Should We Stay or Should We Go?

(1)

Should We Stay or Should We Go?

The Influence of Rebel Governance on Rebel Fragmentation

Lou van Roozendaal

Master Thesis

Spring 2021

Department of Peace and Conflict Research, Uppsala University

Supervisor: Sebastian van Baalen

(2)

Abstract

The fragmentation of rebel groups has received extensive attention from scholars and policy-makers alike. Fragmentation has been linked to an increase in civilian deaths, prolonged conflicts, and difficult peace negotiations. Despite the deliberation given to this field, this study argues that rebel governance serves as an explanatory variable that has been overlooked in the literature. This study aims to close this research gap by examining the influence of rebel governance on rebel fragmentation. The thesis suggests that rebel governance intervenes with the opportunity for rebel lieutenants to find the support of networks to create a splinter, and is successful in pre-venting rebel lieutenants from forming potential splinters. Therefore, it is hypothesized that rebel governance reduces the likelihood of rebel fragmentation. Using data on rebel groups active be-tween 1946-2012, this thesis finds quantitative evidence in support of the hypothesis across all models. Furthermore, support is found that suggests that the more rebel institutions are present, the less likely it is that the rebel group fragments. Additionally, this study employs a survival model of which the results suggest that rebel groups who employ governance take longer to frag-ment compared to rebel groups without governance.

(3)

Acknowledgements

I would like to acknowledge the support and encouragement of some of the people to whom I am deeply grateful throughout the process of writing this thesis. First, my profound gratitude goes out to my supervisor Sebastian van Baalen, for his enthusiasm, critical thinking, and devotion during these last months. To Wera and Tove, thank you for listening, navigating Sweden with me, and always cheering me on and up. Dear Ruth, thank you for your grammar and baking skills. My appreciation and respect go out to my friends who have on- or offline kept me company throughout this strange year.

(4)

List of Figures

Figure 1: Visual Presentation Rebel Governance Count Variable. . . .37

Figure 2: Kaplan-Meier Curves of Rebel Governance. . . .45

Figure 3: Distribution of Secessionist Conflicts: Before and After Matching . . . 64

Figure 4: Distribution of Ethnic Conflicts: Before and After Matching. . . 65

Figure 5: Distribution of Communist Conflicts: Before and After Matching . . . 65

Figure 6: Distribution of the Lifespan of Rebel Groups: Before and After Matching . . . 66

(7)

List of Tables

Table 1: Overview Matched Pairs. . . 33

Table 2: Differences in Means: Before Matching . . . 34

Table 3: Differences in Means: After Matching . . . 34

Table 4: Propensity Scores . . . 34

Table 5: Descriptive Statistics . . . .36

Table 6: Logit I: Rebel Governance and Fragmentation in Rebel Groups between 1946-2012. . . 38

Table 7: Descriptive Statistics: Disaggregated Dimensions of Rebel Institutions . . . 41

Table 8: Logit II: Disaggregated Dimensions of Rebel Institutions . . . 42

Table 9: Kaplan-Meier Model Log Rank Test . . . 44

Table 10: Cox Proportional Hazard Regression. . . .46

Table 11: Cox Proportional Hazard Regression Coefficients of Model 12 . . . 47

Table 12: Overview Variables Rebel Governance . . . 63

Table 13: Logit III: Additional Control Variables . . . 69

(8)

1. Introduction

Why do some rebel groups split while others do not? Empirical evidence shows some rebel groups split into two or more groups, while others stay unified throughout conflicts. Whereas the Eritrean People’s Liberation Front (EPLF) in Eritrea managed to stay united over twenty-one years during the conflict of independence of Eritrea from Ethiopia; the Congolese Rally for Democracy (RCD) in the Democratic Republic of the Congo (DRC) split into seven different splinter groups over the course of only four years (UCDP 2019). The RCD is not unique, as matter of fact, research suggests that over one-third of rebel groups split at least once during their lifespan (Fjelde and Nilsson 2018). Understanding fragmentation is important since it can pose significant challenges due to an increase of actors within a conflict. Multiparty civil wars last longer (Cunningham 2006), are more violent due to inter-rebel fighting (Cunningham, Bakke, and Seymour 2012), complicate finding a satisfactory solution to all the warring actors (Cunningham 2006), and are more likely to relapse after negotiations due to power-sharing disputes (Kydd and Walter 2002; Stedman 1997).

To further our understanding of fragmentation, this thesis proposes a new explanatory fac-tor, namely rebel governance. Rebel governance, in its most basic form, provides some public services, governs civilians through (at least partially) non-violent methods, and in return rebel groups attain access to natural resources, information, and labour forces (Peclard and Mechoulan 2015). However, rebel governance groups can also build an extensive range of different wartime institutions. For example, the Liberation Tigers of Tamil Eelam (LTTE) and Eritrean People’s Liberation Front (EPLF) provided health-care and education to civilians; National Union for the Total Independence of Angola (UNITA), Maoist Rebels of Nepal (CPN-M) and Tigrayan People’s Liberation Front (TPLF) regulated land reform; the Taliban implement a judicial system with mo-bile courts, and the National Patriotic Front of Liberia (NPFL) created their own currency (Arjona 2014a; Mampilly 2011). Despite a growing body of literature that has sought to understand the onset, variations, and characteristics of rebel governance, limited studies examined the effects and consequences of established rebel governance (exceptions are e.g. Breslawski, 2021; Heger and Jung 2017). This thesis sheds new light on the consequences of rebel governance, and specifically in relation to the mitigation of splintering.

(9)

of the fact that civilians can serve as opportunity structures for rebels and therefore the loyalty of civilians matter in examining opportunities for fragmentation. Without understanding the dynam-ics of civilian-rebel relations, we overlook the importance of civilians in creating opportunities for fragmentation and as such, fail to understand the risks leading to splintering and subsequently, multiparty conflicts. Therefore, this study aims to understand the interaction between the field of fragmentation and the field of governance, and combines the two to ask the following ques-tion: how does rebel governance influence rebel fragmentation? This thesis argues that rebel governance will decrease the likelihood of fragmentation.

This study makes several contributions. First, prevailing explanations have overlooked rebel governance as a determinant of fragmentation. Taking previous works on intra-group dynamics (Mosinger 2018; Otto and Salverda 2018; Staniland 2014) as a point of departure, this thesis advances a novel theoretical explanation of rebel fragmentation in which civilians are regarded as opportunity structures for the rebel group; as such the support of a civilian network is important for building a splinter. Splintering occurs when midlevel commanders have the opportunity to start their own splinter group, independent from rebel leadership (Nagel and Doctor 2020). This thesis argues that two factors determine why rebel governance is effective in preventing the occurrence of fragmentation: the mixing of rebels and civilians increases the complexity of social ties as well as interdependency, and rebel governance provides incentives to civilians to remain loyal to the leadership. Therefore, it is harder for mid-level commanders to procure loyalties for the support networks and as such, splinter groups are less likely to arise.

Second, the statistical nature of this study allows for high generalizability of the results and aims to understand the relation between rebel governance and fragmentation while being able to control for confounding variables. In order to test the proposed hypothesis, a sample of rebel groups worldwide between the period of 1946 to 2012 is created based on data from Cunningham, Gleditsch, and Salehyan (2009). The independent variable (rebel governance) is based on the Rebel Quasi-State Institutions (RQSI) dataset (Albert 2020a); the information for the dependent variable (rebel fragmentation) was manually collected. To create a control group a matching algorithm was utilized to reduce potential confounding and to mitigate selection bias. Finally, to test the relation between rebel governance and rebel fragmentation, this study employed logistic regressions and survival models.

The logistic regressions show that rebel governance significantly reduces the likelihood of frag-mentation. Furthermore, the regressions indicate that the more rebel institutions are present, the less likely the rebel group will fragment. Based on the Kaplan-Meier model it can be derived that having rebel governance matters for the longevity of the movements, as the presence of rebel gov-ernance significantly increases the time it takes to fragment. Additionally, the Cox Proportional Hazards model confirms the result of the logistic regressions and indicates that rebel governance groups have a decreased risk of sudden fragmentation.

(10)

(11)

2. Literature Review and Definitions

The first half of this chapter will present the previous literature on rebel fragmentation to examine existing explanations. The second half of this chapter will clarify the definition of several terms used throughout this study.

2.1 Previous Work

The relation between civilians and rebels has been thoroughly examined with regards to the level of violence against civilians (Hultman 2012), the targeting of civilians (Hultman 2009; Kaly-vas 2006; Wood 2010), and the institutionalized interactions between rebel and civilians (Arjona 2014a; Mampilly 2011). However, as this literature review will show, limited attention has been paid to the consequences of the interactions between rebels and civilians in affecting the splinter-ing of rebel groups. This section first provides a general note on the different definitions of rebel fragmentation in current literature; after which it will move on to address three main strands of explanations provided by previous scholars. The main focus of this study will be on the last strand of explanations, with regards to the internal characteristics of rebel groups.

Fragmentation has both been approached from the conflict-level (see Bakke, Cunningham, and Seymour 2012; Fjelde and Nilsson, 2018; Mosinger 2018) and the rebel group-level (see Christia 2012; Staniland 2014). The first refers to studies that have approached fragmentation as the presence of all rebel groups in a conflict, this includes both groups that have originated from each other as well as groups that come into being without any ties to pre-existing rebel groups. The latter approached fragmentation as an in-group process where a part of the original group splits and forms its own rebel group (often referred to as internal fragmentation or splintering) (Fjelde and Nilsson 2018; Perkoski 2015). The scope of this thesis is limited to the latter, group-level processes of fragmentation, as is discussed in section 2.1.2. In contradiction to conflict-level fragmentation, the process of within-fragmentation only focuses on once unified groups that split. As such, within-fragmentation has additional considerations and causes that are not applicable to the independent mobilization of rebel groups (which are included in the conflict-level fragmentation definition). Within-fragmentation requires further examination on a group-based level, which is not considered in studies focusing on conflict-level fragmentation (Fjelde and Nilsson 2018; Perkoski 2015). In this literature review, studies that have used either definition are included. Since the process of conflict-level fragmentation provides explanations relevant for within-group fragmentation alongside inferences about independent mobilization, studies that have examined conflict-level fragmentation are important to discuss for the scope of this thesis. The last strand of literature examined in this section elaborates on processes specific to within-group fragmentation.

(12)

(Perkoski 2015). In the scope of this study, these processes are not considered as fragmentation due to their intentional nature and because they do not create distinct new rebel groups with a separate line of command.

2.1.1 Structural Factors and External Factors

The first strand of fragmentation literature seeks to explain rebel fragmentation by examining structural factors. Structural explanations aim to examine the process of fragmentation through the absence or presence of structural conditions that constrain behaviour or that limit opportunities for fragmentation, such as the presence or absence of natural resources. Several works have relied on Fearon and Latin’s trifecta of (1) weak state influence, (2) terrain layout, and (3) the access to lootable resources, to explain the onset of fragmentation (Mosinger 2018). Although no support has been found for the first two (see Seymour, Bakke, and Cunningham 2016); Lidow (2016) substantiates the third, suggesting that access to lootable resources indeed increases the risk for fragmentation. Fjelde and Nilsson (2018) add to this by arguing that fragmentation is more likely in resource-rich areas since the pay-off for a splinter potentially could be high if the splinter group would gain control over the resources.

The second strand of literature focuses on two different types of actors beyond the rebel group itself, namely the incumbent state and external actors, such as neighbouring states. Outsiders or external state actors that are not part of the main warring parties can increase the likelihood of fragmentation as well as decrease it through their allocation of (financial) support. Ives (2019) finds that co-ethnic external support reduces the likelihood of fragmentation whereas non-ethnic (or non co-ethnic) external support promotes fragmentation as it raises the question of trust with regards to the motivations of the rebel leadership. Furthermore, external support can unify rebel groups if the support keeps the power balance in favour of the rebel leadership. However, if the support is allocated to an internal rival this increases the risk of splintering (Tamn 2016).

(13)

Some incumbent states have additionally used peace processes and mediation as a tool to purpo-sively splinter groups. Paradoxically peace agreements can increase the likelihood of splintering due to diverging opinions within rebel elites that are heightened when mediations occur. For this reason, mediation (Lounsberry and Cook 2011) and the negotiation of peace agreements have been linked to an increase in splintering (Duursma and Fliervoet 2020; Fjelde and Nilsson 2018). This is problematic, as the splinter groups often are difficult to get to the table, thus prolonging the conflict (Stedman 1997).

2.1.2 Internal Characteristics

The third and last strand of literature, and the focus of this thesis, looks at characteristics of rebel groups to specifically explain within-splits. Generally, the overwhelming majority of re-cent studies focus on conflict-level fragmentation, and not on within-fragmentation. However, several studies have focused solely on internal factors and characteristics that cause within-group fragmentation. Internal splits are caused by leadership disputes between the rebel leadership and mid-level commanders (Nagel and Doctor 2020). These disputes can be aggravated by external factors, such as those presented in the previous paragraphs (Nagel and Doctor 2020). This strand of literature has sought to find explanations as to why some rebel groups are more affected by external factors as opposed to others and have accordingly analysed in-group characteristics.

In general, rebel groups that are sizable and commit (sexual) violence are more likely to split (Asal, Brown and Dalton 2012; Mosinger 2018; Nagel and Doctor 2020). Burch and Ocheiter (2020) have shown that having territorial control also increases the likelihood of fragmentation. Territorial control can strengthen the position of the rebel group relative to the government. The survival of the rebel group is thus less threatened and retaining the unity as a rebel group becomes less essential. Therefore, internal disputes that occur in rebel groups with territorial control are likely to result in splinters (Burch and Ocheiter 2020). In opposition, Mosinger (2018) argues that stationary rebels are less likely to fragment since geographic stability facilitates overcoming the collective action problem and benefits the rebel group e.g. by given them access to information from civilian networks. Furthermore, several case studies have focused on the social foundations of rebel groups and have shown that rebel groups founded on social ties (e.g. ethnicity or religion), generally have higher levels of cohesion and are thus less likely to fragment (Burch and Ochreiter 2020; Fjelde and Nilsson 2018; Hafez 2020). In-group cohesion is also strengthened when rebel groups mobilize pre-conflict networks of civilians (Staniland 2014). Mosinger (2018) argues cohesion does not necessarily stem from pre-existing networks but rather shows that rebel groups that are integrated with civilian networks are less likely to experience fragmentation. Lastly, arguments regarding social ties have been integrated into broader theoretical frameworks that combine social ties with the need for clear hierarchical structures (or organizational capacity) that promote control and structural integrity (Otto and Salverda 2018; Staniland 2014).

To this author’s knowledge, no studies have provided a theoretical explanation nor have tested the relation between rebel governance and rebel fragmentation.1 This study contributes to un-derstanding fragmentation through examining the unique interplay between rebels and civilians

(14)

in governance structures, where wartime institutions and the extensive engagement with civil-ians is the norm. Previous studies (e.g. Arjona 2014a; Eck 2014; Mosinger 2018) have shown that the coercion of civilians only works in the short-term, as it may lead to initial compliance but ultimately fails to create new loyalties and can lead to resistance of civilian networks against the rebel leadership. The theoretical framework on which this thesis is based implies that rebel governance, characterized by its dense interactions between the rebels and civilians under its au-thority, can offer incentives to the civilian population and is therefore successful in enlisting their support. In parallel, high levels of organizational control and the ability to manage civilian-rebel interactions contribute to the structural integrity of the rebel group. Rebel governance will thus mitigate the likelihood of fragmentation, which is explained in-depth in Chapter 3. It is important to consider rebel governance as an explanation for fragmentation since it examines the previously overlooked fact that civilians can serve as opportunity structures in terms of fragmentation for rebels and that civilians are paramount in the survival of the rebel group or a potential splinter group. Furthermore, it builds on previously made assumptions that the differences in types of rebel-civilian interactions matter for the onset of fragmentation (Burch and Ochreiter 2020; Ka-lyvas 2006; Peclard and Melouchan 2015). Rebel-civilian interactions are not unitary and should be considered disaggregated and in-depth to understand the creation of opportunities to splinter. In the following chapter, the theoretical framework is introduced. However, before discussing the theoretical framework, several terms are explained in more detail to understand their use through-out this thesis.

2.2 Terms and Definitions

The following section discusses three main terms used throughout this thesis. As the concepts have been used differently amongst scholars it is important to understand how they are defined within the scope of this thesis.

2.2.1 Rebel Fragmentation

The term rebel fragmentation has been used to describe a range of different events and lacks consistent use by scholars (see Perkoski 2015). Throughout this study, rebel fragmentation refers to (unless explicitly mentioned otherwise) the process in which a faction of pre-existing rebel group breaks away to form an independent rebel group, with its own name, chain of command, and leadership.2 This definition exists out of several aspects.

First, the definition limits itself to rebel groups that have splintered from the original group to form an independent rebel group, thus resulting in nascent groups. It is important to note that this definition moves away from the general trend amongst recent studies to consider fragmen-tation as a combination of both the independent mobilization of groups and the splintering from pre-existing groups (see e.g. Fjelde and Nilsson 2018). Combining these groups into one term overlooks group-level dynamics that can account specifically for the formation of nascent groups (Fjelde and Nilsson 2018; Perkoski 2015).

(15)

Second, the newly formed group holds no close ties to the original rebel group and has its own independent leadership (Mahoney 2020; Perkoski 2015; Tamm 2016). It should therefore be noted that in this thesis the term fragmentation does not include organizational specialization (where rebel groups ‘split’ on purpose to create specialized wings, e.g. Irish Republican Army (IRA) and Sinn Fein) or organizational decentralization (where the rebel group has a decentralized leadership existing out of independently operating sub-groups but still identifies as the same group) since both of these processes do not form a separate line of command or a distinct rebel group (Perkoski 2015, 14-21).

Although limiting the definition of fragmentation reduces the scope of this thesis significantly, its importance should not be underestimated; descriptive statistics showed that around 10% of the groups listed by the Uppsala Conflict Data Program (UCDP) splintered from a pre-existing rebel group, and the Minorities at Risk-Organization Behaviour dataset shows that over one-third of all violent actors splinter during their lifespan (Perkoski 2015, 20).

2.2.2 Rebel Governance

In this thesis, rebel governance is defined as “the set of actions insurgents engage in to regulate the social, political, and economic life of non-combatants during civil war” (Arjona, Kasfir and Mampilly 2017, 3). Rebel governance can be extensive, where rebel groups build a wide range of wartime institutions, which can have a wide scope of functions from providing health care and education to the set-up taxation systems or creation of their own currencies (Arjona 2014a; van Baalen 2020; Florea 2020; Mampilly 2011). The term wartime institutions refers to rebel-build institutions with the purpose to regulate the lives of civilians. Rebel governance can also be very limited; Terpstra (2020, 1147) argues that in its most basic form, rebel governance provides at least some services for civilians and in return provides rebels with access to (often material) benefits. This can be informal but also formalized through wartime institutions (Terpstra 2020). It should be noted that rebel governance does not necessarily require territorial control. Some groups like e.g. the Taliban in Afghanistan implemented social services that were portable instead of static within one area (Terpstra 2020). The rebel group thus does not have to be stationary in order to have rebel governance (Terpstra 2020). Lastly, rebel governance does not necessarily regulate the lives of likely-supporters (e.g. civilians that shared aspects with the rebel group such as ethnicity, ideology, or religion) but can additionally or exclusively regulate the lives of civilians in the rebel-held territory that have no initial shared ties to the rebel group (Gowrinathan and Mampilly 2019; Stewart 2018a).

2.2.3 Civilians

(16)

(17)

3. Theoretical Framework

The following sections will explore how rebel governance could theoretically influence the on-set of fragmentation. First, it discusses how internal fragmentation occurs. Second, relevant rebel governance characteristics are elaborated upon. Third, the last section will tie these to rebel fragmentation and show how rebel governance characteristics mitigate the opportunities of frag-mentation to arise.

3.1 Conceptualizing Internal Fragmentation

Internal splits are generated due to power disputes between rebel leaders and the organizational rebel lieutenants3 (Nagel and Doctor 2020, 1235). Often these can heighten when external pres-sures arise from outside factors, including peace negotiations, repression from the incumbent government, or changes in the intensity of the conflict (Nagel and Doctor 2020, 1228-1229). The external pressures can amplify incentives for rebel lieutenants to split if they e.g. feel that the response towards the crisis is not accurately dealt with by the rebel leaders. This can shift the power balance between the rebel leadership and rebel lieutenants (Mosinger 2019, 936). Like other rebel groups, groups employing rebel governance have to face rebel lieutenants that have incentives to split from the original group, e.g. to gain access over resources, potentially get a seat at the negotiation table, or to pursue different objectives (Salverda and Otto 2018, 12). How-ever, the capability or opportunity to split is dependent on two things: (1) access to resources to fund the splinter faction, and (2) a support base that favours the rebel lieutenant as opposed to the rebel leadership (Mosinger 2019). Previous works have been heavily focused on the accessibility to resources, however, it overlooks that rebel lieutenants need the support of networks in order to form a splinter (Mosinger 2019, 941-946). Two main networks of possible support can be identified: (1) civilians living under rebel-rule, and (2) lower-ranking rebels. Both these support networks function as opportunity structures for the rebel group since rebel lieutenants have the opportunity to form their own rebel group when they manage to shift the loyalties of the networks away from the rebel leadership to themselves. The importance of rebel combatants is self-evident (fighters are crucial in challenging the state or other rebel groups), however previous research has shown that civilians can make the survival of the splinter more durable (Arjona 2014a, 9; Mosinger 2018). Civilians can, for example, provide rebels with access to information about e.g. the whereabouts of government soldiers, serve as a pool for recruitment of new soldiers, provide access to population-based natural resources, and support the rebel group with shelter, food, and labour (Breslawski 2021; Ottmann 2017; Sabates and Verwimp 2014; Staniland 2014). Therefore, the cooperation of civilians can strengthen territorial control and increase organizational capacity, which is crucial for the (long-term) survival of a rebel group and can increase the chance of rebel victory (Arjona 2014a, 9; Stewart and Liou 2017, 287).4

3. Which are interchangeably referred to as mid-level commanders in this thesis.

(18)

It should be noted that the networks of rebel combatants and civilians are often at least partially intertwined. Rebel groups that govern over civilians often recruit from the population (Arjona 2014a; Mampilly 2011, 145; Ottmann 2017, 31). The ties to the civilians can leave rebel com-batants with mixed feelings of loyalty to both the rebel leadership and their civilian relatives or friends, as such the support of civilians for a rebel mid-level commander is additionally crucial to understand in relation to their influence on rebel combatants. This thesis aims to consider the influence of civilians (and by extension their influence on the rebel combatants) on the decision-making process of mid-level commanders surrounding the formation of splinters. As such, it is critical to consider the organizational structures and foundations of the civilian-rebel relations, as they “are key factors in explaining divisions of potential sources of power and division, and in particular the (sway) loyalties” (Mosinger 2019, 936). This thesis argues that rebel governance is successful in both unconsciously swaying and actively pursuing the loyalty5 of civilians in favour of the rebel leadership as opposed to supporting the potential rise of midlevel comman-ders. Therefore, the support by the civilian population for the mid-level commander becomes less likely. Since the mid-level commander is unable to gain the support necessary to create a splinter, the rebel lieutenant’s cost-benefit analysis will remain in favour of a unified group rather than facing the risk of a worst-case scenario of a failed attempt to form a splinter (Mosinger 2018, 68). In the following sections, this thesis will first outline how the rebel leadership obtains the sup-port of the civilian network through governance. Thereafter, it will argue how these are connected to fragmentation and the decision-making process of rebel lieutenants (the would-be-leader of the splinter). In sum, the theoretical framework examines how rebel governance shifts the cost-benefit analysis of rebel lieutenants in favour of unity due to the inability of attaining a civilian support base.

3.2 Rebel Governance and Civilians

One of the main arguments of this thesis is that civilians will favour supporting the rebel leader-ship instead of the rebel lieutenant. Therefore, the following two subsections will delineate why civilians would support the rebel leadership in the first place. This thesis argues that rebel gover-nance actively provides both negative and positive incentives to keep the loyalty of the civilians, thus posing the classic carrots-and-sticks model. It starts with an explanation which elucidates why the continued presence of rebels amongst civilians fosters a multivalent of social ties. How-ever, the creation of social ties to the civilian population in itself is not necessarily enough to induce support (Stewart and Liou 2017, 287).6 Therefore, this thesis argues that negative and pos-itive incentives are effective in inducing compliance and maintaining support over time. Negative incentives induce compliance through control and punishments; positive incentives sway

loyal-5. Loyalty here is defined broadly. The term loyalty can be misleading as it implies positive feelings or motivations. However, here it is used to encompass both voluntary acts of support as well as compliance to rebel-made rules. An absence of loyalty would mean that there is active resistance to the rebel group (see e.g. Arjona 2014a; Gowrinathan and Mampilly 2019).

(19)

ties since rebel governance can provide protection and services allowing civilians to have a better quality of life. Whereas the negative incentives allow for control and compliance, the positive incentives encourage cooperation and support. In sum, this thesis proposes potential processes through which rebel governance may sway the loyalty of civilians. After these arguments are laid out, they are explicitly connected to fragmentation in section 3.3.

Before examining the different types of invectives it is important to consider the functions of territorial control in combination with rebel governance on eliciting the support of civilians.7 Rebel groups cannot succeed without gaining the support of civilians (Arjona 2014a). Support can be obtained through regulating civilian life in controlled territories, guaranteeing public order, and providing public goods (Arjona 2014a, 36). Rebel governance constitutes a continued presence of rebels among the civilians within a territory. Local presence allows the rebel group to achieve greater integration over time (Staniland 2014, 46). As such, the presence of the rebel group within the living area of the civilians can, with time, increase the density of social ties through both informal and formal interactions. Informally, rebels that live amongst civilians create social ties to the civilians through social interactions such as through events like local soccer games (Arjona 2014b). Formally, through wartime institutions created under rebel governance, existing norms and especially rules regulating civilian behaviour are taken over by the rebel leadership. The new rules of conduct and social behaviour are, over time, slowly internalized by the civilians, and give shape to a new social structure (Arjona 2014a; Checkel 2017; Kubota 2017). As such, rebel governance establishes new social structures that regulate the behaviour of civilians that are in favour of the rebel group. The existing local institutions become less important – the exchange of information and services is increasingly dependent on rebels (Kubota 2018). Pre-existing social ties amongst civilians are supplemented with social ties to the rebel group and this social interaction is institutionalized (Kubouta 2018). Although this integration over time is important to consider in relation to creating a myriad of social ties, a stationary presence does not necessarily induce support (Mosinger 2018; Peclard and Mechoulan 2015). In order to gain the support of civilians, and thus allow for this integration over time, rebel governance can provide incentives for civilians to support rebel rule.8

3.2.1 Rebel Governance and Positive Incentives

Rebel governance arguably operates on the idea of a social contract (Arjona 2014a, 21-62). Al-though this contract is rarely formally established, informally civilians will learn the rules and live accordingly (Arjona 2014a, 26-27). Civilians might not actively consent to the rules of the rebels or their presence, yet, the wartime institutions mimic a social contract since they provide clear established rules that are in stark contrast with the uncertainty often faced by civilians in conflict zones (Arjona 2014a, 26-27). It is assumed that civilians will generally favour this social contract over living in a state of disorder. In return e.g. for access to resources, rebel governance provides a state of order based on a clear set of rules and can provide services to further induce

7. A scope condition of this theory is that rebel governance groups have territorial control. This is elaborated in section 3.4.

(20)

support (Arjona 2014a, 45-50). Explicit rules of conduct can positively induce support when they are targeted at maintaining public order through intervention in key problems faced by the civilian population. For example, civilians that live in areas where they face rape, extortion, or drug use will welcome rebel groups that punish offenders and intervene to address these problems (Arjona 2014a, 49).

The service provision of rebel governance can range from the basic protection from outside violence to building extensive wartime institutions dealing with healthcare, housing allocation, or media channels (Albert 2020a). However, the most basic provision - the protection from external violence - is key in understanding why civilians support the rebel governance group. Following Kalyvas (2006) line of arguing, civilians prefer protection from violence and security above all else. Therefore, if the rebel governance group upholds this as a basic premise, civilians (even those that were unlikely to have normally supported the rebels) will generally favour or tolerate rebel rule. Being outside of rebel governance areas or living under rebel groups that do not pro-vide governance would leave civilians vulnerable to higher levels of violence (Ottman 2017). Due to the incentive of protection, civilians will interact and participate at least on the most basic level, thus upholding this social contract (Arjona 2014a). Beyond providing protection, service provi-sion (positive incentives) can actively induce support or sway the loyalty of the civilians (Arjona 2014a; Mampilly 2011). The provision of services can improve the quality of life within the con-flict zone drastically. Social services, such as healthcare and education, are deemed desirable to civilians, can thus be paramount in swaying the opinion of the civilian population (Arjona 2014a; Stewart 2018b, 6). Additionally, services such as education can also serve as potential propa-ganda tools that promote a positive perception of the rebel group (Stewart 2018b). This is even more prominent if the state itself does not or is very limited in providing any services (Peclard and Mechoulan 2015; Stewart 2018b). It should be noted that provision of services is not always formally established through wartime institutions but can also be informally provided (as seen by e.g. rebel governance groups in Latin America) (Arjona 2014a, 62).

3.2.2 Rebel Governance and Negative Incentives

(21)

To elaborate, policing institutions can increase the overall level of control of the rebel group over the civilians. The governance structures established over the civilian population in the area under control by rebel governance allows for the monitoring of civilian populations, and are essen-tial in maintaining control (Arjona 2014a; Terpstra 2020). Governance structures can function to police civilians, allowing rebel governance groups to maintain higher levels of control (Mampilly 2011; Staniland 2014). Additionally, the existence of dispute mechanisms is paramount in pre-serving order and enables the coordination of social and political interaction amongst civilians. Dispute mechanisms are observed in the majority of rebel governance structures either formally or informally (Arjona 2014a, 72-74). Although dispute mechanisms do not provide an incentive as such, they can function as a tool to create order as they mitigate (internal) conflicts and can regulate intra-civilian sub-cleavages (Arjona 2014a, 72; Mampilly 2011, 81).

It should be noted that rebel governance does not only increase the level of order amongst civilians, but additionally, it can foster control within the rebel group itself. This is two-folded; first, through informal actions, the rebels are likely to be held accountable by not only the rebel leadership but also by the civilian population within the territory. The process of integration creates a set of norms, and the incentives provided by governance to civilians can induce informal social control from civilians, where they hold rebel combatants accountable for behaviour that damages the (implicit) social contract (Arjona 2014a). As such, rebels do not only face internal policing from the rebel leadership (as seen in the next paragraph) but also from the civilians.

Second, rebel governance has additional incentives to maintain control over the rebel combat-ants. The rebel leadership is likely to favour in-group policing of their own soldiers even more than groups who do not employ governance since the costs of defiance of rebels is higher, as it can mean that civilians will oppose the rebels if they will feel that they are being treated unjustly (and thus suffer audience costs) (Mosinger 2017, 46). Civilians are generally more able to navigate their rights and wishes within rebel governance the longer the wartime institutions are present and can withdraw their cooperation if the civilians feel that the social contract has been violated (Ar-jona 2014a, chap. 3; van Baalen 2020; Peclard and Mechoulan 2015, 20).9 The informal control of rebel combatants can translate into formal control, rebel governance groups have shown to sys-tematically punish transgressions of their insurgents in order to keep the support of the civilians (Arjona 2014a; Kenny 2010; Mosinger 2017). The National Resistance Army (NRA) in Uganda did not only use the wartime institutions to regulate and punish defiance of the civilian popula-tion but also of its own recruits (Kasfir 2005; Mampilly 2011; Tidemand 2013; Weinstein 2007). Rebel groups with rebel governance have incentives to establish clear rules and punish defiance of their combatants as it can hurt the social contract it upholds with the civilian population in the rebel-held area and could cost the rebels the support of the civilians if there is no incentive to comply (Arjona 2014a, 26-27; Mosinger 2017, 46).

In sum, these sections conclude that social order is established through a clear set of predictable rules stemming from the (often) implicit social contract upheld by both civilians and rebels, al-lowing for the enforcement of rules in a systematic and regulated way. Rebel governance can

(22)

formalize these rules through building wartime institutions. The carrots and sticks provided by the rebel leadership elicit civilian cooperation and compliance. This way rebel governance can create incentives to comply and can additionally strengthen mechanisms of control of the mem-bers of the rebel group itself. Through these incentives, rebel governance is often successful in gaining support as well as keeping this support over time (Arjona 2014a, 21-57). Therefore, the rebels are likely to sustain a prolonged presence amongst the civilian population, which in turn allows for the building of robust and reliable social ties and an increasing dependency on the rebel group. It should be noted that this thesis makes the assumption that the rebel leadership provides the incentives, and that civilian’s loyalty is directed at the leadership and not at individual actors within the rebel group. This assumption is not unfounded, since case studies substantiate this (Arjona 2014a; van Baalen 2020).10

3.3 Rebel Governance and Fragmentation

The previous part (section 3.2) showed ways in which rebel governance is often successful in eliciting civilian support. Through territorial control, rebel governance groups have high levels of integration in civilian communities. The positive and negative incentives imposed by a rebel sanc-tion a prolonged presence, and also elicit continued support for the rebel group. The beginning of this chapter concluded that rebel lieutenants require the loyalty of a support network in order to form a splinter (section 3.1). This section is aimed at explicitly connecting the previous parts of this chapter. Therefore, the following paragraphs argue how the functions and characteristics of governance are related to mitigating potential opportunities for rebel lieutenants to gather the support of civilian networks to form a potential splinter. Three main arguments are made: (1) the relative peace and services provided by the rebel group ‘buys’ the support of the civilians in favour of the rebel leadership, (2) higher levels of control deter defiance from civilians and potentially rebels, and (3) integration in civilian networks makes it difficult for mid-level commanders to gain loyalty due to a multitude of social ties and increasing social complexity.

First, governance provides positive incentives (the ‘carrots’) to civilians to support the rebel leadership. The investment of rebel governance groups in their interactions with civilians gener-ally enables them to gain the support of civilians and maintain their obedience over time (Arjona 2014a, 46-62). Rebel governance groups are more successful in creating support since civilians prefer protection from violence and security above all else. Additionally, rebel governance can sway loyalties or induce support by offering services (Arjona 2014a; Terpstra 2020). The pro-vision of security, other services, and relative peace to the civilians will likely mean that they favour obedience to the rebel leadership as opposed to supporting an internal rival that might not be able to provide such services. A splinter is unlikely to have the capacity to maintain the sup-ply of social services, meaning that supporting a midlevel commander would mean a decrease in service provision. In sum, it is assumed that civilians favour the relative peace and services rebel governance provides over the unknown (dis)order under the rebel lieutenant.

Second, rebel governance concerns itself with the creation of both mechanisms of control and social order among civilians (Arjona 2014a; Mampilly 2011). The regulation of civilians in the

(23)

rebel-held areas and governance institutions allow for the monitoring of non-compliance and en-ables intervention in e.g. intra-civilian disputes that can lead to civilian sub-cleavages which can be taken advantage of by the rebel commander (Mampilly 2011, 81). Higher levels of control through policing and the mitigation of disputes by judicial wartime institutions increase the level of control by the rebel leadership (Arjona 2014a). Mid-level commanders will have difficulties in gaining loyalty from civilians, as the civilians will fear repercussions from the rebel leadership (the ‘sticks’ or negative incentives). Moreover, a major deterrent for potential splinter groups is the fear that the rebel group from which the splinter stemmed may retaliate against them and execute attacks on civilians that supported the splinter formation. The increased control can go beyond civilians; rebel governance groups have incentives to force higher levels of control for their combatants since they face non-cooperation by civilians if the combatants break the social contract (Arjona 2014a, chap. 2). As such, rebel lieutenants will have difficulties in convinc-ing support networks (both civilians and rebel combatants alike) to support the formation of a potential splinter.

(24)

68). Therefore, the integration of a rebel group in civilian networks promotes structural integrity and reduces the likelihood of splintering (see e.g. Lilja 2012).

To conclude, these sections have argued that rebel governance groups are successful in both integrating with civilians and staying integrated due to the supply of services (positive incentives or ‘carrots’) or through the creation of high levels of control (negative incentives or ‘sticks’). The motivations for civilians to comply, growing interdependency between rebels and civilians, as well as the increasing complexity of social ties, make it difficult for mid-level commanders to sway the loyalties of the civilians (and arguably rebel combatants), and as such, the opportunity to splinter is lessened. Under these circumstances, the rebel lieutenants cannot be confident that a sizable portion of civilians will support the formation of a splinter and defect with them.

3.4 In Summary

How does rebel governance influence rebel fragmentation? Civilians matter for a rebel group’s (and potential nascent rebel group’s) survival since they can provide access to e.g. information or labour. Additionally, civilians can influence the loyalties of low-ranking combatants as they are often recruited from the civilian population and thus hold close ties to civilians. Rebel governance constitutes a unique interaction between civilians and rebel groups, seen both in the ability of the governance group to integrate into local communities and gather its support as well as through the systematic and regulatory control of the behaviour of civilians. This thesis argues that rebel gov-ernance (independent variable) affects the likelihood of rebel fragmentation (dependent variable). The incentives and increasing social complexity created through rebel governance decreases the opportunity for rebel lieutenants to find civilian support for their splinter and are thus less likely to be able to form a splinter. More specifically, this thesis argues that rebel governance shifts the cost-benefit for mid-commanders to form a splinter (causal mechanism), since the commanders cannot be confident that a portion of the support network will back the formation of their splinter. Furthermore, although not the main focus in the theoretical framework, rebel governance can also mitigate the ability of rebel lieutenants to recruit rebel combatants as a support base for the splin-ter, since the high levels of control and increasing complexity of social ties influence the ability to gain the loyalty of the lower-ranking rebel combatants. Deriving from this theoretical framework this thesis will explore the following hypothesis:

Hypothesis 1: Rebel governance decreases the likelihood of rebel fragmentation

The argument does not claim to explain all the occurrences of fragmentation. Rather, it is seen as an important factor that is assumed to contribute to explaining the variation of fragmentation. The relative strength of the argument in relation to alternative explanations is discussed in Chapter 6.

(25)

(26)

4. Methodology

In the previous chapter it was theoretically outlined how governance could potentially lead to a lower likelihood of fragmentation. In this chapter, the variables and methods used to test the hypothesis are outlined. The chapter is structured as follows; first, the main dataset is introduced as well as the data for the independent and dependent variable. Second, the control variables used in this study are explained and justified. Third, the chapter discusses the creation of a control group through propensity score matching. Last, the statistical models used in this thesis are presented and discussed.

4.1 The Dataset

The universe of cases included in this study are rebel governance groups between 1946-2012 as coded in the Non-State Actor (NSA) dataset (Cunningham, Gleditsch, and Salehyan 2009). The NSA dataset was chosen as base data since the data used to code the independent variable (rebel governance) derives its cases from the NSA dataset. The data used for the independent variable does not allow for the creation of a control group, the NSA dataset was needed as base data to be able to create this reference group, as discussed in section 4.2. The NSA dataset is an extension of the Uppsala Conflict Data Program (UCDP) Dyadic Dataset. The unit of analysis employed in this thesis is the rebel group. In order to achieve this unit of analysis, the NSA dataset was modified. The NSA dataset had double observations of several rebel groups (thus the same rebel group would appear in multiple rows). The NSA data ‘closed’ observation if they had a long period of inactivity (i.e. low conflict intensity). If the rebel group engaged in active combatant again in later years, after this period of inactivity, the rebel group was included in a new row of the NSA dataset. Therefore, to achieve the correct unit of analysis, any observations of the same rebel group that were included more than once, were re-captured within one observation, taking the maximum values. For example, the National Union for the Total Independence of Angola (UNITA) in Angola was registered in NSA in four different observations. In the new data, it is now reduced to one observation, with as start year the earliest observation of a rebel group found in NSA dataset, and as end year the latest observation found in NSA dataset. Since this thesis makes inferences about individual rebel groups and examines group-specific characteristics, it assumes that these characteristics (e.g. ethnic foundation) stay constant over time regardless of periods of inactivity. Therefore, keeping the double observations as initially coded in the NSA dataset would have achieved the wrong unit of analysis.

(27)

their short lifespan or because historical records have not sufficiently reported on the rebel groups due to their small size. Last, any groups with rebel governance11 but no territorial control were excluded since the effect of rebel governance without territorial control is unknown (to understand this effect it was tested separately, see appendix section D). This left the NSA dataset with 311 observations.

4.2 Independent Variable

The data on rebel governance is limited; overall five datasets have been published (Albert 2020a, Breslawski 2021, Heger and Jung 2017, Huang 2016, and Stewart 2018a). This thesis uses the Rebel Quasi-State Institutions (RQSI) dataset by Albert (2020a), since it codes individual rebel groups and captures both rebel institutions as well as service provision by rebels between 1946-2012. The other datasets are limited for the purpose of this study in several ways; Huang (2016) only coded major civil wars, Heger and Jung (2017) and Stewart (2018a) only code for service provision thus excluding other wartime institutions, and lastly, the Breslawski (2021) dataset is limited to a few rebel groups since it captures differences of rebel governance within areas of control by the same rebel group. Therefore, for the purpose of this study, the RQSI was selected since it codes a wide range of rebel institutions and services over a relatively large timeframe. Albert (2020a) uses the NSA to identify her universe of cases and thus is highly compatible to be merged back on the NSA. However, the RQSI dataset was not used as base data since it omitted all rebel groups unrelated to rebel governance, thus not allowing for the creation of a control group.

All institutions and services of the RQSI dataset are inclusive institutions, meaning that all civil-ians under rebel-rule are able to access them and the institutions are not limited to civilcivil-ians that e.g. share the ethnic background of the rebel group (Albert 2020b).12 The RQSI includes informa-tion on twenty-seven different types of rebel instituinforma-tions and services (Albert 2020b). However, in accordance with the territorial framework, this thesis is restricted to only include rebel institu-tions and services that are relevant and related to the theoretical framework. As such, groups that have indicators of rebel governance such as a media channel or a flag but did not build any other institutions are coded 0, as it does not show active engagement with civilians’ lives. Although there are twenty-seven different rebel institutions coded by Albert (2020a) only the ten wartime institutions that were deemed relevant to this study were included. As such, the dichotomous variable is based on the presence of any of these ten institutions. The ten included institutions are further explained below and in appendix section A. Furthermore, as the scope condition is territo-rial control, only rebel governance with territoterrito-rial control is captured in the independent variable. To capture rebel governance in general, the dichotomous variable of RebelGov was coded 1 where there was any rebel institution present and 0 when rebel institutions were absent.13 Furthermore, rebel governance was operationalized in a second way: a count variable that captures the number of institutions created by the rebel group (ranging from 0 to 10).

11. The cases were only excluded if they had any of the 10 institutions elaborated upon in section 4.2. 12. For discussions on differences between inclusive and exclusive governance see Stewart (2018a).

(28)

Additionally, rebel governance is operationalized in a third way; where the ten institutions are disaggregated into four different dichotomous variables. To so account for different dimensions of rebel governance: economic, service provision, political engagement, and security orientated in-stitutions. To be able to test the dimensions, an additional four variables were created to indicate if a certain dimension was present within the rebel governance structures. The dimensions are based on domains of rebel governance identified in previous literature namely: political organization, security, economic regulation, and service provision (van Baalen 2020; Kasfir 2005; Mampilly 2011). The following sections elaborate on how the dichotomous variables are composed. Each dimension is operationalized on the basis of two or three variables from the RQSI dataset.

First, economic engagement with civilians was based on the variables taxation and currency of the RQSI dataset. Taxation indicates that the rebel group regulates the influx of funds from civilians at the group level (Albert 2020b). As such, it captures if civilians in any way fund the rebel group. The funds must be directed to the rebel leadership, not individual combatants (Albert 2020b). It thus excludes extortion on an individual level and serves as an indicator of a larger policy or rule of the rebel group to gain income from the civilian population. Arguably, currency is not an intuitive indicator of the extent to which rebels engage with the lives of civilians and the circulation of this currency can be unsuccessful. However, the creation of a currency serves as a proxy for the economic engagement of the rebel governance group, given that it shows a more extensive level of economic interference and creates an economic system that is no longer depen-dent on the currency of the state as much. When successful, the imposition of a new currency creates another level of dependency of the civilian population on the rebel governance group.

Second, the variable political institutions was operationalized based on combining data from the variables Gov and OrgGov of the RQSI dataset. The variable Gov captures whether rebel groups create a system of government. This can be at a local but also at a national level (Albert 2020b). The Gov variable coded cases of rebel governing administration bodies or parallel gov-ernments that are set up at a national level but actively engage in controlling part of the civilian population in territory held by the rebels. Although this variable also captures rebel governments-in-exile, the political dimension was only coded 1 based on the Gov variable if the rebel group had at least a year where their system of government was either present as parallel government or at the local level. The second indicator OrgGov is coded when the rebel group organizes itself in departments similar to how states operate (Albert 2020b). This indicates the existence of internal legislature or functional departments e.g. for health or welfare distribution, but does not capture the existence of a separate political wing (Albert 2020b). Albert (2020b) only coded a presence of the political departments if the departments are aimed at issues related to organizing the lives of civilians within the area of control. Both variables serve as indicators for a political dimension of rebel governance since it captures the presence of political bodies and the extent of political organization in the rebel group.

(29)

protects civilians from external attacks (Albert 2020b). The last variable, justice, captures whether the rebel group has set-up a justice system that has jurisdiction over the civilians within the rebel group’s territory. This can include courts, prisons, or the enforcement of laws (Albert 2020b).

Fourth and last, service provision was captured through the variables education, health, and welfare/aid. Service provision is important to capture, as it arguably is the most influential in gaining popular support amongst the civilian population. Health and education are the most important indicators of service provision within rebel governance since both are desirable and are deemed greatly beneficial for civilians (Stewart 2018b). Albert (2020b) only coded a presence of education when it went beyond rebel propaganda and taught at least basic skills such as writing and reading. Welfare/aid is additionally included as it captures a variety of tangible goods given to civilians such as food or money.

To conclude this section, the independent variable is operationalized in several different ways. There are two variables that capture rebel governance in general; answering the questions (1) “is there any rebel governance?” (captured in a dichotomous variable), and (2) “how many institu-tions are build by the rebel group?”(captured in the count variable). Both of these variables are based on the ten variables explained within the specific dimensions. Furthermore, the four differ-ent dimensions of rebel governance are captured in separate dichotomous variables to see if the presence of specifically orientated institutions influences fragmentation more than others.14 The rebel governance variables thus allow this thesis to assess three different factors: (1) does the gen-eral presence of rebel governance influences fragmentation, (2) does the number of institutions matter for the likelihood of fragmentation, and (3) last is there a difference in how the dimensions of wartime institutions influence fragmentation.

4.3 Dependent Variable

Currently, datasets on rebel fragmentation either code the origin of rebel groups (responding to: did this rebel group split from another, instead of does this group have a splinter?) or only capture viablesplinters that caused enough battle deaths to be registered. Therefore the existing datasets do not capture what is measured in this thesis for two reasons: the scope of this thesis does not focus on the origin of the rebel group but rather on whether the rebel group splintered, there-fore, the first type of datasets cannot be used. Second, this thesis does not concern itself with the viability of the splinter but rather looks at if the rebel group splintered at all. As such, this thesis collected novel data to code the dependent variable, thereby coding rebel fragmentation manually. Although Cunningham (2006) has argued that groups that form weak splinters (which are excluded from current datasets) should not matter as splinters are not able to influence a war significantly due to their limited military capacity, these smaller groups are often able to use other strategies to influence the conflict, for example through obstructing peace processes (Rudloff and Findley 2016). As such, this thesis has decided to operationalize fragmentation more broadly and includes any splintering that has been reported.

(30)

To code rebel fragmentation news outlets, academic articles, and actor descriptions in databases such as the UCDP (2019)15and Stanford’s Mapping Militants (2018)16were consulted based on several search terms: split(s), fragment(ing), fragmentation, faction, splinter(s/ing), divisions, internal struggle, and breakaway group. The dependent variable is operationalized dichotomously (1 = it splintered, 0 = no splinter found). Fragmentation was only captured if it occurred within the lifespan registered by the NSA dataset. Splintering that occurred after or before the rebel group was active in the NSA dataset was excluded. This decision was made since the years before or after the rebel group was active in the NSA dataset the rebel group often was not yet a rebel group, but generally the groups were political parties that thus either become a rebel group (and thus then became active in NSA) or transformed into a political party post-conflict (and thus stopped being active in NSA). Since this thesis only makes inferences on rebel groups and not on political parties the years surrounding their inclusion in the NSA dataset are not considered as then splintering did not happen when the group was in the capacity of a rebel group but rather when it functioned as a political party.17 In addition, to make the dataset compatible to run a survival model the year of splintering was coded as well.

4.4 Control Variables

This thesis controls for several confounding variables, however, this thesis does not test alternative theories that are not considered as confounders (e.g. when it could only explain variation in the dependent or independent variable).

First, the lifespan of the rebel group influences the likelihood of fragmentation. Logically, the longer a group exists, the higher the likelihood of fragmentation at some point in their lifespan considering that they e.g. have to survive more pressures that could exacerbate a potential rise of splinters (Nagel and Doctor 2020). At the same time, rebel groups that exist longer will be more likely to implement rebel governance at some point, and groups that have rebel governance generally have a longer lifespan (as rebel governance arguably strengthens the groups making them more capable of withstanding repression). For this reason, a count variable lifespan was included; this variable was coded by counting the number of years from the first year the rebel group was active in the NSA dataset until its last year. Thus the unit of time is years. Note that, when a rebel group split in the same year as it became active, in the NSA dataset, the lifespan variable was given the value of 1. Lifespan is thus calculated from the beginning of the first active year (i.e. January 1st) to the end of the last active year (i.e. December 31st).

Second, to account for the goal or nature of the conflict the variable conflict type from the NSA dataset was coded in three different dichotomous variables (coding 1 for the presence, and 0 for the absence): ethnic conflict, communist, and secessionist. Previous studies of rebel fragmentation (Burch and Ochreiter 2020, Fjelde and Nilsson 2018) have argued that ethnic conflicts or commu-nist rebels are less likely to fragment since the social ties within the group forge more cohesive movements. Similarly, due to the ideological nature of communist groups, it is more likely that

15. Available through: https://ucdp.uu.se.

16. Available through: https://cisac.fsi.stanford.edu/mappingmilitants.

(31)

the group is highly structured around political beliefs, and as such, it will have strong internal cohesion (Fjelde and Nilsson 2018). In the case of rebel governance, rebel groups with ethnic ties or a communist orientation are more likely to implement rebel governance than groups that do not, since such rebel groups are aimed at creating a new social or political order (van Baalen 2020, 11; Huang 2016). In addition, a variable capturing whether the conflict was secessionist (or not) was included. Secessionist rebel groups are more likely to implement rebel governance since they aim to build their own state to increase their legitimacy (van Baalen 2020, 11; Huang 2016; Stewart 2018a). At the same time, secessionist groups tend to build more robust organizations, which reduces the likelihood of fragmentation (Mosinger, 2017).

Third, the control variable external support was included. Although the relation between exter-nal support and fragmentation is complex, it has been determined that it can both negatively and positively influence fragmentation (see Ives 2019, Mosinger 2018, Tamn 2016 or section 2.1.1). External support does not promote the implementation of governance structures by rebel groups since groups that receive resources from outside are less reliant on civilian support to sustain themselves. To account for either process, the dichotomous variable external support is coded 1 if the rebel groups either explicitly or allegedly received external support based on information in the NSA variables on external support (Cunningham, Gleditsch, and Salehyan 2012).

Fourth, a variable capturing lootable resources based on the Rebel Contraband dataset (RCD)18 was included (Walsh 2019). If the group at any time used extortion to get lootable resources this variable was coded 1, and 0 if not. Extortion means that the group used violence or the threat of violence to either directly or indirectly profit from a lootable resource (Walsh et al. 2018). Lootable resources are considered resources that are easily accessible and do not require complex or expensive material in order to gain access. Based on the conceptualization of lootable resources by Findley and Marineau (2016), such resources include gold, gems, drugs, and alluvial diamonds (diamonds that are on the surface and therefore easily accessed). Rebel fragmentation is more likely to occur in areas with lootable resources since the pay-off for fragmentation will be higher if the splinter is successful (Fjelde and Nilsson 2018). Rebel governance is less likely in areas with lootable resources since the rebel groups do not require the support of civilians for income and thus tends to use higher levels of violence against civilians instead of governance (Fjelde and Nilsson 2018; Mosinger 2018).

Lastly, this thesis considers territorial control to be an important confounding variable. The re-lation between territorial control and fragmentation remains unclear; studies have found territorial control by rebel groups generally makes them less likely to fragment (Burch and Ochreiter 2020) or more likely to fragment (Mosinger 2018). Although rebel governance can be both with and without territorial control, arguably groups that have territorial control are more likely to imple-ment rebel governance, given that maintaining control over the territory would be eased through governance (Arjona 2014a). Therefore, since a scope condition of rebel governance in this thesis is territorial control, there is no variation on this variable in the treatment group (all observations of rebel governance will automatically have a 1 coded for territorial control). Therefore, the con-trol variable territorial concon-trol was included in an additional regression found in appendix section

(32)

D. There, the scope condition was omitted to understand the influence of territorial control on the found effect.

4.5 The Control Group

Since the data for the dependent variable is collected manually, it was deemed unfeasible to collect data for all the rebel groups in the NSA dataset. Therefore, to test the hypothesis a control group of the same size as the treatment group (n = 77) was created. The treatment group constitutes of rebel groups that are coded with a 1 for rebel governance; the control group does not have rebel governance and thus had the value of 0 for this variable. Given that the control group does not have the treatment (governance), it functions as a comparison group against the treatment group that does have governance.

Although it is possible to randomly assign a control group, a matching algorithm can be instru-mental in reducing bias (Rosenbaum and Rubin 1983; Stuart and Rubin 2007). Previous studies have shown that control groups with very different covariate distributions can lead to misleading results in regressions (see Stuart and Rubin 2007). As such, matching creates a control group with non-treatment observations that have a similar balance on the control variables. Selection bias is thus minimized, since the control group will not systematically differ from the treatment group. As such, the effect of the independent variable governance can be easier isolated since confound-ing is mitigated before the regressions are run. Although randomization can also reduce bias if the sample size is large enough, matching algorithms can guarantee a similar distribution, thereby reducing potential confounding. Furthermore, the matching algorithm allows for more accurate results when confounders cannot be included in a model, as is the case for the Kaplan-Meier method discussed later on.

The matching algorithm this study employs is propensity score matching; propensity scores are the conditional probability of getting ‘the treatment’ (Rosenbaum and Rubin 1983). Thus propensity score matching creates a control group based on the selected control variables that are as similar to the treatment group as possible, and thus theoretically would have had the same or a similar likelihood to have developed the treatment (here the treatment is rebel governance). The control group has a score of 0 for the independent dichotomous variable rebel governance; the treatment group has a score of 1 (indicating that there is rebel governance). Propensity score matching matches every individual observation of the treatment group with its “own” observation from the data without treatment.

Should We Stay or Should We Go?