The Diffusion of Nonviolent Campaigns

Beyond Civil War and the Diffusion of (Violent) Conflict

Although our interest here is nonviolent mobilization, we start with what we know about the diffusion of violent conflict from the extensive research on civil war, in particular over the last two decades (see, e.g., Blattman and Miguel 2010; Cederman, Gleditsch, and Buhaug 2013). Much of the traditional research on civil war has looked for factors and events within countries to understand the origins of violent conflict between non-state actors and governments. There are obviously many important domestic features that may motivate or facilitate resort to violent conflict such as various grievances induced by the state or forms of state weakness. However, the limits to such a domestic “closed polity” model are apparent from merely a casual perusal of actual civil wars, which often display prominent transnational dimensions (see, e.g., Cederman, Gleditsch, and Buhaug 2013; Gleditsch 2002a). The actors themselves are often transnational. Many civil wars involve ethnic groups that transcend national boundaries, and insurgent groups mobilize resources in transnational communities (see Cederman, Girardin, and Gleditsch 2009; Cederman et al. 2013; Salehyan 2009). Moreover, fighting and actors are often not confined within the borders of the main country experiencing civil war. Access to extraterritorial bases or material resources in other states can afford potential insurgents many tactical advantages (see, e.g., Salehyan 2009). States generally have difficulties monitoring and controlling areas outside their sovereign territory, and borders have asymmetric implications for states and insurgents. Although borders are not necessarily difficult to cross in a military sense, states face significant costs for violating the sovereignty of other states that are less relevant to constraining rebels (see Gleditsch, Salehyan, and Schultz 2008). Finally, conflict in other states can have spillover effects that are not actor-specific, such as access to weapons at a lower cost or the spread of ideologies that facilitate insurgent mobilization.

Existing research has found a great deal of evidence consistent with the claims that transnational factors can increase the risk of armed conflict. Studies have consistently demonstrated strong spatial contagion in conflict, or a higher likelihood of civil war when neighboring countries experience conflict, even when considering other potential factors likely to affect the risk of civil war (e.g., Gleditsch 2007; see also Hegre and Sambanis 2006 who highlight this as a robust finding in their sensitivity analysis). Beyond contagion per se, a number of studies have tried to examine the specific proposed mechanisms underlying the tendency for civil wars to spread among neighboring states and the particular constellations that promote conflict. Many studies find evidence that transborder ethnic groups can raise the risk of violent conflict (see Buhaug and Gleditsch 2008; Cederman, Girardin, and Gleditsch 2009; Cederman, Gleditsch, and Buhaug 2013), even though others also find evidence of transnational influences on conflict without shared ethnic ties (see, e.g., Salehyan and Gleditsch 2006 on refugees). Some studies show that the international environment more broadly can affect the risk of civil war; for example, previous specific characteristics of neighboring countries and relations between states such as whether neighboring states are more democratic and the extent of trade in a region can also decrease the risk of civil war onset (see, e.g., Gleditsch 2007). In general, the evidence for diffusion and important transnational linkages seems sufficiently strong that many observers argue that civil wars cannot be fully understood without considering their transnational environment (see, e.g., Checkel 2013).

Conventional civil war or violent uprising is of course only one of many tactics that non-state actors may use in conflicts with the state. The literature on contentious politics has stressed a much wider repertoire of strategies (see, e.g., Tilly 1978; Tarrow 1994), including forms of indirect violence such as terrorism, and various forms of nonviolent direct action, where dissidents try to coerce the state through noncooperation and for various reasons refrain from using violence. By nonviolent direct action, we here mean nonroutine behavior, outside normal political channels, in opposition to the state, and not using violent methods (see, e.g., Bond 1994; Sharp 1973). This encompasses both “acts of commission” (i.e., doing something not sanctioned by the state, such as illegal demonstrations) and “acts of omission” (i.e., refusing to do something ordered by the state, such as disobeying orders to disperse). We examine whether mechanisms conceptually similar to those highlighted for the diffusion of civil war may also lead to diffusion among nonviolent direct action campaigns. We first provide a brief overview of existing research on nonviolent direct action and the relative lack of attention to transnational features. We then proceed to detail how nonviolent and civil wars can differ and develop a theory of how nonviolent direct action can spread and increase nonviolent mobilization in other states.

Existing research on nonviolent direct action is much less extensive than research on civil wars and has generally paid little attention to the potential role of transnational factors and diffusion. Much research on nonviolent direct action has focused on normative aspects of nonviolence or documenting specific historical cases and rarely engaged with more general social science or conflict research. Sharp (1973) and others identify possible mechanisms through which nonviolent direct action may work, but the empirical examples tend to be limited to selected cases, typically only successful outcomes. There has recently been a wave of systematic studies on nonviolent conflict, owing in part to new comprehensive data such as the Nonviolent and Violent Campaigns and Outcomes (NAVCO) data developed by Chenoweth and Stephan (2011). One salient finding is the higher rate of success for nonviolent campaigns compared to violent campaigns. Chenoweth and Stephan attribute this to advantages inherent to nonviolent direct action, including the higher capacity for tactical innovation and mobilizing more individuals, as well as the improved prospects for eliciting defections from the government side (see also Nepstad 2013). Although this research has been very helpful, it tends to be limited to observed campaigns and their outcomes and does not try to account for the emergence of campaigns, or the possible role transnational factors may play in this.

The general literature on contentious politics (both violent and nonviolent) has tended to stress country-specific features in the emergence of dissent, in terms of either motives or factors that generate grievances (e.g., ethnic exclusion, poverty, and economic inequality, see, e.g., Davies 1962; Gurr 1970) or opportunity structures that can facilitate collective action or mobilization (e.g., McAdam 1982; Tarrow 1994; Tilly 1978). There has been some attention to the possibility of emulation or contagion. Tarrow (1994) suggested the term “modular” protest to highlight possible demonstration effects in the spread of dissent (see also Koopmans 1993). Some literature on revolutions has stressed waves of uprisings and their common origins—whether ideological or similar structural conditions—with the 1848 revolutions or the Arab Spring as prominent recent examples (see Weyland 2009, 2012). However, there have been few systematic comparative studies of nonviolent contentious politics that explicitly consider diffusion and transnational linkages. Some studies have looked at how external features may motivate protest or facilitate direct action. ¹ Some recent studies have been attentive to transnational factors and diffusion in protest against dictatorships (e.g., Beissinger 2007; Bunce and Wolchick 2011), but these tend to be limited to a single or few cases rather than broadly comparative. Most relevant for our purposes, Braithwaite, Kucik, and Maves (2012) consider contagion of nonviolent campaigns, but only for autocracies, and do not consider the specific geographic location where events take place.

Many of the mechanisms stressed for why civil wars tend to diffuse or spread may apply to nonviolent conflicts as well. Although it may be tempting to extend existing procedures to new but seemingly similar domains, we also believe that there are many plausible important differences between civil wars and nonviolent direct action campaigns. Many civil wars are motivated by ethnic claims, where the ethnic groups themselves often have a transnational presence or ties to communities in other states. Ethnic groups may of course use nonviolent means instead of, or possibly in addition to, violent means (see, e.g., Shaykhutdinov 2010). However, most instances of nonviolent direct action involve nonethnic claims on the government rather than calls for secession or territorial autonomy. According to the NAVCO data, 80 percent of the nonviolent campaigns over the period 1900 to 2004 involved popular demands for regime change, whereas only 3.7 percent were linked with ethnic claims for secession. Existing research also suggests that nonviolent direct action over territorial claims generally has been less effective than claims on the government or regime change (see Svensson and Lindgren 2011). Since ethnic movements—at least historically—figure less prominently in nonviolent direct action, we are less likely to see inherently transnational actors here than in violent—and frequently ethnic—conflict.

Beyond incompatibility and motive, it is also clear that very different groups tend to resort to nonviolent direct action and violent means. In addition to the link to ethnic claims, civil wars often take place in the periphery and involve small, poor, and excluded groups (see Buhaug and Lujala 2005). By contrast, campaigns that adopt nonviolent direct action as the main tactic tend to take place in urban environments, involve more participants and broad coalitions, and often draw on educated middle classes such as students and white-collar workers. Nonviolent campaigns are unlikely to have much leverage without achieving broad mobilization and at least some influence in the capital (see, e.g., Dahl et al. 2014). Transnational factors can arguably promote resort to violence even in the absence of domestic conditions favoring violence, since many types of violent attacks are feasible with relatively few participants (consider, e.g., outside state support to terrorist attacks to target rival states). However, transnational resources and opportunities alone do not make nonviolent direct action more likely without sufficient initial domestic motivation, and if groups can mobilize large numbers in the first place then transnational support may have little added impact. Violent insurgencies tend to take advantage of potential transnational support, although this often comes with strings attached and may entail a loss of autonomy for the movement (see Salehyan, Gleditsch, and Cunningham 2011). Since nonviolent direct action often appears to succeed through generating defection (see, e.g., Chenoweth and Stephan 2011; Nepstad 2013), the potential resource-autonomy trade-offs may loom much larger if governments find it easier to discredit “foreign-funded” campaigns as illegitimate. This in turn could make campaigns more hesitant to accept or rely extensively on direct outside support.

Finally, the role of spatial proximity emphasized in existing research on civil war linkages could be less relevant for nonviolent direct action if this requires fewer material resources, as there in principle is no inherent reason why, for example, emulation and inspiration needs to be limited to neighboring states.

Our argument here is not that nonviolent direct action campaigns never diffuse or that regional factors cease to be important. But the plausible differences between violent and nonviolent campaigns should sensitize us to how the specific weight of relevant features may differ for nonviolent direct campaigns and highlight the need for further independent research before relying on direct analogies from the diffusion of violent conflict.

Nonviolent Direct Action and Diffusion

Resort to violent and nonviolent direct action against the state requires both motivation and opportunities. Much of the civil war and contentious politics literature has emphasized variation in the latter, based on the notion that plausible motives and grievances are ubiquitous and thus cannot account for direct action (see, e.g., Collier and Hoeffler 2004; Fearon and Laitin 2003; Tilly 1978). We do not agree that motivation is inherently less relevant for onset, and we believe that direct action or nonroutine politics should be much more likely in nondemocratic systems with significant political and economic exclusion than more equalitarian systems or democratic systems that afford individuals opportunities to pursue their interest through routine political channels (see, e.g., Bond 1994; Cederman, Gleditsch, and Buhaug 2013). However, save for extreme cases where campaigns elsewhere draw attention to inequalities and generate new and more salient grievances, campaigns elsewhere are more likely to have a direct effect on opportunities for collective action rather than individual motives for direct action.

Collective action theory highlights many important barriers to mobilization around common interest (see Sandler 1992). Since any benefits from dissent are non-excludable and cannot easily be restricted to participants, many have stressed how the individual costs of participation and risk of government repression should make individuals inclined to free ride on others and discourage active participation (see, e.g., Tullock 1971 on the so-called Paradox of Revolution). Of course, collective action problems can also apply on the government side. Since repression is costly and often increases opposition, state security agents may choose to expend little effort, or at the extreme defect if the tide seems to shift (see Lichbach 1998, 413). Still, the state’s ability to enforce compliance makes the collective action problem loom larger on the dissident side.

Dissident collective action can be achieved in a number of ways (see, e.g., Lichbach 1995), and successful mobilization in other states can lower the barriers for successful collective action. In general, collective action will be easier if mobilization in another state alters the perceived balance of power between an incumbent regime coalition and dissidents. This could conceivably happen through a direct impact on the resource base, for example, if a regime is heavily dependent on support from another state and this regime is toppled. However, such instances of revolutionary change are rare, and relative power is more likely to be affected more indirectly through learning and emulation. By learning we here mean changes in beliefs that result from new information about or changes in the behavior of other actors, whereas emulation denotes adopting specific tactics or efforts through inspiration from examples observed elsewhere (see Simmons et al. 2006, 795).

Information is important, as dissatisfied individuals often lack information about the preferences of others in repressive political systems and cannot easily assess the extent of popular dissatisfaction (see Chwe 1999; Kuran 1990). The MTV slogan reminds us that revolution will be televised in the digital information era, and information about protest in reference states and salient events can provide cues when individuals lack information about the domestic scene. Protest in other states can help generate awareness of the extent of dissatisfaction with the regime, and how dissident sympathies are more widespread than previously anticipated, even in a repressive society without open media (see Lohmann 1994). Moreover, large-scale popular mobilization demonstrates that protest is feasible and that regimes can be more vulnerable to nonviolent tactics, thereby changing individual views in ways that encourage participating in dissent. Increasing participation elsewhere can boost further mobilization, and the costs of individual participation decrease when others are expected to participate (see DeNardo 1985). ²

Beyond informing about preferences and lowering the barriers for collective action, observing mobilization elsewhere also allows dissidents to learn about possible tactics and strategies that are likely to be most effective in challenging the government or generating large-scale mass participation. The specific tactics used in one country (e.g., gatherings in key public spaces) can provide focal points for actions in others and make it easier for people to coordinate on actions and to mobilize quickly without a great deal of prior organization (see, e.g., Schelling 1960). Disciplined nonviolent efforts that make government repression more difficult elsewhere can inspire organizers to take measures to avoid violence against the police. As such, we should expect to see a diffusion of specific tactics between campaigns rather than just dissent or direct action in general. Violent revolution clearly was in vogue during the Cold War (see, e.g., Debray 1967, for an influential text promoting guerrilla tactics). However, efforts to emulate guerrilla tactics in urban settings have generally proven ineffective (see Oppenheimer 1970). Many dissidents have concluded from recent experiences that nonviolent direct action can be an effective form of extracting concessions from governments and can raise the cost of government repression (see Chenoweth and Stephan 2011; Popovic and Miller 2015; Sharp 1973). Sharp’s (1993) handbook From Dictatorship to Democracy has been widely distributed among dissident movements, and sometimes (perhaps in an exaggerated manner) credited with generating the Color Revolutions and the Arab Spring.

Mobilization and nonviolent direct action in other countries can thus provide information that changes individual expectations, facilitates mobilization, and inspires adoption of the specific tactics. Whether the information transmitted is “correct” with regard to regime strength, opposition mobilization capacity, or the effectiveness of specific tactics is less important than the impact on potential dissidents and mobilization. Many researchers stress that misperceptions can facilitate mobilization as individuals are more likely to participate when they overestimate their personal influence, and some survey evidence backs this up (see, e.g., Finkel, Muller, and Opp 1989; Weyland 2012, 920-22). Dramatic events in other states are particularly likely to help mobilization if encouraging stronger—and perhaps exaggerated—beliefs that nonviolent direct action is feasible and able to challenge the government or extract concessions. Misperception can occur at the government side too in the event of salient events in other states. Collective action and mobilization elsewhere can undermine the cohesiveness of regime coalitions, and defections and refusals to repress become more likely as larger numbers are mobilized and when dissent remains nonviolent (Chenoweth and Stephan 2011; DeNardo 1985; Nepstad 2013; Rivera and Gleditsch 2013). ³

The specific outcomes that may arise from learning our emulation will of course depend on the relevant reference countries, and whether these are fully global or confined in particular ways, for example, to neighboring states. We have argued that nonviolent direct action is less likely to rely on direct support and directly connected actors than civil wars. As such, one might argue that events anywhere in the world could spur mobilization efforts, especially in a world more interconnected by information technology. New forms of communication technologies and social networks have been widely used in recent protest in the Middle East and North Africa (see, e.g., Howard and Hussein 2011). ⁴ We will examine the relative weight of global versus regional effects in our empirical analysis. However, we expect to see the strongest impact of diffusion in nonviolent campaigns among proximate countries. Despite the frequent announcements of the death of distance (see, e.g., Cairncross 2000), ⁵ geography remains a key influence on human interaction, and both the extent of information and relevance are likely to be closely associated with distance. We discuss each in turn below.

Although technological change means that distance need not induce delays and increase the cost of communication, people still have limited time and ability to process information. When information becomes overwhelming, individuals start to filter information by other measures of relevance or familiarity that tend to be closely associated with geographical distance. Analyses of e-mail and Facebook data, for example, indicate that new technologies make people communicate more with closer individuals (see Goldenberg and Levy 2009). Thus, individuals are more likely to be aware and familiar with events in neighboring countries than campaigns in remote states.

Likewise, we believe that neighboring and close countries are more likely to be considered relevant or similar, increasing the likelihood of learning and emulation from events. The impact of geography on interactions makes neighboring countries more likely to share similar characteristics, cultural frames, and often face similar political grievances and opportunities (see Beissinger 2007; Buhaug and Gleditsch 2008; Gleditsch 2002a; Lichbach 1998). As such, beyond familiarity, events in neighboring countries are more likely to convey information considered relevant about prospects for mobilization and potential examples for emulation. It is instructive that the perceived relevance of neighboring states can apply even in the absence of similar political institutions or current economic conditions, as seen in the Arab Spring (e.g., Lynch 2012). Moreover, efforts to emulate the Color Revolutions largely failed in distant places such as Togo and Zimbabwe.

The impact of global versus regional events can be evaluated empirically, and we formulate two distinct diffusion hypotheses. Global and regional diffusion are not mutually exclusive, but we expect any added impact of extra-regional campaigns to be limited.

Hypothesis 1a: The likelihood of a nonviolent campaign increases following a nonviolent campaign in a neighboring state.

Beyond location, nonviolent campaigns in other states are unlikely to affect all countries alike, as the impact should be conditional on plausible motivation for nonviolent direct action in the first place. Existing research demonstrates how political freedom is a key motive for nonviolent campaigns. Whereas democracy provides opportunities for making demands through other political channels, lack of political freedom by itself motivates challenges to regimes and their legitimacy and leaves direct action as the only feasible avenue for dissent as individuals are precluded from pursuing claims through routine political channels (see, e.g., Bond 1994; Dahl et al. 2014). As such, we do not expect to see nonviolent campaigns in a democratic state just because there is a campaign in a neighboring state, but neighboring campaigns may help provide a sufficient push for mobilization in a nondemocracy with plausible motivation.

Hypothesis 2: The likelihood of a nonviolent campaign increases following a nonviolent campaign in a neighboring state for nondemocracies but not for democracies.

Our final transnational proposition relates the plausible impact of diffusion to variation in resources available to dissidents in other states. Existing research has shown that transitions to democracy are more likely when a state has a higher proportion of democratic neighbors (see Gleditsch and Ward 2006). One interpretation is that forces seeking to promote change have better prospects in autocracies if they can organize in and access resources in neighboring countries. Nonviolent uprising are sometimes portrayed as spontaneous, but this is misleading as campaigns are unlikely to see much mobilization without planning and organization. Since planning and organizing face many difficulties in closed and repressive societies, the ability to tap on opportunities in other countries can be very helpful. The Otpor! (Resistance!) campaign against Milosevic in Serbia, for instance, convened meetings in neighboring democratic Hungary, and other movements have held similar workshops in more open neighboring countries (e.g., Cohen 2000; Popovic and Miller 2015). Transnational resources can of course help outside campaigns in other states, but we expect the impact of neighboring campaigns to be larger when dissidents can also draw on resources and organize in neighboring democracies.

Hypothesis 3: The likelihood of a nonviolent campaign following campaigns in a neighboring state increases when a state has more democratic neighbors.

Empirical Analysis

We now turn to examine the empirical implications of our theoretical arguments on regional diffusion in nonviolent campaigns. We extract campaigns relying primarily on nonviolent direct action from the NAVCO data (see Chenoweth and Lewis 2013). Campaigns here refer to “a series of observable, continuous, purposive mass tactics” using direct action and must involve maximal demands on the state (either to overthrow the government or to secede). Campaigns may be violent or predominantly nonviolent, and we only consider the second type here. The focus on direct action means that we do not consider nonviolent efforts that rely on conventional political methods, or nonviolent campaigns with non-maximalist aims such as, for example, environmental protest that do not seek to overthrow the government or secede. The NAVCO data focus on campaigns in terms of specific incompatibilities, and each campaign may encompass several organizations with varying degrees of unity and coordination. Our dependent variable is the onset of a new campaign in a country, and our unit of analysis is the country year. Observations with continuing campaign years after onset are dropped. The NAVCO data cover the time period from 1900 to 2006, but the effective range of our analysis is reduced to 1946 to 2006 due to the availability of other data, covering a total of seventy-six country years with campaign onsets.

We first briefly review trends over time in the nonviolent campaigns in the NAVCO data and the evidence for clustering. The gray dashed line in Figure 1 indicates the number of ongoing nonviolent campaigns for each year, while the black bars indicate the number of new onsets. Figure 1 suggests some increase in nonviolent direct action over time, with a notable peak in the 1980s and 1990s. Moreover, although most years do not see any campaigns, outbreaks of campaigns in one country tend to coincide with other onsets—indeed less than half of the nonzero years see just one onset. The observed data do not resemble a distribution with fully independent events, such as a Poisson process, as there are too many simultaneous outbreaks given the total number of events (see, e.g., Coleman 1964). The overdispersion is consistent with what we would expect to see if campaigns diffuse from one country to another. However, it does not by itself prove that diffusion mechanisms promote nonviolent campaigns or speak to whether these are global or regionally confined.

OPEN IN VIEWER

Figure 1.

Number of ongoing nonviolent campaigns in the Nonviolent and Violent Campaigns and Outcomes (NAVCO) data by year (gray dashed lined) and new campaign onsets (black vertical bars).

We create two distinct measures that reflect ongoing nonviolent campaigns in other states to evaluate Hypotheses 1a and b on regional diffusion among proximate states versus global diffusion. For the first, we consider a binary variable indicating an ongoing campaign in a neighboring state. We define states as neighbors if within 500 km of their outer boundaries, using the Cshapes historical maps (see Weidmann, Kuse, and Gleditsch 2010). We consider extra-regional diffusion by a global count of campaigns in other counties J – i, removing country i itself so as to avoid a deterministic nonzero relationship if i sees a campaign onset. ⁶

Beyond the temporal clustering shown in Figure 1, a simple inspection of the data indicates a much higher frequency of nonviolent campaign onsets when there are nonviolent campaigns in neighboring countries. Forty-one of the seventy-six campaign onsets—that is, more than half—take place in cases with an ongoing campaign in a neighboring state. Although campaign onsets are rare events, the share of onsets for observations with at least one campaign in a neighboring state is more than five times greater than the rate of onsets without campaigns in neighboring states. ⁷

These descriptive statistics are highly suggestive but may of course mask other features common to cases with ongoing campaign in neighboring states (see Galton’s 1889 famous comment on the problems of distinguishing between functional independent relationships and diffusion). As such, we also need to consider other features plausibly spatially correlated and related to the onset of nonviolent campaigns. We consider degree of nondemocracy as a source of potential motivation for dissent, measured by a country’s score on the twenty-one-point institutionalized democracy scale in the Polity data as well as a binary measure of nondemocracies based on the conventional threshold of values above six as democracies (see Jaggers and Gurr 1995). ⁸ Beyond controlling for domestic motivation, we also use this measure to evaluate Hypothesis 2 on the conditional effect on diffusion for non-democracies with plausible motivation versus the lack of diffusion effects for democracies.

With regard to other control variables reflecting motives, it is commonly argued that poverty and poor economic performance may increase dissatisfaction with the status quo and encourage dissent and economic fortunes often cluster spatially. We thus include the natural log of gross domestic product (GDP) per capita and growth rate, based on data from Gleditsch (2002b).

Domestic social resources should influence opportunities for mobilization and collection action and may plausibly be spatially clustered. There are a host of individual skills and social network characteristics that can facilitate organization and lower barriers to collective action. One prominent resource indicator that can be assessed relatively easily on a global basis is urban population (see Dahl et al. 2014). Nonviolent direct action appears to be much more common in cities, and efforts to undermine governments through noncooperation tend to be more effective when they have a presence in key cities and urban areas than in the periphery. Furthermore, urban populations tend to have more individual resources, and a higher density of networks among individuals makes it easier to organize. We include a measure of the logged urban population living in areas with population greater than 100,000. ⁹ For facilitating transnational structures beyond individual states and the conditional effects on diffusion, we consider the proportion of democracies among neighboring states within 500 km based on Cshapes.

Repressive capacity could reduce dissent through deterrence and could conceivably be increased in the presence of dissent in neighboring states (see Danneman and Ritter 2014). We thus control for logged military personnel as proxy, using data from the Correlates of War project. The age of a regime may also affect perceived stability and influence the likelihood of challenges. We control for the log of time in years a specific coalition has remained in power, measured by time since any irregular leader exit and entry in the Archigos data (Goemans, Gleditsch, and Chiozza 2009), counting from either the last transition or first observation in the data. We consider time dependence in campaign onsets by including the logged number of years since any previous campaign or independence.

Model 1 in Table 1 presents the results for a purely domestic baseline model, including only country-specific covariates. We comment first on the domestic terms in model 1 before turning to diffusion or the effects of campaigns elsewhere in subsequent models. In terms of political motivation, we find strong evidence that nonviolent campaign onset becomes less likely with greater levels of democracy, consistent with the notion that democracy increases opportunities for dissent through routine political channels and reduces the motivation for direct action. ¹⁰ The baseline odds increase by a factor of two for a fully autocratic country compared to a fully democratic country. Campaign onset is more likely with larger urban populations, consistent with a resource perspective on opportunities for mobilization and some variants of modernization theory, where opposition to autocratic rule is more likely in more urbanized and complex societies with dispersed social power (see Boix 2003; Vanhanen 1990). By contrast, we find less evidence for strong effects of poverty or growth on nonviolent campaign onset. The coefficient for military personnel is negative but far from statistically significant. This offers indirect support for the argument that repressive capacity alone does not suffice to deter dissent and is also consistent with repression being more difficult to apply against nonviolent movements as defections from the security apparatus become more likely. Finally, the two time dependence terms are not statistically significant. ¹¹ Unlike violent conflict, recent campaigns do not seem to make new campaigns more likely, potentially reflecting that successful campaigns generate concessions that remove their original motivation, or that failed campaigns do not leave legacies that facilitate immediate new mobilization.

OPEN IN VIEWER

Table 1.

Logit Estimates of Onset of Nonviolent Campaigns.

	Dependent variable
	Nonviolent campaign onset
	(1)	(2)	(3)	(4)	(5)
Nonviolent campaign in neighbor		1.433***(0.252)		1.128***(0.269)	1.184***(0.275)
Global campaign #—i			0.102***(0.017)	0.074***(0.018)	0.031(0.043)
1989					1.021(0.930)
Ln GDP p.c.	−0.190(0.128)	−0.183(0.127)	−0.188(0.130)	−0.168(0.130)	−0.170(0.130)
Economic growth	−2.112(1.587)	−1.939(1.632)	−1.254(1.613)	−1.380(1.646)	−1.578(1.672)
Ln urban population	0.363***(0.133)	0.343**(0.138)	0.326**(0.136)	0.319**(0.139)	0.334**(0.140)
Polity	−0.054***(0.020)	−0.057***(0.021)	−0.054***(0.021)	−0.057***(0.021)	−0.057***(0.021)
Ln military personnel	−0.044(0.133)	−0.055(0.137)	−0.001(0.137)	−0.019(0.139)	−0.034(0.139)
Ln time at regime	0.180(0.152)	0.095(0.150)	0.121(0.150)	0.072(0.150)	0.085(0.151)
Ln time since nonviolent campaign	−0.009(0.115)	−0.042(0.113)	−0.034(0.112)	−0.053(0.112)	−0.053(0.112)
Constant	−6.230***(1.302)	−6.179***(1.283)	−6.479***(1.328)	−6.509***(1.315)	−6.399***(1.318)
Observations	5,790	5,790	5,790	5,790	5,790
Log likelihood	−348.835	−333.221	−335.143	−326.561	−325.933
Akaike information criterion	713.669	684.442	688.287	673.121	673.866

Note: Ln GDP p.c. = log of gross domestic product per capita.

*p < .10.

**p < .05.

***p < .01.

In model 2 in Table 1, we add a term for the presence of a campaign in a neighboring state to the purely domestic model. The coefficient estimate is positive and significant, and the odds of a nonviolent campaign onset increase by more than four in the presence of a campaign in a neighboring state. The large improvement in the Akaike Information Criterion for model 2 compared to model 1 indicates that the model with neighboring campaigns fits notably better than the purely domestic model, even considering the loss of degrees of freedom. ¹²

The results for model 2 provide clear evidence of interdependence in nonviolent campaign onset but do not prove that neighboring states are the most influential. Model 3 includes a term for the global number of ongoing campaigns in all J – i other countries, and this also returns a positive and significant coefficient. In model 4, we consider both terms in the same model. The fact that we see significant positive coefficients for both could be taken as support for the claim that both regional and extra-regional campaigns can increase the likelihood of a campaign onset over and beyond purely domestic characteristics. However, comparing the actual magnitude of coefficients makes it clear that neighboring campaigns have a much larger impact than additional campaigns not in neighboring countries. ¹³ Even in the presence of as many as five extra-regional campaigns in any single year, the maximum plausible partial effect of global campaigns on the log odds of campaign onset (i.e., 0.074 × 5 = 0.37) remains considerably smaller than the effect of a nonviolent campaign in a neighboring country (i.e., 1.123).

The extreme number of campaigns in 1989 shown in Figure 1 suggests that this year may be exceptional. In model 5, we add a dummy variable for this year, and this reduces the global count term to half its original size and renders it insignificant. The coefficient for 1989 in model 5 is large, but not statistically significant, possibly due to high collinearity. Thus, much of the evidence for global diffusion here seems to rest on data for 1989. Although our data do not include events after 2006 such as the Arab Spring, involving plausibly connected outbreaks in countries that are not geographic neighbors, we see our results as providing strong evidence for Hypothesis 1a on regional diffusion in nonviolent campaigns and at best modest evidence for Hypothesis 1b that global events have an additional effect over and beyond neighboring countries.

Skeptics might wonder whether our results really reflect nonviolent campaigns diffusing or just a greater tendency for political instability to spill over to other countries. To examine explicitly whether the effects are limited to nonviolent campaigns, we reestimate in Table 2 a series of alternative models also considering violent campaigns (based on the data on civil wars from NAVCO). ¹⁴ Model 2 in Table 2 indicates that a term including all campaigns, both violent and nonviolent, also returns a positive but smaller coefficient, with only borderline significance. In models 3 and 4, the coefficient estimates for violent campaigns are actually negative (although not statistically significant), whether submitted alone or with nonviolent campaigns. Thus, there is no evidence that violent instability increases nonviolent campaigns, consistent with our argument that individuals emulate and learn about specific tactics in neighboring countries and respond differently to violent and nonviolent conflict, if the mechanisms through which strategies spread may be conceptually similar.

OPEN IN VIEWER

Table 2.

Logit Estimates of Onset of Nonviolent Campaigns.

	Dependent variable
	Nonviolent campaign onset
	(1)	(2)	(3)	(4)
Nonviolent campaign in neighbor	1.433***(0.252)			1.463***(0.253)
NV or V campaign in neighbor		0.536(0.330)
Violent campaign in neighbor			−0.330(0.269)	−0.437(0.271)
Ln GDP p.c.	−0.183(0.127)	−0.165(0.126)	−0.216(0.131)	−0.221*(0.131)
Economic growth	−1.939(1.632)	−1.900(1.589)	−2.303(1.592)	−2.197(1.630)
Ln urban population	0.343**(0.138)	0.336**(0.134)	0.391***(0.135)	0.383***(0.141)
Polity	−0.057***(0.021)	−0.051**(0.020)	−0.055***(0.021)	−0.058***(0.021)
Ln military personnel	−0.055(0.137)	−0.046(0.135)	−0.047(0.132)	−0.064(0.137)
Ln time at regime	0.095(0.150)	0.170(0.151)	0.185(0.152)	0.102(0.150)
Ln time since nonviolent campaign	−0.042(0.113)	−0.025(0.114)	0.004(0.117)	−0.026(0.114)
Constant	−6.179***(1.283)	−6.507***(1.301)	−6.097***(1.323)	−5.991***(1.312)
Observations	5,790	5,790	5,790	5,790
Log likelihood	−333.221	−347.391	−348.095	−331.950
Akaike information criterion	684.442	712.782	714.191	683.901

Note: NV = nonviolent; V = violent; Ln GDP p.c. = log of gross domestic product per capita.

*p < .10.

**p < .05.

***p < .01.

We now turn to our Hypothesis 2, stipulating that the impact of a neighboring campaign depends on plausible motivation for nonviolent direct action in the first place and thus apply only for nondemocracies. In model 2 in Table 3, we include an interaction between neighboring campaigns and whether a country is a nondemocracy. ¹⁵ We find strong evidence that the diffusion effect appears to be larger for nondemocratic states and indeed largely restricted to these states, as the coefficient for neighboring campaigns becomes small and nonsignificant. This is consistent with the claim that the prospects for contagion hinge on plausible motivation for nonviolent direct action in the first place, but events in neighboring can have an important mobilizing effect when motivation exists. In model 3 in Table 3, we further consider to what extent diffusion effects are conditional on potential opportunities that can be mobilized in other states, proxied by the share of democratic neighboring states. The coefficient estimate for this term is positive and statistically significant, providing strong support that the observed effects of a neighboring campaign are higher when a larger share of the neighboring countries are democracies. Note that this is not simply due to an independent positive effect of democratic neighbors on resources, which when included by itself has a positive but not statistically significant coefficient. ¹⁶ The other results remain similar, save for that the negative coefficient estimate for ln GDP per capita now becomes borderline significant.

OPEN IN VIEWER

Table 3.

Logit Estimates of Nonviolent Campaign.

	Dependent variable
	Nonviolent campaign onset
	(1)	(2)	(3)
Campaign in neighboring	1.433***(0.252)	0.004(0.642)	−0.476(0.683)
Campaign in neighboring × nondemocracy		1.753***(0.677)	1.969***(0.686)
Proportions of democratic neighboring			0.201(0.586)
Proportions of democratic neighboring × neighboring campaign			0.784**(0.335)
Ln GDP per capita	−0.183(0.127)	−0.174(0.129)	−0.236*(0.141)
Economic growth	−1.939(1.632)	−2.005(1.655)	−1.976(1.694)
Ln urban population	0.343**(0.138)	0.343**(0.137)	0.342**(0.139)
Polity	−0.057***(0.021)	0.020(0.024)	−0.021(0.026)
Ln military personnel	−0.055(0.137)	−0.052(0.137)	−0.063(0.140)
Ln time at regime	0.095(0.150)	0.051(0.150)	0.033(0.151)
Ln time since nonviolent campaign	−0.042(0.113)	−0.070(0.113)	−0.068(0.116)
Constant	−6.179***(1.283)	−5.946***(1.294)	−5.427***(1.338)
Observations	5,790	5,790	5,658
Log likelihood	−333.221	−328.945	−320.461
Akaike information criterion	684.442	677.890	664.922

Note: Ln GDP p.c. = log of gross domestic product per capita.

*p < .10.

**p < .05.

***p < .01.

In Figure 2, we plot the implied predicted probabilities of a nonviolent campaign onset based on model 3 in Table 3 for a series of scenarios with differences in regime type, neighboring campaigns, and proportions of democratic neighboring states, keeping other covariates at their median. The black solid line for the predicted likelihood of a nonviolent campaign onset in an autocracy with neighboring campaigns is substantially higher than dashed black line indicating the likelihood for autocracies without neighboring campaigns. For democracies, the likelihood of a nonviolent campaign onset remains low, regardless of whether a county sees a neighboring campaign (solid gray line) or not (dashed gray line). The predicted likelihood of a campaign is consistently lower for a democratic state than in an autocratic state, but the differences are relatively modest across profiles in the absence of a campaign in a neighboring state. The lower bound for the 95 percent confidence interval for the effect of an autocracy in the presence of a neighboring campaign (thin gray line) remains consistently above the upper bound for the confidence interval for an autocracy without a neighboring campaign (thin dashed gray lines—this also applies to the confidence intervals for the democracies profiles, not shown here to reduce clutter). Beyond the effects of neighboring campaigns, we also see substantial variation in the likelihood of nonviolent campaigns across differences in the neighboring context or democratic neighbors. For an autocratic country with a neighboring nonviolent campaign (solid black line), the predicted likelihood of a campaign onset is notably higher when a larger proportion of its neighboring states are democratic. In sum, the results support our claim that neighboring campaigns are particularly likely to diffuse, but their effects hinge on plausible domestic motivation and increase when there are opportunities to organize for mobilization in neighboring democratic states.

OPEN IN VIEWER

Figure 2.

Predicted effects for different scenarios over proportion of neighboring democracies, other variables at the median.

Conclusions and Discussion

There has been renewed interest in direct action against dictatorships and when nonviolent and violent dissident tactics are more likely. A great deal of research on civil war has highlighted how violent mobilization can diffuse and how transnational linkages can increase the likelihood of conflict in other states. We have argued that nonviolent direct action can be contagious as well and that campaigns increase the likelihood of new campaigns among neighboring states, but that the relevant diffusion mechanisms differ for nonviolent and violent mobilization. Whereas contagion of violent direct action generally entails border transgressing actions and direct support to potential insurgents, we argue that mechanisms such as learning about strategies and revealing information that facilitate overcoming collective action problems are more relevant for nonviolent direct action. We find strong support for our claims regarding the diffusion of nonviolent campaigns. Our results indicate that events in neighboring countries are highly relevant but provide little evidence for extra-regional diffusion mechanisms. Our findings also indicate that the diffusion of nonviolent direct action does not reflect “unrest” in general and that actors seem to respond differently to nonviolent and violent events in other states. Our results furthermore support our claim that the effects of international factors on nonviolent direct action are very much contingent on domestic motivations and their implications for resources or opportunities for mobilization. This underscores the limitations of the standard closed polity approach to the study of contentious politics, and we should instead focus on how the transnational dimension of dissent can complement and interact with country-specific motives and wider influences on opportunities.

Of course, much remains to be known about nonviolent direct action both at the domestic level and its transnational dimensions. Future studies of nonviolent direct action can benefit from greater attention to the specific actors involved in dissent and their activities over time. For example, one may argue that whether campaigns are considered successful is likely to be important for potential emulators and that campaigns with a bigger “impact” should have a greater effect on other actors. Studying this raises important issues regarding the conceptualization of “impact” and “success” and assessing, for example, whether campaigns generate momentum in terms of increasing participation or manage to “shake” governments would require more detailed event data and information on responses. ¹⁷ Campaigns that are eventually repressed or ultimately unsuccessful such as the Egyptian revolution can have an important effect on actors in other countries if they seemed promising at the outset. Some stress the “irrationality” of campaigns promoted by diffusion as these uprisings rarely manage to unseat governments (e.g., Weyland 2009). However, this is a narrow conceptualization of success, which disregards how governments often offer various concessions and reforms to try to fend off dissent and the long-term aim of dissidents. The initial protest in Serbia failed but helped foster a series of renewed mobilization and recruitment efforts as well as gradual reforms that eventually paved way for a full transition (see Popovic and Miller 2015). Finally, broad coalitions may succeed in initial common aims such as removing a dictator but fall apart one another once the initial aims are reached, as different factions often disagree on other aims and may turn against. The 1979 Iranian revolution, for example, initially included many factions that wanted democratic reform, later sidelined by the Islamists and repressed after the fall of the Shah. Research on diffusion and interaction in campaigns can be advanced by the development of more comprehensive data, including ongoing efforts to extend the information and scope of the NAVCO data as well efforts to develop protocols for global event data that can help identify nonviolent campaigns, relevant specific actors and events, and interactions with governments see (http://openeventdata.org/).

Research on diffusion in protest and nonviolent direct action can also be developed by examining in more detail potential positive and negative cases of learning and emulation. Direct information and interview with organizers and individual participants can help evaluate how their decisions and tactics were affected by experiences from other countries (see, e.g., Opp, Voss, and Gern 1995, using interview data on individual participation in protests in East Germany). Research can also benefit from studying a wider range of dissent and possible windows for learning and mobilization opened by international events. Although we focus here on nonviolent direct action involving maximal demands on the state, we believe that emulation, learning, and strategic adaptation can play similar roles for a wider range of dissent in different settings, including activities prior to large-scale mobilization and articulated claims that may be non-maximalist for strategic reasons. The Inter-Enterprise Strike Committee preceding the solidarity movement in Poland, for example, was initially careful to emphasize that they did not challenge socialist rules, but the demands obviously grew increasingly more ambitious as the movement gained momentum. The explicit references to the Arab Spring in to use of so-called occupy tactics in anti-austerity protests also suggest diffusion across rather different types of claims (http://www.huffingtonpost.com/2011/10/15/occupy-wall-street-protests-europe-asia_n_1012336.html). The activities of transnational actors seeking to promote the use of nonviolent direct action such as the Belgrade based Center for Applied Nonviolent Action and Strategies also deserve closer study (http://www.canvasopedia.org/; Popovic and Miller 2015).

A better understanding of domestic conflict processes can also inform policy and the appropriate responses to conflict by external actors. Our findings indicate that a favorable international context can help actors overcome barriers to collective action and mobilize nonviolently over grievances. Previous research further indicates that nonviolent direct action promotes subsequent democratization while violent conflict increases the chances of political instability leading to new autocracies and future conflict (see Rivera and Gleditsch 2013). This raises important questions about the wisdom of many Western foreign policy decisions and whether these can be expected to help achieve the intended outcomes. For example, while North Atlantic Treaty Organization (NATO) supported the violent rebellion against Gaddafi in Libya through air strikes, Western countries provided little direct support to nonviolent dissent in other countries in the Arab world and did little to resist the constraints on political reform and subsequent return to authoritarian rule in Egypt (see Goldstone 2012). The turmoil in Libya after the violent overthrow of Gaddafi and the relative success of democratization in Tunisia after the nonviolent revolution against Ben Ali make Chenoweth and Stephan’s (2014) idea of focusing on the “responsibility to assist” look more attractive and highlights the value of efforts to support and strengthen a civil society capable of nonviolent dissent over supporting violent insurgencies or a repressive peace.