Control and Uncertainty in the Delegation of War
A Principal-Agent Explanation of Interrebel Relations
Shawn J. Davies
Shawndavies1992@gmail.com
Master Thesis in Peace and Conflict Studies
Advisor: Erika Forsberg
SPRING 2018
Department of Peace and Conflict Research Uppsala University
i Acknowledgements: First and foremost, I want to thank my advisor at the Department of Peace and Conflict research, Erika Forsberg, for her invaluable help and advice throughout the process. Further thanks go to those friends and classmates whom have taken the time to read and provide feedback on earlier drafts, which has greatly improved the final output. Joel, Therese, Siena, Lovisa and Simon, thank you all for your time and help! Maybe most importantly, thank you all that have made the time writing this thesis a rewarding and, for the most part, fun experience, whether you have done so with wine, memes or encouragement.
ii
Abstract
External support has previously been found to increase both interrebel alliances and interrebel conflict. This thesis uses a modified principal-agent theory to bring these findings together in a common theoretical framework, arguing that external support leads to interrebel alliances when the sponsor’s leverage is high, and to interrebel warfare when the sponsor’s leverage is low. Using data on multiparty conflicts from 1975-2010, it aims to answer why external state support has increased interrebel warfare in some cases and interrebel alliances in other. It finds that the effect of external support isn’t unidirectional on interrebel alliances or interrebel warfare. This thesis finds that when controlling for forms of support that allows a sponsor to monitor the rebel group, external support is positively correlated with interrebel warfare and negatively with interrebel alliances. It also finds that the monitoring capacity of the sponsor decreases interrebel warfare, whilst the effects on interrebel alliances are inconsistent.
Further, strong rebel groups, groups active in areas of drug cultivation and groups that share ideational ties with other groups in the same conflict, are found to be more likely to engage in interrebel warfare, and less likely to engage in interrebel alliances.
Key words: External Support, Sponsor, Interrebel Warfare, Interrebel Alliance, Leverage, Principal-Agent, Monitoring
iii
Table of Contents
List of Figures and Tables ... iv
1. INTRODUCTION ... 1
1.1 Previous Research on External Support ... 3
1.2 Previous Research on Interrebel Relations ... 4
2. PRINCIPAL-AGENT THEORY FOR EXTERNAL SUPPORT ... 6
2.1 Sponsor Leverage and the Conflict Environment ... 7
2.2 High Sponsor Leverage and Interrebel Alliances ... 8
2.3 Low Sponsor Leverage and Interrebel Warfare ... 10
2.4 Hypotheses ... 12
3. RESEARCH DESIGN ... 15
3.1 Data ... 16
3.1.1 Interrebel Alliances ... 17
3.1.2 Interrebel Warfare... 18
3.1.3 Independent Variable ... 18
3.1.4 Control Variables ... 21
4. RESULTS ... 23
4.1 The Effect of External Support ... 24
4.2 Interrebel Alliances and Sponsor Leverage ... 27
4.3 Interrebel Warfare and Sponsor Leverage ... 31
4.4 Hypothesis tests ... 34
4.5 Robustness Checks ... 39
4.5.1 Alternative Models ... 39
5. ANALYSIS ... 42
5.1 Implications for the Theory ... 42
5.2 Further Observations... 45
5.3 Alternative Explanations ... 47
5.4 Limitations ... 48
6. CONCLUSION ... 49
List of References ... 52
Apendix ... i
1. Multicollinearity ... i
2. Influential Observations ...ii
3. Alternative Models ... iii
iv
LIST OF FIGURES AND TABLES
Figure 1: Causal framework for sponsor leverage and interrebel relations ... 12
Figure 2: Marginal Effects on Interrebel Alliances for Troop Size and Monitoring ... 31
Figure 3: Marginal Effects on Interrebel Warfare for Troop Size and Monitoring ... 34
Table 1: Summary Statistics for Variables of Interest ... 21
Table 2: Summary Statistics for Control Variables ... 23
Table 3: Logit Regression on the Effect of External Support ... 25
Predicted Probabilities of Interrebel Alliances and External Support ... 27
Predicted Probabilities of Interrebel Warfare and External Support ... 27
Table 4: Logit Regression on Interrebel Alliances and Sponsor Leverage ... 29
Predicted Probabilities of Interrebel Alliances ... 30
Table 5: Logit Regression on Interrebel Warfare and Sponsor Leverage ... 32
Predicted Probabilities of Interrebel Warfare ... 33
Table 6: Logit Regression for Alternative Model ... 41
1
1. INTRODUCTION
The ongoing Syrian civil war has seen the opposition to the Assad-regime engaging in coalitions and infighting in ever changing compositions. Analysts have observed how external support caused the opposition to fragment and weaken, as groups aligned themselves with different external supporters (Baylouny & Mullins 2017; Lynch 2016). Considering that external sponsors are generally assumed to be interested in weakening their state rivals or increasing their regional influence through external support,1 a weak and fragmented opposition appears to contradict the interests of the external sponsor (Akcinaroglu & Radziszewski 2005; Byman et al. 2001; Maoz & San-Akca 2012). A paradox emerges where external support causes dynamics that runs counter to the interest of the sponsor.
Previous research has found two contradicting empirical results connecting external support to interrebel relations. The first is that external support increases the likelihood of a rebel group engaging in interrebel warfare as the groups capacity to fight expand (Fjelde &
Nilsson 2012; Burton 2015). The second set of findings show that external support improves the prospects of interrebel cooperation (Bapat & Bond 2012; Popovic 2017). This thesis aims to address this contradiction by providing a common framework in which some forms of external support encourages cooperation, whilst other lead to conflict.
Faced with empirical and theoretical support for external support both increasing interrebel warfare and interrebel alliances, this thesis asks, why does external state support for rebel groups increase interrebel warfare in some cases while fostering interrebel alliances in others? To answer this question, this thesis introduces a novel theoretical framework, built on the logic of principal-agent theory, which explains why some forms of external support increase interrebel alliances and other lead to interrebel warfare.
This thesis argues that external sponsors with leverage over a rebel group increases the likelihood of that group entering in interrebel alliances, whilst the absence of leverage encourages interrebel warfare. It argues that when the sponsor can impose considerable costs upon the rebel group it has a large degree of leverage, which causes the autonomy of the rebel movement to decrease. Sponsors want to limit rebel autonomy in order to direct their activities towards the goals of the sponsor. Because sponsorship generally is targeting the government, sponsors with a high degree of leverage over a rebel group is expected to make it join forces
1 External support and sponsorship are used interchangeably to denote the provision of sanctuary, financial assistance, logistics and military support, including, but not limited to, troops to assist in an ongoing conflict, from a foreign state to a rebel group active in another country’s territory.
2 with other groups to fight the target government. When sponsors have limited ability to monitor and sanction an agent’s behavior, and when the group is less dependent upon the sponsor’s resources, its autonomy increases. As the rebel group is able to act more autonomously, it will prioritize its own interests above those of the sponsor. In line with previous research, rebel groups are assumed to seek to maximize individual gains, rather than collective gains, and face incentives to hinder any partner from becoming too powerful (Christia 2012; Fjelde & Nilsson 2012). External support increases the capacity of rebels to engage in interrebel warfare, while simultaneously diminishing its need to cooperate with other groups to defeat the government (Akcinaroglu 2012; Burton 2015). Therefore, when the sponsor is unable to apply leverage over the rebel group, the provision of external support makes the rebels more likely to engage in interrebel warfare.
I test this argument through statistical analysis using dataset which contains rebel-year observations of all groups involved in multiparty conflicts in the period 1975-2010. It contains observations for interrebel alliances, interrebel warfare and external support provisions. It combines data from the Dangerous Companions dataset, the UCDP Non-State Conflict dataset, and Popovic’s replication data, amongst others (San-Akca 2016; 2009; Sundberg, Eck & Kreutz 2012; Allansson, Melander & Themnér 2017; Popovic 2017). The leverage of the sponsor is proxied as its monitoring capacity, the strength of the rebel group, the rebel groups access to alternative resources, and the presence of multiple sponsors. I find that when controlling for forms of support that allows a sponsor to monitor a rebel group, external support is positively correlated with interrebel warfare, and negatively with interrebel alliances. I also find weak support for the presence of monitoring capacity correlating with a decrease in interrebel warfare, and an increase in interrebel alliances. Further findings indicate that stronger rebel groups, groups active in areas of drug cultivation, and groups that share ideational ties with other groups in the same conflict engage more in interrebel warfare, and less in interrebel alliance.
Next section will review the previous findings from the field of external support and how external support affects interrebel relations, before presenting other important determinants for interrebel relations. Thereafter, I present the principal-agent theory shortly, and develop it to explain why external support increases interrebel alliances in some cases, and lead to interrebel warfare in other. In the third chapter, I present the dataset which was developed to test the argument, after the results of a series of logit regressions are presented in depth. Finally, I discuss the findings and their implications for further theory development and future research.
3
1.1 Previous Research on External Support
The sponsor is the foreign state which supplies external support in the form of financial or logistical aid, troops, weapons, safe havens, training or operational management to a rebel group active in another states territory.2 The target is the state which is subject to violence from the non-state group that receives support from a sponsor. Much of the attention to external support has gone to determine when and why states support rebel groups. Superpowers, bordering countries and rivals are the most common supporters of rebel groups (Byman et al.
2001). States have been found to support rebels to hurt their rivals, for geopolitical benefits and to increase their regional influence (Akcinaroglu & Radziszewski 2005; Byman et al. 2001;
Maoz & San-Akca 2012). Others have found ethnic, religious and ideological ties to be important (San-Akca 2016; Salehyan, Gleditsch & Cunningham 2011).
Considerable research also exists on the effect of external support on conflict dynamics.
It has been found that external support prolongs conflict by introducing more actors and decreasing the cost of fighting, especially if both parties receives support or the sponsor has an agenda independent from that of either conflicting party (Regan 2002; Cunningham 2010). It makes negotiated settlements harder to reach and increases the risk of interstate conflicts (Cunningham 2010; Salehyan 2009). It also has negative effects for post-conflict democratization and increases the risk of conflict recurrence (Colaresi 2014; Karlén 2017a).
Military intervention on the side of a rebel group or a relatively weak government increases their likelihood of victory (Sullivan & Karreth 2015). Though, this effect might be dependent upon timing; early support for a rebel group increases its likelihood of victory, while states are mostly benefited when the conflict has become protracted (Jones 2017). External support increases the risk of conflict outbreak and can help nascent rebel groups, which have not yet built up resilience to state repression, mobilize beyond government reach and to have access to resources beyond its capacity to mobilize on its own (Regan & Meachum 2014; Salehyan 2009).
While important, the weakness with this research has been its tendency to treat the rebel opposition as a unitary actor, or only determining the state level effects of external support.
Some rather limited research has looked at how external support affects rebel groups. The effect of external sponsors on rebel groups is debated and appears to be able to cause either increased
2 The same terms are used in the literature for actions supporting a state involved in a civil war, but this analysis limits itself to exclusively focus on support provided to rebel groups. It is also limited to including only state support for rebels, not rebel groups supporting each other.
4 cohesion or fragmentation, by affecting the internal balance of power within rebel groups (Tamm 2016; Olson Lounsbery 2016; Baylouny & Mullins 2017). Fragmentation in turn also negatively affect the sponsors control of the rebel group, increases rebel engagement in interrebel fighting, and violence against civilians (Popovic 2015; Cunningham, Bakke &
Seymour 2012; Bakke, Cunningham & Seymour 2012). Sponsorship involves an investment, and the danger of rebel defection makes sponsor opt for groups that share their preferences, which is more common amongst groups sharing ideational ties (Salehyan 2010; San-Akca 2016;
Byman & Kreps 2010). Some findings indicate that rebel groups might at times adopt ideologies to attract sponsors (Walter 2017; Baylouny & Mullins 2017). Support for rebel groups also increases their use of violence against civilians as the group becomes less dependent on their local constituency, though democratic states have been shown to restrain their agent if there are not multiple sponsors active (Salehyan, Siroky & Wood 2014; Wood 2010).
Other findings relate to how external support can provide help for groups to credibly commit to an alliance by promising continued support for cooperation and punishment for defection. External support can decrease uncertainty between rebel groups and increase the benefit of an alliance (Salehyan 2010; Bapat & Bond 2012). Popovic (2017) finds support for an increase in interrebel alliances from sharing the same external sponsor, though notes that this can be dependent on the sponsors capacity to monitor and sanction non-compliance.
External support can also affect rebel capabilities and incentives, which might motivate them to engage in fighting with other groups (Fjelde & Nilsson 20012; Burton 2015). External support can also encourage competition between groups, similar to competition over lootable resources, causing groups to fight to control the inflow of external resources (Baylouny &
Mullins 2017, 3).
1.2 Previous Research on Interrebel Relations
Previous research has primarily viewed non-state actors through the lens of a single, unitary rebel group in a dyadic conflict with the government. Conflicts are more complex than this, and non-state actors often spend as much time fighting each other as they do fighting the government (Cunningham, Gleditsch, & Salehyan 2009; Wood & Kathman 2015). In recent years, a research field is emerging to address this complexity by including interactions between non-state actors. The field now covers intrarebel fractionalization, interrebel alliances and interrebel competition (Bencherif & Campana 2017; Bapat & Bond 2012; Fjelde & Nilsson
5 2012; Pischedda 2015; Walther, Leuprecht, & Skillicorn 2017). In multiparty civil wars, lone rebel groups are often unable to enforce their demands upon the government. Such groups can seek to improve their material position by taking resources from other groups, and thus over time gather sufficient strength to enforce concessions (Fjelde & Nilsson 2012; Cunningham, Bakke, & Seymour 2012). Otherwise, they can join in alliances with other rebel groups, so that their combined strength is sufficient to gain concessions (Christia 2012).
Multiparty civil wars are complex phenomena where rebel groups are involved in “dual contests” against the host government and against other rebel groups (Cunningham, Bakke &
Seymour 2012). Rebel groups fight each other more in territories beyond government control, where the state has limited control, in areas with drug cultivation, and when there is a power imbalance between a group and other rebel groups (Fjelde & Nilsson 2012). Competition for resources, recruits, territory, authority and concessions have been identified as major motivations for interrebel warfare (Piscedda 2015; Fjelde & Nilsson 2012; Wood & Kathman 2015; von Uexkull & Pettersson 2013). Fighting other groups can also be a means of signaling resolve and relevance for rebel groups, forcing observers to alter their perception of the group (Salverda 2017).More rebel groups entail increased competition over limited resources and a decreasing possibility of reaching a deal mutually acceptable to all parties, as the bargained for concessions need to be divided to more parties (Wood & Kathman 2015). Interrebel warfare is defined as the use of violence between two formally organized non-state armed groups that have an incompatibility with the state. As such, it includes only rebel groups, not communal or ethnic groups, or criminal networks.
Sometimes, rebel groups join in alliances to jointly fight against the government.
Interrebel alliances are defined, following Bapat and Bond (2012, 811), as resource-sharing or tactical co-ordination between two rebel groups (i.e. sharing weapons or personnel, conducting joint attacks or communiqués). Allied rebel groups are less likely to end conflict through a negotiated settlement and are less likely to be defeated by the government (Akcinaroglu 2012).
Yet, while some alliances last long, others regularly break down as the involved groups struggle to achieve a “minimum winning coalition” and to hinder any group from becoming too powerful (Christia 2012, 21). Groups at times attack their allies or even switch sides to the government (Nygård & Weintraub 2014). Power relations between groups can therefore determine both the breakdown of interrebel alliances and be a driver for interrebel violence (Christia 2012; Burton 2015; Nygård & Weintraub 2014). Otto (2017) also found that armed groups switch sides as the number of other armed groups with the same alignment increases and when the state is weak.
6 Despite these studies, the field is still nascent with major gaps. While there is a growing literature on the conditions during which rebel groups ally or fight each other, it does little, with a few exceptions, to explain the interaction between external support and interrebel relations.
The findings thus far have however been interesting, and contradictory. External support has been shown to increase interrebel warfare (Fjelde & Nilsson 2012; Burton 2015). Yet, external support also increases the probability of interrebel alliances (Bapat & Bond 2012; Popovic 2017). Below, a theoretical framework is developed to explain these contradictory findings.
2. PRINCIPAL-AGENT THEORY FOR EXTERNAL SUPPORT
This chapter first introduces the principal-agent theory and how it applies to the sponsor-rebel relationship. Second, this theory is developed through the introduction of leverage, which is thereafter argued to determine whether external support lead rebel groups to engage in alliances or warfare with other groups. Finally, the empirically testable hypotheses, based on the prior discussion, are introduced.
Many of the theorists on external support view the sponsor-rebel relationship in terms of a principal-agent framework, in which the principal (sponsor) delegates to its agent (rebel group) to act on its behalf within a conflict (Byman & Kreps 2010; Karlén 2017a; Salehyan 2010; Salehyan, Gleditsch & Cunningham 2011; Baylouny & Mullins 2017). In line with previous research on external support, this paper adopts a principal-agent framework to explain the effects of external support on rebel behavior, applying it to interrebel relations. In principle, delegation entails that the agent sets aside its own interests, and acts as would the principal in the given situation (Byman & Kreps 2010, 6). The benefit of a sponsor for the rebel group is the resources it brings, but it also limits the autonomy of the rebel group (Salehyan 2010). Rebel groups are therefore interested in getting the resources that sponsorship brings, whilst minimizing the cost in autonomy loss. Principal-agent relationships are therefore fraught with tensions.
There is almost always some conflict between the interests of those who delegate authority (principals) and agents to whom they delegate it. Agents behave opportunistically, pursuing their own interests subject only to the constraints imposed by their relationship with the principal.(Kiewiet & McCubbins 1991, 5).
Such opportunistic behavior is called agency slack (also called agency loss or, for some types, moral hazard). Agency slack is when agents pursue their own interests in ways that are
7 inconsistent with the preferences of the principal, whom has delegated to the agent to act within the conflict on their behalf (Salehyan 2010). Such agency slack can take many forms. For instance, the Abu Nidal Organisation (ANO) claimed responsibility for attacks that embarrassed its sponsor, Syria, diplomatically. In a further instance ANO shirked from missions that Syria wanted, preferring to target civilians, rather than Israeli soldiers, as this carried less risks (Byman & Kreps 2010, 7). This thesis models interrebel warfare as a form of agency slack, based on a theoretical model, developed below, which theorizes that state sponsors encourage rebel groups to cooperate against the target state. The next section introduces the novel concept of leverage in the sponsor-rebel relationship and develops on the setting of sponsor-rebel interaction, and how this affects the leverage of the sponsor. Thereafter the effects of leverage on rebel group behavior is developed.
2.1 Sponsor Leverage and the Conflict Environment
Leverage is defined as the capacity of the sponsor to control its agent by imposing costs on the agent for non-compliance to the interests of the sponsor. Rebel groups are expected to be opportunistic and act according to its own interest, even when these contradict those of the sponsor, when the sponsor has insufficient leverage to constrain it. The principal must be able to credibly threaten sanctions against an agent to hinder agency losses (Salehyan 2010; Popovic 2015). The capacity of the principal to impose costs upon a rebel-agent is often limited and costly, as it entails sacrificing some of the benefits of delegation; the sponsor requires a presence with the group, which undermines plausible deniability, and it limits the rebel groups autonomy, which can undermine its efficiency (Byman & Kreps 2010). There are several more factors which makes agency problems severe, from the perspective of the principal, in the case of state sponsorship of rebel groups.
First, there are often considerable information asymmetries between rebels and their sponsors, as conflict zones do not lend themselves easily to monitoring (Byman & Kreps 2010;
Nygård & Weintraub 2014). This limits the leverage of the sponsor, as it is unable to determine whether the group it sponsors is loyal or not. Second, options through which to punish a rebel group can be limited, unless the sponsor is willing to get more involved itself. Such punishment is often limited to the withdrawal of future benefits, or, more effectively, by transferring funds to a rival organization (Byman & Kreps 2010; Popovic 2017). While costly, the withdrawal of future benefits might not be devastating for a rebel organization, especially if it has access to other sources of income, or if the gains of acting contrary to the interest of the principal are large.
8 Third, the cost of cancelling support can in some cases be higher than retaining support to a rebel group, even if it is acting independently. While it is more intuitively to think that the rebels would become dependent on a sponsor providing it with vital resources, a sponsor also has considerable interest in the agent. Often a sponsor supports a rebel group to increase its regional influence, to hurt a target government, to spread a certain ideology or due to shared ideational ties with the rebel group (Akcinaroglu & Radziszewski 2005; Salehyan, Gleditsch &
Cunningham 2011; San-Akca 2016). Cancelling support for a group without any other group to delegate the job to cancels these benefits, and wastes resources already spent.
The principal in sponsor-rebel relations often have few alternative agents to delegate to. When the principal-agent theory is used in its classic field of economics, a shareholder (principal) can invest elsewhere if the manager (agent) exploits them (Kiewiet & McCubbins 1991). In conflict zones, on the other hand, sponsors often have few suitable agents available if one exploits them. For example, the USA has had considerable difficulty finding any major rebel group that could credibly claim to share their interests in Syria, because of the unpopularity of American foreign policy in Syria and American distrust of active groups. Fear of Islamists led the USA to adopt strict vetting for groups they supported, which led most groups to prefer to rely on other sponsors (Lynch 2016, 30-32).
In some cases, the principal might also be gaining domestic benefits from supporting a specific agent. Cancelling support for the agent, even if it misbehaves, might then be costlier than retaining it. For example, Syria had difficulty controlling Fatah, the Palestinian movement, as it was well perceived by the Assad-regime’s key Arab nationalist constituency. Fatah was capable of incurring costs upon the regime if it provided insufficient support, by delegitimizing its claim to fight for the Arab nationalist cause (Byman & Kreps 2010, 5).
All these factors weaken sponsors’ leverage over their agents, which allows agents to act according to their individual interests, thus imposing agency costs on the principal without consequences for themselves. The next section discusses the effect of sponsor leverage on rebel behavior in the ideal form, when leverage is high. Thereafter the effect of external support when leverage is low is developed.
2.2 High Sponsor Leverage and Interrebel Alliances
When the principal has a large degree of leverage, i.e. the principal can impose considerable costs upon the agent, the autonomy of the rebel movement decreases, and it has to act according to the wishes of the sponsor. This is how the literature on external support usually envisions the relationship, with the principal delegating authority to a loyal proxy (Salehyan 2010; Baylouny
9
& Mullins 2017; Karlén 2017a). Sponsorship of rebels is often viewed as a way for the sponsor to indirectly fight its rivals, thus implicitly expecting the rebel group to adopt the enemies of their sponsor and act against them on behalf of their sponsor (Maoz & San-Akca 2012; 2016;
Akcinaroglu & Radziszewski 2005). From this perspective agents are expected to be highly responsive to the principals wishes, and to act on their behalf. There is good reason to expect sponsors to be interested in encouraging rebel cooperation, as this increases the sponsors leverage by being able to shift resources to groups that perform better, decreases the sponsors costs by making the agents monitor each other, and increases the probability of rebel victory (Popovic 2017; Byman & Kreps 2010; Akcinaroglu 2012).
While external support is a cheaper option than direct intervention, it is still a costly act, which not only entails financial costs, but also risks causing interstate conflicts as the target state retaliates (Salehyan 2009). It is also likely to entail diplomatic and reputational costs for a state, as sponsorship of rebels is an intervention into another state’s domestic affairs, and thus violates international norms on state sovereignty (San-Akca 2009; Salehyan, Gleditsch &
Cunningham 2011). Sponsoring rebel groups is thus not an action made lightly, which implies that sponsors should have specifics goals when getting involved. It would thus be reasonable to suppose that the sponsor would attempt to pool the resources available in the conflict into an alliance to achieve this goal, not having the resources squandered on interrebel fighting. Foreign sponsors are also seldom interested in their agent spending resources on interrebel warfare, as the resources that a group can gain by attacking other rebels already are in the hands of a group opposed to the target government.3 Therefore, this paper expects increased sponsor leverage to increase rebel propensity to join alliances and decrease the likelihood of them engaging in interrebel warfare. Sponsors do this by imposing costs for groups which defect from the sponsors goals by fighting other groups or shirk from their alliances.
To summarize, the above outlined theory leads to the expectation that influential sponsors can, and want to, limit rebel autonomy, in order to direct their activities towards the goals of the sponsor. Because sponsorship generally is targeting the government, sponsors with a high degree of leverage over a rebel group is expected to make it join forces with other groups to fight the target government. Five hypotheses, aimed at capturing different aspects of sponsor leverage, will be presented further down.
3 There are a few cases in which foreign sponsors have been supporting groups to target other rebel groups, most notably the anti-IS coalition, but they hardly represent the norm. In most cases where foreign supporters target other rebel groups they do so via support for the government, rather than other rebel groups.
10
2.3 Low Sponsor Leverage and Interrebel Warfare
Proxy organizations are seldom as loyal to their sponsors as researchers and sponsors assume:
without accountability mechanisms the agents pay lip service to the principal but act as they please (Baylouny & Mullins 2017). This shouldn’t be surprising, as rebel groups are organizations with their own interests and goals, which seldom perfectly align with those of the principal. While principals might attempt to control agents via monitoring and sanctioning mechanisms, monitoring is often difficult and costly. Unless a principal invests heavily in monitoring, often requiring it to get more involved in the conflict, it will often have to rely on the information its agents provide and information from third parties (news agencies and NGOs). Agents therefore tend to have an information advantage over the principal, giving them opportunities to shirk (Byman & Kreps 2010; Bapat & Bond 2012).
The principal might have limited cost-free means of punishment as well, having invested resources into a movement to secure influence (Salehyan 2010). Ending support for the rebel group will end the benefits the state derived from supporting the group in the first place, whether this be regional influence or helping a group sharing ideational affinity. Ending support might also significantly sour relations with the rebel group, and can even cause a group to turn against the sponsor. For example, the Kashmiri militant group Jaish-e-Mohammad attempted to murder their former sponsor, Pakistani President Pervez Musharraf, when relations between the two broke down (Popovic 2015). Thus, a principal with limited alternatives should have increased difficulty in controlling its agent, as the agent will know that the principal is unlikely to alter its support over minor issues.
When principals have limited ability to monitor and sanction an agent’s behavior, and when the group is less dependent upon the sponsor’s resources, its autonomy increases. As the rebel group is able to act more autonomously, it will prioritize its own interests above those of the sponsor. Based on the literature on interrebel warfare, it is assumed that rebel groups have two overarching interests. First, to maximize political concessions, and, second, to maximize the material spoils that the leader can distribute (Fjelde & Nilsson 2012). Therefore, rebel groups aim to achieve a “minimum winning coalition” to minimize the concessions and spoils they have to share with other groups. As the power relations between groups shift it creates “a process of constant defection, alliance reconfiguration, and group fractionalization” (Christia 2012, 21 & 32). This makes rebel cooperation fraught with commitment problems, as neither group can credibly commit not to strike at their allies when they are engaged with the government. All groups seek to maximize individual gains, rather than collective gains, and face incentives to hinder any partner from becoming too powerful (Christia 2012). For instance,
11 if a group is successful against the government, or engaged in negotiations they are more likely to be engaged in fighting by other groups (Salverda 2017, 34). Knowing that every group in the alliance faces considerable commitment problems makes cooperation difficult between rebel groups. Foreign sponsors can help groups overcome these problems by functioning as guarantors for cooperation between rebel groups, helping them to overcome their commitment problems, if it can credibly monitor and sanction non-compliance (Bapat & Bond 2012).
Popovic (2017) argues that a sponsor that supports only one group in the dyad can worsen relations as the balance of power between them shifts, where the groups receiving support grows stronger while the one not receiving support might perceive the need to act preemptively. External support increases the capacity of rebels to engage in interrebel warfare, while simultaneously diminishing the number of allies they require to win. Based on this, the provision of external support when the sponsor is unable to apply leverage over the rebel group, makes the rebels increasingly interested in engaging in interrebel warfare. The effect of external support is thus expected to be dependent upon the leverage within the sponsor-rebel relationship. When sponsor leverage decreases rebel groups are expected to act increasingly according to their individual interests, and therefore compete with other rebel groups to control the inflow of external resources. Less dependent rebels thus engage more in interrebel warfare.
To summarize, this thesis uses the concept of leverage to explain why external support sometimes increases interrebel alliances, while in others it increases interrebel warfare. This thesis argues that sponsors have an interest in facilitating their agents to pool their resources into interrebel alliances and in hindering their agent to waste resources in interrebel warfare.
Rebel groups have largely opposite interest to their sponsors in this regard, as they want to minimize any coalition they join, while they can gain much by eliminating rival organizations.
In line with the presented arguments, the autonomy of a rebel group is not always decreased because of external support. When the sponsor lacks leverage, external support creates incentives for a group to engage in interrebel warfare, as a form of agency slack. In the following section several hypotheses are developed, each aimed at capturing different aspects of sponsor leverage.
12 Figure 1. Causal framework of how sponsors leverage over rebel groups affect interrebel relations
High
Low
2.4 Hypotheses
In this section, empirically testable hypotheses are developed to test the theory. Each hypothesis repeated for both interrebel warfare and interrebel alliances, but in opposite directions, as the effect of each is expected to be the opposite for interrebel alliances and interrebel warfare. A factor which encourages a group to engage in cooperation should not simultaneously be able to encourage a group to engage in conflict. Whilst hypothesis 1 relates to the effect of external support, the four further hypotheses relates to how leverage affects interrebel relations in the presence of external support. They aim to capture four aspects of sponsor leverage, namely the presence of multiple sponsors, the monitoring capacity of the sponsor, the strength of the rebel group, and the availability of alternative sources of income.
The first hypothesis of this paper aims at capturing the effect of external support on interrebel relations in relations to the theory. This paper has argued that the effect of external support on interrebel relations is dependent on the leverage within the principal-agent relationship. High leverage allows the sponsor to focus the efforts of the rebel group, whilst low leverage creates the incentives for the group to act according to its own interests. In situations with low leverage, external support increases the capabilities of a rebel group, while decreasing its need to cooperate with other groups, thus encouraging rebel groups to engage in interrebel warfare. This implies that external support devoid of leverage decreases the likelihood of a group engaging in interrebel alliances whilst it increases interrebel warfare.
H1a: External support correlates negatively with interrebel alliances, when sponsor leverage is low.
Sponsor’s Leverage over Rebel Group
Interrebel Alliances
Increased Rebel Autonomy Decreased Rebel Autonomy
Increased Interrebel Competition for External Resources
Interrebel Warfare Increased Responsiveness
to Sponsor’s Interests
13 H1b: External support correlates positively with interrebel warfare, when sponsor leverage is low.
Multiple principals can create a heterogeneity in sponsor preferences, as well as decrease agent dependence on a single sponsor, both of which decrease the sponsors leverage over the agent (Salehyan 2010; Baylouny & Mullins 2017). When multiple sponsors supporting the same rebel group lack coordination or compete with each other it allows said group to play the sponsors against each other by moving loyalties between sponsors, and to act in accordance to the sponsor whose preferences least infringe upon the rebel group (Baylouny & Mullins 2017). For instance, multiple sponsors have been shown to undermine the restraining effect on civilian abuse democracies otherwise can excerpt, because the presence of multiple sponsor means no sponsor can restrain the agent (Salehyan, Siroky & Wood 2014).
H2a: A rebel group having multiple sponsors, indicating low sponsor leverage, correlates negatively with interrebel alliances.
H2b: A rebel group having multiple sponsors, indicating low sponsor leverage, correlates positively with interrebel warfare.
Rebel groups often operate covertly, and in areas with limited outside contact, which creates an information asymmetry between sponsor and agent, giving the rebel groups opportunity to shirk (Byman & Kreps 2010). Byman and Kreps (2010) highlight how the sponsors’s capacity to monitor a rebel group is often in direct relation to the degree of presence the sponsor is willing to have on the ground in the conflict. A larger presence costs more and negatively impacts their possibility to plausibly deny involvement, though it increases their control of their agents. Bapat and Bond (2012) argue that rebel groups can cooperate and ally if a foreign sponsor can offer security guarantees for the group, by monitoring and sanctioning non-compliance between their agents by withdrawing or shifting support between groups. Sponsors that can monitor their agents have greater possibility of punishing wrongdoing by the agent, and thus have greater leverage over it.
H3a: Increased sponsor monitoring capacity, indicating high sponsor leverage, correlates positively with interrebel alliances.
14 H3b: Increased sponsor monitoring capacity, indicating high sponsor leverage, correlates negatively with interrebel warfare.
Rebel groups that face a relatively strong state are in greater need of additional resources, and more willing to accept foreign support (Salehyan, Gleditsch & Cunningham 2011). Their chances of winning without sponsor are much smaller than relatively stronger groups. Stronger rebel groups are in general less dependent on their sponsors, as they are likely to have the capability to secure resources anyway and can better absorb losses, and therefore doesn’t need to fear abandonment (Salehyan 2010; Cunningham, Bakke & Seymour 2009). This might explain the previously observed pattern that state sponsors prefer to sponsor relative weaker rebel groups, as such groups are more likely to act according to the wishes of the sponsor (Byman et al. 2001; Bapat & Bond 2012). Strong groups that receive external support should thus be more likely to engage in interrebel warfare than weak groups that receive external support, as they have the capability to do so and face less restrictions to furthering their own interests. Weak groups have the least to lose from being regulated (as their weakness already regulates their freedom of action), and the most to gain through better possibilities to survive through external regulation. Smaller groups should therefore be more likely to engage in interrebel alliances, especially if a foreign sponsor is able to overcome the commitment problems of larger groups in the alliance (Nygård & Weintraub 2014; Bapat & Bond 2012).
H4a: Strong rebel groups, indicating low sponsor leverage, correlate negatively with interrebel alliance.
H4b: Strong rebel groups, indicating low sponsor leverage, correlate positively with interrebel warfare.
Resources provided by external state sponsors have been shown to make rebel groups less dependent on local resources, including their local constituency (Salehyan, Siroky & Wood 2014; Weinstein 2007). The reverse can be expected to hold true as well, that the availability of local resources makes a group less dependent on their foreign sponsor. These resources can either be local in the form of natural resources, or be available in the form of financial, or other, support from within foreign states. This includes private sponsors and support from diaspora communities. Access to alternative sources of finance means a rebel group can better absorb the loss of resources from a foreign sponsor, and therefore has less reason to fear abandonment.
15 Because they are less dependent upon external resources, they are likely to be less willing to sacrifice their autonomy for it, and therefore more likely to engage in agency slack (Salehyan 2010). A group that has access to alternative sources of income should be less dependent on their patron state, and thus more likely to act according to their self-interest (Baylouny &
Mullins 2017).
H5a: Access to alternative sources of income, indicating low sponsor leverage, correlates negatively with interrebel alliances.
H5b: Access to alternative sources of income, indicating low sponsor leverage, correlates positively with interrebel warfare.
3. RESEARCH DESIGN
The hypotheses are tested using statistical analysis. Because my dependent variables are binary, I’m using logit regression, and calculating the odds ratios when of interest. One of the problems with odds ratio and logit coefficients is the difficulty in interpreting the results. The nonlinearity of logit models means that the probability of a positive outcome varies depending on the magnitude of change in the independent variable, its starting value, and the values of other variables in the model (Zelner 2009; Long & Freese 2006, 171). Interpretation of logit results can be improved by reporting marginal effects a variable when other independent variables are kept at different, theoretically or empirically, interesting values (Hoetker 2007; King, Tomz &
Wittenberg 2000). Zelner (2009, 1344) suggests we “report differences in predicted probabilities associated with theoretically meaningful or empirically relevant changes in key independent variable values, when the other independent variables are also set to theoretically meaningful or empirically relevant values.” Predicted probabilities for theoretically meaningful values on the independent variables have are reported at key points during the results section to facilitate interpretation, and to see whether the findings indicate substantial effects.
Because observations of the same group over time violate the assumption of independence, all models are using robust standard errors clustered by rebel group, unless otherwise specified (Chen et al. 2003). Most models have been tested with alternative clustering of the standard errors on country level, rather than by rebel group, as observations are not independent on the country level either; this generally results in higher levels of significance.
All independent variables also control for time order by lagging observations by one year, except for the measure of duration.
16
3.1 Data
The dataset used for testing my hypotheses was built on Belgin San-Akca’s Dangerous Companions Dataset (San-Akca 2015; 2016).4 The dataset holds observations of all rebel groups active in multiparty conflicts between 1975- 2010, and includes data on a variety of forms of external support and characteristics of the rebel group. The unit of analysis in my dataset is rebel-years within multiparty conflicts. The Dangerous Companions dataset is in a triadic (target-sponsor-rebel) format to closely examine sponsor-rebel relations, whilst both the UCDP Non-State Conflict Dataset and Milos Popovic’s data on alliances, which are used for my dependent variables, use rebel dyad-years as their unit of analysis to capture rebel interactions (San-Akca 2015; 2016; Sundberg, Eck & Kreutz 2012; Popovic 2017). Rebel-years was chosen as the unit of analysis for this study, as it appears most able to combine the need to closely examine rebel groups behavior towards other groups, while simultaneously being able to closely examine external support provisions.
The Dangerous Companions Dataset was chosen for this project over alternative databases for external support, such as the UCDP External Support – Primary Warring Party Dataset, due to its inclusion of ideational characteristics for rebel groups and because it holds observations for years when a rebel group doesn’t reach the 25 battle-related deaths activity threshold with the government (Högbladh, Pettersson & Themnér 2011; San-Akca 2015; 2016).
All groups that at some point have reached 25 battle-related deaths with the government are included from their formation until they have been inactivated through a peace agreement or the conflict is terminated. The exclusion of lower activity observations can imply a serious bias when measuring instances of interrebel violence, because only including groups that can keep up high activity against both the government and other rebel groups during the same period is likely to catch primarily strong groups and intense civil wars. Including observations below this threshold allows a more careful analysis of interrebel interaction for a greater variety of groups.
The dataset used in this thesis is limited to multiparty conflicts and includes groups active in the same year and territory with at least one other rebel group with which it could either fight or ally. Groups active in the same country but in different territorial conflicts are excluded from the dataset, as such groups have limited possibility to interact (Bapat & Bond 2012). The only known bias in the data is the exclusion of cases where groups have consistently
4 I am grateful to Dr. Belgin San-Akca for sending me the dataset.
17 stayed below 25 battle-related death with the government, or when a group remains active against other groups after reaching an agreement with the government. While the first bias mostly involves minor rebel groups, it also means that failed state conflicts are excluded if the government is too weak to be active against the rebels. The exclusion of observation where there is no state-based conflict could potentially be problematic, but because of the absence of a common enemy in these cases, we could likely also expect different dynamics for interrebel interaction and for the intervention of external actors. The issue of groups that have signed an agreement with the government is more problematic to exclude, as such groups may remain active in interrebel conflict for a long time after it has signed an agreement with the government.
This also means that this paper does not capture the dynamics of groups that align themselves with the government in multiparty civil wars, better captured by others (i.e. Otto 2017;
Staniland 2012).
3.1.1 Interrebel Alliances
The first dependent variable – interrebel alliance – is a dummy taking the value of 1 if a rebel group is engaged in resource-sharing or tactical co-ordination (i.e. sharing weapons or person- nel, joint attacks or communiqués) with another rebel group at some time during that year, and 0 otherwise. The data is from Popovic (2017)5 which has coded all occurrences of interrebel alliances between rebel groups active within the same territory and year between 1975 and 2009. It builds on a dataset from Bapat and Bond (2012), which gathers evidence from historical accounts, domestic and foreign media reports, case studies of individual conflicts, and conflict actor and event encyclopedias. Certain reliability issues which may exist due to the complexity of coding alliances, incomplete coding criteria, and the uncertainty of information available, which hinders this thesis from expanding on the available data on alliances to better fit the main dataset used. These reliability issues are considered to be outweighed by the benefit of a much more valid term for alliances than available otherwise.
No systematically coded database for interrebel alliances exist, making it a choice be- tween replication datasets of different authors that all suffer from similar limitations. The da- taset by Popovic (2017) is the largest of the currently available datasets on rebel alliances, to
5 The replication dataset for Popovic (2017), accessed 2018-02-22, is available at:
https://www.researchgate.net/publication/320556769_Dataset_for_Inter- Rebel_Alliances_in_the_Shadow_of_Foreign_Sponsors
18 my knowledge. The main issue with this data is that it includes cases where there are no alli- ances as well, limiting my observations to those in the Popovic dataset, which has more limited inclusion criteria than the rest of my dataset. The Popovic dataset is limited through its reliance on the UCDP Armed Conflict Dataset coding of armed conflict as being cases with 25 battle- related deaths against the government in a calendar year. The intermediate years in the Danger- ous Companions dataset is therefore excluded when testing the relationship to interrebel alli- ances. As such, only 670 rebel-year observations of potential alliances are included in the da- taset, of which 402 are cases of active external support.
3.1.2 Interrebel Warfare
The second dependent variable – interrebel warfare – is a dichotomous variable taking the value 1 if a rebel group is involved in fighting reaching 25 battle-related deaths in that year with another rebel group, and 0 otherwise. The data is taken from the UCDP Non-State Conflict Dataset in which non-state conflict is defined as “the use of armed force between two organized armed groups, neither of which is the government of a state, which results in at least 25 battle- related deaths in a year” (Allansson & Croicu 2017, 2; Sundberg, Eck & Kreutz 2012). In this thesis only formally organized groups are included, which means that they have announced a name for their group and are using armed force against another formally organized group (Ibid.). The UCDP Non-State Conflict Dataset is limited to post-1989, while the external support data is limited to 2010, which means that interrebel warfare is tested for the period 1989-2010. While this limits my observations, no equally comprehensive dataset for non-state violence is available to my knowledge, thus limiting the data to post-1989. Once transformed from dyadic to rebel-years and integrated with the main dataset, it includes 1588 observations of potential rebel warfare, of which 671 are cases of active external support.
3.1.3 Independent Variable
My independent variable is more complex, due to the difficulty in finding valid measurement for leverage within a sponsor-rebel relationship. Instead, several indicators for a sponsor’s lev- erage over a rebel group has been created: multiple sponsors, the monitoring capacity of the sponsor, the strength of the rebel group, and the availability of alternative sources of income.
The combination of these indicators is argued to represent a valid measure for the definition of leverage used in this paper, meaning the capacity of the sponsor to control its agent by imposing costs on the agent for non-compliance to the interests of the sponsor. All indicators capture an
19 important factor influencing how effectively a sponsor is able to force compliance of its agent by imposing costs on said agent. Multiple sponsors, alternative resources and rebel strength are measures of how well a rebel group can absorb the costs imposed by the sponsor, while moni- toring measures whether the sponsor can credibly threaten punishment for non-compliance.
The availability of multiple sponsors allows a group greater independence from the control of the sponsor and makes the group less dependent on any single sponsor, decreasing the sponsor’s leverage over the group (Salehyan 2010; Baylouny & Mullins 2017). Therefore, a variable for multiple sponsors is created. It’s a dummy variable taking the value of 1 when several states support the specific rebel group during that year, and 0 otherwise. The variable is coded from data drawn from the Dangerous Companions Dataset (San-Akca 2015; 2016).
For a sponsor to be able to credibly punish agency loss it is dependent on its capacity to monitor the agent; relying on third party reporting conveys considerable autonomy to the rebel group. To be able to monitor a rebel group the sponsor needs a presence on the ground or local intelligence capacity (Byman & Kreps 2010). To test whether the sponsor has the capacity to monitor and sanction a rebel group a variable is created where types of support that allows the sponsor direct access to the rebel group, compared to sponsorship agreements where the spon- sor only provides resources. The forms of support that would increase the sponsor’s monitoring capacity by providing the sponsor with a presence on the ground is the provision of supporting troops, safe haven and training. For this, the Dangerous Companions variables for all forms of safe havens, training and provision of supporting troops are merged into a single dummy vari- able, taking the value of 1 if any of these provisions are provided, and 0 otherwise.
The validity of this measure needs to be discussed, as it is not a perfect indicator for monitoring, and could even be argued to capture entirely different concepts. It could for exam- ple be argued that safe havens will decrease interrebel warfare from a group having at least part of its capacity at a distant from the conflict, rather than this indicating that the supporting state is allowed a greater degree of control. This is somewhat offset by it then being unlikely to show any support for interrebel alliances, which is also required for it to receive support. This meas- ure can only claim to capture some aspects of the sponsor having any form of capacity to mon- itor, not actually observed forms of monitoring or degree to which monitoring was applied. It is therefore at the best of times a blunt measure for the phenomena of interest. With no alterna- tive data available to capture the concept, this paper argues that the variable is a good start in capturing the phenomena in a quantitative setting. It does to a large extent follow the empirical forms of monitoring that Byman and Kreps (2010) found in their qualitative work, aiming to capture the physical presence of the sponsor.
20 Weak rebel groups are less able to mobilize resources domestically, are more willing to sacrifice autonomy in exchange for vital resources and are more likely to need external support to survive (Salehyan, Gleditsch & Cunningham 2011; Bapat & Bond 2012). This increases the sponsor’s leverage over the group, as it is more dependent upon the sponsors support. Strong rebel groups are less willing to give up their autonomy and are less dependent on the foreign resources, which decreases the sponsor’s leverage over stronger groups (Salehyan, Gleditsch &
Cunningham 2011). To capture this, the troop size of the group, taken from the Non-State Actor dataset, is included (Cunningham, Gleditsch & Salehyan 2013).6
Rebel groups that have access to local sources of income are less willing to give up their autonomy because they are less dependent upon the sponsor’s resources, which reduces the sponsor’s leverage over the group (Salehyan, Gleditsch & Cunningham 2011). The variable for income is a measure of the number of resources from which a “group earned a significant frac- tion of funding” in a given year, according to the Rebel Contraband Dataset (Walsh et al. 2018, 3). A further measure is included which controls for whether a group had access to nonselective support from within the territory of a foreign state, which includes private sponsorship and other instances where a state is de facto aiding a rebel group without actively selecting to do so. The data is from the Dangerous Companions Dataset and is coded 1 for observations where a group received nonselective support from within at least one foreign state in a given year, and 0 oth- erwise (San-Akca 2015).
The variable for external support is from the Dangerous Companions dataset. It shows a state intentionally supporting a rebel group by providing weapons, logistic or financial aid, transport of military equipment or advice, providing troops, training, training camps, or safe havens for members, leaders or headquarters.7 It takes the value of 1 if “there is clear evidence for states creating channels to abet certain groups and this evidence is confirmed by multiple reliable sources as incidents where states select the NAGs to provide support to”, and 0 other- wise (San-Akca 2015, 20).
6 The measure is the log of the absolute number, rather than a measure of relative strength. This is because the relevance of relative strength as compared to the government or other rebel groups varies between the independent and dependent variables. The relative strength against other rebel groups is a more relevant measure for interrebel alliances and warfare, whilst the relative strength against the government is likely to better capture the leverage of a sponsor.
7 Issues of multicollinearity with the variable monitoring capacity is addressed in appendix 1. Both had VIF values below 10 in all models.
