a major deterrent of counterrevolution. There is also a negative relationship between the end year of the revolution and counterrevolution, indicating a secular decline in rates of counterrevolution over the last twelve decades.

Table 1. Revolutionary Violence and Counterrevolution (Penalized Logistic Regression)

Note: *p < 0.10, ^**p < 0.05, ^***p < 0.01.

Table 2 lays out models that split the dependent variable into two parts—counterrevolutionary emergence and counterrevolutionary success—which helps us to understand why revolutionary violence is negatively associated with counterrevolution (i.e., H2 and H3). The first three models analyze whether a counterrevolution emerges, and the second three models analyze whether a counterrevolution succeeds, conditional on one having emerged. Again, we see consistent results regardless of how we operationalize violence. Revolutionary violence has no significant relationship with the emergence of counterrevolution. These results provide little support for either H2a or H2b (though the fact that the coefficients are negative, even if not statistically significant at normal levels, provides slightly more support for H2b). Instead, we see that the negative relationship between revolutionary violence and counterrevolution is due primarily to a superior ability to weather counterrevolutionary challenges when they emerge (H3).

Table 2. Revolutionary Violence and Counterrevolutionary Emergence/Success (Penalized Logistic Regression)

Note:*p < 0.10, ^**p < 0.05, ^***p < 0.01.

Table 3. Contingency Table of Revolutionary Violence and Counterrevolution

Note: Consistency: 0.92; Coverage: 0.57.

Figure 3 plots the marginal effect of deaths on each of the three outcomes. We see that the probability of a counterrevolution following revolutions involving very few deaths is about 20%, whereas in revolutions where hundreds of thousands of people die the probability of counterrevolution falls to 8% or lower. Furthermore, when we look at the two disaggregated outcomes, we see clearly that though the relationship with counterrevolutionary challenges is negative there is a much stronger relationship with counterrevolutionary success. The probability of a counterrevolution succeeding is about 10% following very violent revolutions, whereas it approaches 50% following nonviolent revolutions.

Figure 3. Marginal Effect of Deaths on Likelihood of Counterrevolution

We see similar patterns in Figure 4, which plots the marginal effects of the two binary variables. The probability of a counterrevolution is 16% lower following revolutions that devolve into civil wars. But a civil war only slightly lowers the probability of a counterrevolution emerging (6% lower), whereas it lowers the probability of counterrevolutionary success by 23%. We see an even stronger relationship with the measure capturing whether the new regime has its own military organization. Counterrevolutions are 22% less likely under these regimes and, again, the relationship stems primarily from the way in which these armies lower the likelihood of counterrevolutionary success (26% lower) rather than emergence (1% lower).

Figure 4. Marginal Effect of a Revolutionary Militia and Civil War on Likelihood of Counterrevolution

Overall, these findings support the idea that revolutions in which challengers take up arms and deploy violence yield regimes that are less likely to be overturned by counterrevolutions. Moreover, the analyses shed light on the mechanism through which these outcomes are related. Violent revolutions tend to survive because they yield regimes with the tools to put down counterrevolutionary challenges when they emerge. However, violence does not seem to affect the likelihood of counterrevolution emerging in the first place—either because of the desperation it instills in potential counterrevolutionaries or because of the damage it inflicts on their coercive might. Without further analysis it is difficult to know what is behind this null result. One possibility is that violence simply does not have much bearing on counterrevolutionaries’ decision to launch a challenge or not. Another, and perhaps more likely, possibility is that the causal processes captured by H2a and H2b are both at work—and thus cancel each other out. In other words, violent revolutions might leave old regime forces with few coercive resources but more incentive to launch a challenge. Conversely, nonviolent revolutions leave more of the old regime intact but give their agents less reason to attempt a risky counterrevolutionary gambit. With these two dynamics pulling in opposite directions, we might see little relationship in statistical analyses between revolutionary violence and the emergence of counterrevolution.

Although these statistical findings are striking in their strength and consistency, there are obvious limitations to the inferences we can draw from observational analyses of a relatively small cross-sectional dataset. The two biggest issues are the deeply endogenous nature of the causal process we are trying to understand and the small number of observations. On endogeneity, the main issue is that the factors that push a revolutionary movement down a violent versus nonviolent path may also shape the likelihood of counterrevolution. Although I have tried to identify and control for some of the most obvious variables that could be related to both outcomes, there may be other confounding variables that I could not account for. One point that should give readers some reassurance: many of the variables that would seem to affect both revolutionary violence and counterrevolution are most likely to affect counterrevolutionary emergence because they shape the capacities of old regime elites (e.g., mountainous terrain) or their interests in returning to power (e.g., former regime type). The fact that I find a strong relationship between violence and counterrevolutionary success, but not emergence, should give readers some confidence that an unobserved confounder is not biasing the findings.

As for issues of sample size, beyond the concern that estimates may be biased upward (and thus the need to use a penalized logistic regression), there are two additional concerns: one is that the models are overfitted (given the ratio of predictors to observations), and the second is that results may be driven by certain especially high-leverage observations. Therefore, in Appendix A5 I show that results are robust to different combinations of controls and to the removal of the highest-leverage observations. However, a more fundamental critique might be that such a small dataset requires an entirely different analytical setup—that is, a set-theoretic approach that examines whether violence/nonviolence are necessary or sufficient conditions for counterrevolution. In the next section, I show that the findings above are robust to this alternative approach.

Set-Theoretic Analysis

An alternative to statistical modeling for datasets with relatively small sample sizes is the use of Boolean set theory. The logic of this approach is different from that of statistical analysis. Whereas, statistical modeling is directed at identifying probabilistic relationships, Boolean approaches are directed at finding necessary or sufficient relationships between conditions (Goertz and Mahoney Reference Goertz and Mahoney2012; Ragin Reference Ragin1987).

Here I show partial results from a set-theoretic evaluation of my hypotheses about revolutionary violence and counterrevolution. Though more complex Boolean analyses involve examining multiple causal pathways with different combinations of conditions, I am concerned with a single independent variable—revolutionary violence—and so conduct only a simple analysis of how this condition is related to counterrevolution. The clearest way to show set-theoretic relationships between two variables is with 2 × 2 contingency tables that represent distributions in cases according to a causal condition (for our purposes, revolutionary violence) and an outcome condition (counterrevolution). In Table 3 I show the relationship between one of my violence measures (the revolutionary militia variable) and the aggregate counterrevolution outcome. The rest of the tables, using the other violence measures and the disaggregated counterrevolution outcomes, can be found in Appendix A6. For each table, I also calculate the consistency and coverage scores, which are useful for specifying necessity and sufficiency.

The table and resulting consistency and coverage scores tell a clear story, which is in line with the statistical analyses above. There are only five counterrevolutions that topple revolutionary regimes with their own coercive organizations and 17 that topple those lacking a revolutionary army. The consistency score of 0.92 further suggests that nonviolence is a virtually necessary condition for a successful counterrevolution and that revolutionary violence is a nearly sufficient condition for no counterrevolution. Moreover, when we produce the same tables using the disaggregated counterrevolution outcomes (which appear in Appendix A6), we find that there is no relationship of necessity or sufficiency between revolutionary violence and counterrevolutionary emergence (consistent with the null results in the statistical analyses) but a strong relationship with success. Overall, then, the findings from both statistical and set-theoretic analyses point in the same direction: when revolutionaries come to power through violent resistance it is virtually impossible for the old regime to return through counterrevolution.

Revolution in Cuba: 1933 versus 1959

In this final empirical section, I use qualitative evidence from two carefully paired cases to trace the mechanisms behind the relationships identified above, which can be a good way to establish causality in observational studies (Goertz Reference Goertz2017; Seawright Reference Seawright2016). Specifically, I compare Cuba’s 1933 revolution, a nonviolent revolution that was toppled by a counterrevolution, and its 1959 revolution, a violent revolution that successfully survived multiple counterrevolutionary threats. These cases offer a fairly unique opportunity for a “most-similar-systems” paired comparison. Normally with this design one would compare two similar countries with different trajectories or outcomes (Lijphart Reference Lijphart1975). But these comparisons still have to contend with the fact that even very similar countries usually have a host of differences in their cultures, demographics, political economies, etc. With these two revolutions I am able to compare two episodes from within the very same country that occurred within relatively close temporal proximity, therefore mitigating some (but not all) of the “degrees of freedom” problem that can be an issue with the comparative method. In other words, I am able to focus on differences between the revolutions themselves, bracketing other background or underlying variables like culture or economy. Nevertheless, as with all paired comparisons, the power of the inferences and the strength of the analysis comes from the combination of comparing like cases and tracing causal processes within these cases (Falleti and Mahoney Reference Falleti and Mahoney2015; Slater and Ziblatt Reference Slater and Ziblatt2013), with the latter helping to adjudicate between rival explanations that cannot be eliminated through the structure of the comparison.