The politics of nonviolent mobilization: Campaigns,competition,and social movement resources

Abstract

Nonviolent resistance is considered one of the most effective methods of bringing about political change. Yet empirical research on how nonviolent campaigns emerge is limited. This article considers two sets of considerations which influence the strategic decision of a campaign to use nonviolent or violent methods: interactions with the state and campaign resources. Campaigns are concerned about the risk of repression from the state and are more likely to choose nonviolence when they believe the state will accommodate their demands. Open political competition signals likely accommodation. Campaigns’ nonviolent tactics are more effective when they possess social movement resources, though high levels of movement resources threaten the state. The argument implies political competition and social movement resources are associated with a higher likelihood of a campaign using nonviolence, though the relationship with social movement resources diminishes at high levels. Implications of the argument are tested on a sample of contentious campaigns from 1945 to 2006. Political competition and social movement resources are related as expected to the use of nonviolence. However, an extension shows these factors do not account for the advantage of nonviolent methods, particularly in triggering backlash protests, once a campaign is under way. An original ordinal measurement strategy for campaign methods suggests the intensity of violence and nonviolence are likewise unrelated to competition and resources.

Keywords

civil resistance mobilization nonviolence political opportunity repression social movements

Introduction

Contentious campaigns’ use of nonviolent resistance against states is an effective method of bringing about political change (Chenoweth & Stephan, 2011). Nonviolent resistance campaigns better mobilize citizens, especially through their ability to create ‘moral cascades’ and backlash protests which draw people into participation through attachment to the campaign’s cause (Pearlman, 2018; Aytaç, Schiumerini & Stokes, 2018). Yet scholars’ understanding of what causes political movements to use nonviolent resistance – over other methods such as violence – remains incomplete. Explanations have focused on demographic (Dahlum, 2019; Zeira, 2019) or ideologically extreme (Ryckman, 2020) characteristics of the participants, the activities of other movements in the same country (Cunningham, Dahl & Frugé, 2017), structural grievances (Chenoweth & Ulfelder, 2017), and different combinations of these explanations (Cunningham, 2013b).

This article considers a critical actor which shapes the behavior of contentious campaigns: the state. In so doing, it extends the role of political opportunity in shaping contentious campaigns (Tilly, 1978; McAdam, Tarrow & Tilly, 2001) to their choice of resistance method. The argument treats the selection of violence or nonviolence as a strategic interaction between states and campaigns. It begins with the assumption that campaigns are rational actors which maximize their chance of achieving their demands. Campaigns choose violence or nonviolence conditional on the anticipated outcome of employing either resistance method: how the state would respond to their choice and how effectively the campaign could mobilize resources for the chosen resistance method. Campaigns evaluate these considerations by (1) the likelihood the state would accommodate nonviolent resistance, and (2) whether the campaign has social movement resources for such a resistance method.

There are two hypotheses derived from the argument. First, campaigns are more likely to choose nonviolence when a state signals it will accommodate dissent through permitting open political competition. Second, campaigns are more likely to choose nonviolence when they possess more resources conducive to social movement activity. However, the relationship between resources and nonviolence diminishes at higher levels of resources: campaigns’ increasing mobilization ability poses a threat to the state and reduces the state’s willingness to accommodate. I test the argument on a sample of contentious campaigns and find support for the hypotheses. Extending the main results, I show nonviolent resistance remains effective at generating backlash protests – a key argument in literature on the nonviolent advantage (Chenoweth & Stephan, 2011) – even when accounting for the factors which the argument identifies as leading to the choice of nonviolence.

To further extend the analysis, I take a random sample of 64 contentious campaigns from the original data and code the ordinal level of violence and nonviolence used during each campaign-year according to the coding criteria in the Minorities at Risk data. I then test the relationship between political competition and social movement resources and ordinal levels of violence and protest. This coding reveals more variation in the strategic choices campaigns make, compared to the conventional use of dichotomous measures of violence or nonviolence. I find that competition and resources do not appear to explain the intensity of resistance, pointing to the conclusion that these features of the strategic environment are at best weakly related to the ongoing dynamics of contentious campaigns.

This article contributes to related literature in four ways. First, it adds to literature on nonviolent resistance by pointing out the systematic differences in campaign attributes and attributes of the state which distinguish the environments in which violent and nonviolent resistance occur. I treat the resistance methods campaigns use as a dependent variable, and show the choice of resistance method is part of interaction with the state. Rationalist arguments which consider resistance methods as a dependent variable are not uncommon (Lichbach, 1987; Moore, 1998, 2000; Young & Dugan, 2011; Cunningham, 2013b; Aksoy, 2014; Heger, 2014), but the literature on contentious campaigns is more thin. Second, it shows nonviolent mobilization originates in an interaction between campaigns and the state. In particular, political competition reduces the use of violence consistent with the ‘domestic democratic peace’ (Davenport, 2007a,b) and theories of political opportunity (Carey, 2009). Third, it affirms literature which shows social movement resources – such as organizations, networks, and skills of participants – explain the choice of nonviolent resistance (Thurber, 2019; Dahlum, 2019; Clarke, 2014; McAdam, Tarrow & Tilly, 2001). Finally, it makes an empirical contribution by extending data on contentious campaigns to include more than a dichotomous measure of resistance methods (Cunningham, Dahl & Frugé, 2017).

The origins of nonviolence

There is extensive debate over the effectiveness of the resistance methods contentious campaigns use to make demands.¹ While these methods vary along many dimensions (Pressman, 2017; Cunningham, Dahl & Frugé, 2017), they can be collapsed to broad categories of violent and nonviolent tactics.² Using violent tactics such as armed insurgency can empower weak actors and signal their capabilities (Kydd & Walter, 2006). Campaigns may use violent resistance as a form of coercion against a rival campaign or the state (Pape, 2003). On the other hand, nonviolent tactics such as protest reduce barriers to participation in campaign activities, and give moral authority to participants (Stephan & Chenoweth, 2008). Nonviolent tactics generate sympathy among the public, reducing support for state repression of protests (Edwards & Arnon, forthcoming) and motivating increased participation in resistance when nonviolent campaigns are targeted (Pearlman, 2018; Aytaç, Schiumerini & Stokes, 2018). As a result, nonviolent campaigns may grow and eventually overthrow repressive regimes (Chenoweth & Stephan, 2011).

Given the many advantages of nonviolent campaigns for mobilizing participants and achieving their objectives, it is important to understand the conditions under which they emerge. Chenoweth & Ulfelder (2017) group explanations for emergence into grievances, modernization, resource mobilization, and political opportunity. In their analysis, the latter two categories have the strongest relationship with the emergence of nonviolence. Resource mobilization explanations comport with recent findings that the education level of campaign participants affects their ability to engage in nonviolent resistance (Zeira, 2019; Dahlum, 2019). For political opportunity, Cunningham (2013b) finds that political variables such as non-democracy and exclusion from power affect the use nonviolence as opposed to violent civil conflict. A lack of progress toward campaign objectives can also lead the abandonment of nonviolence for violence (Ryckman, 2020).

A broad literature speaks to the importance of specific human and material resources for sustaining nonviolent resistance (McCarthy & Zald, 1977). Nonviolence requires a certain portfolio of human and material resources to be effectively mobilized. Tilly (2008) explains how contemporary nonviolent campaigns needed special purpose associations, infrastructure to organize mass marches, the use of media, and communication technology. These necessary conditions for mass mobilization first emerged among urban networks of British activists in the 18th century, and continued to develop into the 20th century. The implication from the study of campaigns is that they must have such resources for nonviolence. For example, in the 1953 East German uprising, networks and communication among construction workers were important in driving nonviolent mobilization (Thomson, 2018; Crabtree, Kern & Pfaff, 2018). In Nepal, the country’s Congress Party built connections across social groups and with the state, turning to nonviolent resistance, whereas the Communist Party lacked such connections and turned to violent resistance (Thurber, 2019). Similar social connections promoted nonviolence among Palestinians in the West Bank (Gade, 2020).

This article also builds on the political opportunity literature by considering how campaigns assess the state’s likely response to their expression of demands. That is, the existence of opportunities depends on how campaigns express demands: the state may be receptive to some methods and not others. Rationalist theories suggest campaigns choose resistance methods conditional on the likely state response to each method (Lichbach, 1987). In the context of elections (Dunning, 2011; Heger, 2014), and non-state actors such as terrorist groups and rebels (Aksoy, 2014; Young & Dugan, 2011), a bargaining framework explains how the state responds to non-state actor demands. In this framework, the choice of resistance method may result from uncertainty: incomplete information about the capabilities of the state or non-state actor affects whether there is a violent or nonviolent outcome. Uncertainty undermines bargains, leading to violence if the state lacks information about non-state actors’ capabilities. Accommodation is more likely if these capabilities are known (Cunningham, 2013a). The willingness of the state to repress the non-state actor is also a source of uncertainty which shapes non-state actor choices (Cunningham & Beaulieu, 2010).

Regime type and institutions are one way for campaigns to assess likely state response. Authoritarian regimes tend to limit political competition and exclude, repress, or co-opt potential challengers to the state (Levitsky & Way, 2010; Gandhi, 2008). Democratic systems are designed for power-sharing and permit more political participation by non-state actors (Carey, 2009; Davenport, 2007a), offering institutional channels, such as constitutional guarantees of civil liberties, free and fair elections, and representative government, for campaigns to express their demands (Young & Dugan, 2011; Cunningham, 2013b; Heger, 2014). These institutional channels in democratic systems suggest that states will accommodate campaigns whose methods of resistance are consistent with the methods permitted within those channels. The following section explains how state response and campaign resources shape campaigns’ choice of resistance method.

Explaining the choice of nonviolent resistance

In this section, I outline the structure of contentious campaigns’ interaction with the state and use the structure to identify two influences on campaigns’ choice of nonviolent resistance. I focus on campaigns which make a maximalist demand against the state, calling for regime change, significant institutional reform, territorial secession, autonomy, or the expulsion of a foreign occupier (Chenoweth & Lewis, 2013).³ Campaigns may express these demands through violent or nonviolent means, and the state may respond with either accommodation or repression of the campaign (Klein & Regan, 2018; Stephan & Chenoweth, 2008; Lichbach, 1987; Moore, 2000).⁴

If the campaign chooses nonviolent resistance, it mobilizes mass marches, demonstrations, and other comparable actions such as economic, social, or political non-cooperation (Cunningham, Dahl & Frugé, 2017). An accommodative state response means mass marches and demonstrations pressure the state to come to the negotiating table with the campaign and address its demands. A repressive response, on the other hand, could cripple future nonviolent resistance if the campaign’s leaders are eliminated and organizing capacity diminished (Sullivan, 2016). Alternatively, repression could trigger backlash and galvanize more nonviolent resistance (Pierskalla, 2010) as domestic and foreign audiences respond with outrage to the violation of campaign members’ human rights.

If the campaign chooses violent resistance, it uses tactics such as sabotaging infrastructure, targeting state military capabilities, attacking members of the security forces or even civilians. An accommodative state response could be motivated by potential loss of public support over a protracted conflict against violent resistance (Kydd & Walter, 2002, 2006). Thus violent attacks could coerce the state to address campaign demands (Pape, 2003). States may also accommodate violent campaigns even without being forced to the negotiating table: sustaining violent campaigns’ activity allows the state to justify further repression. However, accommodations under violence are limited: states tend to funnel resources to security forces when violent campaigns are present, leading to greater ‘militarization’ of governance (Fergusson et al., 2016). Public opinion is also less supportive of violent resistance than nonviolent resistance, pushing for less accommodating responses to violent campaign demands (Edwards & Arnon, forthcoming). A repressive state response to violence is similar in its intent to a repressive state response to nonviolence: it involves the use of security forces against a campaign in order to raise the costs of the campaign’s resistance (Davenport, 2007a; Carey, 2009).

Table I summarizes the structure of the interaction. Each cell characterizes a pair of campaign decision and state response. The nonviolent-repressive outcome I label ‘political repression’, in which the campaign could face destruction or potentially trigger backlash. The nonviolent-accommodative outcome I label ‘political participation’, in which the campaign negotiates with

Table I.

Campaign–state interaction

		State
		Repress	Accommodate
Campaign	Nonviolent	Political repression	Political participation
Campaign	Violent	Conflict	One-sided violence

the state over its demands. The violent-repressive outcome I label ‘conflict’, in which the campaign defends itself against destruction but exchanges attacks with the state in a war of attrition. The violent-accommodative outcome is labeled ‘one-sided violence’, in which violence may bring the state to the negotiating table.

Campaigns prefer to choose a resistance method – nonviolence or violence – which maximizes the probability of achieving their demands (Lichbach, 1987; Cunningham, Dahl & Frugé, 2017). Based on the structure outlined in Table I, the campaign’s two main considerations are whether the state will repress or accommodate, and how the campaign – based on its available resources – may be better suited for nonviolent or violent resistance. I describe two strategic problems, uncertainty and mobilization, which correspond to these considerations and describe how they shape campaigns’ choice of nonviolent resistance.

The problem of uncertainty

The campaign’s first strategic problem is anticipating the state’s response. A campaign cannot be perfectly informed about state policy in advance and thus the state’s response – accommodative or repressive – to a particular resistance method cannot be precisely predicted. One source of uncertainty is inconsistency in past response, limiting the campaign’s ability to accurately forecast future responses (Cunningham & Beaulieu, 2010). Another source of uncertainty is whether the state, even if consistent in past responses, may unexpectedly shift its future response as an effort to ‘outwit’ the campaign and gain an advantage either through repression or at the negotiating table (Davenport, 2015). Finally, campaigns tend to lack the centralization and specialization necessary to gather intelligence on the state, facing ‘formidable difficulties in forecasting’ (Lichbach, 1995: 289).

Yet correctly anticipating state response is critical for the campaign to achieve its demands. If the campaign anticipates an accommodative response, it knows nonviolent resistance is more likely to achieve concessions than if the state response was repressive (Klein & Regan, 2018). Accommodation of nonviolent campaign demands facilitates such campaigns by increasing the likelihood that demands will be met, thereby encouraging participation in campaign-organized collective action (Tilly, 1978). Furthermore, violent resistance is less likely to be effective than nonviolence under an accommodative state response. Violence introduces opportunity costs, as the campaign allocates superfluous effort protecting itself with arms and threatening the state when the state would have neither used force against the campaign in the absence of violence nor required threats to give concessions.

If the campaign anticipates a repressive response, it knows nonviolent resistance is likely to lead to a crackdown which could cripple future nonviolent actions (Sullivan, 2016). Violent resistance blunts the effects of repression. To illustrate, consider if a campaign chooses nonviolent resistance and security forces massacre demonstrators, meaning the campaign may be unable to mobilize future demonstrations and sustain pressure for its demands. This outcome is what befell the Mexican student movement during the 1968 Tlateloco Massacre, in which violent repression decimated nonviolent resistance, and the the anti-apartheid resistance campaign during the 1960 Sharpeville Massacre in South Africa, in which police opened fire on peaceful protesters. Violent resistance – the preferred method in Mexico and South Africa after these massacres – helps the campaign blunt the damage repression could cause: weapons caches, veteran militants, and secrecy give violent actions ‘operational security’ under a repressive response (Finkel, 2015). Nonviolent actions, being public and with identifiable participants, lack such security.

Therefore, campaigns are more likely to choose nonviolent resistance when they are more certain the state’s response will be accommodative rather than repressive. Campaigns could arrive at this certainty in different ways. One possibility is whether repression of the campaign has decreased, suggesting accommodation is possible. Yet the above discussion indicates that changes in state response are a poor predictor of future response (Cunningham & Beaulieu, 2010; Moore, 2000). The most credible signal for campaigns that the state will accommodate is a competitive political system in which opposition is allowed to freely express demands (Carey, 2009). Here the state does not ban the activity of political groups, does not imprison opposition leaders, and shares or transfers power without conflict. With a competitive political system, the campaign has a clear indication the state has institutional barriers to repression (Davenport, 2007a). Rules of political competition also mark the boundary where accommodation ends and repression begins. This discussion produces Hypothesis 1:

Hypothesis 1: A campaign is more likely to choose nonviolent resistance if the state permits open political competition. A decrease in repression has no relationship with the choice of nonviolent resistance.

This hypothesis does not imply nonviolent resistance is only more likely in democracies. Non-democracies’ level of political competition also varies. Since the latter decades of the 20th century, a number of ‘competitive authoritarian’ regimes have tolerated political opposition and even on occasion relinquished power when the opposition wins elections (Levitsky & Way, 2010). Hypothesis 1 implies that the greater the extent of permitted opposition in a state, the more likely a campaign in that state is to choose nonviolent resistance. As an example, Trejo (2012: 11) argues that in Mexico under the PRI regime, ‘expand[ing] political and civic rights and liberties to allow dissident campaigns to continue expressing their grievances through the ballots or in the streets, protest […] will likely remain peaceful’.

The problem of mobilization

The campaign’s second strategic problem is mobilization. Making demands requires the campaign to identify which method of resistance – violence or nonviolence – best leverages its existing resources for mobilization against the state.⁵ Secure weapons caches and veteran militants would have greater marginal return with use in violent resistance than with nonviolent resistance (Finkel, 2015). Networks of activists spanning local populations and dissident organizations have greater marginal return for nonviolent resistance than violent resistance (Gade, 2020), as these underlie the growth of the contemporary social movement. Linkages with international activists and NGOs favor nonviolence by bringing international organizations, and normative pressure to support the campaign (Murdie & Davis, 2012). The campaign’s ability to generate written or broadcast content promoting its demands and to distribute messages quickly to mass audiences also would have greater marginal return with nonviolence. I label these social movement resources.

If a campaign possesses few social movement resources, it is unable to convert those resources into large-scale nonviolent action. These actions require many participants who use a focused repertoire of tactics (Tilly, 2008). A lack of social movement resources such as linkages between activists and communication technology exacerbates the collective action problem (Lichbach, 1995) – restricting the number of participants – and the coordination problem (Clarke, 2014) – restricting the ability of those who do participate to use the same repertoire. As a result, it is more likely the campaign will face internal divisions and participants who use violence against the wishes of campaign leadership (Ryckman, 2020). The smaller remaining number of participants who use nonviolent action undertake ‘high personal risk’ and ‘put members in harm’s way’ with little means of defense (Cunningham, Dahl & Frugé, 2017: 595). Violence thus appears more likely to achieve a campaign’s demands when social movement resources are low.

If a campaign possesses higher levels of social movement resources, it is able to generate a focused repertoire of nonviolent tactics which display successful collective action – a large number of participants – and successful coordination – unity of tactics (Tilly, 2008). With effective collective action and coordination, a campaign can deploy repeated nonviolent actions to sustain pressure against the state and achieve its demands. It may also help trigger backlash protests if the campaign is repressed, as the campaign remains resilient under repression (Sullivan, 2016). Large and tactically consistent nonviolent actions increase ‘disruption costs’ for the state – increasing the burden if the state fails to address campaign demands and thereby end the political, social, and economic penalties nonviolent action incurs (Klein & Regan, 2018). Nonviolence thus appears more likely to achieve a campaign’s demands when social movement resources are higher.

Yet social movement resources at the highest levels do not exclusively promote nonviolent action. If a campaign becomes capable of generating consistent, large-scale and well-coordinated nonviolent action against the state, it may be able to push for concessions from the state which exceed what the state is willing to offer at the negotiating table. As a result, the campaign poses a threat to the state, limiting the extent to which the state’s political institutions will accommodate campaigns’ demands (Dower et al., 2018). Therefore, the threat high levels of social movement resources pose to the state limits the benefits these resources provide to the use of nonviolence. This leads to Hypothesis 2:⁶

Hypothesis 2: Increasing social movement resources have a diminishing positive relationship with the likelihood a campaign chooses nonviolence. The likelihood of nonviolence increases, but the rate of increase decreases as social movement resources reach high levels.

While I have presented these two strategic problems separately, they are likely connected. Open political competition could increase the availability of resources available to campaigns for nonviolent mobilization such as organizations, networks, communication, and linkages to international human rights organizations. Conversely, the presence of these resources could, over time, pressure states to increase the level of political competition and permit campaigns’ participation in the political process. The interdependence of competition and resources presents a challenge for empirically estimating their relationship with nonviolence. I address this possibility in the analysis which follows.

Research design

Data

To test the hypotheses, I use the NAVCO 2.0 data (Chenoweth & Lewis, 2013). NAVCO 2.0 contains panel data on 250 contentious campaigns running from 1945 to 2006, including information on campaign attributes measured each year of the campaign. A longitudinal snapshot of a campaign is important, because it allows for the estimation of campaign behavior given past and current conditions: comparing year-over-year changes is more precise than taking a single peak year from a campaign to interpret events and campaign decisions (Stephan & Chenoweth, 2008). For example, if there is a mass campaign with more than one million participants, severe state repression, and a switch in resistance methods over the course of a campaign, the sequence of these events is unknown by examining only a peak year. I leverage the sequence of campaign events in the analysis.

To be included in the NAVCO 2.0 data, a campaign must reach at least 1,000 observed participants in two coordinated events in less than one year. If participation dips below 1,000, then the campaign is excluded from the data. Another inclusion condition for campaigns is the presence of maximalist goals which include regime change, secession, or removal of a foreign occupier. In the argument, campaigns sustain collective action to achieve a maximalist goal which could include a reformed government or territory for an ethnic minority. Campaigns which seek specific policies or the election of specific politicians are not included in the data.

Variables

The dependent variable of interest is whether a campaign employs violence or nonviolence as its primary resistance method in a given year. To measure resistance method, I use the prim-method variable in the NAVCO 2.0 data, which is coded 0 for a primarily violent campaign and 1 for a primarily nonviolent campaign. The prim-method variable can change from year to year in a campaign, unlike in previous iterations of these data. Because the dependent variable is dichotomous, empirical estimations of relationships with this variable are interpreted as a likelihood that a campaign chooses nonviolence, with 0 being certain violence, and 1 being certain nonviolence. A potential concern with this and other variables from NAVCO is measurement error: information on historical campaigns is often limited. To address this concern, I present results in the Online appendix, dropping observations rated as low reliability in the NAVCO data.

There are two independent variables of interest for Hypothesis 1, each measured in year t, relative to the dependent variable measured in year $t + 1$ . For Hypothesis 1 I include the campaign’s experience of repression and the state’s level of political competition, To measure a campaign’s experiences of repression, I use the repression variable in NAVCO 2.0, which ranges from 0 to 3. A score of 0 means an absence of repression, while a score of 3 means severe repression in the given campaign year. I operationalize a state’s openness to political competition in period t using a freedom of association variable from the Varieties of Democracy data (Coppedge et al., 2017). This freedom of association index assigns a value from 0 to 1 in response to the question: ‘to what extent are parties, including opposition parties, allowed to form and to participate in elections, and to what extent are civil society organizations able to form and to operate freely?’

For Hypothesis 2 I include an independent variable for social movement resources. Increased social movement resources means a campaign sustains or enhances its collective action through organizing coordinating mass actions, communicating the campaign’s message using media infrastructure, and building a broad-based coalition of support across networks of activists. I therefore operationalize social movement resources as a bundle of social movement repertoires at time t. I measure this bundle with the variable SM Assets based on the NAVCO 2.0 variables media-outreach, pi-tradmedia, camp-orgs, and ingo-support. These variables measure if the campaign invests in public relations, creates an alternative media system, has new organizations join the campaign, and receives support from international nongovernmental organizations, respectively. A coding of 0 means the campaign had none of these resources in a campaign-year, while a coding of 4 means the campaign had all of these resources in the campaign-year. Other campaign features in NAVCO 2.0 are poor indicators of social movement resources and are not included.⁷ Because the hypothesis expects the relationship to be positive but diminishing, I square the variable and estimate coefficients for both the linear and quadratic terms.

Several covariates are included which could potentially correlate with both the independent variables and the outcome of nonviolence. Stephan & Chenoweth (2008) include a Polity score of the state lagged one year from the end of the campaign. However, given that the measure of political competition used before is related to an overall democracy score, the resulting multicollinearity in the model would bias estimates of the association between Polity score and likelihood of choosing nonviolence. Another implication from Stephan & Chenoweth (2008) is the expected positive relationship between campaign size and nonviolence. Therefore, I also include campaign size at t as a control variable. The NAVCO variable camp-size is an ordinal scale ranging from 0 (less than 1,000 participants) to 5 (greater than 1,000,000 participants). The ideology of a campaign could also affect its choice of resistance method. In particular, more extreme campaigns may be more likely to be violent, attract repression, and induce the state to close off avenues of participation. To account for this possibility, the variable extreme goal is included. This is a dummy variable which takes the value of 1 if the campaign’s stated goal coded in NAVCO 2.0 is regime change, and 0 if a

Figure 1.

Imbalance of violent and nonviolent campaign characteristics

campaign’s demands are still maximalist but do not involve regime change.

Finally, country-year-level control variables are included: the log of GDP per capita in year t and log of total population in year t are each included in model specifications. Data are from the World Bank. Wealth and population could each contribute to a country’s political system, as well as the strategies used by contentious campaigns based in that country.

Comparing violent and nonviolent campaigns

Before presenting results, I present descriptive evidence to motivate the variation explained in the analysis: violent and nonviolent campaigns differ systematically in their political context and internal attributes. Figure 1 presents results of a balance test for different covariates between violent and nonviolent campaigns. Each standardized difference in means estimate has 95% confidence intervals around the point estimate. Intervals which do not overlap zero represent statistically significant differences between violent and nonviolent campaigns on those covariates. A positive difference in means indicates higher values of that covariate among nonviolent campaigns, while a negative difference in means indicates higher values of the covariate among violent campaigns. The first and fourth variables in the plot, campaign size and use of media by campaigns, are greater in nonviolent campaigns. The second, third, and fifth variables in the plot, level of repression against the campaign, use of indiscriminate repression against the campaign, and internal conflict within the campaign are all significantly more common in violent campaigns.

Statistical estimation

I test the first two hypotheses with both a linear probability model including campaign fixed effects and standard errors clustered at the campaign level, and with a conditional logistic regression.⁸ Conditional logistic regression models a choice of a campaign among a discrete number of alternatives – in this case violence and nonviolence – using campaigns as their own controls. I specify campaigns as the strata in the conditional logistic model to control for unmeasured stable characteristics which could explain their propensity to choose violence or nonviolence. I estimate the relationship between repression, political competition, and social movement resources and the likelihood a campaign chooses nonviolence in the following campaign year.

Coefficients for the linear probability models represent the estimated change, positive or negative, in the probability of a campaign choosing nonviolent resistance in year $t + 1$ given a unit change in the independent variable of interest. Positive coefficients for the conditional logistic model represent an increase in the log odds of a campaign choosing nonviolence for a one-unit increase in the independent variable, while negative coefficients represent a decrease in the log odds of a campaign choosing nonviolence (or an increase in the log odds of the campaign choosing violence) for a one-unit increase in the independent variable.

Results

The choice of nonviolence

Results are displayed in Table II and Table III. For Hypothesis 1, political competition shows a strong positive relationship with the likelihood that a campaign chooses nonviolence in the following campaign year. Repression is negatively associated with nonviolence in the baseline specification and in the model with controls and with political competition but not significant. These results are consistent across the linear probability models and logit models. This supports the reasoning of Hypothesis 1: political competition is associated with the choice of nonviolence, but repression is not. This result lends support to the theoretical mechanism of political competition being a signal of accommodation,

Table II.

Correlates of the choice of nonviolence

	Dependent variable:
	Nonviolence, t+1
	(1)	(2)	(3)	(4)	(5)
Repression, t	$-$ 0.007				0.009
	(0.018)				(0.009)
Political competition, t		0.201 $^{* *}$			0.239 $^{* *}$
		(0.056)			(0.068)
SM assets, t			0.031 $^{*}$	0.088 $^{* *}$	0.085 $^{* *}$
			(0.014)	(0.032)	(0.030)
SM assets sq., t				$-$ 0.017 $^{†}$	$-$ 0.015 $^{†}$
				(0.009)	(0.009)
Campaign size, t	$-$ 0.003	$-$ 0.006	0.019	0.015	0.011
	(0.012)	(0.012)	(0.012)	(0.011)	(0.011)
Extreme goal, t	0.003	$-$ 0.006	$-$ 0.004	0.001	$-$ 0.003
	(0.016)	(0.006)	(0.015)	(0.022)	(0.019)
Log(GDP per capita), t	0.022	0.003	0.050 $^{* *}$	0.054 $^{* *}$	0.034 $^{*}$
	(0.013)	(0.014)	(0.018)	(0.018)	(0.016)
Log(population), t	0.021	$-$ 0.027	$-$ 0.002	$-$ 0.002	$-$ 0.067 $^{* *}$
	(0.016)	(0.018)	(0.020)	(0.020)	(0.026)
Constant	0.706 $^{* *}$	1.195 $^{* *}$	0.603 $^{* *}$	0.564 $^{* *}$	1.191 $^{* *}$
	(0.157)	(0.160)	(0.176)	(0.184)	(0.175)
Observations	778	779	555	555	555
Adjusted R $^{2}$	0.713	0.724	0.709	0.711	0.725
Fixed effects	Y	Y	Y	Y	Y

$^{†}$ p $<$ 0.1; $^{*}$ p $<$ 0.05; $^{* *}$ p $<$ 0.01.

compared to a campaign’s experience of repression. Among covariates, GDP per capita and campaign size are occasionally positive and significantly related with nonviolence.

Social movement resources are statistically significant and in the expected direction in the linear probability models: the linear and quadratic terms for social movement resources are signed in the expected direction, indicating a curvilinear relationship between a campaign’s nonviolent resources and its choice of resistance method. However, these results do not maintain conventional levels of significance in the logit models. These results show support for Hypothesis 1 and offer some support for Hypothesis 2. To illustrate the logit estimates in Table III, I convert the log odds ratios into predicted probabilities for political competition and social movement resources in Figure 2. The left panel shows the strong, positive, and significant relationship between political competition and the predicted probability of choosing nonviolence. The right panel shows the estimated relationship between social movement resources, including the linear and quadratic terms, and predicted probability of nonviolence.

The results in Figure 2 suggest the likelihood of choosing nonviolence increases in political competition. The increases in likelihood of nonviolence are consistent across the political competition variable. Moving from the first quartile to the mean of the political competition variable is associated with an increase from 52% to 61% in the likelihood a campaign chooses nonviolence in the subsequent year. Moving from the mean to the third quartile of the competition variable is associated with an increase from 61% to almost 74% in the likelihood a campaign chooses nonviolence in the subsequent year. The results also show an increasing relationship between the level of social movement resources and the likelihood of choosing nonviolence, fitting with the expectations in Hypothesis 2, though not always with statistical significance.

Robustness tests

To increase confidence in the results, I conduct several robustness tests. I address the possibility of endogeneity between the two main independent variables, political competition and social movement resources, by

Table III.

Correlates of the choice of nonviolence (logit models)

	Dependent variable:
	Nonviolence, t+1
	(1)	(2)	(3)	(4)	(5)
Repression, t	$-$ 0.200				$-$ 0.037
	(0.298)				(0.173)
Political competition, t		6.199 $^{* *}$			3.839 $^{* *}$
		(1.777)			(1.101)
SM assets, t			0.465 $^{* *}$	1.452 $^{* *}$	0.664
			(0.143)	(0.452)	(0.459)
SM assets sq., t				$-$ 0.227 $^{*}$	$-$ 0.081
				(0.094)	(0.105)
Campaign size, t	0.024	$-$ 0.153	0.278 $^{* *}$	0.341 $^{* *}$	0.165 $^{†}$
	(0.099)	(0.102)	(0.077)	(0.081)	(0.099)
Extreme goal, t	$-$ 0.351	0.178	$-$ 0.587	$-$ 0.826 $^{*}$	$-$ 0.410
	(0.236)	(0.254)	(0.366)	(0.401)	(0.332)
Log(GDP per capita), t	$-$ 0.597	0.642	1.190	1.423 $^{†}$	1.651 $^{*}$
	(0.591)	(0.539)	(0.829)	(0.808)	(0.811)
Log(population), t	6.957 $^{* *}$	$-$ 1.839	7.066 $^{*}$	5.188 $^{†}$	0.887
	(2.549)	(3.305)	(3.238)	(3.004)	(2.922)
Observations	778	779	555	555	555
R $^{2}$	0.012	0.035	0.043	0.046	0.055
Log likelihood	$-$ 69.086	$-$ 59.965	$-$ 49.401	$-$ 48.600	$-$ 45.946

$^{†}$ p $<$ 0.1; $^{*}$ p $<$ 0.05; $^{* *}$ p $<$ 0.01.

Figure 2.

Predicted probability plots

including model specifications in Table II and Table III with competition and resources separately and then together in the same model. This addresses the possibility that competition and resources might cause each other in two ways. First, if political competition fuels social movement resources, for example, then the model specification for resources without the competition variable could suffer from omitted variable bias. Including the competition variable in each table’s fifth column addresses this possibility. Second, if social movement resources generate political competition, for example, then the estimate for resources in the specification with political competition could suffer from ‘post-treatment’ bias. The model with resources and without political competition in each table’s fourth column addresses this possibility. Results across these specifications, particularly in the linear probability models, are consistent and robust.

Another concern about endogeneity is that an unmeasured variable affects competition, resources, and the choice of nonviolence and therefore biases the estimated relationships between them. To account for this possibility, the results are subjected to a sensitivity analysis. This procedure determines the extent to which a statistically significant estimate of the relationship between two variables could be the result of omitted variable bias and retain its significance. The result for political competition is strong: 44% of the estimate would need to be the result of bias from an omitted variable to invalidate the statistical significance. For social movement resources, the results are slightly weaker: 38% of the estimate would need to be the result of omitted variable bias to invalidate the statistical significance. Full results are in the Online appendix.

Finally, I conduct a robustness test in which I replace the measure of political competition used in the main results with one from Polity IV data (Marshall, Gurr & Jaggers, 2016). The parcomp variable from the Polity IV data constructs a five-point ordinal scale which measures the ‘extent to which alternative preferences for policy and leadership can be pursued in the political arena’. This measurement matches the underlying concept of a signal of likely accommodation of nonviolent resistance methods. Results with this alternative measurement remain robust and are reported in the Online appendix.

Extension: The continued nonviolent advantage

While the argument implies political competition and campaign resources shape campaigns’ choice of nonviolent resistance methods, it does not suggest these features eliminate the advantages of nonviolence, such as its ability to generate backlash protests when repressed (Pearlman, 2018; Aytaç, Schiumerini & Stokes, 2018; Chenoweth & Stephan, 2011). However, the argument does suggest that studying the relationship between nonviolence and campaign outcomes such as backlash protests should account for the possibility of selection into nonviolence based on political competition and resources. That is, nonviolent campaigns could be more effective than violent ones not because they are nonviolent, but because they emerge in environments systematically more favorable for achieving their demands.

Even after accounting for the selection of nonviolence, I expect nonviolence to retain its advantage in generating backlash protests. Backlash emerges in response to the repression of nonviolent campaigns because participants in those campaigns are targeted with force considered disproportionate to the means of the resistance (Edwards & Arnon, forthcoming; Pearlman, 2018). While increased political competition may affect the likelihood of nonviolent resistance, there is nothing about such competition which changes the comparative advantage of nonviolent campaigns in generating backlash compared with violent campaigns.

The estimation for this extension proceeds in two parts. The first strategy treats the correlates of nonviolence as omitted variables. I estimate a series of linear probability models with backlash protests as the dependent variable and specifications including nonviolent resistance methods alone and while controlling for political competition and social movement resources. The dependent variable in this analysis is whether repression of a campaign triggered backlash in a given year. This variable is a dummy variable created from the camp-backlash variable in NAVCO 2.0, and takes the value of 1 if repression of the campaign increased domestic mobilization of the campaign in a given year and the value of 0 otherwise. Models also include the controls indicated for tests of the first two hypotheses as well as campaign fixed effects. There should be a positive relationship between nonviolent tactics and backlash protests when only nonviolence is included, as well as when political competition and resources are included.

The second strategy estimates a selection model, treating the correlates of nonviolence as shaping whether nonviolent campaigns appear in the sample in the first place. The use of the selection model would be justified if there was a concern that nonviolent campaigns appear in the sample because of more open political competition, and that it was because of this competition – rather than nonviolence itself – that drove the emergence of backlash protests. By adjusting for the selection of nonviolence, I account for such a possibility. To this end, two models are specified. The first stage estimates the choice of nonviolence as a function of political competition, while the second stage estimates the occurrence of backlash protests as a function of the choice of nonviolence, with the latter weighted according to its probability of selection.

Table IV.

Nonviolence and backlash

	Dependent variable: Backlash protests, t+1
	Linear probability model				Selection
	(1)	(2)	(3)	(4)	(5)
Nonviolence, t	0.408 $^{* *}$			0.468 $^{* *}$	0.615 $^{* *}$
	(0.085)			(0.091)	(0.106)
Political competition, t		0.350 $^{* *}$		0.086
		(0.111)		(0.129)
SM assets, t			0.030	0.031
			(0.039)	(0.037)
Campaign size, t	0.041	0.038	0.037	0.025	0.040
	(0.029)	(0.029)	(0.031)	(0.032)	(0.030)
Extreme goal, t	$-$ 0.199	$-$ 0.211	$-$ 0.262	$-$ 0.255	$-$ 0.198
	(0.144)	(0.146)	(0.183)	(0.184)	(0.144)
Log(GDP per capita), t	$-$ 0.172 $^{* *}$	$-$ 0.197 $^{* *}$	$-$ 0.095 $^{†}$	$-$ 0.122 $^{*}$	$-$ 0.177 $^{* *}$
	(0.046)	(0.047)	(0.051)	(0.052)	(0.047)
Log(population), t	0.041	$-$ 0.036	$-$ 0.035	$-$ 0.059	0.039
	(0.113)	(0.116)	(0.122)	(0.126)	(0.113)
Constant	1.216	2.388 $^{*}$	1.943 $^{*}$	1.843 $^{†}$	1.059
	(0.923)	(0.952)	(0.975)	(1.027)	(0.920)
Observations	727	727	555	555	727
Adjusted R $^{2}$	0.378	0.376	0.378	0.399
Fixed effects	Y	Y	Y	Y	Y

$^{†}$ p $<$ 0.1; $^{*}$ p $<$ 0.05; $^{* *}$ p $<$ 0.01.

Table IV displays results from each estimation strategy. The first four columns depict results from linear probability models according to the first strategy, while the fifth column depicts results from the second strategy. There is a strong, positive relationship between nonviolence and backlash protests in the following year, and this relationship remains substantively and statistically significant even when controlling for political competition and social movement resources. Nonviolent campaigns have a probability of generating backlash protests 40% higher than violent campaigns. Political competition is positively related with backlash when not accounting for nonviolence, but this relationship disappears when including nonviolence. Results from the selection model suggest similar conclusions: nonviolence has a relationship with backlash protests which is independent of political competition, and political competition is related to backlash protests to the extent that it contributes to nonviolent resistance.

Ordinal measurement of violence and nonviolence

To this point, I have treated a campaign’s choice of resistance method as dichotomous. However, campaign strategies fluctuate over time, employing different intensities and combinations of tactics (Cunningham, Dahl & Frugé, 2017; Pressman, 2017). The previous analysis has suggested that political competition and social movement resources relate to whether a campaign chooses nonviolence, but do not relate to other aspects or actions related to campaigns such as backlash protests. If this is the case, then we should also not expect a relationship between political competition and social movement resources and the intensity of campaign tactics. Intensity of tactics, particularly the number of participants, is another advantage of nonviolence (Chenoweth & Stephan, 2011). If there is no relationship between political competition and intensity and social movement resources and intensity, this increases confidence that the explanation advanced in this article complements existing research on the advantage of nonviolent mobilization.

To test this additional implication, I take a random sample of 64 campaigns and all their associated campaign-years from the NAVCO 2.0 data, then extend the measurement of resistance method in each campaign-year to an ordinal scale. I hand-code each campaign-year according to criteria in the Minorities at Risk (MAR) data (MAR, 2009). The MAR rebellion variable ranges from 0, with no violent incidents

Table V.

Results for intensity of protest

	Dependent variable:
	Protest, t + 1
	(1)	(2)	(3)	(4)
Repression, t	$-$ 0.272 $^{* *}$			$-$ 0.345 $^{* *}$
	(0.081)			(0.114)
Political competition, t		0.341		$-$ 0.409
		(0.465)		(0.258)
SM assets, t			0.195 $^{*}$	0.181 $^{†}$
			(0.094)	(0.096)
Time since start	$-$ 0.052 $^{* *}$	$-$ 0.028 $^{*}$	$-$ 0.0001	0.015
	(0.018)	(0.012)	(0.110)	(0.107)
Extreme goal, t	$-$ 1.356	$-$ 1.381
	(0.994)	(0.992)
Log(GDP per capita), t	$-$ 0.222	$-$ 0.222	$-$ 0.143	$-$ 0.250
	(0.190)	(0.194)	(0.174)	(0.165)
Log(population), t	3.913 $^{* *}$	2.472 $^{* *}$	1.512	1.428
	(1.154)	(0.577)	(5.007)	(4.953)
Observations	296	300	150	150
Adjusted R $^{2}$	0.110	0.071	$-$ 0.045	$-$ 0.013

$^{†}$ p $<$ 0.1; $^{*}$ p $<$ 0.05; $^{* *}$ p $<$ 0.01

reported, to 7, with a full-scale civil war. Intermediate values include campaigns of terrorism and guerrilla activity. The MAR protest variable ranges from 0, with no nonviolent incidents reported, to 5, with protests or strikes exceeding 100,000 participants. Details of the sample and coding criteria are in the Online appendix.

First, I use linear regression with campaign fixed effects and standard errors clustered at the campaign level to estimate the relationship between the correlates of nonviolent resistance and intensity of resistance. Next, I conduct a similar test except with violent rebellion as the regressor. Specifications are identical to the tests of Hypothesis 1 and Hypothesis 2, except I also include a covariate time since start. This counts the number of years since the campaign onset, as campaign intensity could follow a temporal pattern (Truex, 2019). Instead of estimating the determinants of choice of resistance method, here I estimate the determinants of intensity of resistance method. The results are reported in Table V and Table VI.

Table V indicates no consistent relationship between political competition and the intensity of protest. However, repression is consistently and significantly related with a decrease in the intensity of protest. Campaigns which occur in states with larger populations also tend to have more intense protests, primarily because this is measured in terms of protest size. In Table VI, repression is associated with an increase in the intensity of rebellion. These results contrast with results explaining the binary choice of campaign tactics, suggesting that tactical choice depends on political commitments such as open competition, but participation and momentum in a campaign depend on state use of force against campaign participants. State violence appears to be associated with smaller protests and more inflamed violent rebellions. Taken together with the results for backlash protest, the results suggest that nonviolent resistance generates backlash when repressed, yet protests following greater repression tend to be smaller. In neither case do political competition or resources relate to the ongoing dynamics of campaigns.

The main implication from these tests is that structural characteristics, such as political competition and social movement resources, may explain the type of resistance a campaign employs, but poorly explain the ongoing dynamics of resistance. Rather, the tactical interplay – the use of repression, emergence of backlash protests, or shifting types of violent or nonviolent resistance – between states and campaigns shape resistance dynamics and outcomes (Moore, 1998, 2000; Davenport, 2015; Cunningham, Dahl & Frugé, 2017). However, a challenge with interpreting association of repression with the intensity of protest and rebellion is endogeneity (Carey, 2009; Ritter & Conrad, 2016). Repression could shrink protest, or protest could decrease in anticipation of repression.

Table VI.

Results for intensity of rebellion

	Dependent variable:
	Rebellion, $t + 1$
	(1)	(2)	(3)	(4)
Repression, t	0.658 $^{*}$			0.032
	(0.324)			(0.036)
Political competition, t		$-$ 1.345		0.600
		(1.210)		(0.454)
SM assets, t			$-$ 0.014	$-$ 0.064
			(0.027)	(0.040)
Time since start	0.137 $^{†}$	0.076	$-$ 0.016	$-$ 0.020
	(0.072)	(0.058)	(0.042)	(0.038)
Extreme goal, t	1.537	1.639
	(1.376)	(1.385)
Log(GDP per capita), t	$-$ 0.454 $^{*}$	$-$ 0.443 $^{*}$	0.093	0.197
	(0.227)	(0.251)	(0.090)	(0.149)
Log(population), t	$-$ 3.927	$-$ 0.102	0.582	0.009
	(2.439)	(2.290)	(1.957)	(1.692)
Observations	296	300	150	150
Adjusted R $^{2}$	0.151	0.079	$-$ 0.185	$-$ 0.144

$^{†}$ p $<$ 0.1; $^{*}$ p $<$ 0.05; $^{* *}$ p $<$ 0.01.

Discussion

Through the analysis of 250 contentious campaigns from 1945 to 2006 in the NAVCO 2.0 data, I find support for my hypotheses that political competition and resources shape the choice of campaigns’ resistance methods. Hypothesis 1 and Hypothesis 2 receive support across different estimation strategies for campaign choice of violence or nonviolence. Several robustness tests address potential concerns about endogeneity and measurement. I also find that nonviolence continues to have an advantage in generating backlash protests even when accounting for political competition. An extension of the NAVCO binary indicator for violent and nonviolent resistance methods to an ordinal scale of resistance intensity reinforces the findings for backlash protests: political competition and campaign resources do not explain the ongoing dynamics of resistance.

The findings suggest two different processes at work in the interaction between campaigns and states. First, the political environment and social movement resources are related to a campaign’s resistance method. Political competition signals the state is willing to accommodate campaign demands, making violence an inefficient means of winning concessions. Social movement resources offer greater marginal benefit to the use of nonviolence, though this benefit diminishes at higher levels of resources. Second, phenomena such as backlash protests and shifts in the intensity of campaign tactics relate more to dynamic factors such as the state’s use of repression and the campaign’s use of certain repertoires of action. Each of these processes points to a rationalist approach to campaign decisionmaking: campaigns choose the resistance method which best ensures future mobilization and the ability to achieve their demands. Theories of how campaigns choose such methods should focus on conditions that affect whether campaigns select violence or nonviolence, and integrate these theories with the dynamic processes of repression and resistance.

Overall, the findings are consistent with two different strands of literature on contentious campaigns and civil resistance. First, by showing that open political competition corresponds with a higher likelihood of choosing nonviolence, the findings fit with work on the ‘domestic democratic peace’ (Davenport, 2007a,b) and political opportunity (McAdam, Tarrow & Tilly, 2001; Carey, 2009; Young & Dugan, 2011; Heger, 2014): when campaigns are more easily incorporated into the political process, it is more likely they forego violence as an inefficient means of achieving their demands. Second, by showing that social movement resources correspond with nonviolence, the findings fit with a literature which suggests the social movement resources of these campaigns – in particular the organizations, networks, and skills of participants which permit connections within and between social groups – drive nonviolent resistance (Thurber, 2019; Zeira, 2019; Dahlum, 2019; Clarke, 2014; Gade, 2020). Importantly, the findings complement both literatures: state characteristics and social movement resources shape campaign decisions. Future research should consider how these two features interact to shape resistance methods and campaign dynamics.

The findings also contribute to cross-national empirical work on contentious campaigns using NAVCO data (Chenoweth & Lewis, 2013). By showing that political competition and movement resources contribute to nonviolence, they add to a list of other determinants of nonviolence tested with NAVCO data. In particular, Dahlum (2019) identifies the education level of participants as a key factor in campaigns’ turn to nonviolence, Karakaya (2018) demonstrates that globalization contributes to campaign nonviolence, and Schaftenaar (2017) finds a connection between gender equality and nonviolence. Future empirical work should continue to engage with these data to test plausible correlates of nonviolent resistance.

Understanding how campaigns choose resistance methods also has policy implications. Normatively, nonviolent resistance causes less harm than violent resistance. Yet I have shown nonviolent campaigns are tenuous, and become violent if the conditions which support them disappear. In Egypt, for example, opposition groups’ fragmentation inhibited the development of organizations, networks, and connections among activists, making sustained nonviolent resistance difficult after the Arab Spring uprising (Nugent, 2020). A faction of one opposition group, the Muslim Brotherhood, soon turned to violence after the Sisi dictatorship took power in 2013. Developing the connections among activists which could reverse fragmentation and encouraging open political competition would do more for nonviolent resistance than promoting the method itself.

Footnotes

Replication data

The dataset, codebook, and R-files for the empirical analysis in this article, along with the Online appendix, can be found at .

Acknowledgments

The author is grateful to Jennifer Gandhi, Danielle Jung, William Wagstaff, Sophia Wilson, and two anonymous reviewers for thoughtful comments.

ORCID iD

Pearce Edwards

Notes

References

Aksoy

Deniz

(2014) Elections and the timing of terrorist attacks. Journal of Politics 76(4): 899–913.

Aytaç

S Erdem

Schiumerini

Luis

Stokes

Susan

(2018) Why do people join backlash protests? Lessons from Turkey. Journal of Conflict Resolution 62(6): 1205–1228.

Carey

Sabine C

(2009) Protest, Repression, and Political Regimes: An Empirical Analysis of Latin America and Sub-Saharan Africa. New York: Routledge.

Chenoweth

Erica

Lewis

Orion A

(2013) Unpacking nonviolent campaigns: Introducing the Navco 2.0 dataset. Journal of Peace Research 50(3): 415–423.

Chenoweth

Erica

Stephan

Maria

(2011) Why Civil Resistance Works: The Strategic Logic of Nonviolent Conflict. New York: Columbia University Press.

Chenoweth

Erica

Ulfelder

Jay

(2017) Can structural conditions explain the onset of nonviolent uprisings? Journal of Conflict Resolution 61(2): 298–314.

Clarke

Killian

(2014) Unexpected brokers of mobilization: Contingency and networks in the 2011 Egyptian uprising. Comparative Politics 46(4): 379–397.

Coppedge

Michael

Gerring

John

Lindberg

Staffan I

Skaaning

Svend-Erik

Teorell

Jan

Altman

David

Bernhard

Michael

Steven Fish

Glynn

Adam

Hicken

Allen

, et al. (2017) V-dem dataset v7 (https://www.v-dem.net/en/).

Crabtree

Charles

Kern

Holger L

Pfaff

Steven

(2018) Mass media and the diffusion of collective action in authoritarian regimes: The June 1953 East German uprising. International Studies Quarterly 62(2): 301–314.

10.

Cunningham

Kathleen Gallagher

(2013a) Actor fragmentation and civil war bargaining: How internal divisions generate civil conflict. American Journal of Political Science 57(3): 659–672.

11.

Cunningham

Kathleen Gallagher

(2013b) Understanding strategic choice: The determinants of civil war and nonviolent campaign in self-determination disputes. Journal of Peace Research 50(3): 291–304.

12.

Cunningham

Kathleen Gallagher

Beaulieu

Emily

(2010) Dissent, repression, and inconsistency. In: Chenoweth

Erica

Lawrence

Adria

(eds) Rethinking Violence: States and Non-State Actors in Conflict. Cambridge, MA: MIT Press, 173–195.

13.

Cunningham

Kathleen Gallagher

Dahl

Marianne

Frugé

Anne

(2017) Strategies of resistance: Diversification and diffusion. American Journal of Political Science 61(3): 591–605.

14.

Dahlum

Sirianne

(2019) Students in the streets: Education and nonviolent protest. Comparative Political Studies 52(2): 277–309.

15.

Davenport

Christian

(2007a) State Repression and the Domestic Democratic Peace. New York: Cambridge University Press.

16.

Davenport

Christian

(2007b) State repression and the tyrannical peace. Journal of Peace Research 44(4): 485–504.

17.

Davenport

Christian

(2015) How Social Movements Die: Repression and Demobilization of the Republic of New Africa. Cambridge: Cambridge University Press.

18.

Dower

Paul Castaneda

Finkel

Evgeny

Gehlbach

Scott

Nafziger

Steven

(2018) Collective action and representation in autocracies: Evidence from Russia’s great reforms. American Political Science Review 112(1): 125–147.

19.

Dunning

Thad

(2011) Fighting and voting: Violent conflict and electoral politics. Journal of Conflict Resolution 55(3): 327–339.

20.

Edwards

Pearce

Arnon

Daniel

(forthcoming) Violence on many sides: Framing effects on protest and support for repression. British Journal of Political Science. https://doi.org/10.1017/S0007123419000413.

21.

Fergusson

Leopoldo

Robinson

James A

Torvik

Ragnar

Vargas

Juan F

(2016) The need for enemies. Economic Journal 126(593): 1018–1054.

22.

Finkel

Evgeny

(2015) The phoenix effect of state repression: Jewish resistance during the Holocaust. American Political Science Review 109(2): 339–353.

23.

Gade

Emily Kalah

(2020) Social isolation and repertoires of resistance. American Political Science Review 114(2): 309–325.

24.

Gandhi

Jennifer

(2008) Political Institutions under Dictatorship. Cambridge: Cambridge University Press.

25.

Heger

Lindsay L

(2014) Votes and violence: Pursuing terrorism while navigating politics. Journal of Peace Research 52(1): 32–45.

26.

Karakaya

Süveyda

(2018) Globalization and contentious politics: A comparative analysis of nonviolent and violent campaigns. Conflict Management and Peace Science 35(4): 315–335.

27.

Klein

Graig R

Regan

Patrick M

(2018) Dynamics of political protests. International Organization 72(2): 485–521.

28.

Kydd

Andrew H

Walter

Barbara F

(2002) Sabotaging the peace: The politics of extremist violence. International Organization 56(2): 263–296.

29.

Kydd

Andrew H

Walter

Barbara F

(2006) The strategies of terrorism. International Security 31(1): 49–80.

30.

Levitsky

Steven

Way

Lucan A

(2010) Competitive Authoritarianism: Hybrid Regimes After the Cold War. Cambridge: Cambridge University Press.

31.

Lichbach

Mark I

(1987) Deterrence or escalation? The puzzle of aggregate studies of repression and dissent. Journal of Conflict Resolution 31(2): 266–297.

32.

Lichbach

Mark I

(1995) The Rebel’s Dilemma. Ann Arbor, MI: University of Michigan Press.

33.

MAR (2009). Minorities at Risk Dataset. Center for International Development and Conflict Management (http://www.mar.umd.edu/mar_data.asp).

34.

Marshall

Monty

Gurr

Ted Robert

Jaggers

Keith

(2016) Polity IV Project: Dataset Users’ Manual. Center for Systemic Peace (https://www.systemicpeace.org/polityproject.html).

35.

McAdam

Doug

Tarrow

Sidney

Tilly

Charles

(2001) Dynamics of Contention. Cambridge: Cambridge University Press.

36.

McCarthy

John D

Zald

Mayer N

(1977) Mobilization and social movements: A partial theory. American Journal of Sociology 82(6): 1212–1241.

37.

Moore

Will H

(1998) Repression and dissent: Substitution, context, and timing. American Journal of Political Science 42(3): 851–873.

38.

Moore

Will H

(2000) The repression of dissent: A substitution model of government coercion. Journal of Conflict Resolution 44(1): 107–127.

39.

Murdie

Amanda

Davis

David R

(2012) Shaming and blaming: Using events data to assess the impact of human rights INGOs. International Studies Quarterly 56(1): 1–16.

40.

Nugent

Elizabeth R

(2020) The psychology of repression and polarization. World Politics 72(2): 291–334.

41.

Pape

Robert

(2003) The strategic logic of suicide terrorism. American Political Science Review 97(3): 343–361.

42.

Pearlman

Wendy

(2018) Moral identity and protest cascades in Syria. British Journal of Political Science 48(4): 877–901.

43.

Pierskalla

Jan Henryk

(2010) Protest, deterrence, and escalation: The strategic calculus of government repression. Journal of Conflict Resolution 54(1): 117–145.

44.

Pressman

Jeremy

(2017) Throwing stones in social science: Non-violence, unarmed violence, and the first intifada. Cooperation and Conflict 6(4): 519–536.

45.

Ritter

Emily Hencken

Conrad

Courtenay R

(2016) Preventing and responding to dissent: The observational challenges of explaining strategic repression. American Political Science Review 110(1): 85–99.

46.

Ryckman

Kirssa Cline

(2020) A turn to violence: The escalation of nonviolent movements. Journal of Conflict Resolution 64(2–3): 318–343.

47.

Schaftenaar

Susanne

(2017) How (wo)men rebel: Exploring the effect of gender equality on nonviolent and armed conflict onset. Journal of Peace Research 54(6): 762–776.

48.

Stephan

Maria J

Chenoweth

Erica

(2008) Why civil resistance works: The strategic logic of nonviolent conflict. International Security 33(1): 7–44.

49.

Sullivan

Christopher M

(2016) Undermining resistance: Mobilization, repression, and the enforcement of the political order. Journal of Conflict Resolution 60(7): 1163–1190.

50.

Thomson

Henry

(2018) Grievances, mobilization, and mass opposition to authoritarian regimes: A subnational analysis of East Germany’s 1953 Abbreviated Revolution. Comparative Political Studies 51(12): 1594–1627.

51.

Thurber

Ches

(2019) Social ties and the strategy of civil resistance. International Studies Quarterly 63(4): 974–986.

52.

Tilly

Charles

(1978) From Mobilization to Revolution. New York: McGraw-Hill.

53.

Tilly

Charles

(2008) Contentious Performances. Cambridge: Cambridge University Press.

54.

Trejo

Guillermo

(2012) Popular Movements in Autocracies: Religion, Repression, and Indigenous Collective Action in Mexico. Cambridge: Cambridge University Press.

55.

Truex

Rory

(2019) Focal points, dissident calendars, and preemptive repression. Journal of Conflict Resolution 63(4): 1032–1052.

56.

Young

Joseph K

Dugan

Laura

(2011) Veto players and terror. Journal of Peace Research 48(1): 19–33.

57.

Zeira

Yael

(2019) From the schools to the streets: Education and anti-regime resistance in the West Bank. Comparative Political Studies 52(8): 1131–1168.