Better the Devil You Know? How Fringe Terrorism Can Induce an Advantage for Moderate Nonviolent Campaigns

ABSTRACT Fringe terrorism is common during nonviolent campaigns. We examine how this can modify the strategic environment between dissident groups and the state in ways that present both challenges and opportunities to moderate factions. Terrorism is intended to promote violent escalation in a conflict, but we argue that fringe terrorist activities in a nonviolent campaign under certain conditions can induce an advantage for well-organized moderate factions. The risk of escalation following terrorism can give the government more incentives to offer concessions to moderate campaign leaders if the movement can credibly prevent armed escalation. The ability to control and prevent violence is more likely when nonviolent movements have a hierarchical structure and a centralized leadership, as such campaigns are better able to prevent shifts by supporters towards violent fringes. Using new data on terrorist attacks by factions sharing the same overall objectives as ongoing nonviolent campaigns, we show that nonviolent campaigns are more likely to see substantial gains in spite of fringe terrorist activities when a movement has a hierarchical structure and a centralized leadership.


Introduction
Recent research has argued that nonviolent dissent tends to be more effective than armed violence and highlighted its many strategic advantages. 1 There is a conceptually clear distinction between resistance campaigns that primarily rely on organized violence or nonviolence, and shifts from the former to the latter as dominant tactics are relatively rare. However, many primarily nonviolent campaigns still see much smaller-scale fringe political violence such as riots or terrorist attacks. We posit that fringe terrorism can modify the strategic environment between nonviolent organizations and the states through a credible threat of escalation. 2 We argue that even though states are generally reluctant to concede to dissidents, they also have incentives to make concessions to more moderate nonviolent factions, if this can help undermine violent factions that pose a credible threat. 3 Accommodation to moderate factions may be a lesser evil to states when moderate nonviolent organizations can credibly limit radicalization and a drift towards support for violence in their support base. 4 Using new data on terrorism in nonviolent resistance campaigns between 1946 and 2006, we find a higher likelihood of substantial political gains when fringe terrorism occurs during nonviolent campaigns with a hierarchical structure and a centralized leadership, consistent with our argument that states have more incentives to accommodate moderate campaigns when this can help sideline violent fringe groups.

Radical flanks: Harmful or helpful?
There is a large body of research studying the overall effects of "radical flanks" on the outcomes of political dissent, and in particular whether fringe violence makes it more or less likely that nonviolent campaigns see progress towards their goals. 5 Yet, a clear answer has remained elusive; some studies argue that violent flanks undermine nonviolent resistance campaigns, 6 while others hold that violence by radical flanks can strengthen moderate groups relative to the state. 7 The disparate findings across the existing literature may reflect that there simply is no general or systematic relationship between violent flanks and campaign outcomes. However, in our view, existing research also suffers from key theoretical and methodological limitations. The mechanisms through which fringe violence could affect political outcomes of civil resistance campaigns often remain unclear. It is also unclear what types of violence could affect the strategic environment surrounding a nonviolent campaign. Finally, with some exceptions, 8 most research has tended to focus on individual case studies, often selected to illustrate a specific argument, rather than systematic evaluation or identification of the most common patterns and outcomes.
With regard to mechanisms, some literature emphasizes how violent flanks can generate crises that "are resolved to moderate advantage," by favoring greater acceptance of the goals and methods of moderate organizations. 9 Herbert Haines, for example, suggests that radical violent activities strengthened the power of moderate organizations in the US civil-rights movement through increased donations. 10 Similarly, Jo Freeman shows that violent activities within the US women's liberation movement increased the bargaining power of the reformist organizations. 11 Carol McClurg Muller finds that assaults on civilians and property during the US civil-rights movement increased the authorities' willingness to provide short-term symbolic reassurance. 12 Other scholars, however, emphasize how violent activities can discredit a nonviolent movement and drive away potential supporters. Erica Chenoweth and Kurt Schock find that fringe violence is generally associated with lower active participation in nonviolent campaigns. 13 Their study of four civil resistance campaigns suggests that radical flank violence at best has highly varied and unpredictable effects on political outcomes.
Existing work tends to define radical flanks in nonviolent campaigns rather generally in terms of either extreme positions and ideology, 14 or advocacy for more militant tactics. 15 Most studies thus fail to differentiate between different types of violence, even though distinct forms of violence can have very different effects on the strategic environment between a government and a nonviolent dissident campaign. We argue that different violent activities have distinct implications for the government's threat perception, depending on the degree of organization involved as well as the potential for conflict escalation. We focus on fringe terrorist activities in civil resistance campaigns, carried out by groups that try to appeal to the same aggrieved population as the dominant nonviolent organizations. Fringe violence can come in many forms, ranging from spontaneous riots, which are often emotionally driven responses to police violence, to more organized forms of violence-such as guerrilla warfare-that require military training and planning. We focus on terrorism as a form of smaller-scale organized violence particularly likely to occur in dissident campaigns. Unlike largely disorganized forms of violence such as riots and police clashes, terrorism entails deliberately planned and organized attacks against intermediate targets to coerce the government and its supporters. Terrorism is less labour and equipment intensive than conventional armed violence, and can be carried out with limited active participation and material resources. Terrorism is often seen as a means to larger ends, where small groups use attacks to raise their profile and attract supporters, in the hope of eventually mounting organized violence against the state. There is of course an extensive literature on whether terrorism in general is "effective," without a clear scholarly consensus. 16 However, a very broad set of groups perpetrates terrorism at some point, and "success" depends on either the objectives of specific actors or the criteria applied by observers. We have a more specific focus on the potential effects of fringe terrorism on the political outcomes in ongoing dissident campaigns. We also consider a specific measure of success in terms of the objectives of the larger moderate campaign, which often may not fully correspond to the more ambitious goals of the individual terrorist cells.

Terrorism as a "game changer"
We posit that terrorism can modify the strategic environment between dissident campaigns and the government by generating a credible threat of conflict escalation. 17 Radical fringes may resort to terrorism due to ideological and strategic disagreements with competing factions within a nonviolent campaign, and divergent beliefs on the efficacy of violent attacks. 18 Radical fringe factions typically lack sufficient resources to engage in direct conventional military violence against the government, but can carry out indirect terrorist attacks that can impose significant costs on the government through their broader economic and political consequences. Terrorist attacks reflect organized activities and deliberate planning, unlike disorganized forms of violence such as riots and clashes with the police. Threats to launch an organized insurgency are unlikely to be credible at the outset, 19 but a salient terrorist campaign could make future escalation more realistic, by allowing groups to grow support and resources. 20 Terrorism can help demonstrate resolve or commitment to the cause, and raise the profile of a faction and attract more supporters, with the hope that the group may eventually be in a position to mount more sustained organized violence against the state.
Terrorist campaigns often fail to realize their dreams of instigating successful popular revolts, but there are several examples where fringe groups have grown from small-scale terrorist activities to large-scale violence. Existing research shows that terrorist attacks are a leading indicator of subsequent outbreaks of civil war. Michael Findley and Joseph Young highlight how groups resort to terrorist attacks "at the beginning of a violent conflict or before civil war to provoke the state, build support and capacity, and thus challenge the state." 21 For example, the Red Brigades, a violent flank of the New Leftist movement in Italy, clearly expressed the logic of terrorist attacks as a strategy to attract greater support in their 1978 strategic manifesto: In this phase, the struggle must assume, by the initiative of the revolutionary vanguards, the form of war… . we … want war! … revolutionary violence pushes the enemy to face it, … it forces the enemy to react, to operate on the terrain of war: we intend to mobilize and to flush out the imperialist counterrevolution from the folds of the 'democratic' society where it has comfortably hidden in better times! 22 Terrorism can also be used as a provocation or outbidding tactic. The former aims to provoke the state to respond with repression, based on the expectation that repression will encourage greater popular mobilization against the state. 23 As an example, Andy Truskier highlights how efforts to prepare for armed struggle in the campaign against the Brazilian government in the 1970s actively sought to provoke the government: "[b]y firing on workers, the army makes the people angry and brings them to the point of understanding action on yet another level-that is, action directed against the military." 24 Outbidding strategies aim to win over support, and to raise the profile of a group relative to competing dissident factions. 25 Fringe groups often highlighted terrorist attacks as helpful for increasing support and recruitment. As an example, Massoud Ahmad-Zadeh, a leader in the Iranian People's Fadaee during the uprising against the Shah, notes how "[g]roups [that] take up arms and extend the struggle, thereby [increase] the possibilities of material support from the people." 26 The government and the moderate leadership of nonviolent campaigns clearly have opposing interests on the main issues in a conflict, but they can also have some degree of shared interest in avoiding escalating violence. On one hand, even if governments seek to minimize overall dissent, violent extremism may be deemed a worse prospect than nonviolent dissent. On the other hand, moderate nonviolent organizations have an interest in avoiding violence that can undermine the nonviolent campaign, and preventing a loss of support to violent fringe groups.
The first response of a government is normally to repress dissent, 27 but indiscriminate repression often backfires, potentially increasing popular support and alienating the security forces. 28 When terrorism is used as "provocation" and to radicalize supporters, indiscriminate repression can reduce the opportunity costs for supporting violence relative to nonviolence. 29 More selective counterterrorism and repression against violent factions is often difficult and will require substantial time and efforts. As such, it is not surprising that successful counterterrorism often combines coercive policing with accommodation strategies, aiming to win hearts and minds through concessions. Concessions to moderate factions in nonviolent mass dissident campaigns may help isolate extremist groups and avoid escalation. Governments are unlikely to accommodate factions using terrorism, which may set a dangerous precedent. 30 By contrast, offering concessions to nonviolent movements is less unattractive, since "there is more space for negotiations, compromise, power sharing and even complete accommodation when regime members do not fear that losses of power will directly translate into rolling heads." 31 Thus, governments facing both large-scale nonviolent campaigns and fringe terrorism have greater incentives to offer concessions to moderates if this can help to avoid escalating violent conflict and to secure a more controlled transition.

Hierarchical structure and centralized leadership
Simply offering concessions to moderate organizations does not guarantee decreased support for violent fringe groups, and governments may fear that it could encourage more violence.
Nonetheless, concessions to nonviolent campaigns become more appealing when a government can expect moderate campaigns to have sufficient capacity to prevent increasing support for violent flanks and escalation after concessions. 32 A nonviolent campaign with a hierarchical structure and centralized leadership is better able to implement coherent strategies and control its support base. 33 Thus, civil resistance campaigns with such characteristics have a credible capacity to limit violence for governments. Hierarchical structures help strengthen the leadership's capacity to regulate dissent and contentious politics, 34 and coordination can decrease the costs of participation in dissent, which increases effectiveness. 35 Formal lines of communication and command structures can allow leaders to persuade participants to refrain from violence, even in emotionally charged situations. 36 Moreover, a centralized leadership provides recognizable leaders that can negotiate with governments on behalf of the movement, which further advantages moderates to leverage resources and political organization. 37 Our argument is not that centralized leaderships and hierarchical structures alone motivate the government to offer concessions. In contrast, we claim that, combined with a credible risk of conflict escalation, a centralized leadership and hierarchical structure can confer an important advantage to moderates as governments are more likely to believe they stem violent fringe groups and radicalization. Without followers, violent fringe groups cannot grow and become vulnerable to counterterrorism efforts. 38 In sum, nonviolent dissident campaigns that are hierarchically structured and have a centralized leadership should be more likely to be offered political concessions when fringe terrorism and a demonstrated potential for escalating violence occur.
Our theory has two important scope conditions. First, since many strong movements can prevent the emergence of potential violent organizations perpetrating terrorist attacks in the first place, we will not observe systematic terrorism. Although such groups may get concessions more often, this is driven by the direct weight of the movement itself rather than the threat of escalation. Still, even capable moderate organizations often see the emergence of fringe terrorism outside their control. 39 Second, if terrorism could provide a potential advantage in terms of a credible threat, one might wonder if nonviolent campaigns could have incentives to encourage fringe terrorism. Note that fringe terrorist groups also threaten the moderate leadership, and that violence in general undermines the broad participation that maximizes a nonviolent campaign's prospects for success. 40 Thus, efforts to encourage violent factions normally generate more disadvantages for a nonviolent campaign than possible benefits, especially if violent fringes threaten to take over the leading role in the movement. Appearing to either tolerate or failing to condemn violent activities often undermines the legitimacy of moderate groups. It can be difficult to establish the precise relationships between violent fringes and civil resistance campaigns' leaders, especially as the latter have incentives to deny ties with violent fringe groups. However, in our data (described in more detail below) we do not find any cases where mainstream nonviolent organizations either directly organized or colluded with fringe terrorist groups.

Illustrative cases
Before turning to a large-N comparative analysis, we consider a series of illustrative cases to emphasize the causal mechanisms. We start by examining two "most likely" cases, where the mechanism would have to be discernable, if the theory could work at all. 41 We first consider the 2005-6 mass civil resistance campaign against Thai Prime Minister Thaksin Shinawatra, which involved both fringe terrorism and hierarchically organized campaigns. In line with our argument, the government made major accommodations to the moderate organizations that remained committed to nonviolence and these concessions were explicitly framed as attempts to limit the escalation and radicalization that might otherwise result. Since the Thai case could be seen as a case where the support of established institutions plays a central role in ensuring that the movement reaches its goals, we consider a second case with more deep-seated upheaval against the existing order, namely the 1990-91 mass civil resistance campaign against the Communist regime in Albania. Here, even though the government formally retained control of the army and coercive apparatus, we also find concessions to moderate organizations, and clear references to how the alternative could be worse and entail dramatic escalation. To show that terrorism alone does not produce concessions in the absence of a strong moderate faction, we then consider the case of the 1972-2009 dissident campaign for Tamil independence in Sri Lanka. The initial weakness of the moderate factions in the movement allowed the violent fringe movement known as the Liberation Tigers of Tamil Eelam (LTTE) to effectively displace the nonviolent umbrella campaign Tamil United Liberation Front (TULF). 42 This started the transformation from a largely non-violent campaign to a civil war that was to last decades, ultimately resulting in a decisive win for the Sri Lankan government.

2005-6 Thai crisis
The 2005-6 anti-government campaign in Thailand demonstrated how Thai Prime Minister Thaksin Shinawatra had become unpopular, due to increasingly autocratic rule, with severe human rights and civil liberties violations, and unpopular privatization measures. In 2005, the government removed from TV a talk show by publishing mogul Sondhi Limthongkul, which often denounced government corruption and abuse of power. Sondhi Limthongkul then mobilized a large nonviolent dissident campaign calling for Thaksin Shinawatra to resign. The People's Alliance for Democracy (PAD) established in February 2006 successfully brought together a broad coalition, including unions opposed to the privatization of state enterprises, human rights activists, factions in the Thai military who objected to nepotism, as well as groups objecting to corruption. 43 In November 2005, Thaksin Shinawatra sued Sondhi Limthongkul, who was drawing thousands of individuals to weekly rallies. 44 Observers believed that violence could arise during the protests in Bangkok, 45 and January 2006 saw terrorist attacks directed at Thaksin Shinawatra and the Ministry of Justice, and threats of violence by radical university student factions. 46 There is no evidence that PAD orchestrated these violent attacks, and Sondhi Limthongkul repeatedly called for peaceful rallies and to avoid violence. 47 Thaksin Shinawatra stepped down on February 24, 2006 and called new elections. In a resignation speech, Thaksin Shinawatra noted that national intelligence reported that "ill-intentioned people" were trying to infiltrate the dissident campaign in order to instigate violence and stressed how this had made him decide to dissolve the government and seek to resolve the crisis through elections. 48 The 1990-91 Albania crisis The events leading to the end of Communist rule in Albania 1990-91 also illustrate the possible moderate advantage in crises with fringe terrorist violence. When longstanding dictator Enver Hoxha died in 1985, his successor Ramiz Alia took over a country with massive economic challenges. 49 Mass demonstrations emerged throughout the country during the second half of 1990, fuelled by economic scarcity and decreasing fear of repression after the fall of the Berlin wall and end of many Socialist regimes earlier in the year. 50 November 1990 saw a series of violent attacks, including lynched police officers and attacks on public buildings. 51 Alia highlighted the risk of escalating violence, and how "the country threatens to be destabilized and cast into anarchy." 52 In December 1990, moderates formed an umbrella organization called the Democratic Party of Albania, and this organized base for the opposition was met "with the tacit consent of the Communist leadership." 53 The Democratic Party of Albania demonstrated its willingness and ability to maintain non-violent discipline and avoid further escalation of violence. Its leader Arben Imame denounced violence by "hooligans and provocateurs" and stressed how "the Democratic Party was for continuing peaceful dialogue with Communist leaders." 54 Alia subsequently met with the Democratic Party leaders. He initiated a process of concessions, "sacking five politburo members, [and] promising to hold multi-party elections next February and vowing to make changes in the government." 55 Over the following months, the communist party joined the opposition parties in a coalition government to secure political stability and isolate violent fringes. 56 The 1972-2009 Tamil secessionist campaign The fate of the initially non-violent Tamil independence movement in Sri Lanka illustrates how fringe terrorism can generate problems for civil resistance movements when moderate factions and the central leadership are weak. It was evident from the outset that the moderate leadership in the Tamil movement was ineffective in mobilizing and coordinating a coherent nonviolent mass movement. 57 After several failed attempts to secure Tamil autonomy within a larger federal state over the previous decade, the Sinhalese-dominated Sri Lankan government de facto excluded the mainstream Tamil political parties from access to political influence in the early 1970s. As a reaction to the increasing discrimination and repression, various Tamil parties and groups funded the Tamil United Front (TUF) in 1972. However, the TUF failed to gain a clear position as leading representative of the Tamil population. 58 A leading individual in what would later emerge as the Liberation Tigers of Tamil Eelam (LTTE) argues that TUF leaders failed "to offer leadership and guidance to carry out an effective programme of action." 59 Although more militant groups advocating terrorist violence such as the Liberation Tigers of Tamil Eelam (LTTE) were not formally part of the mass movement, the TUF failed to isolate the fringe group and maintain support for a non-violent strategy. 60 By the late 1970s the militant groups eventually became much more than violent fringes, and the LTTE assumed the role as the leading Tamil organization with increasingly mass support. 61

Comparative analysis
In this section we move beyond individual case studies to illustrate the divergent effects of terrorism in non-violent campaigns depending on the strength of the moderate central leadership and consider explicit comparisons to success rates across different campaign profiles. We provide a comparative empirical analysis of whether hierarchically organized nonviolent campaigns have been able to gain concessions at a higher rate in situations with fringe violence across a large number of cases. We identify nonviolent campaigns based on the Nonviolent and Violent Campaigns and Outcomes (NAVCO) 2.0 dataset. This records 109 primarily nonviolent resistance campaigns between 1946 and 2006, and it classifies whether campaigns see varying degrees of "success." 62 For the primarily nonviolent NAVCO campaigns, we then identify terrorist attacks by groups with related political goals as the resistance campaign, using a procedure explained in more detail below. Finally, we use information from NAVCO to identify whether the civil resistance campaign has a hierarchical structure and centralized leadership.
The NAVCO data classify mass dissident campaigns as nonviolent if this is the primary resistance method and participation is limited to unarmed civilians. 63 Note that this does not exclude the use of low-level violent tactics by other dissident organizations, as long as these do not become dominant or primary tactics. 64 Civil resistance campaigns are defined as a "series of observable, continuous, purposive mass tactics or events in pursuit of a political objective, … taking place outside the institutional realm of politics, and confrontational in nature." 65 They must entail shared maximalist goals, have at least 1,000 observed participants, and there must be evidence of coordination among participants. 66 Our main outcome variable is whether civil resistance campaigns see substantial gains. This is based on the ordinal progress variable in NAVCO, which measures the achievements of a campaign on an annual basis and specific type of political concessions from the targeted government. If a state does not change its position at all, the "status quo" prevails (with a value of 0). If the state does not make formal concessions, but changes its behavior to accommodate the opposition, for example allowing greater political openness, we have "visible gains short of concessions" (with a value of 1). Verbal statements of conciliation or changes in the stated position of the regime without additional action constitute "limited concessions" (2). Real actions short of ultimate capitulation, such as policy changes, the removal of state leaders, or the initiation of negotiations with the opposition, constitute "significant concessions" (3). When the campaign entirely achieves its stated political objectives, we have "complete success" (4). Erica Chenoweth and Orion Lewis note that coders often had highly subjective judgments as to whether a campaign achieved a value of 3 or 4. As a consequence, they recommend relying on a "dichotomous variable indicating 'strategic success' (3 & 4) or 'otherwise' (0, 1, & 2)." 67 Based on this suggestion, we operationalize substantial gains as a dichotomous variable, identifying whether civil resistance campaigns reach significant gains or complete success or not. We also consider an alternative measure restricted to only campaigns that see complete success, ignoring all lower level concessions. This provides a stricter measure of success since full accommodation can only be coded in the final years and not during ongoing campaign years.
We identify terrorism occurrence by flagging systematic fringe terrorism by factions sharing political goals related to the objectives of the nonviolent campaigns. We consider terrorist violence as systematic if there are at least three terrorist attacks by organizations sharing the same political goals as the civil resistance movement in a given year. We believe that a dichotomous indicator is more appropriate than the count events, since we are not interested in the intensity of terrorism, and the severity of attacks in any event is not unambiguously measured by the number of attacks. Our main tests use a threshold at three or more terrorist attacks, since single individual attacks are often flukes that may not reflect organized groups or systematic terrorism.
We extract terrorism attacks during civil resistance campaigns from the Global Terrorism Database (GTD) for nonviolent movements between 1970 and 2006 and we code independently terrorist attacks in campaigns before 1970 (see Online Appendix Table 1). We use the GTD's three basic coding rules and three additional criteria to identify terrorist events: 1) Attacks must be intentional; 2) Attacks entail the use or the threat of violence; 3) Perpetrators are non-state actors; 4) Attacks must be aimed at political, economic, or social goals (the exclusive pursuit of economic profit does not satisfy this criterion); 5) Attacks must have intention to coerce, intimidate, or transmit the same message to a larger audience than the immediate victims; 6) Attacks must be outside the context of legitimate warfare activities, and violate humanitarian law with regard to targeting civilians or non-combatants. 68 We only consider terrorist attacks carried out by groups and actors sharing the broad political claims of the civil resistance campaigns (e.g., regime change, independence). To verify that the goals are similar, we used information on the perpetrators and audience of terrorist activities using the GTD search tool (https://www.start.umd.edu/gtd/search/) as well as news reports from Lexis Nexis. We follow the protocol suggested by Walter Enders, Todd Sandler, and Krushav Gaibulloev to extract domestic terrorist attacks from the GTD, 69 which requires that terrorist strategies must have direct consequences for the country only, its institutions, citizens, property, and policies. Some civil resistance campaigns included in the NAVCO data involve terrorist groups that target foreign states, considered to be occupying forces. Although these normally fall under what Enders and co-authors define as transnational terrorism, 70 we include all actions by violent groups pursuing goals that are similar to a civil resistance campaign, focusing on the location where attacks take place. For example, a terrorist attack by Palestinian nationals targeting Israeli or Palestinian nationals in Israel would be counted as a domestic attack, but we do not include a terrorist attack perpetrated by Palestinians against US nationals outside Israel, even if aimed at furthering the liberation of Palestine. We traced the timeline of terrorist activities and concessions in campaigns using secondary sources to ensure that attacks occurred prior to any political progress or concessions within the year. We do not include any cases of terrorism after full accommodation to nonviolent campaigns, which could reflect terrorism used to spoil or undermine existing settlements.
Our measure of hierarchical structure (hierarchy) is a dichotomous variable from NAVCO 2.0, identifying whether nonviolent civil resistance campaigns have "a clear centralized leadership structure, hierarchically organized and with clear lines of authority-often but not necessarily focused on a single leader." 71 This excludes cases where different organizations participate in the same civil resistance campaign with their own individual leaders, each of which has influence over the contentious political tactics and strategies of their individual organizations, or if there is no identifiable leadership in a campaign. 72 Our final dataset in the analysis includes 307 annual observations. We first examine the distribution of cases with and without major concessions across presence/absence of systematic terrorism and hierarchical campaign organization. Our argument is not that we should expect to see more concessions from either feature alone, but it is important to consider first whether there is a clear unconditional relationship with either factor before we can consider the conditional relationship. In Table 1 we compare the share of civil resistance campaigns with systematic terrorism occurrence against whether the campaign sees significant political gains. As can be seen, we observe systematic use of terrorist attacks in 30.6% of the campaign years included in the table. We have a marginally higher share of campaign years with substantial gains in instances where we see systematic terrorism than in campaigns without. However, the majority of campaigns both with and without terrorism do not see concessions. Table 2 displays the share of campaigns with substantial progress by hierarchical structure. As can be seen, 31.33% of the campaign years included in the table have a hierarchical structure. We see a slightly higher share with substantial progress in campaigns with a hierarchical structure than in campaigns without, but the difference is modest, and the modal outcome remains no substantial gains even for campaigns with hierarchical structure.
Taken together, Tables 1 and 2 do not provide any support for the hypothesis that either systematic terrorism or hierarchical campaign structures by themselves are clearly associated with differences in the prospects for concessions. In the analysis below, we examine the evidence for the interactive effect implied by our argument, and we also consider a number of control variables possibly associated with either systematic terrorism or hierarchical campaign structures. Erica Chenoweth and Maria Stephan find that nonviolent dissident campaigns are more likely to succeed in democratic regimes, 73 and some studies argue that democracies provide a favorable environment for terrorist groups. 74 We thus control for democracy, using a dichotomous variable flagging if a country has a score of 6 and above in the "Polity 2" index of Polity IV data. 75 Although we expect democracies to be more responsive to political demands and provide more avenues for dissent, it should be kept in mind that the NAVCO data are limited to maximalist claims on the state. 76 Since such maximalist campaigns are uncommon in democracies in the first place, any cases that we actually observe in democracies are likely to be atypical cases where states are particularly reluctant to offer concessions, as in ethnic separatist claims that threaten an existing polity.

Erica Chenoweth and Maria
Stephan also find some evidence that longer nonviolent campaigns are less likely to obtain political gains. 77 Campaign duration may be correlated with terrorism as factions may adopt terrorist tactics when nonviolent campaigns have gone on for a long time without delivering the desired political outcomes. 78 We thus control for the duration of the dissident campaign, using the log of years elapsed since the beginning of the mass civil resistance mobilization for each campaign (after adding 1 to the base).
Larger mobilization increases the likelihood of success for civil resistance campaigns. 79 Terrorism is also plausibly associated with campaign size, since factions are more likely to become disillusioned with the effectiveness of nonviolence and turn to terrorism when a campaign remains small. 80 We thus include a measure from NAVCO indicating the order of magnitude of campaign sizes measured by the number of participants, ranging from 0: 1-999; 1: 1,000-9,999; 2: 10,000-99,999; 3: 100,000-499,999; 4 = 500,000-1 million; to 5 > 1 million. 81 Chenoweth and Stephan argue that nonviolent campaigns are more likely to be resilient in the face of repression than violent campaigns, 82 and repression is also believed to encourage resort to terrorism. 83 We thus control for repression against campaigns, using data from NAVCO 2.0, capturing "the most repressive episode or activity perpetrated by the state" against mass dissidents in a given year, on a four-point scale, ranging from no repression, to repression with the intention to kill. 84 Chenoweth and Stephan find that nonviolent campaigns have been more often successful after the Cold War. 85 We have a possible period difference for terrorism here as well, as terrorism has become more common by many estimates, especially after the mid-2000s. We thus add a post-Cold War dummy for the period after 1992.
Low income may be associated with grievances that can motivate both more organized nonviolent dissent and resort to terrorism. 86 We control for a country's GDP per capita (logged) using data from Gleditsch's dataset. 87 In order to test more formally that organized fringe terrorism constitutes a perceived threat of conflict escalation for governments in a way that less organized violence does not, we compare systematic terrorism and riots during campaigns in alternative analyses. Riots can also be contagious and have a potential mobilizing effect, which some argue promotes concessions. 88 We create a measure of Riots, using a binary measure based on information from the "Cross-national Time Series Data Archive," flagging if there is at least one riot or clash with the police of more than 100 citizens involving physical force. 89 It is possible that our findings reflect a more general effect of any type of violent action perpetrated by radical flanks, including conventional military strategies and guerrilla attacks. To examine this, we consider alternative analyses with a measure from NAVCO, indicating whether a "radical flank" using any type of violence is active at the same time as the non-violent campaign. 90 We present descriptive statistics of all variables in the Online Appendix. Multicollinearity is a possible concern, but neither the correlation matrix nor Variance Inflation Factor (VIF) tests show any specific danger.

Empirical results
In Table 3 (below) we report probit regressions estimates of the effects of the covariates on substantial gains. We report standard errors clustered by campaigns, since the variance may differ systematically across cases. We also control for time dependence, using the log of campaign time without prior substantial gain. 91 Before turning to the main features of interest, we comment on the control variables, based on the estimates in Model 1. In line with previous research, we find that larger civil resistance campaigns are more likely to see substantial political gains. We also find that repression appears to make substantial political gains less likely, and concessions are less likely during the Cold War period. Mass civil resistance campaigns in democratic states also appear to be less likely to see political gains, but the coefficient estimates are not consistently significant. We stress again that maximalist nonviolent Robust standard errors in parentheses ***p < 0.01, **p < 0.05, *p < 0.1.
campaigns are unlikely in the first place in democratic systems, precisely because these provide greater possibilities for dissent to be pursued through regular political channels. The bulk of the maximalist campaigns take place in non-democracies, which prevent meaningful opportunities for voicing political demands. 92 A longer duration appears to increase the likelihood that mass dissident campaigns will see political gains. This is consistent with the idea that more capable moderate organizations that can sustain nonviolent discipline throughout a campaign are more likely to see success. Finally, a country's GDP per capita does not seem to have a significant effect on substantial political gains. In Model 1 we consider the likelihood of substantial political gains by systematic terrorism occurrence and campaign hierarchical structure individually, with the control variables. We find some evidence of a modestly positive and weakly significant effect of terrorism occurrence on substantial political gains, and little evidence that hierarchical campaign structure by itself has any consistent effect on the likelihood of gains. In Model 2 we introduce the interactive term implied by our argument. As can be seen, the small individual coefficient for terrorism suggests that fringe terrorism in the absence of a hierarchical campaign does not increase prospects for campaign success. However, the large estimated positive interactive term for terrorism and hierarchical campaign structure is consistent with the idea that fringe violence can spur crises resolved to moderate advantage, if the moderate factions have credible prospects of preventing escalation.
In Figure 1 we plot the predicted probabilities of success for four profiles on the key variables with 90% confidence intervals, keeping other values at the median. We can see that campaigns with a hierarchical organization have a considerably higher likelihood of substantial gains in the presence of fringe terrorism, and that neither hierarchical structure nor terrorism notably increase prospects for concessions.
In Models 3 and 4 we consider whether effects are specific to terrorism, using the radical flanks indicator from the NAVCO data. We find no evidence that the radical flank indicator generates similar results, either in having a significant positive influence by itself (Model 3) or significant interaction with hierarchical campaign structure (Model 4). Finally, Models 5 and 6 include fringe violence in the form of riots. Again, the apparent effects of systematic fringe terrorism for hierarchical campaigns do not seem to generalize to disorganized violence such as riots without clear coordination or organization. Hence, the findings are consistent with the idea that only organized violence can constitute a credible threat of conflict escalation and induce a state to make concessions.
We have also conducted a number of additional robustness tests reported in the supplementary appendix. Our main findings do not change when clustering standard errors by countries rather than campaigns and remain also robust to alternative control variables and measures (see Appendix). The results are robust to an alternative more restrictive measure limited to full success; when estimating a multinomial logit models with graded measures of the success outcomes and when including anti-colonial campaigns (see Appendix). 93

Conclusion
Much of the existing research on fringe violence in civil resistance campaigns has looked for unconditional positive and negative effects of violent activities on political outcomes, irrespective of type of violence or the characteristics of the movement. We have identified specific conditions under which a particular type of low-level organized violence, namely terrorist attacks, can induce crises that are resolved to the advantage of moderate leaders with strong organizations. We have argued that terrorism can induce a plausible threat of escalation and encourage governments to concede to moderate organizations when these have more organized structures and can credibly prevent radicalization and escalation to more extensive violence. Fringe terrorism generates a credible threat of conflict escalation in a way that disorganized violence during mass civil resistance activities does not, as terrorism implies organized attempts to escalate the conflict that are a plausible precursor to major organized violence. Our empirical findings are consistent with this argument, and highlight the value of focusing on how specific types of fringe violence and the characteristics of campaigns can affect the strategic environment and incentives of the state.
Although we have highlighted a set of very specific conditions that can induce a credible threat of escalation and resulting advantages for stronger moderate groups, we have not examined the more general effects of terrorism, and the many predominantly negative consequences that fringe violence may have on nonviolent civil resistance movements. In particular, it is likely that fringe violence can undermine participation in nonviolent campaigns and alienate potential supporters, and possibly also undermine at the outset the emergence of precisely the type of strong organizations that could withstand the potential challenges from fringe violence at a later stage.
Our research suggests many potentially promising extensions to understand how features, tactics, and organizational structures can condition the prospects of short-term success and failure in civil resistance campaigns. For example, it may be possible to identify specific types of strategies movements use to retain support and prevent fringe violence. Plausibly, nonviolent movements emerging from membership-based organizations such as trade unions or traditional structures such as religious institutions may be better able to maintain support than movements emerging spontaneously from for example student activism, without prior organizational structures. Studying support dynamics and how this responds to the activities of movement leaders and the state is difficult with current data, which rarely provide over-time information on participation. However, some researchers have suggested ways to identify protest size or participation in specific cases. 94 It may be possible to use experimental methods to understand the impact of fringe violence and terrorism on individual decisions, or explore computational models of participation in protest.
Moreover, to understand responses to conflict dynamics and their short and long-run outcomes, it would be helpful to consider more explicit measures of success, identifying the actual political concessions as well as the distributional consequences for specific groups and segments. Tactics and competitions could affect the likelihood of specific political outcomes such as powersharing, or changes in political institutions such as leadership change or a transition process. For example, both the ability and willingness of nonviolent campaigns to accept powersharing agreements could be affected by whether groups face competition with violent fringe groups, and a dominant campaign that faces no challenges may be less likely to accept powersharing proposals and less willing to settle for smaller concessions, in ways that can ultimately undermine pluralism and diversity. 95