WHEN ETHNICS REBEL NONVIOLENTLY

(1)

WHEN ETHNICS REBEL NONVIOLENTLY

EVALUATING OPPORTUNITIES FOR CIVIL RESISTANCE IN

ETHNO-EXCLUSIVE REGIMES

MICAELA WANNEFORS

Master's Thesis

Spring 2020

Department of Peace and Conflict Research, Uppsala University

Supervisor: Corinne Bara

(2)

Acknowledgements

Thank you, Corinne, for your inspiration and excellent advice. Thank you also, Rickard and Emelie, for your encouragement and support, and to all who have shared this journey with me.

(3)

Abstract

This thesis seeks to understand which opportunities make excluded ethnic groups use nonviolence. The research question is: Under which conditions do politically-excluded ethnic groups initiate

nonviolent resistance rather than violent resistance? The study challenges an existing assumption

that ethnic exclusion creates opportunities for violence and hinders nonviolence, by exploring if constraints to violence favor nonviolence. I hypothesize that two opportunities from nonviolence theory – autocracy and mass participation – and two constraints to violence – the state having a powerful ally and high territorial outreach – constitute alternative pathways to nonviolence. I make a joint evaluation of these contrasting theoretical views with Qualitative Comparative Analysis (QCA), using a comprehensive dataset of 79 campaigns initiated by excluded ethnic groups. The theory evaluation is complemented by three informative within-case studies: the Lithuanian

Sajudis campaign (1988), the Druze resistance in Israel (1981) and the Malawian Anti-Banda

campaign (1992). I find that opportunities from both theoretical strands work in combination. One pathway – mass participation in a state with high outreach and no powerful ally – leads to nonviolence in almost 71% of the cases occurring in that setting, and explains roughly 63% of all nonviolent campaigns initiated by excluded groups.

(4)

LIST OF TABLES AND FIGURES

1. Figure 1. Model of hypothesis ... 18

2. Figure 2. Descriptive statistics ... 31

3. Table 1. Truth table for base model ... 34

4. Table 2. QCA solution for nonviolent-campaign onset ... 35

5. Table 3. Truth table for violent-campaign onset ... 37

6. Table 4. QCA solution for violent-campaign onset ... 37

7. Table 5. Intersection of theory (T) and solution term (S) with types of cases ... 46

8. Table 6. Cases of politically-excluded groups initiating nonviolent or violent campaign ... 63

9. Table 7. Complete truth table from sufficiency analysis for nonviolent-campaign onset, with covered cases ... 65

10. Table 8. Results from robustness tests (alternative operationalizations) ... 67

(7)

1

1. Introduction

In 1970, Ted Robert Gurr published his landmark study Why Men Rebel, which became the starting point of a whole field of research focusing on causes of rebellion and civil war, particularly in ethnic settings. For a half-century, the focus has been on actual cases of armed conflict, despite such instances being relatively rare compared to all those contexts where groups suffer injustices and resentment but still do not take up arms. In civil-conflict research – as in peace and conflict research generally – studies of violence have dominated the field, at the expense of “non-cases”. Bartusevičius and Gleditsch observe that “civil conflict research has traditionally focused on armed conflicts and paid little attention to incompatibilities that do not see violence… grievances can also generate nonviolent responses, and mobilization for armed action may fail” (2019: 226). As an alternative to violence, nonviolence can also be highly contentious (Sørensen and Johansen 2016); in this regard, ethnic conflict should be no exception.

Against this backdrop, two research gaps motivate my study. First, explanations to ethnic conflict generally focus on grievances or the interplay between risk factors. My study delves into opportunities for ethnic conflict. Second, nonviolence theories are rarely applied to ethnically-divided societies. This may be because ethnic groups are sometimes assumed to face contextual constraints to nonviolence. For example, Thurber claims that the structures that favor violent uprising in ethno-exclusive regimes also disfavor nonviolent resistance, and finds that nonviolent campaigns are less likely than violent campaigns to involve excluded ethnic groups or ethnic claims (2018, 255-259). The assumption that ethnic exclusion creates opportunities for violence and hinders nonviolence often remains untested, however, while begging the question whether the reverse of Thurber's logic holds: do constrained opportunities for violence favor nonviolence? The ideas about excluded groups being conditioned in their choice between nonviolence and violence clearly merit a more fine-grained examination. The lacking knowledge about exactly what opportunities encourage nonviolent instead of violent tactics in ethnic contention forms the second motivation for this study.

The purpose of this thesis is to understand which opportunity conditions enable ethnically-excluded groups to use nonviolence. To the best of my knowledge, no study has yet determined whether any specific opportunity structure is the primary cause of ethnic nonviolence, or if this phenomenon is the result of a combination of opportunities. This is the goal of my examination, which is guided by the following research question: Under which conditions do

(8)

2

I will address this research problem by departing from a context where grievances are already assumed, and evaluate theories about opportunities for nonviolence and constraints to violence. The aim is to clarify whether excluded ethnic groups engage in nonviolence because such opportunities are in place or because they lack opportunities to use violence. Put differently: do excluded groups protest because they can protest, or because they cannot fight? These lines of thought are normally dealt with in isolation, with one theoretical strand explaining onset of nonviolent resistance and the other examining causes of violent conflict (or its absence). Here, these perspectives are brought together and jointly evaluated in the context of ethnic nonviolence. In this way, my study builds on, bridges and hopefully contributes to both fields.

I hypothesize that excluded groups opt for nonviolence when they either have opportunities for nonviolence or constrained opportunities for violence. I explore two opportunities for nonviolence – autocracy and mass participation – and two constraints to violence – the state having a powerful ally and high territorial outreach. Autocracy sets the stage for nonviolence under political exclusion by restricting more conventional tactics and creating a context favorable to broad-based mobilization. Mass participation offers the ability to increase costs for governments and collectively erode the institutions upholding the state, especially if the excluded group engages in cross-ethnic mobilization. A powerful ally on the government's side makes violent insurgency too risky, but is a suitable external audience for nonviolent resistance. State outreach limits ethnic rebels' ability to organize militarily and makes the state institutions more accessible to target nonviolently.

(9)

3

Lithuanian Sajudis campaign (1988), assess a possible theory extension in the Druze resistance (Israel 1981) and explore alternative explanations for the Malawian Anti-Banda campaign (1992).

The short answer to whether excluded ethnic groups protest because they can protest or because they cannot fight is: both. My key finding is that one single pathway – mass participation in a state with high outreach and no powerful ally – leads to nonviolent resistance instead of violent resistance in almost 71% of the campaigns occurring in that context. This path alone explains roughly 63% of all nonviolent-campaign onsets among ethnically-excluded groups. It combines one opportunity for nonviolence, one opportunity for violence and one constraint to violence, thereby indicating that opportunities from both theoretical strands work in tandem.

(10)

4

2. Theory

2.1 Previous research

Given the abundance and breadth of literatures asking and answering “why men rebel”, which constitute the broader scholarly tradition of my thesis, I will focus on recent research dealing specifically with opportunities for nonviolence and constraints to violence. The civil-war and nonviolence literatures can be categorized based on whether they concern ethnic conflict, since ethnicity may have specific implications for how opportunities unfold. Below, I discuss opportunities along ethnic and non-ethnic lines, to highlight what is (un)known about these structures in these different domains. While the ethnic subfield's narrower analytical focus speaks more to my topic, several aspects in the general civil-conflict research also appear in ethnic contention and thus broaden the perspective on opportunities.

Much of the relevant research relates to the contentious-politics literature. Contentious politics is commonly situated between collective action, politics and contention, where intra- and extra-institutional actions combine. It relates closely to protests and social movements and often regards political violence (Tilly and Tarrow 2015; Vráblíková 2017; Tarrow and Tilly 2009; Hutter 2014; Nulman and Schlembach 2018). Its framework is relevant as it recognizes the whole scope of inaction/action as a continuum, where conventional politics, nonviolence and violence constitute repertoires with differing intensity and characteristics in “the total range of options available to people engaged in conflict” (McCarthy 1990, 108). It thus comes close to how contentious behavior unfolds in reality. Yet, I will not primarily relate to this continuum, since the groups of interest here are both excluded and do resist violently or nonviolently, meaning that they are already outside the low-intensity/institutional domain. While the blurred lines between modes of action capture dynamics of ongoing contentious processes, I contrast factors preventing violence to factors favoring nonviolence in the strategic choice per se. Therefore, I specifically focus on theories distinguishing violence and nonviolence as two distinct, separate choices, while referring to contentious politics when relevant and reflecting upon this theoretical and methodological choice in section 4.5.

Constrained opportunities for violence

(11)

5

Wimmer (2018) are two excellent examples of how this interaction can be studied using QCA, where the latter is particularly relevant here since it incorporates a causal story of why ethnic violent conflict does not take place even when risk factors are present.

Lindemann and Wimmer examine three grievance conditions and two opportunity conditions to identify pathways leading excluded groups to civil war. The opportunity conditions are limited territorial outreach of the state's security apparatus and/or the ruling party, and external support involving financial/military resources or external sanctuary (2018, 309f). Their results show that only 27 of 92 of their cases (25 of 58 ethnically-excluded groups) experience armed-conflict onset (over 25 fatalities). Two causal pathways lead to violence: indiscriminate state violence and low territorial outreach; and socio-economic marginalization, indiscriminate repression and external support. They also identify one single temporal sequence producing ethnic war – limited repressive capacity of the state, indiscriminate state violence and external sanctuary – which they interpret as a downward spiral, where escalation spins off only if any of the opportunity factors is present (2018, 306, 313, 315-317). No condition is found necessary for ethnic war (ibid, 313), suggesting both that conflict contexts are more complex than individual conditions causing violence to erupt and that risk factors may lead to other outcomes than violence, which appears to happen exactly when opportunity factors are constrained. The

absence of Lindemann and Wimmer's theorized opportunities could thus be conceptualized as

constraints to violence that stall their escalation spiral.

Importantly for this thesis, Lindemann and Wimmer's appendix includes an analysis of “ethnic peace” (the absence of their outcome), where the pathways leading to peace involve both high outreach and absence of external support, while not mirroring the paths to ethnic war. This is an example of the asymmetric causation assumed in QCA, where “the conditions explaining the occurrence of an outcome can differ from those explaining its nonoccurrence” (Thomann and Maggetti 2017, 5; see section 3.2).

(12)

6

External pressure has also been researched in the broader civil-conflict literature, with recent theories on how external influence can simultaneously constrain violence and be an opportunity for nonviolence. Cunningham (2016) and Breslawski and Cunningham (2019) study the effect of external influence on the likelihood of using nonviolent and violent tactics. Cunningham (2016) examines the threat of large-scale external intervention in states with hierarchical links to the United States. This threat is found to deter insurgents from rebelling violently, which according to the author also allows for state repression, triggering other forms of dissent:

The empirical analyses show that states that receive US protection are more repressive toward their populations. This repression, in turn, makes these states prone to other forms of dissent – such as nonviolent campaigns and terrorism – that are less likely to provoke US intervention (2016, 337, italics added).

Yet, as the statistical analysis does not show the temporal order, Cunningham asserts rather than demonstrates this causal story. Breslawski and Cunningham (2019) statistically test the effect of external influence on violent and nonviolent contention in African elections. They find that military aid from the US or being a former French colony has a direct positive effect on lower levels of electoral violence, suggesting that this constraint is because rebels are aware of this influence and adapt their cost calculus accordingly, while the effect on nonviolence is statistically insignificant. Membership in highly-structured intergovernmental organizations (HSIGOs) in turn makes both nonviolent and violent contention more likely (ibid, 639f, 645, 655, 658). Other research on HSIGOs indicate a significant capacity to prevent intra-state violence by raising costs, e.g. by pressuring member states not to escalate conflict (Tir and Karreth 2018; Karreth and Tir 2012).

Opportunities for nonviolence

(13)

7

non-democracies. Civil war is instead more likely when groups are larger, have ethnic kin in neighboring countries, face internal fragmentation and act in states with lower economic development. Mirroring Lindemann and Wimmer's findings (2018), Cunningham shows that state outreach has restrictive effect on secessionist civil war, though no significant effect on nonviolent campaigns (2013, 295).

Cebotari and Vink examine traditional factors to explain nonviolent protest mobilization among 29 ethnic minorities in Europe with similar risk of discrimination. Their conditions are level of democracy, degree of national ethnic fractionalization, degree of political discrimination, territorial concentration and sense of national pride among members of the minority group (2013, 300f, 304). The authors seek to explain both strong and weak protest, motivating their use of fuzzy-set QCA, which assigns levels to variables. Territorial concentration and national pride are found to be necessary conditions for strong and weak protests, while the absence of strong national pride is sufficient for strong protest (ibid, 298, 310f, 313). Their opportunity condition, level of democracy, only explained ethnic-minority protest when combined with other conditions, reaffirming Lindemann and Wimmer's (2018) findings.

Turning to ethnic barriers to nonviolence, extant theories suggest they arise in relation to two key concepts in nonviolence strategy: mass mobilization and the movement's ability to penetrate the state's “pillars of support” – the societal institutions upon which the state rests – so that they eventually erode (Sharp 1973). In Svensson and Lindgren's words, “unarmed insurrections build on the logic that all power ultimately rests on the consent of the people and that withdrawing this consent basically implies challenging the legitimacy of the state” (2011, 98).

Thurber presents “a theory for how ethnic structures and practices of political exclusion shape opportunities for civil resistance” (2018, 256). He posits that politically-excluded ethnic minorities face constraints to key mechanisms of nonviolent resistance, even without ethnic claims. First, these groups may not have the broader support among other groups necessary to make repression backfire. Secondly, given their political exclusion, they likely do not have the connections necessary to achieve loyalty shifts and defections within the political and security elites (ibid, 255-259).1 His hypotheses are corroborated as he finds a statistically significant

1_{In relation to nonviolence success in non-democracies, Svensson and Lindgren (2011) find similar ethnic}

(14)

8

negative relationship between ethnic groups' political exclusion and the likelihood of initiating a nonviolent campaign and twice the likelihood of violent-campaign onset, even across group sizes. Yet, the relationship between political status and violent-campaign onset is weaker than in other studies examining only violent conflicts (and even statistically insignificant in his base model) (ibid, 263-265). This detail indicates that we are not dealing with mirror images, and puts into perspective the ground logic that the same structures that favor ethnic violent uprising inhibit nonviolent uprising. We simply cannot know whether the groups that did not initiate violent campaigns were limited in their use of violence and/or opted for nonviolence.

In a statistical analysis of extraordinary events during protests in the Ethiopian Oromia region, Arriola (2013) finds a negative effect of ethnic heterogeneity and intra-ethnic divisions on the likelihood of both protest onset and protest violence, which he attributes to two mechanisms. First, ethnic heterogeneity neutralizes personal motivations for protest (such as grievances or available resources) by influencing the expectations on mass participation and the individual risk of being targeted by repression (2013, 148-150, 156f). Secondly, he finds that in-group policing reduces protest violence since co-ethnics have information on political preferences, whereas out-group policing lacks such information and thus fails to target repression (ibid, 151, 163f). According to this logic, nonviolence is more likely in ethnically homogenous societies, as is protest violence if policing is carried out by out-groups.

(15)

9

Abbs' (2019) stands out by challenging both the use of traditional violent-conflict conditions and the notion of ethnic barriers to nonviolent mobilization. He argues that cross-cutting grievances “ease horizontal mobilization in ethno-politically exclusive environments, and clear short-term incentives to join in antigovernment protests, which fosters vertical mobilization” (ibid, 5, italics in original). He uses spikes in food prices at subnational level as such a “symbolic issue”, and in line with his hypothesized causal mechanism finds statistically significant effects on the emergence of nonviolent action in sites with political exclusion and ethnic diversity in African states. As Thurber, he also finds that excluded groups are less likely to protest than included groups (ibid, 1, 10, 12f). While focusing on a conflict issue of minor nature, he points to a key mechanism for nonviolent resistance: the opportunity to mobilize across social segments when there are broad-based grievances.

Leaving the exclusively ethnic domain, White et al. (2015) identify resource mobilization as an opportunity for collective action in governmental disputes. They suggest that actors counting with mass-mobilization capacity are more likely to opt for nonviolence, while focused military capacity favors the choice of violence, and provide evidence from the former Soviet Union (2015, 472, 474, 486). In a theoretically and methodologically impressive study, D. E. Cunningham et al. (2017) divide the mobilization process in governmental disputes into two stages – claims and contentious outcomes (civil war or nonviolent campaigns) – and offer a complex factor interplay based on nonviolence and civil-war theories. As opportunity factors for nonviolent-campaign onset, they consider regime type, economic development, urban population, social networks, and transnational support. For civil war, they consider rural population, low economic development, ongoing territorial conflict and transnational support. Similar to K. G. Cunningham (2013, 299), they most crucially find that both autocracies and anocracies are more likely than democracies to see claims and nonviolent campaigns (conditional on claims). Regime type is found insignificant for civil-war onset and incidence, and they attribute anocracies' proneness to civil war to the higher likelihood of claims (2017, 468, 477f). Interestingly, their results return scarce support for economic development (a proxy for state outreach), ongoing territorial conflict and demography.

(16)

10

collective protest in the absence of structural conditions favorable to collective action” (ibid, 87). This causal mechanism is crucial, as it involves a constraint to violence encouraging nonviolence even without structures favorable to nonviolence. As Breslawski and Cunningham, they find evidence that threats of economic sanctions instigate anti-government protests, especially multi-sender threats, and include case-study evidence of the causal mechanism from Zimbabwe (ibid, 87, 93).

Summarizing, research on ethnic and non-ethnic conflicts largely coincide in identifying state territorial outreach and external influence as constraints to violence, where the former appears more in ethnic-conflict research and the latter receives some support as opportunity for nonviolence from the broader civil-conflict literature. As for nonviolence theories, there has been ambiguous evidence on traditional opportunity factors such as transnational support, economic development, democracy, group size and territorial concentration, leading both to and away from nonviolence. Theories on ethnic nonviolence have until recently focused on barriers, while new research highlights opportunities arising from cross-cutting societal issues. These recent contributions feed into my own theory.

2.2 Key concepts and definitions

Before relating my causal story, I review key concepts in this thesis. To facilitate connections between my study and previous research, I adhere to well-established definitions from theories on ethnic conflict and nonviolence.

Beginning with ethnicity, I define this concept as a subjectively-identified commonality based on shared ancestral or cultural features, such as language, religion or ethno-somatic aspects (Vogt et al. 2015). An ethnic group is a community self-identifying in such commonalities, and

ethnic claims as demands with bearing on that community. To define ethnic political exclusion,

I use the categories “discriminated”, “powerless” and “self-exclusion” in the Ethnic Power

Relations (EPR) Core Dataset (ibid), which provides data on ethnic groups' political relevance,

share of population and access to state power in countries where ethnicity is politicized, thereby aligning with Thurber (2018, 260f). Likewise, an ethno-exclusive regime is characterized by state-driven or sanctioned exclusion of specific groups from the political arena.

(17)

11

limited to the nation's borders, where at least one of two parties is the government. Key to this definition is the incompatibility, which I understand as generally incompatible positions stated in the form of claims (Cunningham et al. 2017, 474). Crucial to this thesis, I do not see ethnic

conflict as both involving an organization that claims to represent an ethnic group and recruiting

predominantly from this group (Bara 2014, 700). While many cases meet both criteria, the latter would go against a central logic of nonviolence: the broader attraction of participants (Chenoweth and Stephan 2011). Likewise, I do not define ethnic mobilization as driven by ethnic claims but that an ethnic group has the agency to initiate such mobilization.

A campaign is “a series of observable, continual tactics in pursuit of a political objective” that is maximalist (challenging government authority in terms of regime change, secession, or anti-occupation), with discernable leadership and oftentimes names (Chenoweth and Stephan 2011, 16; Thurber 2018, 259). Campaigns are defined as violent or nonviolent based on the nature of the tactics predominantly used (Chenoweth and Shay 2019b, 5f; Svensson and Lindgren 2011, 106). Here, I rely on distinctions made by Chenoweth and Stephan in their seminal work Why

Civil Resistance Works (2011) that are widely accepted and used ever since. A violent campaign

is characterized by the primary use of violent means implicitly intended to harm the physical wellbeing of the opponent (Chenoweth and Shay 2019b, 5). Armed state-based conflicts often imply that violence results in 25 fatalities per calendar-year. While violent-campaign onset does not necessarily entail fatalities, armed conflicts meeting the UCDP criteria are categorized as campaigns in the Nonviolent and Violent Campaigns and Outcomes (NAVCO) 2.1 dataset (Chenoweth and Shay 2019b; 2019a). Nonviolent campaigns primarily involve extra-institutional actions defined by non-use of violence. The repertoire of nonviolent tactics is extensive, ranging from persuasion (e.g. protests) to non-cooperation (such as labor strikes) and nonviolent interventions (e.g. sit-ins) (Svensson and Lindgren 2011, 100; Sharp 1973).

(18)

12

established scope to obtain the desired outcomes given specific configurations of existing conditions at a given time” (2013, 299). Finally, in parallel to considering whether constraints to violence trigger nonviolence, I have to establish whether my conditions also trigger violent campaigns; if so, they cannot ultimately determine the variation.

2.3 Causal story and hypothesis

Below, I outline the specific conditions I will examine and the causal story underlying each of them, based on which I formulate my hypothesis. First, I recapitulate my departure point and make some general comments about my selection of conditions. Thurber (2018, 268f) and Rørbæk (2019, 489) suggest that ethnic barriers to nonviolent resistance may explain why ethnic groups choose violence although nonviolence should be the preferred option. Assuming like Thurber that ethnic groups “engage in a rational or quasi-rational decision-making process” (2018, 257), I examine which opportunity structures ethnic groups without political influence rely on when initiating nonviolence.

I evaluate one structural condition and one contextual/agency condition from each strand. From nonviolence theory, I evaluate the structural condition of autocracy and the contextual condition of mass participation. As constraints to violence, I use state outreach as the structural condition, and the state having a powerful ally as the contextual condition. The conditions have mutual implications that I describe below – this arrangement will be useful when exploring how the conditions interact. I have purposefully disregarded opportunity structures that either have been found to, or theoretically could, influence the strategic choice in both violent and nonviolent direction. Such conditions include group size, territorial concentration, state membership in HGISOs and external support to groups (Abbs 2019; Bara 2014; Cebotari and Vink 2013; D. E. Cunningham et al. 2017; K. G. Cunningham 2013; McLauchlin 2018; Lindemann and Wimmer 2018; Thurber 2018; White et al. 2015; Breslawski and Cunningham 2019; Rørbæk 2019).

The constraints to violence correspond to possibly existing opportunities negated to the groups. These conditions not only have to be relevant for constraining violent uprising, but should also be a theoretically plausible explanation for nonviolent uprising. Based on theoretical propositions from previous research, I chose one constrained opportunity for violence (powerful

ally) and one absent opportunity for violence (state outreach), and develop causal stories for

(19)

13

When establishing opportunities for nonviolence, I considered three “comparative advantages” that nonviolence literature has found conducive for relatively higher rates of success in resistance campaigns: “(1) nonviolent movements create higher costs of repression; (2) they can win over defections from elites within the regime; and (3) they generate higher rates of participation” (Thurber 2018, 257). The third feature, mass participation, qualifies for evaluation since it is a key mechanism for nonviolence, while not being necessary for violent insurgency, given that these tactical choices rely on distinct repertoires to impose costs and achieve government concessions. One of the clearest findings in nonviolence literature is that nonviolence requires mass participation to effectively put pressure on governments. By contrast, violent rebellion does not depend on mass participation, as it primarily relies on military capability and violent means to induce costs. White et al. (2015, 472, 474, 486) and D. E. Cunningham et al. (2017, 472, 480) provide theoretical arguments and evidence that groups anticipating mass participation are more likely to use nonviolence, whereas focused military capacity and organization increase the likelihood of violent insurgency.

The other two features would however be problematic to include for evaluation. Higher costs of repression would be difficult to classify as either opportunity for nonviolence or constraint to violence, while being visible only after campaign onset and also depending on other factors relating to movement quality. This feature is instead considered as an implication of the state having a powerful ally. Loyalty shifts, in turn, should be less likely for politically-excluded groups, as described below in relation to my first condition, autocracy.

Opportunity for nonviolence. Argument 1: Autocracy

The structural opportunity for nonviolence concerns the state being an autocratic regime, based on D. E. Cunningham et al.'s finding that autocracies and anocracies are more prone to both claims and nonviolent resistance (2017, 477f). Thurber's study backs the idea that nonviolent contention becomes more likely in autocratic regimes (2018, 265), while Cebotari and Vink found democracy to favor nonviolence in ethnic settings (2013, 302). Despite the ambiguous empirical basis, several reasons warrant the evaluation of autocracy in ethno-exclusive settings. First, political exclusion of ethnic groups is arguably more likely to occur in autocratic regimes, and according to McCarthy, nonviolence becomes a more relevant strategic choice under political exclusion:

(20)

14

nonviolent struggle as the choice of those members of a society who are potentially strong but currently excluded from exercising power (1990, 107, 110).

Nonviolence should thus be attractive to excluded ethnic groups, especially if they can become strong by mobilizing widely. Secondly, the nature of anocracy enables power struggles, where excluded ethnic groups could also have a stake, or even rise into power if conditions were right. According to Thurber's (2018) logic, such conditions should not be equally present in ethno-exclusive regimes, though this proposition remains to be tested. In the context of politically unstable anocracies, ethnic groups could also have more elite connections, at least periodically, if their group representatives become involved in the power struggles. Elite links make loyalty shifts – another key mechanism for nonviolence – more likely. By contrast, an autocratic regime probably maintains ethnic exclusion more efficiently, which would theoretically make it more difficult to achieve loyalty shifts in favor of excluded groups. On the other hand, an opportunity for nonviolence may present itself if other social groups also have (similar or other) grievances against the autocratic regime, and thereby have reason to join a movement that voices claims (Cunningham et al. 2017). Finally, D. E. Cunningham et al. found that "anocracy provides an opportunity structure for initial mobilization that might become violent or nonviolent" (ibid, 478), suggesting that anocracy is more likely than autocracy to be an ambiguous condition.

In sum, I expect autocracy to favor nonviolence by restricting more conventional (and violent) tactics for the politically-excluded groups and creating a setting where their movement could involve broader dissent. Autocracy has implications for mass participation if the broader society is affected by the autocratic practices, and autocratic states should have relatively high outreach, since they are characterized by one-party rule and military dominance. To not completely disregard anocracy, it will be considered at later stages of the empirical investigation.

Opportunity for nonviolence. Argument 2: Mass participation

(21)

15

whether campaigns include ethnic groups or ethno-political demands, which does not perfectly equal ethnic groups using violence or nonviolence. It may simply be that certain demands are more effectively claimed through certain campaign types.

Based on my assumption that ethnic groups do not need to raise ethnic claims, I argue that cross-cutting claims can unify ethnicities in nonviolent resistance. As Abbs contends, “while large groups facilitate ethnic-based protest, this is unlikely to facilitate mobilization across ethnic lines to engage in intergroup action” (2019, 10). My theoretical scope extends to excluded groups raising cross-cutting claims to attract participants more widely. There are some empirical findings that excluded groups raising non-ethnic claims may still be mistrusted by outgroup members, producing commitment problems that constrain broad mobilization and instead fosters in-group cohesion prolonging violent conflict (McLauchlin 2018). Yet, there is scant systematic knowledge about the implications surrounding broad-base mobilization by ethnic groups at the onset of nonviolent resistance. Building on Abbs' (2019) finding that minor issues can give rise to a short-term opportunity of mobilizing across social segments, I extend his argument and posit that groups may unify under major issues, such as maximalist claims. This scenario deviates from Arriola's theory regarding ethnic heterogeneity:

uncertainty over the extent to which anti-regime preferences were shared locally, across ethnic groups as well as within the ethnic majority… individuals holding antigovernment sentiments could interpret a larger number of local cleavages as representing greater variation in protest thresholds (2013, 148, 150).

Arriola's reasoning, which he also found empirical support for, includes an implicit assumption that the specific grievances or demands must be shared. Instead, I suggest that excluded groups can gather several claims when initiating nonviolent campaigns. Indeed, a range of claims could constitute a common ground for overarching, maximalist claims to be voiced and launched. Here, specific ethnic grievances may or may not be part of the maximalist umbrella. Such a scenario is thinkable not least in an autocratic regime. I thus identify cross-ethnic mobilization as causal mechanism for mass participation.

Constrained opportunity for violence. Argument 1: Powerful ally

(22)

16

part of a movement's leverage to erode the state's pillars of support (Svensson and Lindgren 2011, 101).

The causal story goes as follows: when states count with backing from powerful allies, excluded groups can no longer anticipate that use of violence would cause serious costs for the state (K. G. Cunningham 2013, 294) and they also risk serious costs themselves if their violence triggers external intervention (Beardsley, Cunningham, and White 2017). From here, previous studies suggest two possible scenarios. The first links back to Cunningham's (2016) suggestion that threats of external intervention constraining violent uprising facilitate repression and trigger nonviolence. An alternative scenario arises from the ethnic group's cost calculus, whose premises are altered if considering nonviolence instead of violence: if the group rebelled nonviolently and was met by repression, that cost would be relatively lower than an external intervention, while the state's ally would be in a harder position to support a state repressing part of its population on the basis of nonviolent dissent. This pathway thus connects to the higher costs of repression imposed from nonviolence (Thurber 2018). Breslawski and Cunningham (2019) tested but did not find support for this idea, while Grauvogel, Licht and von Soest (2017) found increased likelihood of protest.

Both pathways involve repression. In the first, repression enabled by hampered violent uprising triggers nonviolence. In the second, the group anticipates that a repressive response to nonviolence would backfire if triggering international pressure or reprisals to the state. Key to both is that nonviolent resistance could take place even if opportunities for nonviolence were absent, because of the excluded group's expectations about external engagement.

Constrained opportunity for violence. Argument 2: State outreach

The final condition is a structural opportunity factor, limited state outreach, being absent. State outreach is usually defined as territorial outreach and/or presence of the ruling party in remote or rural areas (Lindemann and Wimmer 2018; Cunningham et al. 2017). The inclusion of an absent opportunity for violence enables an evaluation of asymmetric causation: limited state outreach may explain ethnic violence, but unlimited state outreach may not explain absence of violence or, for that matter, nonviolence.

(23)

17

state outreach to equally restrict nonviolence, despite some studies finding nonviolence to be more common in urban environments and discussing challenges to nonviolence in rural areas (Abbs 2019, 7, 9f, 14; Cebotari and Vink 2013, 303, 308, 312; Cunningham et al. 2017, 472f; White et al. 2015, 478, 486). I argue that the infrastructure being established under higher state outreach would logically be an appropriate target for nonviolence. Studies comparing the choice of violent and nonviolent tactics have relatedly referred to nonviolence being more feasible in developed countries with a more advanced division of labor (Cunningham et al. 2017, 472f, 480; White et al. 2015, 478). I simply argue that greater state outreach constitutes a platform where nonviolence makes more sense.

First, the state's pillars of support of actually present, making it more difficult to ignore voices of dissent. Secondly, as states increase their outreach, they probably have higher capacity to rely on state institutions, e.g. taxation. In that context, citizen noncooperation or withdrawal of consent can become more damaging (White et al. 2015, 478). If so, nonviolence targeting exactly those pillars and structures should be perceived as more useful, favoring nonviolence rather than violence. Thirdly, higher state outreach may bring better infrastructure and service provision, including technology and other means of communication, that could facilitate mass participation and be used to effectively organize mass rallies (as during the Arab Spring). I thus expect state outreach to expose the state's pillars of support as an easier target for nonviolence.

Construction of hypothesis

I will evaluate the four conditions using QCA, which examines combinations of conditions.2 Just as regression analysis per se does not say anything about the theoretical value or feasibility of the statistical results, QCA is insensitive the “theoretical belonging” of each argument (nonviolence vs. violent-conflict theory). Instead, QCA analyzes conditions based on an expectation that individual conditions exert their effect in the presence of other conditions. It is exactly this combination of opportunities that will be put to test in my study. To reflect this conjunctural causality typical for QCA (further explained in section 3.2), my four conditions are articulated in the same hypothesis. To still capture the contrasting theoretical expectations – conditions are either right for nonviolence, or not right for violence – I divide the conditions into two separate paths. My hypothesis then spells out as follows:

2_{I use the terms combination, conjunction, path and pattern interchangeably, always referring to the combination}

(24)

18

H1: Politically-excluded ethnic groups initiate nonviolent resistance instead of violent resistance when the state is autocratic and the groups are able to mobilize large masses, or when the state has a powerful state ally and high territorial outreach.

This is the “extreme” hypothesis, in the sense of being the most exact and including the fullest theoretical expectations, as visualized in Figure 1. In reality, the two opportunities from each category do not need to combine – the constraints to violence may substitute each other, as may the opportunities for nonviolence. I also expect combinations across the two theoretical strands. For instance, an autocratic state with high outreach could be a suitable target for broad-based protests across a region or country, or state outreach can provide better infrastructure for mass mobilization. The valid conjunction(s), if any, will be demonstrated in the empirical analysis. In the research design, I go into depth on how evidence from QCA will confirm or reject my hypothesis, since it does not involve standard statistical methods, while the procedure is demonstrated in detail in the empirical chapter.

Figure 1. Model of hypothesis

(25)

19

3. Research Design

In this chapter, I present the methodology used to understand under what conditions ethnically-excluded groups initiate nonviolent rather than violent resistance. First, I describe my case selection and operational definitions of variables. In connection to the operationalization, I discuss choice of data, sources and measurements. I also explain how I address pitfalls and alternative explanations with robustness tests and control variables. Finally, I introduce QCA as a comparative method, which resembles regression analysis in how it presents causal relations but specifically analyzes combinations of conditions, and outline what I will be looking for in the analysis in order to evaluate my theoretical expectations.

3.1 Case selection, operationalization and data Case selection

My population is politically-excluded groups initiating nonviolent or violent campaigns (irrespective of size); this is also my unit of analysis. The variation in my study is observed in the type of campaign excluded groups initiate. I intentionally focus on excluded groups because they represent the vast majority of ethnic conflicts and are more likely to start violent campaigns than other groups (Lindemann and Wimmer 2018, 307; Thurber 2018, 257). Since the empirical evidence speaks against these groups initiating nonviolent campaigns, it is important to clarify under what conditions they do.

(26)

20

claims and are somewhat organized, since they are politically discriminated – a key grievance factor in the literature – and have arguably succeeded in turning incentives into mobilization to initiate campaigns significant enough to be included in the NAVCO dataset.

Thurber's (2018) EGC dataset on ethnically-excluded groups' participation in and initiation of nonviolent or violent campaigns 1948-2006 is ideal for the purposes of my study. Yet, there are now newer versions of the EPR and NAVCO datasets that Thurber (and Rørbæk) used – the EPR Core Dataset (2019) includes data until 2017 and NAVCO 2.1 (2019) until 2013. Following Thurber’s conceptual and operational logic, I update his dataset with new campaigns from the period 2007-2013. With this enlarged temporal domain, I can include a higher number of relevant cases, though 2014-2019 is unfortunately still excluded.

Practically, I select cases as follows. First, I combine variables from Thurber’s EGC dataset and the replication data for his article “Ethnic Barriers to Civil Resistance” (EBCR), and make sample checks against NAVCO 2.1 and EPR 2019 to ensure that they correspond. I then merge Thurber's data with selected variables from NAVCO 2.1 in the campaign-onset year, to find out which campaigns in NAVCO are not covered by Thurber. When variables mismatch, I have trusted the updated NAVCO 2.1 data, since Thurber's dataset is based on an older version of NAVCO, and triangulated with secondary sources. Using the new EPR data, I confirm that 51 new campaigns took place in a country with at least one excluded group in the year of campaign onset, which therefore constitute plausibly relevant cases.

(27)

21

Despite using more recent datasets, no campaigns 2006-2013 met my selection criteria, meaning that my sample does not cover the Arab Spring and other campaigns that probably relied more on social media and other technologies. It is intriguing that no more recent campaigns were initiated by ethnically-excluded groups; while this observation is not further explored within the scope of my study, I reflect upon its implications for theory in section 4.5.

Operationalization of variables

Here, I describe my operationalization of the dependent and independent variables, and elaborate on alternative operationalizations for robustness tests. Both the opportunities for nonviolence and the constraints to violence are operationalized in terms of their presence. QCA enables an analysis that also examines the absence of the theorized conditions, so while all possibilities are scrutinized in the analysis, the focus lies on their expected presence. For example, a powerful ally constraining violence is absent when no such powerful ally exists. In case of missing data, I have consulted secondary sources and manually added the value.

Dependent variable

My dependent variable is dichotomously defined as a nonviolent or violent campaign. Since my DV is binary, I use crisp-set QCA where all variables are operationalized dichotomously, where 1 means present and 0 absent (Schneider and Wagemann 2012, 277). Theory evaluation in QCA is more straightforward with binary variables, making the results easier and more practical to handle at the moment of interpretation. The operationalization into binary variables also requires me to be specific about the boundaries I choose to apply. My DV is obtained from the NAVCO 2.1 dataset's binary variable prim_method, indicating the primary type of resistance method used in a campaign year: 1 for primarily nonviolent campaign and 0 for primarily violent campaign. The reasons for comparing nonviolent campaigns to violent campaigns instead of no campaign have been outlined in section 2.2.

Autocracy

For autocracy, I use Polity IV’s ordinal polity2 variable, ranging from -10 to +10. I use the Polity IV’s thresholds, where -10 to -6 correspond to autocracy, -5 to +5 (and the special values -66, -77 and -88) equals anocracy, and democracy ranges from +6 to +10 (Marshall 2019).3_I am also informed by Basedau and Richter’s treatment of the same polity2 indicator to define democracy in a QCA study on oil dependence and internal-conflict onset: they use the mean of

3_{Cunningham et al – whose discussion and findings on regime-type effects on nonviolent- and violent-campaign}

(28)

22

the combined values throughout their dataset or until the conflict-onset year (2014, 560). Yet, in my study I only examine cases where campaigns take place, which makes the mean for a longer time period less relevant. I therefore consider a delimited time period and take the mean from the onset year and the two preceding years.4 Though not capturing long-time regime developments, it provides a more accurate representation of how the autocratic practices manifested in the years preceding the campaign, while accounting for any drastic changes and controlling for reversed causality (i.e. that a regime becomes more autocratic due to campaign onset). In sum, the condition of autocracy is present (1) if the mean for the country’s polity2 values in the onset year and the two preceding years ranged from -10 to -6, while being absent (0) in all other cases.

Mass participation

The condition of mass participation is also obtained from NAVCO 2.1, which reports the size of all included campaigns. I adhere to NAVCO's threshold to demarcate campaigns as small when they have an estimated size of hundreds to thousands participants (indicated under

camp_size_cat). Applied to its ordinal scale camp_size, the values 0-1 (1-9,999 participants)

denote absence of mass participation (0), whereas all values over 10,000 participants (2-5) represent its presence (1). I only measure mass participation in the onset year, informed by Cunningham et al.’s assertion that individuals in autocracies become more inclined to participating en masse once they have observed one group starting (2017, 427).

I evaluate the presence of my suggested causal mechanisms (broad-based mobilization and cross-cutting claims) in later analyses, using NAVCO's div_ethnicity and Thurber’s variables

MOVEMENT, SOLO and CLAIM. NAVCO's binary indicator and Thurber's ordinal indicator MOVEMENT measure to what extent other ethnic groups participate. SOLO dichotomously

indicates whether or not the group was the sole participating group at campaign onset. These indicators allow me to include a component of broad-based mobilization to account for whether the group succeeded in involving other groups at campaign onset (while only being indicative of the causal chain). CLAIM dichotomously indicates whether or not the campaign explicitly involves an ethnic claim – its absence (0) could be a sign of the excluded group using cross-cutting claims to gather different social segments under the same campaign umbrella. For these

(29)

23

indicators, I obtain data for new cases from UCDP, ACD2EPR, SGNAD and secondary sources listed by NAVCO.

Powerful ally

The first step in determining the presence of a powerful ally is to define which states should be considered powerful. In relation to violent and nonviolent contention around African elections, Breslawski and Cunningham list the United States due to its international position, France and the United Kingdom as former colonial powers, and Russia and China as influential states. They only include the US and France in their study, arguing that these states generally act abroad in support of governments (2019, 642, 650f). I adhere to this assertion, but also include the UK, Russia, China, Saudi Arabia and Iran as powerful states, since their external influence and/or track record of involvement in intrastate conflicts cover greater regions.

Drawing on previous literature, I identify an alliance by combining two indicators. I code whether the state targeted by the campaign was in a military alliance with any of these powerful states in the year prior to campaign onset. This measure is developed based on the Correlates

of War (COW) Formal Interstate Alliance Dataset, 1816-2012 (Gibler 2009). Specifically, I

use the directed-dyad alliance data, which captures promises made by one specific country to another country, and set the first country to be one of the abovementioned powerful states. This distinction captures the hierarchical relationships theorized by Cunningham (2016); crudely speaking, I am only interested if the state subject to the campaign counts with the support of a powerful ally, not if that state would “return the favor”. Moreover, I only use the dataset’s definition of a defense alliance, meaning that the powerful state signed terms to protect the other state, since other types of alliances (neutrality, nonaggression and entente) are irrelevant to my study. Based on Breslawski and Cunningham’s reasoning (2019), I also code whether the state is a former French colony, using the Colonial Dates Dataset (COLDAT) (Becker 2019), which develops dummy variables of colonial history by combining several sources and definitions of colonialism. I define the condition of having a powerful ally as present (1) if any of these two criteria is met. I lag the values for defense alliance, since these may change due to internal instability in the onset year, while deeming the colonial past static.

State outreach

(30)

24

located, and code the conditions as present if at least one of the two criteria is met. While Lindemann and Wimmer have data for the years 1946-2005, their dataset is arranged by ethnic-group-periods. When transforming their dataset into country-year and merging their variable into my data frame, it covers only 15 of my 79 cases. This bleak coverage may be a result of their selection criteria – group size 10-44%, territorial concentration, and lower-middle income countries (ibid, 310) – which excludes cases relevant to my study. Reviewing alternative indicators, relevant research usually combines military capacity (e.g. military personnel in thousands divided by population) with bureaucratic capacity (social services and workforce mobilization) (cf. Hendrix 2010). While this definition aligns with Cunningham et al.'s (2017) and White et al.'s (2015) focus on the state having a more developed economy and distribution of labor, I argued in section 2.3 that this conceptualization is too narrow for my purposes. Since I am interested in the state’s (institutional) presence, I opt to code Lindemann and Wimmer's missing data, using their coding rules and source indications. State outreach is coded 1 if the state is present in the rural areas of the ethnic group's region, and 0 otherwise.

Discussion on data and sources

(31)

25

sources sufficiently valid and reliable: they indicate existing defense alliances (COW) and a colonial past (COLDAT), both relying on extensive triangulation of secondary sources.

I also use Thurber's EGC/EBCR data and Lindemann and Wimmer’s replication data. Thurber's datasets are considered trustworthy since he uses NAVCO and EPR (see discussion above) and also states very clear coding rules for his new variables, to which I adhere when coding new cases. Thurber's heavy reliance on these databases is symptomatic of several of my sources: the researchers developing these data relate to each other's work, thereby increasing the risk of redundancy in the data. Still, I trust that Thurber, and other referenced scholars, adhere to good research practices, which is warranted by their works being published in some of the field's most influential journals. Lindemann and Wimmer's (2018) qualitative assessment of state outreach is straightforward and broadly defined: any sign of a state's political and/or military presence is seen as a valid indicator and coded as 1. Their reliance on multiple secondary sources increases the reliability of their data, and I follow the same coding rules and procedures.

For secondary sources, I have used references listed in the databases, combined with reports and articles from renowned organizations and media institutions. UCDP and SGNAD, which are regularly referenced in the above datasets, are considered reliable since they are produced by academic institutions and widely used in research. I have triangulated information to the highest extent possible. Where the presence of a condition has not been possible to confidently confirm, I have consistently made conservative assessments and coded the value as 0.

(32)

26

cases where ethnics demands are enough for other groups to campaign, or that cross-cutting issues may be included in ethnic claims. In this regard, my findings will be only indicative.

In terms of external validity, the extensive time period and the use of global datasets warrant for relatively high generalizability. My temporal domain was established based on data availability for the time period 1946-2013, and by expanding Thurber’s dataset with recent data, I ensured that all reported cases of campaigns initiated by politically-excluded ethnic groups were included. My universe should thus be well represented in the sample. Yet, it is not possible to know how many campaigns are excluded from the NAVCO dataset for never reaching 1,000 participants. Any inferences I may draw based on my analysis extend to excluded groups that initiate campaigns meeting this NAVCO threshold, while I cannot confidently generalize to broader sets of campaigns. Moreover, I only investigate four conditions that I expect to be conducive to nonviolence, without considering other conditions that may appear in different campaign settings and affect the variables. These issues naturally reduce the external validity of my results, though they will still highlight conditions that deserve further scrutiny.

Robustness checks and control variables

(33)

27

and military expenditures per soldier (Hendrix and Young 2014, 341-343; D. E. Cunningham et al.. 2017, 480).

As for control variables, QCA does not offer the same possibilities as regression analysis to include controls in different models, because QCA treats each added variable as a condition that combines equally well with other conditions to produce new patterns ready for interpretation. In other words, while theory guides the interpretation of variables as independent or control variables in both regression analysis and QCA, the former does not penalize for added complexity, since the effect of an IV can be analyzed “holding all else constant”. QCA, by contrast, does not hold anything constant, but instead tests all possible combinations of conditions. Complexity thus increases exponentially with each added condition in QCA, meaning that the inclusion of CVs can yield very cumbersome results that cannot be easily interpreted in a substantively meaningful way.

(34)

28

3.2 QCA

As method, I will employ Qualitative Comparative Analysis (QCA). The name alludes to a comparison of qualitative features, while the method is quantitative in nature and the analysis is conducted with R software (Thiem 2018; R Core Team 2017). What distinguishes QCA from conventional statistical analyses is the focus on conjunctions of conditions and the measures of fit used to assess evidence and counterevidence.

As for the analysis of conditions, QCA aims to determine necessary or sufficient conditions for the outcome to occur.5 The procedure is based on three assumptions of causal complexity: conjunctural causation, causal asymmetry and equifinality (Schneider and Wagemann 2012, 78). Conjunctural causation refers to the combined causal effect of conditions. In contrast to regression analysis, which involves a baseline assumption about individual factors having the same effect across different cases, no independent effects are assumed in QCA. Instead, the effect of one condition unfolds in tandem with other conditions. Using a practical example, I may take the bicycle to work if I am in a hurry and missed the bus. The notion of how a condition's presence bears relevance to other conditions resembles how we treat preconditions and triggers in case studies. As a consequence, one condition may lead to either the outcome or the non-outcome, depending on how it combines with other conditions in different paths.

Relatedly, causal asymmetry points to the idea that explanations of the outcome do not necessarily mirror explanations of the non-outcome. A close-to-hand example is hunger: when you are hungry you usually eat, but this does not imply that you do not eat when you are not hungry. Asymmetry is particularly interesting for my study, since I posit that existing explanations of non-occurrence of violence may explain the onset of nonviolence – not as the direct opposite of violent conflict, but as a theoretically plausible pathway away from violence. Finally, equifinality suggests that several pathways may cause an outcome. Returning to the example with hunger, this may be one reason why you eat, but another reason could be that you are at a social event, or that you are bored – all these “paths” may lead to eating.

Departing from all logically possible combinations of my conditions (listed in a truth table), the analysis will yield a solution including those combinations that are found empirically in my

5_{A condition is necessary if the outcome cannot be achieved without the condition being present, whereas there}

(35)

29

cases of nonviolent-campaign onset. For example, mass participation may occur under specific circumstances of autocracy or high state outreach. This feature of QCA is useful for my study, since I explicitly ask which paths lead to nonviolence.

The second difference compared to regression analysis is that QCA does not report insecurity of results in statistical parameters such as significance or strength, nor counterevidence in the form of insignificant or opposite statistical results. Instead, QCA communicates the reliability and relevance of the necessary and sufficient conditions that are found. To this end, consistency and coverage parameters are used as measures of fit.Consistency indicates the degree to which

the empirical data aligns with the assertion that specific conditions are sufficient for the outcome, where 1 would be a perfect subset relation (corresponding to a perfect correlation in regression analysis). Coverage, in turn, points to the empirical importance of specific conditions by indicating how much of the outcome is covered by the sufficient combination – i.e. the relative amount of cases explained – where 1 corresponds to all cases with the outcome (ibid, 119, 122, 128f). This feature of QCA also rhymes well with my aim to evaluate what opportunities are actually in place when excluded groups initiate nonviolent campaigns. Both parameters of fit fill a similar function to that of R2 in regression analysis, though I will explain their interpretation in detail when taking the reader through the analysis.

Evidence and counterevidence

So, how will I go about using QCA to answer my research question under which conditions politically-excluded ethnic groups initiate nonviolent rather than violent resistance? The theory evaluation will not take the regular form of a statistical test of my hypothesis and interpreting the results as refuting or confirming the theoretical expectations. Instead, I will evaluate my theory in two steps, as outlined below and further described in the empirical chapter.

(36)

30

were found present in the sufficient paths, or if any condition is present solely in combinations under the thresholds for consistency and coverage. Such results are particularly interesting from a theory-evaluating perspective, as they would point out which opportunities are more important for nonviolence and should be further explored, and if any of the contrasting theories should be revised, e.g. by extending or delimiting scope conditions.

As for counterevidence, the most straightforward falsification of a condition as necessary for the outcome is if there are cases where the condition is absent but the outcome is present. Conversely, a condition is refuted as sufficient if the condition is present but the outcome is absent (Schneider and Wagemann 2012, 58). A second form of counterevidence is if the conditions do not meet the parameters of fit, which will speak against these opportunities being conducive for nonviolent-campaign onset. Finally, a condition cannot logically be sufficient (or necessary) for both the occurrence and the non-occurrence of the outcome (ibid, 198). If parallel analyses for violent-campaign onset produce a solution term that involves the presence of conditions initially theorized (and found) to lead to nonviolence, I will interpret these results as counterevidence of my theory and evidence that other factors play in. Obviously, there are many other conditions that can be considered opportunities for nonviolence and constraints to violence, which were not included in the analysis. Low consistency and coverage levels are also an indication that there are likely alternative explanations to be explored.

(37)

31

4. Empirics and Analysis

To first get a grip of the variable distribution in my sample, Figure 2 offers some descriptive statistics.6 There are 60 cases of violent-campaign onset and 19 cases of nonviolent-campaign onset, i.e. there are three times as many cases without the outcome of interest. As for the conditions, 46 campaigns did not take place in an autocracy while 33 did; similarly, 46 cases did not reach mass participation while 33 did, though these are not the same cases. In 62 cases, the state did not have a powerful ally, while only in 17 cases it did. In 25 cases, the state did not have high territorial outreach, whereas in 54 cases it did.

Source: Author.

Main results

The QCA yields results that partly support my hypothesis. One of the “extreme” theoretical pathways formulated in my hypothesis, constraints to violence in place without opportunities for nonviolence, had perfect consistency with nonviolent-campaign onset among politically-excluded ethnic groups, but only covered one case, thereby reducing its theoretical importance. The other “extreme” path, mass participation under autocracy, only covered violent campaigns. These results suggest that neither opportunities for nonviolence nor constraints to violence in

isolation are enough to make excluded groups act nonviolently. Instead, the two other consistent

combinations from the sufficiency analysis involve conditions from both theoretical strands,

WHEN ETHNICS REBEL NONVIOLENTLY