The Pennsylvania State University The Graduate School College of the Liberal Arts

Transcription

1 The Pennsylvania State University The Graduate School College of the Liberal Arts DON T STOP TILL YOU GET ENOUGH: REBEL GROUP WAR AIMS AND CONFLICT RESOLUTION IN CIVIL WAR A Dissertation in Political Science by Jakana L. Thomas 2012 Jakana L. Thomas Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy August 2012

2 The dissertation of Jakana L. Thomas was reviewed and approved* by the following: Douglas Lemke Associate Professor of Political Science Dissertation Advisor Chair of the Committee Glenn Palmer Professor of Political Science Phil Schrodt Professor of Political Science John Horgan Associate Professor of Psychology Director of the International Center for the Study of Terrorism Lee Ann Banaszak Professor of Political Science Director of Graduate Studies *Signatures on file in the Graduate School ii

3 ABSTRACT This dissertation examines whether government concessions to rebel group demands have an effect on conflict resolution in civil wars. This dissertation also examines whether the credibility of the rebel group being conceded to moderates the effect of these concessions on bargaining outcomes. Previous literature suggests rebel group demands, in general, are not credible. These arguments lead to the expectation that granting concessions to rebels will have no positive effect on the probability that a dyad will reach a peaceful settlement. From existing arguments we should expect, instead, that concessions to rebels will be followed by additional, escalated demands as rebels will attempt to gain as many concessions from their government opponents as possible. If all rebel demands are incredible and all rebels act on the same incentives to gain the greatest number of concessions in bargaining, concessions to these demands should always be negatively related to measures of conflict resolution such as the signing of peace agreements and the cessation of hostilities. Existing literature, however, does not consider that rebels, or some types of rebels, are actually responsive to government concessions. Absent from existing literature is any explanation as to why some rebel groups end their conflicts when their demands are met while others do not. I offer these explanations here. In this project, I argue that such variation in settlement exists and can be explained by rebel group credibility. Specifically, more credible rebel groups are more likely to stick to their demands and stop fighting when they are offered concessions that address their aims. Non-credible groups, however, will not be responsive to concessions as their demands are not credible indicators of their requirements for conflict termination. Empirically, this means that rebel groups that follow through with their threats to continue fighting until their aims are met should sign agreements and stop fighting when governments offer them significant concessions on their demands. Groups whose aims are not credible should not be any more likely to agree to settlements when they are granted concessions addressing their demands. I argue that the variation in the credibility of rebel groups (and their demands) can be explained by the presence of an internal mechanism binding these groups to their publicly announced demands. The argument in this dissertation explains why a civilian support base is this credibility mechanism for groups that have them. Variation in rebel credibility, then, iii

4 can be attributed to the fact that not all rebel groups rely on a civilian support base to operate. In Chapters 5 and 6, using original data on rebel group demands and government concessions in a sample of 108 African rebel-state dyads involved in civil conflicts from , I test this argument and find that rebel groups operating with support bases are in fact more likely to sign formal agreements and terminate their conflicts when their demands are met than are groups that do not rely on civilian supporters to mobilize resources. These results demonstrate that credibility is variable across types of rebel groups and furthermore that a key characteristic distinguishing credible rebels from non-credible rebels is readily discernible to governments. This dissertation is important because it demonstrates that for groups with support bases, publicly made demands represent credible information that can help dyads negotiate peace deals. Substantial concessions to these groups, then, should lead to conflicts with successful peaceful terminations. In addition to examining how concessions to demands affect conflict settlement, this project also examines the types of rebel groups to which governments tend to offer concessions during negotiations. The results here show that although groups with support bases are more credible negotiating partners, governments are significantly more likely to negotiate with groups that do not have support bases. Together, these findings suggest a convincing alternative explanation for why civil conflict tends to be so difficult to settle. Overall, this project offers new arguments about the importance of demands, concessions and rebel credibility in intrastate conflict bargaining. iv

5 TABLE OF CONTENTS List of Tables. vii List of Figures... viii Acknowledgments.. ix Chapter 1: Introduction... 1 Dissertation Roadmap 3 Chapter 2: Signaling Commitment How Rebel Support Bases Help Form Credible Commitments in Civil War. 5 Uncertainty in Battle 5 Uncertainty in Negotiations.11 Rebel Credibility and the Costs of Bluffing.16 The Rebel Support Base 19 Cost to Rebel Leaders 26 Public Demands- A Public Good. 30 Compromise.. 31 Signaling Commitment 32 Chapter 3: Demand and Concession Data: An Investigation of What Rebels Want and Receive in Civil Wars. 37 What Rebels Want and Receive in Civil War.. 44 Demand- Level Codebook 55 Dyad-Month- Level Codebook.. 57 Chapter 4: Rewarding Bad Behavior: How Governments Respond to Rebel Group Tactics in African Civil Wars..60 Introduction: Putting the Horse Before the Cart 60 Why Not Negotiate With The Terrorists?...65 Terrorism, Terror and the Power to Hurt.. 69 Research Methods Results v

6 Rebel Claims...85 Discussion.. 85 Chapter 5: The Effect of Concessions on Negotiated Settlements. 87 Agreements in Civil War Research Design Results Substantive Effects Conclusion Chapter 6: Stopping When They Get Enough: The Effect of Concessions to Rebel Demands on Civil War Termination.109 Literature Research Design Results Substantive Effects Conclusion Chapter 7: Conclusion. 121 Results Future Work Data Issues and Solutions Conclusion 130 References 132 Appendix A: Figures and Tables for Chapter Appendix B: Statistical Tables from Chapter Appendix C: Statistical Tables for Chapter Appendix D: Statistical Tables for Chapter vi

7 LIST OF TABLES Table 3.1: List of Demands..140 Table 3.2: Coding Rubric.144 Table 4.1: Logit Regressions of the Effect of Group Characteristics on Civil War Negotiations 150 Table 4.2: Negative Binomial Regressions of the Effect of Group Characteristics on the Number of Government Concessions..151 Table 4.3: Rebel Groups Coded as Using Terrorism in the Global Terrorism Database (GTD)..153 Table 5.1: Probit Regressions of the Effect of Government Concessions on the Probability of Agreements (Parsimonious Models).155 Table 5.2: Probit Regressions of the Effect of Government Concessions on the Probability of Agreements (with Support Bases and Many Controls). 156 Table 5.3: Probit and Bivariate Probit Regressions of the Effect of Government Concessions on the Probability of Agreements 157 Table 6.1: Probit Regressions of the Effect of Government Concessions on Conflict Termination Table 6.2: Probit Regressions of the Effect of Government Concessions on Termination (Lagged Dependent Variable) Table 6.3: Bivariate Probit Regressions of the Effect of Government Concessions on Negotiations, Agreements & Termination Table 6. 4: Cox Regressions of the Effect of Government Concessions on Time until Termination vii

8 LIST OF FIGURES Figure 3.1: Frequency of Rebel Demands (Repeated Demands) 145 Figure 3.2: Frequency of Rebel Demands by Group 146 Figure 3.3: Moderate Government Concessions by Demand Type Figure 3.4: Substantial Government Concessions by Demand Type 148 Figure 3.5: Maximal Government Concessions by Demand Type.149 Figure 4.1: Predicted Probabilities of Negotiations by Terrorism, Rebel Strength and Time.152 Figure 5.1: Figure 5.1 Predicted Probability of Agreement by Ethnic Claim and Percentage of Concessions.159 Figure 6.1 Predicted Probability of Episode Termination by Ethnic Claim and Percentage of Concessions. 164 viii

9 ACKNOWLEDGEMENTS While he did not necessarily play the largest role in the evolution of my dissertation, I must do things in decency and in order and thank Michael Jackson for providing the title for this project. If it had not been for the musical genius of the king of pop my dissertation title would have been something like Concessions to Rebel Demands Help End Civil Wars. Had this been the title, I m positive people would have had less fun in the lead up to my talks and fewer political scientists would know the appropriate usage of the words til and till. I consider that Michael Jackson and I have done the political science community a great service. You can thank us later as it is my turn to thank you. First, I must thank my advisor and dissertation chair, Doug Lemke for his support every day since the Fall of Thank you for your attendance at every single meeting from my candidacy exam, to my master s defense, to my prospectus defense, all the way to my dissertation defense. I am thankful for your advice, the pep talks, the ideas, you trusting me as a co-author, the lunches, and most of all for reading and providing extensive comments on my dissertation. You have been a terrific, outstanding advisor and you deserve more thanks than I have space to give in these pages. The entire political science department at Penn State deserves mention. Thank you to everyone who attended practice presentations and job talks. Your comments and suggestions were incredibly helpful and were taken seriously. Individually I must thank Glenn Palmer who helped professionalize me, Lee Ann Banaszak who was been supportive in every way throughout the years, Burt Monroe, Jim Piazza, Scott Bennet and Donna Bahry for the extensive job market advice, Errol Henderson for believing in me and giving me so much advice, and Sona Golder for being awesome. Special thanks go to Phil Schrodt and John Horgan for their help in guiding this project. I am so appreciative to everyone at the CSCW at PRIO, especially Scott Gates. To my dear friends, Kanisha Bond and Martha Thomas, thank you for being there when I started this journey. Kanisha, thank you for making me dinners when I was too crazy to eat, keeping me sane, being a good friend, letting me bum in your room at conferences and for offering advice and resources. Martha, thank you for being such a supportive friend, for allowing me to talk through every paper I have ever written, and for reading and providing comments on so many things. To my newer but no less dear friends in the BCG, I love you all. Thank you for being my escape from reality when I needed it most. Thanks to Matthew Lane, Matthew Wilson, Andrew Carr, Christopher Ojeda and Dallan Guzinski. To my mentee Jessica Maves, thank you for sometimes mentoring me, for food and shoe therapy, and for always being down for a good celebration. Molly Ariotti, thank you for our craft nights, secrets, shopping and our fun times down on College. What would I have done without you? To my loving family, thank you so much for everything. To my amazing mother, thank you for supporting everything I ve ever done, with this being no exception. Thank you for supporting my creativity, for listening to me gripe, complain and talk incessantly, and for never getting upset with me for being absent. Grandma, thank you for being so loving and caring, for setting such a good example for all of us. To my lovely older siblings, thank you all for always supporting me and believing I could do everything, even when it wasn t true. To Diana, thank you for listening to all of my big plans and understanding me over the last two decades. To my favorite boys, Amir, Malachi and Aiden, you are my little blessings. You make life so much more colorful and give me much to look forward to. I especially thank God from whom these and all blessings flow. There are so many people that deserve mention for their impact on this dissertation. Without my family, friends, classmates and acquaintances I could not have finished this dissertation. It is to the people who love me, those who have allowed me to be inattentive in my personal life for the past five years, that I dedicate the good parts of this project. ix

10 Chapter 1: Introduction Civil wars do not arise randomly. Conflicts are usually initiated by organized groups articulating a willingness to use violence if their governments do not meet some set of demands. Rebel groups are often clear about what they want and about their readiness to use violence to attain their stated goals. If rebel demands motivate conflict onset, it is reasonable to expect them also to play an important role in rebels decisions of when they will stop fighting. Despite the obvious importance of rebel aims to conflict resolution, to date there has been no study examining what rebel groups demand, what rebels get, and how these two are related to conflict resolution. Rebel groups usually make many demands over the course of a conflict. While some groups demands may lack coherence, other groups demands will be effective in revealing information about their propensity to settle with the government. Demands can indicate the location of rebels ideal points as they can represent exactly what groups hope to get out of settlements. Together with government concessions, rebel group demands can symbolize the bargaining space within a dyad. Thus, as governments offer concessions closer to what rebels demand, the probability of settlement should increase. Ayres (2001) characterizes the distance between the positions of the belligerents in violent intrastate nationalist conflicts as bargaining space. He suggests that as bargaining space becomes wider, conflicts will be more likely to end when both parties pursue peaceful ends to their conflict. I expect a similar relationship when examining the effect of government concessions to rebel group demands. As government s concession offers approach rebel demands, the likelihood a bargain can be struck to stop the fighting will increase 1. 1 Reiter (2003) explains how the size of the bargaining range between a dyad affects conflict settlement (also see Morrow (1985) and Fearon (1995)). Reiter (2003:34) writes the bargaining model would propose that, ceteris paribus, as two sides reveal information about their capabilities and intentions, bargaining space will open up to permit a war avoiding bargain. He similarly suggests that an enlarged bargaining range can increase the probability of conflict termination after 1

11 Addressing rebel demands should help warring dyads locate acceptable settlements but should also have broader impacts on the frequency and duration of conflict in general. As rebel demands often represent underlying grievances in societies, addressing them with concessions should eliminate some of the conditions that made violence possible and permissible. As most rebel groups are only able to wage successful wars when they are supported by a subset of the civilian population, the existence of civilian grievances is key to rebellion. Addressing rebel demands, then, should increase the probability that the present rebellion will end and should also ensure that future conflicts are avoided. Even if a particular rebel group is defeated during war or agrees to settle before attaining concessions, civilians are likely to support other violent groups that will pursue concessions on their behalf. Thus, addressing rebel demands will be essential for peace when demands represent broader societal grievances. Allaying these grievances deprives actual and potential rebels of the support they need to execute conflicts. Ostensibly, those supporting rebel demands are enablers of continued conflict. Jenne (2006) examines how nationalist group demands covary with external support. She argues that non-state nationalist groups escalate their demands when their external supporters are supportive of those demands. When they perceive widespread external support, they are more apt to pursue violent means of change. When external supporters decrease their support for the aims of nationalist groups, these groups are most likely to abandon violence. In this dissertation, I also examine the role rebel support bases play in supporting rebel demands and furthering the violence rebels employ. a war begins (Reiter 2003:34). The reduction of uncertainty should create bargaining space. As belligerents engage in battle they become more certain, a range of bargains both sides prefer to war is revealed and locating a mutually preferable bargain within that that range becomes easier. Although Reiter s argument suggests refers to bargaining between two states, the argument should also apply to wars involving non-state actors. Ayres (2001), for example, also expects a larger bargaining range to facilitate agreements between non-state nationalist groups and their governments. As the number of acceptable bargains increases, the probability a dyad will be able to sign a bargain should increase reciprocally. In this dissertation, I argue that revealing rebels demands should help reduce uncertainty over what rebels will accept in conflict. As uncertainty is reduced, a bargaining range should widen (or appear) and that conflict should be more amenable to settlement. I do not argue more bargaining space alone will increase the probability of settlement, but I expect peace deals to be more likely as belligerents preferences converge. 2

12 Instead of examining external supporters to nationalist groups, I focus on internal supporters of all types of rebel groups. I expect that when rebels have civilian support bases, they will be highly likely to continue fighting until they gain concessions that satisfy these support groups. When support bases stop backing rebels, or when the demands rebels made on behalf of their supporters are met through concessions, fighting will end. Rebels will only fight as long as they are able to retain support for their aims and for their efforts to prosecute their conflicts. While the title of this dissertation, Don t Stop Till You Get Enough, may seem a bit tongue-in-cheek, it is actually an accurate description of my argument. Rebels will not agree to settle their conflicts peacefully until they are granted a sufficient amount of concessions that address their grievances and those of the civilian populations they represent. Rebels are constrained by their support bases not to stop fighting until they gain enough concessions from governments. As such, rebels should not stop until they feel they have gotten enough. In the following chapters I develop and explore this argument further. Road Map My dissertation will proceed in the following manner: The next chapter, Chapter 2, lays out the argument. Specifically, I explain why existing conflict resolution literature does not deal with the question of rebel demands and why it is remiss in failing to do so. I explain the mechanism by which rebel demands can reveal credible information that helps belligerents locate settlements. I offer hypotheses that I test in subsequent chapters. Chapter 3 features the data I collected to conduct the original statistical tests in later chapters. Chapter 3 contains detailed information about how I collected and coded data on rebel group demands and on government concessions. I end the chapter by describing the types of demands rebels make most frequently and the types of concessions governments are most likely to offer in response. Chapter 4 examines the characteristics 3

13 that make rebel groups more likely to be included in negotiations and to gain concessions of any kind from governments. Chapter 5 and 6 test the hypotheses set forth in Chapter 2. Specifically, I examine how concessions to rebel group demands affect the probability of formal agreements and conflict termination, respectively. I offer concluding remarks in chapter 7 and discuss how I plan to deal with problems confronted in executing my research thus far, in hopes of moving further forward in future research. 4

14 Chapter 2: Signaling Commitment How Rebel Support Bases Help Form Credible Commitments in Civil War Uncertainty in Battle The problems inhibiting the settlement of conflict are often classified as information problems in bargaining models. Particularly, scholars argue that parties in conflict have incentives to misrepresent privately held information in order to extract better bargains from their opponents (Fearon 1995). While incentives to misrepresent information are present in war between states, the literature on bargaining in civil war suggests that these information problems are more acute when conflict dyads are composed of states and rebels. One reason for this, offered by Walter (2009), is that rebel groups not only have incentives but also have the ability to conceal private information, while governments are usually not able to do so. A defining feature of democracy is its transparency, and this feature persists through periods of war. Ramsay (2004) demonstrates that democratic states domestic politics influence crisis bargaining by revealing information on their resolve or true bargaining positions through partisan competition (also see Shultz 1998, 2001; Smith 1998; Guisinger and Smith 2002 for similar arguments). For both democracies and non-democracies alike, information about governments military strength is often a matter of public record. In interstate conflict, this transparency may minimize disagreements over which state has more resources to devote to military contests, thus decreasing the uncertainty surrounding the probability of victory for either side. Rebel groups, however, are generally not transparent and information on them is particularly difficult to observe. This creates a severe information asymmetry where rebel groups are generally informed about government capabilities while governments lack any reliable evidence about rebels fighting capacity. Forest cover and mountainous terrain, for example, are frequently cited reasons for governments inability to access information on rebel capabilities, as rebel troops 5

15 are usually located in inaccessible areas especially difficult to survey (Walter 2009). When not hidden in forests, unlike conventional troops, guerrillas are able to blend in with the surrounding community, making it nearly impossible for governments to distinguish between rebel and civilian. Thus, it is quite difficult for governments to gain a true account of rebel group strength. Since information about capabilities factors heavily into a government s calculus when deciding whether to offer concessions or ignore rebel group demands, rebels are able to benefit from uncertainty if they can convince their opponent they are stronger than they are in fact. Uncertainty, however, is often problematic for both governments and rebels, as it inhibits and ultimately delays a government s decision to offer a settlement. If a government was certain that the rebels were both strong and committed to attaining a settlement, they would likely offer some concession immediately to avoid paying the costs that accrue with violent conflict. That governments do not know with certainty whether rebels are strong or weak in military power or will prevents expedient settlements forcing both sides to suffer the costs of war. Given that governments can settle to avoid the heavy toll of war, why do they not err on the side of caution and assume their armed opponents are strong when they say they are? The reason is rooted in this information asymmetry. Governments are reluctant to believe rebel group claims because they know they lack accurate information on rebel group capabilities and more importantly, rebels are aware of this. Bargaining arguments suggest that rebel groups have incentives to use information asymmetries to their advantage. If groups are able to convince their government they are strong and committed, even if they are not, they may also be able to convince them to grant large concessions and do so much more quickly than if they were weak. 1 Rebel groups are expected 1 Incentives to misrepresent should only exist for groups that are weak or unresolved. If a group is clearly stronger than the government and committed to seeing a settlement through, they need not pretend at all. These groups, then, have little incentives to misrepresent. Therefore, information issued by strong and committed groups should more often be accurate than that issued by groups that are neither of these things. 6

16 to take advantage of governments inabilities to gather accurate information and verify claims made by rebels regarding their strength. Since governments expect rebels to pretend to be strong when they are not, they likely heavily discount information freely offered by these groups. Thus, governments are likely to be doubtful about information on rebel strength even when it is truthful. Information problems do not only affect calculations of capabilities but they also affect estimations of rebel resolve. States are resistant to giving concessions to groups that do not have the means to inflict a settlement on them. Rebel groups may make demands and want to inflict penalties until they are met, but may be physically incapable of sustained fighting. Additionally, governments have a hard time ascertaining how committed groups are to seeing a settlement through. Governments do not know whether rebel groups actually plan to fight for the concessions they demand or instead are hoping to bluff their way to settlement; Rebels may demand concessions with no real intention of fighting until their demands are met. Rebels may threaten to inflict great penalties on government interests until the government offers concessions, but may be unwilling to follow through with such a threat. Governments fear conceding to groups like these. Because governments imagine that groups will attempt to attain concessions by any means including bluffing and misrepresenting information about themselves, they are generally dubious about rebel claims. Governments would rather verify claims made by rebel groups about their strength than believe what they say outright, even if confirmation comes at a significant cost. Slantchev (2003:622) argues that information revealed during negotiations can be highly manipulated, whereas the information gathered through battle has less potential for manipulation 2. Therefore, when belligerents, particularly states, are presented with two conflicting sources of information they are 2 Also see Wagner (2000:478) and Powell (2004:349) for discussions on the non-manipulability of information that is garnered through fighting. 7

17 more likely to rely on the source less subject to manipulation: information garnered through battle. Consider, nothing stops belligerents from making unreasonably large demands to signal strength, even while they are consistently losing battles. Such signals provide contradictory information to a receiver. 3 In such a case, a state should be far more likely to update its beliefs about a potential victory with information from recent battle outcomes than with information offered freely by rebels. This argument suggests that incentives to misrepresent information about rebel group strength and resolve is the main inhibitor of settlements. In other words, the fact that rebels are expected to misrepresent the amount of resources they can devote to a war forces states to enter into battles with their opponents to gain a clearer picture of the conflict s outcome. This uncertainty makes conflict inefficient because it prevents belligerents from agreeing to ex-ante settlements equal to an eventual postwar division without paying the costs of war. While numerous studies corroborate that uncertainty over capabilities is sufficient to spark conflict onset and to influence the duration of conflict, newer literature suggests that it does not play such a large role in delaying the settlement of conflict as previously believed (Filson and Werner 2002; Slantchev 2003; Smith and Stam 2004; Powell 2004). Scholars argue that learning is expected to alleviate much of the uncertainty that plagues belligerents involved in bargaining at the outset of conflict. Specifically, as rebels and governments spend more time engaging one another on the battlefield, uncertainty surrounding the probability of victory decreases. Thus, after some time, both sides can estimate more accurately the likelihood that either of them will win a contest. However, this argument tends to be limited to uncertainty surrounding military capabilities. 4 Although governments may become more certain that their own victory is not assured after successive rounds of fighting, a greater 3 Here Slantchev is talking about information gathered during conflict between states. However, there is no reason to believe that the process of information gathering is substantially different in conflict between a state and a non-state actor. 4 Filson and Werner (2002), Slantchev (2003) and Powell (2004) examine alternative models where combatants update their beliefs about their opponent s type based on events that transpire during negotiations. 8

18 number of military encounters does not produce certainty about the kind of negotiating partner a state is faced with. 5 This kind of learning can only occur after explicit bargaining between government and rebels. This suggests at least two interrelated bargaining processes affecting patterns of settlement: bargaining on the battlefield and bargaining in conference rooms. Each process reveals important information about belligerents resolve and both affect the probability, size, and content of a settlement. Most importantly, these two processes are interrelated as states and rebels use information from battle to gain leverage in the conference room and vice versa. Despite the importance of both processes for conflict settlement, most existing literature on bargaining tends to focus on uncertainty surrounding one process, the process of battlefield bargaining. Newfound clarity about battlefield outcomes rarely allows belligerents to achieve settlement seamlessly. Unless a clear victor emerges, belligerents still have to come to some agreement on the division of the issue in contention. Understanding how strong an opponent is only allows governments to get to the point of agreeing to offer concessions. While certainty about relative capabilities must affect whether states decide to make offers, it does not really determine what states ultimately decide to offer. 6 A final agreement is largely dependent upon how much of a concession 5 Filson and Werner, for example, argue that during negotiations, combatants update their beliefs about whether their opponent is a weak or strong type when an opponent accepts or rejects an offer. If an opponent rejects an offer, the attacker becomes more pessimistic and if it is accepted, the attacker is more optimistic about the prospects of settlement. Filson and Werner note that war does not terminate because one side is incredibly committed to fighting, but when belligerents are able to come to some agreement on the division of the issue. In their model, when the attacker makes a strong defender an offer that any strong defender will accept, conflict ends. If the attacker makes a strong defender an offer that only a weak defender would accept, conflict continues for another round. In this model, there are only two types of offers that can exist, a low offer and a high offer. There are also only a finite number of rounds of fighting and bargaining that each round can sustain. 6 Consider a hypothetical situation where rebels and governments have been fighting long enough such that both sides acknowledge, barring some major change in circumstances, the rebels would neither win nor lose and continued war would prove unbearable for the government. Realizing that continued conflict is much too costly, this government would likely decide to pursue negotiations. Even after deciding peace is a viable option, there would still be marked uncertainty preventing belligerents from agreeing to end their conflicts right away. Despite both parties expectations on the battle field having converged, a sufficient amount of uncertainty would still remain as neither side knows how far they can push the other in negotiations. 9

19 governments are willing to offer together with how much of a concession rebels require in order to accept an agreement. The specific agreement in the space that lies somewhere between these two points will only be determined through negotiations. The catch is that governments generally offer negotiations only once they have decided they are willing to make concessions. Governments normally decide to make concessions only to groups they believe they have to make concessions to, that is groups with the ability to impose high costs on their opponents. Dyads, then, will only begin negotiations once the state has decided that the rebels have the capacity to wage a successful or painful campaign against them. 7 Negotiations, therefore, serve as a costly signal from the government, as negotiations, themselves, are a kind of concession in their own right. When governments agree to negotiate with rebels they send a signal that they have gathered sufficient information on the group s capabilities to begin dialogue. It is significant when governments decide to hold these formal talks with groups, as doing so gives an organization legitimacy they might not have gained otherwise. A Government s decision to negotiate serves as a signal because it is made before an audience. Particularly, the audience, which is composed of groups actual and potential supporters, observes negotiations as a stamp of approval on a rebel organization. Specifically, the extension of negotiations means that the rebels are serious contenders such that they pose a substantial threat to the government. Individuals that were marginally The government s knowledge that rebels would never emerge from war the victor, might suggest rebels do not deserve full concessions. However, the fact that those same rebels can maintain a lengthy war absent a settlement should force a government to moderate its stance. And although an offer of negotiations can be taken by rebels as a signal indicating the government is willing to compromise, rebels do not know how much they can demand before the government will prefer war to such a costly settlement. Without any prior knowledge of their opponent s bargaining profile, the government does not know what kind of offer their rebels will accept nor do rebels know what kinds of demands the government would reject. Here, the distribution of power would not help the belligerents agree on the division of the issue as each side still possesses an outside option of continued aggression. Both sides want to maximize their gains without altering their opponents preference for war but absent any bargaining history neither side can make an informed decision. 7 Governments might offer negotiations to groups they do not believe will inflict any significant damage on the state if they are trying to save face in the presence of an outside actor or under the pressure of the international community. 10

20 supportive of an organization and its cause become much more supportive after the government s informal recognition of that organization. This additional support may provide substantial tactical advantages for an organization which, in turn, represents a setback for the government. Offering negotiations to a rebel group might even have the effect of increasing rebel group strength by facilitating recruitment. In order to avoid unnecessarily strengthening weak challengers, governments refuse negotiations with groups until information about their ability to inflict costs is verified. This formal bargaining is only likely to occur after substantial fighting. After multiple rounds of battles, governments are able to observe how rebel groups fight and thus can revise their subjective estimate of rebel success. If rebel groups are consistently winning battles, capturing towns or rapidly advancing toward the capital, the government is likely to revise upward their beliefs about the probability of rebel success. Around this time, they should be most willing to offer negotiations, if not outright concessions. Uncertainty in Negotiations Finding common ground on a settlement is as difficult as achieving negotiations in the first place. If governments do decide to offer concessions at some point, they are likely to be fearful of being taken advantage of in bargaining. Specifically, states are concerned they will be gulled into providing concessions greater in size and scope than rebels will actually accept. As they are with information regarding rebel capabilities, governments are generally wary of misinformation with respect to rebel perceptions of the bargaining space. While the inefficiency of conflict is often attributed to uncertainty about capabilities, uncertainty in negotiating may also lead to inefficient outcomes. Because governments are mistrustful of rebel demands in bargaining, conflicts may last longer than they should given the amount of accurate information belligerents hold about the probability of victory for either side. Although, the government and rebels might both be reasonably 11

21 certain about their probability of victory, conflicts may still persist as uncertainty in explicit bargaining prevents governments from making an acceptable offer despite any knowledge it might hold about what a rebel group will accept in the conflict. Governments are likely to come out of these prolonged negotiations worse off than if they had settled in a previous period as the more costs they suffer, the less favorable a settlement they might be willing to accept. Knowing rebels have both the incentives and the ability to misrepresent information during negotiations, governments are generally dubious about information provided by rebel groups regarding the concessions they require. Governments are concerned that rebels will demand far greater concessions than they will realistically accept. Both sides know that rebels want to gain the greatest amount of concessions in a settlement and governments want to provide the least amount of concessions. If governments believe that rebels always demand more than they will accept, governments will always offer less than rebels demand. This produces a cycle, where rebels demand more than they are willing to settle for because they believe governments will offer less than they will finally concede. Rebels know they have to start high to get a settlement close to their ideal point, while governments know they have to make an initial offer lower than rebel s demands to attain a settlement close to their ideal point. This kind of haggling increases the duration of negotiations (and concurrently of conflict) especially if either side is inflexible. As conflict is extended each side must pay costs of conflict, as war infrequently ends when negotiations begin. Each side, but the government especially, wants to minimize the rounds of bargaining so they can minimize the costs of delay for any settlement. Governments really want to locate a settlement while at the same time minimize the amount they pay out in concessions. In some cases, the government will be willing to accept a delay in settlement if they believe the cost of settling is much greater than the cost of continued conflict. 12

22 Most governments are apprehensive about offering any concessions to any rebel group. Since concessions tend to be zero-sum, governments try to minimize the number of concessions they offer to rebels in order to maximize their share of what is being divided. 8 Given they believe they have to settle, governments are interested in gaining as much information as possible about their opponents bargaining behavior to minimize the size of that initial offer. If a government s initial offer is high, it is unlikely that they will be able to downgrade that offer even if they later realize their bargaining partner would have accepted less. If governments make high offers initially, it signals they have the propensity to be generous. This leads rebels to believe, perhaps correctly, perhaps erroneously, that they can continue to push for large settlements even if the government later revises their offer downward. If there are third parties involved in negotiations, it will be even more difficult for governments to offer less in subsequent rounds of bargaining, as they do not want to appear unfaithful, even if they are. Thus while governments are concerned with ending conflict as soon as possible, they are also concerned with avoiding terrible settlements that might be more costly in the long run. This means that early on, governments may be willing to accept costs associated with delayed settlements in order to uncover information about their opponent that might allow them to locate a more beneficial arrangement in the future. As governments learn more about their opponents over time settlement should be more likely after each subsequent round of bargaining. States typically know very little about the bargaining profile of the opponents they face at the onset of conflict. Governments generally do not know whether rebel groups are obstinate or flexible or whether their demands represent reservation points or ideal points. If rebels demand independence, for example, does it really mean they will settle for no less than independence, or are 8 Even when offering more to rebels does not leave the government with any less, governments may still be resistant to offer concessions because they are not sure if there will be a future challenger that is emboldened by government concessions to a group presently involved in conflict (Walter 2004) 13

23 they actually willing to settle for some form of political autonomy? Or would allowing them representation in an eventual reconciliation government be sufficient to gain peace? Rebel groups may make extreme demands, like independence, only to attempt to signal that they are strong. These same groups, however, may be willing to scale down their demands once in negotiations. The government is unsure. The possibilities exist that a rebel group is actually weak, uncommitted, and flexible in their demands but the converse is also possible. A group bearing a combination of these characteristics is also conceivable. For governments, a weak and uncommitted opponent is ideal. Whether rebels are flexible or obstinate when they are weak in bargaining does not matter as governments are unlikely even to initiate negotiations. If rebels are strong and uncommitted, governments may be able to offer much less than they demand in take-it-or-leave-it offers as rebels will not fight to force greater concessions. With strong, committed groups whose demands represent their ideal point, governments may benefit by engaging in serious negotiations in order to uncover the group s reservation point for a particular settlement. This enables the government to offer the optimal amount of concessions that lie between the rebels original demands, which represented their ideal point, and their reservation point. The worse scenario yet is if rebel groups are strong, committed and unyielding in negotiations. An indefinite stalemate is likely to occur if the government resists the demands of a group of this type, especially if demands are made at their reservation point. The government would eventually be forced to give maximal concessions if they discovered that the group was strong enough to inflict unbearable costs or ultimately win the contest. Before deciding to make any offer, governments should take care to gather more information about these characteristics. Governments should be particularly interested in a group s demands and more specifically how flexible and committed they are to them. Quite a bit of uncertainty exists around all 14

24 of these parameters, however, as rebels have incentives even to misrepresent this kind of information. Governments do not know what type of group they are dealing with, but they do know that groups have incentives to manipulate information about their resolve and their demands to appear to be the type of group that should gain the most concessions in bargaining. Knowing that all groups want to appear to be the strong, committed, and obstinate type, governments regard any information freely offered by rebels as cheap talk. Rebels are not only expected to pretend they are stronger than they are, but also to demand greater concessions than they could ever possibly achieve through battle. Governments anticipate that rebels will pretend to be completely inflexible in bargaining and hope that by appearing obstinate themselves, they can call weak uncommitted groups bluffs, and force them to stop their fight. As new bargaining dyads share no history or bargaining reputation, both sides must learn about each other before they can form a credible agreement. Although governments and rebels may not have had any previous interaction to warrant this belief, governments still tend to think their rebel counterparts inherently lack credibility. Governments know that great incentives exist for rebels to try to gain as much as possible in settlements through bluffing and this fact alone makes rebels less credible. That these incentives exist is reason enough for governments to regard information provided by rebels as meaningless cheap talk not to be taken seriously. Despite this widely held conception, all rebels are not incredible. In fact, rebel credibility varies widely by group. The validity of information provided by rebels varies in the same way. And while discerning a group s credibility may prove quite arduous in the absence of bargaining history, there are actually some observable characteristics that may make it possible for governments to make distinctions between groups. One way of delineating between credible and non-credible groups is by examining 15

25 how likely groups are to be punished if they do bluff about the information they provide publicly. Groups that will face significant costs for misrepresenting a certain type of information will be less likely to misrepresent that information. These groups should be expected to avoid the costs of misrepresenting information, especially when the costs of bluffing outweigh the gains they can expect to make from misrepresenting information. Literature on conflict bargaining suggests there are certainly incentives for rebel groups to misrepresent information in civil war. If groups can convince their opponents that the information they offer is true, even when it is not, there is a real opportunity to extract better concessions in negotiations. Scholars, however, have not considered whether there are costs associated with misrepresenting information that might deter some rebels from doing so. Similarly, no scholarship has examined whether all rebels have the same propensity to misrepresent information. In the next section, I argue that the costs of bluffing for rebel groups are not uniform. Moreover, the penalties of bluffing for certain types of groups are likely to be too high for it to be a profitable endeavor. As such, bluffing becomes increasing unlikely as the costs of doing so mount. Information provided by groups likely to bear these high costs is more credible than information provided by groups facing no costs for misrepresenting. I distinguish between groups with high and low costs for bluffing by their ability to generate audience costs, such that groups with a mechanism to generate costs are more credible than groups lacking one. Specifically, I argue that the rebel support base is the mechanism to ensure rebel credibility. Finally, I offer expectations about how establishing that some types of groups have credibility might affect our understanding of bargaining outcomes. Rebel Credibility and the Costs of Bluffing Since bluffing about certain information is likely to gain rebel groups an edge in bargaining, it is reasonable, prima facie, to expect them to do it. If rebel groups can convince their opponents they 16

26 might eventually win or at least inflict significant costs, states may be more likely to offer negotiations that may lead to concessions. If rebels can convey to governments that they are really committed to their cause such that they will continue fighting until their aims are attained, governments may be more likely to yield to some of their demands. This suggests that for rebel groups, misrepresenting information is quite reasonable. Surely, if bluffing bore no costs, rebel groups would be remiss if they did not take advantage of the opportunity to mislead their government opponents. Extant arguments, however, fail to offer any insight into what one should expect when there are significant costs associated with misrepresenting information. This literature does not speak of costs that may be attached to bluffing nor does it explain whether we still expect rebels to bluff to gain greater bargains in the face of a consequence. Although it has not garnered any attention in the literature, there are, in fact, costs associated with misrepresenting certain types of information, specifically demands. Although public statements issued by rebel groups are often regarded as cheap talk by governments, publicly stated demands, for instance, can actually constitute costly signals revealing what rebels are willing to accept in settlements with their governments. Governments assume that any information provided freely by rebels must be cheap talk as rebels have motivations to falsify such information to do better in bargaining. Publicly stated demands, however, cannot always be cheap talk. Cheap talk is talk that is costless for the sender. While it is not directly tied to the issuer s payoffs, cheap talk can be used to alter a receiver s beliefs. For it to be cheap talk, no penalty can be attached to sending the signals, even if they are found to be false (Farell and Gibbons 1989). In the context of civil war, cheap talk can be information rebels provide to governments to alter what the state thinks about its opponent. Cheap talking rebels suffer no cost if what they say is found to be completely fabricated. Statements about capabilities, for example, can be considered cheap talk, as these statements are not made in front of audiences with incentives to inflict penalties for providing 17

27 false information. 9 Information about demands or resolve, however, should not be considered cheap talk in the same way, as there is generally an audience ready and willing to inflict costs on a rebel group that provides incorrect information about their demands or resolve. 10 More precisely, supporters investing in rebel campaigns serve as the audience that may inflict penalties on rebel groups when they issue inaccurate information about what concessions they are fighting for and how resolved they are to attain them. Supporters often offer to back rebel groups expressly because of the demands they espouse. Because their support is only contingent on rebels pursuing a particular aim, supporters have incentives to punish groups for claiming to fight for something for which they do not really intend to fight. In other words, if a support base decides to aid a rebel group because they believe the group is fighting for demand x, they are likely to withdraw their support if the rebels change their aims, are found not to be fighting for aim x, or if they back down before attaining x. Making public statements about demands or resolve in front of these audiences effectively ties rebel groups hands. In more certain terms, if rebel groups publicly profess to be fighting for a particular aim, that statement binds them to attaining concessions on the issue. If they fail to attain the concessions, the organization s supporters will levy heavy penalties. This effectively excludes demands from the type of information that can be manipulated without punishment, thus also excluding it from accurately being considered cheap talk. Audience costs rebels expect to face if they are caught bluffing must be 9 It is conceivable that bluffing about relative strength might also elicit a negative reaction if groups gain supporters solely based on beliefs about their strength. If supporters choose to back a group based on their strength alone, they might be upset to find the group was actually weaker than they claimed to be. This scenario, however, is different in scope than a case where rebels misrepresent the demands that gained them support. In the former situation, loss of support is likely the only punishment that would occur. While a loss of support would be detrimental in any other situation, it is not so in the case, as a group only loses support they would not have gained if they were honest about their strength. This makes bluffing, in this case, still the most logical strategy, as punishment is the same result as not bluffing. 10 There is not always an audience that can impose costs if demands turn out to be incredible. In cases where rebels do not need supporters to effectively execute war, groups should be more willing and able to engage in cheap talk. Signals about strength, resolve and demands from these groups, therefore, should be uninformative as there is no mechanism to generate costs if the information is untrue. I elaborate on this in a following section. 18

28 real for them to be effective. I argue that costs inflicted by their supporters are among the highest costs that can be suffered by rebel groups, namely loss of support for an organization or movement and deposition for a leader. The Rebel Support Base It is plausible to assume that like any organization, it is the paramount objective of a rebel group to maintain itself and ensure its continued survival (Crenshaw 2001, Siqueira 2005). If they cease to exist prematurely, groups are unlikely to see their aims realized. Moreover, if the organization does not thrive, the rebels within the organization forfeit the right to any benefits that may accrue from potential or eventual government concessions. By perishing prematurely, groups also fail to receive any benefits from existing as an entity, such as the ability to tax, loot, or exercise control over the civilian population. Since groups fight for a purpose, whether to extract resources, or to realize some political objective, they want to continue to survive at least until their ultimate aim is achieved. Basic survival for most rebel groups is guaranteed through a support base (Weinstein 2005; Fearon & Laitin 2003). This base consists of actual and potential recruits and external supporters or financial guarantors. 11 In order to maintain their survival, rebel groups must satisfy these constituencies to ensure their continued support. Rebel groups maintain their supporters only as long as they continue to provide some goods to them. Goods may be ideological (providing civilians with a means to air their complaints against the government or hope in believing that justice will be served) or something more tangible (for investors, providing goods to be sold or profited from, like drugs or minerals (Ross 2004)). 11 The civilian population can provide both types of support for groups. Civilian populations may support groups by providing troops or they may provide basic funding, like food, shelter and money for organizations. Money and weapons may be provided by domestic political groups while state sponsors may contribute by providing shelter, training, weapons, and funds. 19

29 Groups are representatives acting on the behalf of the community from which their support and supporters are drawn. Funding for groups political and armed programs come from their financial backers while soldiers are recruited from supportive communities. Since both are required to win in war, pleasing their supporters should take precedence even over winning battles and shortterm concessions from governments. Rebel groups should not compromise their long-term survival by making statements that alienate their supporters solely to make gains on the battlefield or in negotiations. Since their continued existence requires them to assume the role of agents, they are constrained to do their supporters bidding. In conflicts where there are multiple organizations, groups are most likely to suffer from defecting from their stated aims as supporters can easily pull out and join other organizations. If groups initially claim to fight for a particular issue and later abandon these claims, they risk disaffecting supporters who maintained the organization s original aims. Organizations tend to attract supporters from the causes they profess to fight for. When groups alter their aims such that they vary significantly from the positions of their constituents, they run the risk of alienating their support base. When supporters feel alienated from the groups they support they either withdraw support from the movement as a whole, or remain within the movement but transfer allegiance to another organization still maintaining aims within the scope of their own. Either way, the effect is that the organization changing aims suffers a loss of support that is crucial for the group s continued operation. The Sudan People s Liberation Movement/Army (SPLM/A) provides a good example. Although the SPLM/A predominately fought for the marginalized Black Christians in the south of Sudan, they also attracted the support of other ethnic groups in the country, including some Muslim groups in the North. The rebel organization was particularly attractive to communities in the South 20

30 Kordofan and Blue Nile states because they claimed to be fighting for all marginalized peoples of Sudan, many of which were concentrated in these provinces. Because of their commitment to ensuring that the conditions of all disregarded people in Sudan were improved, thousands of people from the Nuba Mountain region joined the SPLM/A in their fight against the oppressive Arab regime. Nuba support was not only important to the organization because of its strategic location between the Northern and Southern strongholds, but tactically as well, as the Nuba mountain region s backing lent heavily to the organization s military capabilities. In 1996, the SPLM/A entered into a formal alliance with the Umma party, the Islamic government ousted in 1989, as well as with other northern and southern opposition parties. The resulting Asmara agreement, signed by the groups of the new National Democratic Alliance (NDA), laid the framework for a future peace settlement and interim government of Sudan. Under the agreement, the new government would recognize Southern Sudan s right to self-determination and would set up a regime in Sudan explicitly separating politics from religion, both main demands of the SPLM/A. This agreement, however, failed to address any of the grievances held by SPLM/A s longtime allies in Southern Kordofan or in Blue Nile. Particularly, the agreement failed to include any provisions for self-determination for nonsouthern states. The Nuba felt both used and alienated and as a result withdrew their support from the mainstream SPLM/A group and started their own SPLA/Nuba Mountains faction. The emergent leader of the Nuba faction claimed that their departure was due to the main SPLM/A faction abandoning its initial objective of fighting for a democratic Sudan in favour of only liberating southern Sudan from domination by the Islamic north. 12 The emergence of this splinter proved problematic for the main group as both factions had to compete for the same scarce resources: monetary and popular support. A new group meant fewer 12 Sudan-Politics: Nubas Break Ranks With Main Rebel Group IPS-Inter Press Service July 2, Nuba Supporters To Separate From Rebel SPLA Group Agence-France Presse (English). June 28,

31 resources for the SPLM/A to draw on. Perhaps even more disadvantagous was the fact that the mainstream SPLA derived so much of its manpower from this community. Consequently, the split affected the group s ability to conduct its war against the state. Although the SPLM/A seemed to have a clear route to achieving its aims, it had to reconsider the strategy by which those aims were achieved. While the NDA plan provided the SPLM/A a way to see all of its demands realized, it would only work if the alliance was successful in ousting Omar al Bashir. This would only be possible if the SPLM/A and its violent allies remained militarily viable long enough to see that this happened. For the SPLM/A, this would be highly unlikely without the support base in the Nuba Mountains. Realizing the impact that their changing aims had on their ability to prosecute war, the SPLM/A had no choice but to reconsider the promises they made to their supporters. Out of necessity, the SPLM/A became reinvested in the plight of their supporters in the Nuba region. When the SPLM/A began negotiations with the government of Sudan, they decided to act as official representatives for the Nuba Mountains and the Blue Nile state. Although the question of selfdetermination for these areas remained one of the most contentious issues throughout their negotiations with the government, the SPLM/A maintained their aims until the end of the war. In the final comprehensive peace agreement inked in 2005, the government of Sudan and the SPLM/A signed a protocol on wealth- and power-sharing for the South Kordofan and Blue Nile States (United States Institutes of Peace 2004). Although the issue of South Kordofan and the Blue Nile were responsible for upsetting negotiations on numerous occasions, the SPLM/A was stuck in their bargaining position. The military threat posed by the organization was certainly among the reasons the northern groups wanted to ally and the government decided to offer negotiations and subsequent concessions to the SPLM/A. The organization knew that their ability to achieve meaningful concessions would be hampered if they failed to pose a military threat to the government of Sudan. The group recognized 22

32 they would not be able to credibly threaten the continuation of a painful war without their support bases in the Blue Nile and South Kordofan and the Nuba Mountains, in particular. Because the support of these regions was paramount to the organization s survival, the rebels were willing to stall negotiations indefinitely to stick to their aims. Without their supporters they would likely get nothing in the end, with their supporters they might have to endure a longer war, but they would eventually gain the concessions most important to them. That there were factions competing with the main SPLA group exacerbated the organization s need to maintain their aims at all cost. Because a real outside option existed for their supporters, the SPLM/A knew that they had to be particularly steadfast in their aims. Supporters could always transfer allegiance to another faction, like the SPLM/A-United, South Sudan Independence Movement (SSIM) or SPLA- Bahr el-ghazal faction. The presence of these alternatives not only made it more likely that the SPLM/A s survival would be jeopardized if they abandoned their aims, but it also increased the probability that any potential peace with the government of Sudan would be interrupted as the other groups would be more than willing to spoil peace if their interests were not represented in a settlement. That the SPLM/A was literally locked in their bargaining position and had no choice but to maintain aims in line with those of their supporters may have actually worked in the group s favor. If the government of Sudan was even the least bit attentive to the mêlée with the Nubas, they would have recognized that the SPLMA s supporters were not reluctant to punish the organization and its leadership for pursuing a set of aims that were incongruent with their own. After the group s initial experience with its supporters defecting, the government of Sudan should have gotten the impression that the SPLM/A was, in fact, beholden to its support base. This may have allowed the group to take a more hard-line stance in negotiations. If the government was convinced that the 23

33 rebel organization was truly stuck in their bargaining position, the government might have been convinced, similarly, that they were the only ones in the dyad able to compromise on certain issues during the negotiations. Thus, if they were genuinely interested in reaching a peaceful settlement to the conflict, the government may have been much more inclined to concede on certain points as they believed the rebel group had no ability to do so. Groups using their support bases as bargaining leverage might be incidental or it might be strategic. Rebels may use the fact that they have strong support bases to convince governments they cannot diverge from their bargaining positions, even if their support bases are not that constraining. 13 In general, it is reasonable to assume that governments are at least a little aware of the politics occurring within the organizations they are fighting. This is especially likely when events, like the Nubas splitting from the mainstream faction of the SPLM/A, make international news. It is also reasonable, then, to believe governments have some capacity to tell when groups have support bases and are able to estimate the probability that their rebels support bases can place constraints on their bargaining activity. Whether a group has a strong support base may factor into the government s initial calculations about a group s strength but, it should also enable it to estimate how flexible a group can be in bargaining as a group with a strong support base may be constrained in its behavior. While this may be detrimental in some respects, it should also be helpful for the government in deciding what kind of group to offer concessions to. Although constraints on rebel bargaining may cause immediate problems in trying to locate an acceptable settlement, it can also serve as a commitment mechanism to get rebels to adhere to agreements they sign. Particularly, the presence of a strong civilian support base ensures rebels will stick to the settlements they sign. Just as 13 Having a support base may cause a moral hazard where rebels are able to demand greater concessions because they know they can credibly claim to be bound by their constituents. I believe the moral hazard effect is attenuated by the very support base that allows the condition to arise. Because rebels would be bound to attain greater concessions if they demand them, groups may be less likely to demand exorbitant concessions, especially when they believe they will be unable to attain them. 24

34 supporters are likely to withdraw from groups that fail to advocate on their behalf, supporters can also withdraw from groups threatening to continue fighting after basic aims are met. Since these groups often need their supporters to remain viable in war, groups that lose support will be forced to stop fighting when they gain settlements favorable to their supporters. Mukherjee (2006:490) explains why having a support base often causes rebel groups to abandon their fight when they receive significant concessions. Speaking specifically of power-sharing arrangements, he says Because the insurgents political demands have been satisfied by the maximal political power-sharing concessions offered by the weak type, the insurgent group clearly does not have incentives to continue fighting because it is too costly to do so. Second, if insurgent leaders choose to fight even after their demands are met, as the defeated government provides maximal concessions, they will risk alienating their civilian supporters. This is because fighting is costly and civilian supporters of the insurgent group often face the brunt of the costs of fighting the government. Therefore, once their demands are met, insurgent leaders cannot justify the decision to continue fighting to their civilian supporters Mukherjee s point demonstrates that once rebels gain the concessions they had been fighting for, supporters are unlikely to continue providing valuable support. If a rebel group advocating on behalf of a civilian population gains significant concessions that address the grievances that lead to their initial support of violence, they no longer have an incentive to support continued fighting. Presumably, governments know this. If supporters act to constrain a group s bargaining behavior, they are likely also to constrain their conflictual behavior. Thus, groups with support bases may be considered somewhat more predictable and certainly more credible than groups that do not have 25

35 substantial support bases. Discerning governments, then, should be more inclined to offer concessions to groups with substantial civilian supporters than to those without them. Cost to Rebel Leaders Losing financial or logistical support are but two ways rebels can lose support from constituencies. Losing office is another way rebels may get punished for failing to deliver collective goods to their supporters. Like leaders of states, leaders of rebel groups fear being deposed by dissatisfied supporters. Supporters, both internal and external to the group, may have influence in deciding whether the leader of a movement is retained or replaced. Since rebel leaders often derive great benefits from heading groups, they should be reticent to engage in behavior that may jeopardize their positions. Through settlement, leaders often obtain important positions in eventual power-sharing government, like the special status conceded to National Union for the Total Independence of Angola s (UNITA) Jonas Savimbi by the Angolan government. In another instance, Foday Sankoh, the leader of the Revolutionary United Front (RUF) in Sierra Leone, was accorded the position of Chairman of the Strategic Resources Commission (which included control over the diamond mining industry) as well as privileges associated with the office of the vice president during negotiations with the government. Outside of negotiations, leaders also enjoy special privileges, including access to extractable resources, access to money, drugs and sex slaves, control over a large part of the population, and a national and international profile. Leaders of these groups know that there is a long line of subordinates waiting and willing to take their places in order to enjoy some of the things that make the position profitable. If these benefits are meaningful, leaders, like organizations, should avoid making controversial decisions that might inspire supporters to defect. This should include making demands that are controversial to the organization. 26

36 An ongoing internal disagreement over whether the SPLM/A s primary aim should be independence or self-determination lead to an attempted coup of John Garang, the group s leader, in September A faction within the SPLM/A demanded full-fledged secession while the main group advocated only for the right of the south Sudanese to determine their own future. Garang was reticent to oblige the dissidents, as he knew shifting the group s aims to independence would lead to a loss of valuable popular support. The organization may have prevented that split if it radicalized its aims, but such a move would have alienated the majority of the members of the organization. Moreover, the organization would have lost the support of a great number of their civilian and eventual political allies. On top of the detrimental affects to the movement, Garang would have been even less likely to remain the group s leader as he only survived the attempted coup because the majority of the group s members supported a more moderate position. If he took the more radical stance, he would have certainly been successfully deposed. The fractionalization of the organization proved painful for the main faction as it now had to compete with other groups vying for the same support. The split was especially difficult on the group as it also led to intense skirmishes with its new rivals. Surely, it would have been easier for Garang to change the aims of the organization. Such a stance, however, would have alienated any supporters interested in maintaining a unified Sudan, especially the northern opposition groups it eventually allied with to form the NDA. We can infer that a call for independence would have been a major source of contention in 1991 from its affect in While the SPLM/A did not adopt an official position advocating secession that year, a rift within the ranks of the NDA formed as the Sudan National Alliance (SNA) criticized the SPLM/A for advocating partition. In 2000, the Umma Party officially departed the alliance over the SPLM/A s position on secession The SPLM/A did not explicitly adopt a position of secession, but they maintained their demand for self-determination through a referendum on independence. The divergence in position came about when it became clear to the Northern groups that the results of such a referendum would likely be independence. 27

37 John Garang s predicament demonstrates the difficulty that rebel leaders face. While they are always concerned with their own immediate self-interest, they must simultaneously be concerned with the interest of the group. Paradoxically, even though faced with a coup, Garang was forced to put the interest of the group before his own as prioritizing the group enabled him to further his own interest. More pointedly, the illustration shows that controversial decisions by a leader, especially when regarding aims, may paralyze an organization s ability to maintain unity as members become divided about whether the group s leader is steering the organization in the right direction. These divisions tend to lead to shake-ups within the leadership, organization fissure or both. The Congolese Rally for Democracy s (RCD) leader, Ernest Wamba dia Wamba, fared much worse with internal dissention. In fact, Wamba was ousted as leader of two RCD groups. In May 1999, he was ousted from the main RCD group (RCD-Goma) for a failure to garner support from the Congolese people. 15 After he was deposed from the main group he started the Congolese Rally for Democracy-Liberation Movement (RCD-ML/Kisangani) from which he was ousted in November of Again, the elite members of the group claimed that Wamba was divisive and unable to maintain unity among the organization s members and supporters. Particularly, his supporters claimed he was attempting to separate from his main allies in Uganda. In a somewhat odd and unlikely twist of fate, the Ugandan People s Defense Forces (UPDF) swooped in militarily to restore power to Wamba dia Wamba. The internal dissention, however, did not dissipate. Coups of numerous other rebel leaders, including Sekou Damate Conneh of Liberians United for Reconciliation and Democracy (LURD), Foday Sankoh of RUF, Mahamat Saboum of Chadian National Front (FNT), Jonas Savimbi of UNITA and Sheikh Sharif Sheikh Ahmad of Alliance of the Garang: We're going to activate the SPLA in Khartoum proper Mideast Mirror. May 22, Bertrand Rosenthal. Rivalries within Sudanese opposition hang heavy over future. Agence-France Presse (English). Sudanese Drift from the Road to Peace The Independent (London). June 4, Ousted Congolese rebel leader vows to retain position. Associated Press Worldstream. May 20,

38 Re-liberation of Somalia (ARS), were attempted with varying levels of success. That so many African rebel leaders faced challenges to their tenure, suggests that like the groups they run, rebel leaders are also beholden to their supporters. Rebel leaders, then, should be doubly concerned with keeping their supporters happy as they risk stifling the organization s success as well as their own personal ambition if they fail to do so. Social movement resource mobilization literature examining the effect of radicalization on insurgencies suggests that to attract the greatest amount of support and minimize the probability of splintering over extreme and diverse preferences, groups have strong incentives to appear moderate rather than extreme (McAdam 1992, 1999; Haines 1984; Gupta 2002). Groups are encouraged to make and maintain demands that are likely to appeal to the greatest proportion of potential supporters. Groups are also encouraged to retain these aims over time as organization fissure is often the result of internal disagreements over changing aims. Splinter factions of existing groups may crop up to fill the void left when organizations divert their attention from important issue areas. Dissidents in existing groups may exit to form new organizations, taking with them valuable support. In the absence of a substitute, the entire movement may lose their backing when a group becomes too radical or begins to profess aims out of line with constituents preferences. Rebel groups, then, have strong incentives not just to appear moderate, but only to make demands favored by their supporters. 16 If rebel groups become too radical or extreme or simply deviate from the agreed upon program, they face alienating their support base which increases the probability that they will lose in conflict with the government. 16 If they forcibly recruit and they support themselves through resource extraction or drug sales, this will likely not apply at all. This will be discussed in greater detail later in this chapter. 29

39 Public Demands- A Public Good In order to maintain them, rebel organizations and their leaders must be responsive to their supporters. As most rebel organizations need support to function, they take special care to provide goods to their supporters that ensures their continued support. Groups provide agency, the collective good supporters expect in exchange for their backing. 17 To that end, rebels often demonstrate they represent a subset of a population by the demands they adopt as these petitions are often closely related to the grievances of their support bases. In order for demands to be effective at generating and maintaining support, they must be made publicly. Public demands enable rebel groups to retain old supporters, as these supporters become convinced that their backing is being translated into a tangible outcome. They also serve to broaden a group s support base, as latent supporters cement their support after realizing the group has policies congruent with their preferences. Therefore, once rebels overtly issue their demands, they are committing to the bargaining positions these demands represent. Groups become bound by these demands as an array of punishments exist for groups that renege on commitments to their supporters. While rebels issue public demands to attract support, they are not only accessible to rebels supporters but also to their government opponents. Rebels want governments to observe this information they provide. After all, how can rebel groups expect to gain concessions without first asking for them? In order for them to reach their targeted audiences, groups often make their claims known through statements given to the media, on their websites, through printed materials, hostages and, even by word-of-mouth. Neither governments nor rebel supporters can be excluded from this information and are therefore privy to the same information. That both types of audiences access 17 Kalyvas and Kocher (2007) argue that another collective good rebels offer their supporters is protection. Although this is a very important reason for civilian participation in rebellion I do not talk about this type of support here. 30

40 the same information further suggests that rebel groups with support bases are bound by their demands. 18 Compromise By publicly making demands, rebel groups effectively tie their own hands and lock themselves into a bargaining position. Schelling (1956) argues that making loud declarations is sufficient to create a commitment to a bargaining position. Groups accomplish this by making demands through public channels. When they make their demands public, they commit themselves to continued bargaining on the issue. While rebels do become bound to their bargaining positions when they issue demands publicly, this does not imply rebels have no ability to compromise in negotiations. In discussing commitments to bargaining positions, Schelling (1956:291) suggests an important distinction between qualitative and quantitative commitment. This distinction is not trivial. Qualitative commitment implies dedication to the principle of a demand while quantitative commitment obligates rebels to seek concessions corresponding to their demands to the letter. Qualitative demands are more important and should be considered binding while discussions over quantitative divisions of an issue are much less so. As rebels are expected to negotiate in negotiations during civil war, supporters do not expect them to be able to attain concessions without compromise on an issue. While rebel groups have a modicum of leeway when bargaining over the quantitative division of an issue, the principle of the issue cannot be compromised without penalty, especially when compromise entails dropping an issue completely. The Burundian Hutu CNDD-FDD, which demanded the complete disarmament of the Tutsi army in December 2002, later settled for a government concession stipulating the rebel group 18 Surely it is possible that rebels provide some of their supporters with private information and it is probable that rebels make separate demands in closed negotiations with the government. However, since demands are not only made to maintain old supporters but also to attract new supporters, rebel groups that only make their demands in private lose the capacity to gain new supporters through their demands. 31

41 would make up 40 percent of a new, reformed army where Hutus and Tutsis would be equally represented without any apparent backlash from their Hutu supporters (Watt 2008:80,96). In principle, the rebel group and its supporters wanted the Tutsi army restructured to ensure that the military were no longer able to commit atrocities against the Hutu population. Although the CNDD-FDD accepted less than they originally sought, the concession of military integration addressed the principle of the matter and was considered a sufficient gain for their supporters. Conversely, abandoning the issue of military restructuring altogether would constitute compromising the principle rather than the division of the issue. Such compromise is more likely to lead to penalties imposed by supporters than is give and take in the normal course of bargaining. 19 Signaling Commitment The rebel support base constitutes a mechanism by which rebel groups can credibly commit to their aims and settlements. More pointedly, the presence of a support base is a signal to the government that rebel group aims are credible. When groups issue public demands, they are observed by both their opponents and their supporters. These aims connote credibility because once they are announced rebels are bound to them as reneging carries the threat of considerable punishment. After observing that a rebel group has a strong base of supporters, opponents can be assured of two things. First, when rebels announce demands publicly, they are resolved to attain concessions on those issues. Governments can infer, then, that rebels with support bases will be more resolved in bargaining. Second, if governments offer earnest concessions to groups with entrenched supporters, they can be assured that rebels will accept them and not pursue more as the rebel s supporters have no reason to continue their support after the government makes substantial 19 If rebels were not equipped to make concessions, there would be no need for dialogue at all. Groups could simply make demands publicly and fight on until the concessions were attained. 32

42 concessions. This partially alleviates the commitment problem as governments can be more confident that rebels will not renege on agreements once they are signed. A rebel support base should facilitate bargaining between governments and rebels as these groups are more resolved and committed and their demands are more credible. The argument advanced above applies to a subset of rebel groups, albeit the majority. There are groups that may not generate audience costs and may not be punished for changing their preferences or failing to attain meaningful settlements. Rebel groups that do not rely on a support base to operate, or self-sustaining groups, are less constrained in their behavior and are more likely to bluff about their demands and resolve. Groups that do not rely on a base of supporters are not bound by any preferences external to the group. Groups like the LRA and the RUF, for example, that forcibly recruit and exploit natural resources, can sustain rebellions without a substantial support base as they generate their own funds and maintain continuous supplies of fighters without relying on any support base. Groups that have the ability to fund their activities with extracted natural resources are less likely to need to rely on civilians for support. In a similar fashion, groups that forcibly recruit, particularly through abduction, are less likely to have to rely on the surrounding community s support to remain viable. Rebels that use violent coercion as a recruitment tool are able to maintain a steady stream of recruits without offering any benefits to the communities that provide those recruits. Together with groups that can exploit natural resources, these organizations should be less responsive to any aggrieved subset of a population. Because they are not beholden to the preferences of any supporters they are also not bound by their aims as there is no one to inflict punishments for deviating from them. These groups are more likely to be disingenuous about their aims as there is no negative consequence for such behavior. When groups have either, but especially both traits, they should be less likely to receive offers of concessions from government, and less 33

43 likely to sign formal agreements. Conversely, they are much more likely to renege on arrangements when they do sign them. If publicly stated demands are, in fact, rebel organizations true aims, we can expect that when their demands are met rebel groups will be more likely to agree to settlements than when governments do not offer concessions consistent with their demands. 20 This relationship is predicated on the credibility of rebels or their ability to generate audience costs for bluffing. If rebel groups have strong bases of support, they should be much more likely to sign formal agreements and agree to terminate their conflicts when governments offer substantial concessions on their demands. When rebel groups do not rely on a local support base, they should be no more likely to sign agreements or terminate their conflicts. Finally, if extant bargaining literature is correct in its failure to distinguish between types of rebel groups, then distinguishing groups with support bases from groups without them should be a fruitless exercise. Moreover, examining government concessions to rebel demands should offer no purchase in understanding patterns of conflict resolution. If all rebel groups have the same incentives and propensity to bluff about privately held information, as is implicit in existing literature, we should expect to gain no useful information from statements about rebel demands. Therefore, looking at how governments address those claims is likely to be useless. Empirically, this means that inserting information representing the extent to which rebel group demands are addressed in statistical models should yield null results. In this case, addressing rebel demands should neither increase nor decrease the likelihood that belligerents locate acceptable agreements. 21 As I believe examining government concessions vis á vis rebel demands is likely to explain a great deal about belligerents propensities to settle, I hypothesize the following: 20 Concessions alone should not matter as much as concessions to demands. The latter should be more likely to bring belligerents closer to agreement in conflict. 21 This should be an easy test for the bargaining literature to pass as results can be null for a variety of reasons. 34

44 H1: Rebel groups that have support bases should be much more likely to sign formal agreements when governments make meaningful concessions on their demands. H2: Conflicts should be much more likely to terminate when governments provide meaningful concessions to groups that have support bases. Interestingly, neither the expectations implicit in extant bargaining literature nor the aforementioned expectation have been tested. In chapters 5 and 6, I empirically examine the effect that making concessions to rebel demands has on conflict resolution in civil war. Specifically, I examine whether rebel groups are any more likely to agree to a decisive end to their conflicts when they attain important concessions in negotiations with governments. If rebel groups regularly misrepresent their demands, the coefficients on the variables that represent the governments response to group demands should not have a statistically differentiable effect. In contrast, if rebel groups are sincere about their demands then there should be a positive relationship between the amount of concessions they receive on their demands and the different measures of conflict resolution. 22 In the next chapter, I describe the data I collected to test these arguments. In the proceeding chapter, I look at the effect that characteristics of rebel groups have on whether groups are invited to participate in negotiations and whether they gain concessions. In Chapter 5, I test whether rebels are more likely to sign agreements when governments make substantial concessions on their demands. The last empirical chapter, Chapter 6, examines how government concessions to rebel group aims affect the probability that conflicts will terminate after belligerents sign an agreement. I 22 Another option, which I do not expect, is that making concessions to rebel groups signals government weakness and therefore makes rebel groups more likely to forsake settlement and attempt to extract larger concessions during bargaining. Concessions may suggest that the government is weak or weak-willed and may inspire rebels to hold out for a better bargain or for eventual battlefield success. If concessions have this effect, rebel groups should be less likely to agree to settlement when their demands are met. 35

45 demonstrate that there is a separation between group types such that groups with support bases are much more likely to agree to settle than are groups without them. 36

46 Chapter 3: Demand and Concession Data An Investigation of What Rebels Want and Receive in Civil Wars Although data on rebel groups are generally sparse, information about rebel group demands is not. Since rebels most often use the media as a vehicle to publicize their demands, there is quite a bit of information about them. By issuing statements through the media, rebel groups try to credibly commit themselves to bargaining positions. Rebels may also channel their demands through the media to ensure that they reach the government. When rebel statements make the front page, governments must take notice of the issuer s complaint. Finally, groups might use the publicity gained from making public statements to engage the civilian population and ratchet up support for their cause both nationally and internationally. In some cases, attracting attention to a conflict may convince foreign governments and NGOs to apply pressure on states. Such pressure may eventually induce settlement. Because there are benefits attached to announcing demands publicly, rebel groups are often eager to do so. Information related to bargaining activity, however, is chronicled by the media even when it is not intentionally supplied by one of the belligerents. News services generally provide good coverage on negotiation processes in civil wars. Newspapers occasionally publish snippets of position papers issued by belligerents during negotiations. The BBC, in particular, issues transcripts of world broadcasts and often publishes records of radio programs issued by rebel groups. Therefore, there is a wide array of open source information on rebel group aims. However, while this information is widely available it is, unfortunately, quite tedious and time consuming to collect. It is, nevertheless, extremely useful information so it is a worthwhile undertaking. 37

47 Although some groups issue very few demands over the course of their conflicts with governments, others issue multiple demands within a single week. Some groups may maintain the same demand for years while others groups demands change much more frequently. This presents a challenge in coding. While most existing conflict datasets are coded on a yearly basis, constructing an annual rebel demands dataset proved impractical. By looking at the process at a higher level of abstraction, yearly, one would miss all of the notable variation as important bargaining activity is generally isolated to and clustered across weeks or months rather than years. In order to exploit all of the variation in demands and concessions over time, I collected rebel demands data on a monthly basis. The argument set forth in the preceding chapter is general and can be applied to any subset of rebel groups. More offers of government concessions to rebels should make settlement more likely in any geographic region. However, in this dissertation I focus solely on Africa to test the argument. As a practical matter, I choose a single region to facilitate data collection. However, Africa is an appropriate region to test this argument for a few reasons. First, many recent civil wars have taken place in African countries. Because of this, scholars have a renewed interest in Africa as a political region. Theoretically, Africa provides an ideal test case as it is often claimed to be a region where conflicts cannot be settled peacefully. For example, Rothchild and Hartzell (1993:63) write Africa s civil wars and especially those which hold out the prospect of the break-up of the state, are not normally amenable to political compromise and negotiation. Thus, African civil wars should represent hard cases for any study examining conflict resolution. Moreover, my argument relies on the expectation that rebel demands are credible. I argue that being self-sufficient, or not relying on a civilian support base should make rebels 38

48 less credible. Ross (2004) notes that many observers of Africa's recent civil wars have suggested that combatants are fighting for control of a resource. As controlling resources is a measure of self-sufficiency, using a sample comprised of African cases can be considered a conservative test of my argument. Since African conflicts are expected to be less likely to end peacefully and African rebel groups are expected to be less credible, the statistical relationship between rebel group demands and settlement outcomes should be weakest when tested on a sample of just African conflicts. Therefore, if a statistical relationship is established with the present sample, the relationship should hold elsewhere. Thus, while using data made up of only African dyads does pose some limitations (to be discussed in later chapters), doing so also provides clear benefits. Using the UCDP dyadic Armed Conflict Dataset (ACD), I identified all African dyads involved in intrastate conflict from Using open source information, mainly news articles in Lexis Nexis, Keesings, and Janes Security and Terrorism Monitor, I coded concessions demanded and offered for each conflict month. I bolstered the information gathered from news sources with data from historical accounts and narratives as well as from statements, manifestos, and political programs posted on or reproduced from rebel group websites. Specifically, I coded each time a rebel group made a claim against the government in a month. I then coded whether the government made a concession to that particular demand and if so how much they conceded in that month. 1 Although governments may make concessions later, I only coded whether a concession was made in the same month it was demanded. If a rebel group demanded x in April but the government offered part of x in May, a concession was not coded in April. However, if the group makes the same 1 I coded offers of concessions. I do not distinguish between those offers that culminate in implemented concessions or empty promises. This is important, however, for actual conflict termination and recurrence and will be addressed in later chapters. 39

49 demand in May, the offer would be coded for that month. In practice this coding rule is not really problematic as rebel groups often repeat their demands until they are adequately addressed. The SPLM/A, for example, demanded a referendum for self-determination and the repeal of Sharia law for the entire duration of its conflict with the government of Sudan. The concessions variable was coded on an ordinal scale. I recorded whether demands were ignored/rejected, or whether the government made minimal, moderate, substantial or maximal concessions. When a demand was announced, I recorded whether and how it was addressed by the government. The government could have ignored or outright rejected the demands. Alternatively, the government could have offered a range of concessions. The most preferred outcome for rebels is a maximal concession. A maximal concession is coded when the government gives a rebel group what it demanded in its entirety. A substantial concession is also a favorable outcome for rebels as it is a concession that falls just short of everything. If rebels asked for the vice-presidency in a power-sharing deal, for example, a substantial concession would be a post that came with the power, prestige and benefits of the vice-presidential post, but without the title. This is the type of concession offered to the Revolutionary United Front (RUF), the Justice and Equality Movement (JEM) and the Sudanese Liberation Movement/Army (SLM/A). When the RUF demanded the vice presidency in July 1999, the government of Sierra Leone offered Foday Sankoh a position that encompassed all of the privileges that would accrue to an actual vicepresident. In May 2006, the JEM and SLM/A movements demanded that the vice-president position be given to a Darfuri. Because of an earlier peace deal signed between the government of Sudan and the SPLM/A, the government could not oblige their demand, but offered instead that a Darfuri could occupy the position of Special Assistant to the President. Again, this position was reported to be equivalent to a vice-president position in everything 40

50 but name. In both of these cases I recorded a substantial concession as the government did not offer full concessions, the vice-presidency, but made an offer that was quite close. Governments could also offer moderate and minimal concessions. Moderate concessions are modest offers. They are neither very large nor very small offers. These are offers that might be considered sufficient for settlement, but just as easily insufficient. In most cases, I used the moderate category as the midpoint for the ordinal coding of the concessions variable. This proved quite useful when a concession on a demand could be quantified. For example, in January 1994, the Angolan rebel group UNITA demanded that 40 percent of its troops be integrated into the existing 30,000 troop army. In that month, the government offered to integrate 5,500 rebel troops, which constituted a little more than an 18 percent share. I coded this concession as a moderate concession, because it was approximately midway between no concession and a maximal concession. When divisions of demands cannot be considered quantitatively, coding is a bit more difficult. For example, among SPLM/A s main demands was the repeal of Sharia law in Sudan. In May 1989, the government countered this demand by making a provision in the constitution that guarantees religious freedoms for all Sudanese. While the decision was not much in the way of a concession on Sharia, the government did concede something relevant, no matter how trivial. This concession was coded as minimal as it was almost insignificant, but more than an offer of nothing. In response to the same demand, in July 2002 the government offered to exempt the majority Christian South from Sharia while the laws were retained in the Arab North. Again, this is not a maximal concession, but it was substantial and was coded as such. 41

51 Because I do not expect that individual concessions on demands have independent effects on the probability of agreement, the demand-level is not the appropriate unit of analysis for me. Instead, I expect the bundle of concessions in a period will most influence settlement. Therefore, I begin by choosing the dyad-month as the unit of analysis. I aggregate the demand-level data into a dyad-month dataset by combining the individual demands into monthly data. I calculate and report the proportion of significant concessions governments make to rebel demands by first counting the number of significant concessions a government made to rebel demands during a month of conflict. I then divide that number of significant concessions by the number of demands rebels made in that month and multiply the quotient by 100. I create two measures that record whether the government makes major concessions to rebel demands. The first codes a more relaxed measure, weak concessions, that represents the percentage of demands the government responds to with moderate, substantial or maximal offers. I also code a more strict measure of concessions that records the percentage of government offers that fall in the maximal or substantial concessions. This is labeled strict concessions. 2 Coding both of these variables is important for at least two reasons. While the strict measure should represent concessions that are considered most valuable to rebels and their supporters, it is possible that even moderate concessions are sufficient to encourage an agreement. Second, including two measures serves as a robustness check for my analyses. It ensures that the results I obtain are not an artifact of subjective coding rules about how substantial a concession was. If the categories that make up the strict concessions variable are coded imprecisely, a significant amount of measurement error may be introduced into my analyses. Measurement error can lead to increased standard errors which attenuates the significance of the coefficients. By 2 To ensure the robustness of my results, I also ran models that included the number of concessions governments offered rebels in a month as the main independent variable. Results do not change substantively. 42

52 aggregating a greater number of categories, I can reduce the effects that imprecisely measured covariates may have on the results. In negotiations, some demands are certainly more important for settlement than others. If a government makes a substantial concession by offering to redistribute wealth and hold elections (assuming these to be main rebel demands), it is hard to believe that not conceding to a change in the structure of negotiations would inhibit settlement. I attempted to code a measure that represented whether governments made concessions on main rebel demands. However, I found such a measure quite difficult to code. A measure that accounts for how important a demand is to a rebel group is likely to be biased and unreliable as it would require me to make too many subjective judgments about whether a demand was truly important to rebels. While in some cases coding could be relatively easy, as some rebels are explicit about which of their demands are most central to their conflict with the government, most rebels are not. I attempt to proxy such a measure by examining whether a demand is political. Concessions on political demands should have a greater effect on the probability rebels and governments sign deals than would concessions on other types of demands. To test this assumption, I include two variables that record the extent to which governments address rebels demands that are political in nature. These variables are labeled weak political concessions and strict political concessions. I code whether governments make any additional concessions because, contrary to my expectations, rebel groups may be sensitive to concessions in general and not only to those that address their demands. Alternatively, additional concessions may demonstrate good faith by the government and inspire a group to sign an agreement. Methodologically, 43

53 this variable may be important because it should pick up on concessions made to demands I missed while coding and will serve as another robustness check on my coding. A more detailed account of the variables I collected is at the end of this chapter. I provide codebooks for the demand-level and dyad-level datasets. In the section that follows, I examine the types of demands rebels tend to make and the concessions governments tend to offer in response. What Rebels Want and Receive in Civil War Surely, whether rebels and government are able to come to agreements is largely dependent on what rebels ask for in exchange for peace. Certain aims are likely to be more amenable to settlement than others. Toft (2006) argues that settlements should be more difficult to come by in conflicts where stakes are indivisible. We should be least likely to observe bargains when the main issue in contention cannot be compromised such that one party gains everything while the other losses everything. Walter (2004) describes absolute or total goals as rebel aims that call for complete control of a state. This includes rebel demands for the violent overthrow of a sitting regime or an entrenched political system. Toft (2006) proffers that a quintessential indivisible issue is territory, particularly the homeland for a particular ethnic group. Territorial conflicts are likely to be most intractable when they are fought over land inhabited by groups that believe the territory in contention is tied to or representative of their identity. Both Toft (2006) and Walter (2004) argue that conflicts surrounding absolute or indivisible issues are likely to lead to deadlocks in 44

54 bargaining. Toft believes conflicts over indivisible issues will last much longer, while Walter believes conflicts over absolute aims are also to lead to the recurrence of conflict. Here, the data show that governments are, in fact, less given to compromise on absolute or indivisible issues. I consider demands to be either absolute or indivisible when there are explicit calls for independence or secession, or when rebels demand the complete resignation of the head of state or the government. Allowing for repeated demands, there are 98 cases of rebels demanding independence and 166 cases of rebels demanding the complete resignation of the leader or the overthrow of the regime. In the data, governments offered concessions on demands for independence about 9% of the time. Governments compromised on less than 8.5% of the demands calling for the unseating of their regime. When just considering significant concessions, or concessions of either the substantial and maximal types, governments are especially reticent to give into absolute demands in order to stave off conflict. In the sample, governments never compromised by offering complete secession to any territory. Governments, however, made substantial concessions three times to two different groups. In 1989, Morocco agreed to allow a referendum for independence in Western Sahara and Angola offered FLEC-R autonomy in While governments did not offer full concessions on independence, there are two cases in the sample where governments fully obliged rebels demands for the head of the state to step down. In 1991, Siad Barre, the President of Somalia, agreed to leave at the USC/SNA rebels behest and in 2003, Liberian President Charles Taylor yielded to LURD s demands that he resign. 45

55 Absolute or indivisible demands consist of only a fraction of the issues between rebels and governments in civil wars. I coded 3258 demands across 89 African rebel groups. Absolute demands make up a little more than 8% of the demands rebels made. If issue indivisibility makes it particularly difficult for governments and rebels to come to agreements, it should be easier for rebels and governments to locate acceptable bargains since about 92% of demands do not fall in the absolute or indivisible category. Yet governments only made compromises on 14% of rebels demands in the sample, (that is, they conceded something to 459 demands). It is notable that cultural autonomy demands occur most infrequently in the data. Demands for cultural autonomy often arise when a state represses the expression of certain ethnic identities. Rebels mobilize in response to these policies of exclusion to demand their culture, frequently language rights, be respected by the state. Concessions on these issues often require governments to alter their policies disqualifying or excluding certain ethnicities or attributes of ethnic groups from state institutions. Examples of demands that fall under the category of cultural autonomy are CRA demands for the Nigerien government to allow the institutional teaching of Tamashek, EPRDF calls for the restoration of culture and language rights to marginalized groups in Ethiopia, OLF bids for the Ethiopian government to respect the rights of the Oromo nation, specifically, with respect to the teaching of the Oromo alphabet and language in schools, and the BDK demands that the government allow the Kongo tradition to flourish. It is noteworthy that demands for cultural autonomy are the only demands related to ethnicity in the sample. This is curious since there has been so much attention given to ethnicity in civil war literature. Many scholars argue that the majority of civil wars of late are 46

56 fought over ethnic contentions as opposed to being waged over political and ideological goals. Scholars, like Kaldor (1999) and Huntington (1992/3) for example, expect conflicts over ethnicity to be more severe, more deadly and certainly more frequent than other types of conflicts. In light of these claims, it is difficult to understand why there are so few demands related to ethnicity. That ethnicity is not frequently the subject of rebel demands suggests that existing explanations for the intractability of civil wars are likely inaccurate. Toft (2006) suggests that most civil conflicts are intractable because they are fought over issues of group identity, demands that are indivisible and less amenable to compromise. That cultural autonomy is the only type of demand expressly related to ethnicity and that these demands represent only about a quarter of 1% of all demands, suggests that ethnicity is not a salient explanation for the difficulty in settling civil wars. Together, so-called indivisible issues make up less than 8.5% of the demands rebels made in the sample. If we think of demands for independence, full control of the government and ethnicity as the demands that are least amenable to compromise, it becomes clearer that the nature of the issues in civil wars is neither what causes its duration or the failure of peace settlements. It must be another explanation and I believe the explanation is the degree to and manner in which rebel demands are addressed as opposed to the constitution of the demands themselves. People rarely fight over ethnicity. Only in very extreme cases do ethnic groups actively seek to eradicate each other, and even when they do, as in Rwanda, they do so to consolidate political power. Scholars, however, often expect that in order to be settled, conflicts involving groups of different ethnic backgrounds require compromises on ethnicity. Since ethnicity is often thought to be a fixed attribute, compromises on ethnicity 47

57 are thought to be impossible. Therefore, in ethnic conflict a win for one ethnic group means a loss for the other ethnic group(s). However, it is important to consider that even in conflicts with a significant ethnic component, the demands are likely expressed in political terms. The Burundian civil war is an exemplary case of a conflict mobilized along ethnic lines where the demands and expected solution were nevertheless cast politically. The conflict in Burundi could be considered an ethnic conflict insofar as it was fought between Hutu rebels and a Tutsis government/military and the violence was distributed along ethnic lines. The solution envisaged by the Palipehutu- FNL, CNDD and CNDD-FDD, however, was political. While the conflict was initiated by rebels disputing the Tutsi army s use of violence against the Hutu population, the solution sought by the rebels was not about whether Tutsi s and Hutu s could remain ethnically different and co-exist in the same state, but it concerned the distribution of power in the country that contributed to the army s abuses of Hutus. The main concession required of the government was the disbandment and the reconstitution of the Tutsi dominated army. This solution would allow both parties to share power, as power was concentrated in the military rather than the executive. While there were certainly conflicting ethnic groups, the conflict could be considered political as the demands and grievances were expressed in political terms. This is an important distinction as the expression of issues in this manner, politically rather than ethnically, allows parties to locate and agree on terms of settlement. In fact, the conflicting parties in the Burundian civil war were able to come to a negotiated end of this conflict by compromising on the issue of military restructuring and integration. Fittingly, the Burundian rebel s primary aims were coded as demands for military integration. 48

58 While some scholars argue that ethnic diversity, fractionalization or heterogeneity cause civil war onset (Ellingsen 2000), other scholars have tested and found no relationship between ethnic variables and conflict onset (Collier and Hoeffler 2004; Fearon and Laitin 2003). This suggests that ethnicity, as an explanation, is not so salient in explaining where conflicts are likely to occur. While having competing ethnic groups may not be an adequate explanation of when rebellions are likely to be launched, a sizeable amount of literature focuses on how rebels mobilize participation around issues of ethnicity and identity (Sambanis 2000; Sambanis 2001; Fearon and Laitin 2003; Humphreys and Weinstein 2008; Blimes 2006). Ethnicity may not be the main issue in conflict, but it may explain who fights on either side. Kalyvas (2001) explains that reasons for participation in civil war are not always congruent with the motives of the rebel elite. Political entrepreneurs may use ethnic rhetoric as a recruitment tool even when ethnic grievances are not the reason for war (Fearon and Laitin 2003; Humphreys and Weinstein 2008). In other words, conflicts may be mobilized along ethnic lines, even if the issues at stake are not divided along ethnic lines. This explains why even though scholars expect ethnic war to be so prevalent, very few demands in my sample are related to ethnicity. 3 <Insert Figure 3.1 Here> Before discussing further how governments tended to respond to rebel demands, it is prudent to first consider what they responded to. There is wide variation in rebel demands in my data. I coded 32 different types of demands which are listed in Table 3.1 and described in the codebook. Aims ranged from highly political, collective demands (calls for independence or the reorganization of the state) to apolitical demands (requests for food, 3 Ethnic war is expected to be especially prevalent in Africa. Therefore, the finding that few demands are related to ethnicity is unlikely to be an artifact of the region the data are from. 49

59 medicine, supplies or personalistic benefits) that only satisfy members actively involved in these organizations. Figure 3.1 displays the distribution of rebel demands in the data. From the distribution, it is clear that rebels are most likely to demand governments make substantial changes to their structure in exchange for the abatement of violence. The government change category contains a range of demands related to substantial changes in the government. While the most extreme version of this type of demand is the complete resignation of the head of state or government, there are more moderate demands frequently issued that fall within this category. Demands rebels make in this category range from vague demands such as broad political reforms to demands detailing specific changes the government ought to adopt such as federalism or a bicameral legislature. Rebels have demanded their governments change from presidential to parliamentary system and from one party to multiparty states. A sizable proportion of demands in this category are for democratization or some form of a coalition government. Also included in this category is the small subset of demands calling for the departure of the head of state. This category, government change, constitutes about 12% of all of the rebel demands in this sample. Second most frequent to changes in government, rebels demanded concessions regarding the peace process. This category includes changes in negotiators, protocols, and procedures as well as calls for negotiations. As outlined in the preceding chapter, negotiations are not automatically offered to all groups, thus a government s decision to negotiate serves as a signal of legitimacy for groups that attain them. As negotiations with the government alone are a significant concession, it is hardly surprising that rebel groups frequently demand them. In addition to demands for explicit bargaining with the government, rebels also demand accommodations during negotiations. They demand different negotiators, new venues, different negotiating protocols, and sometimes they 50

60 demand the neutrality of third parties already involved. This category also includes claims regarding previous peace processes. A frequent source of contention in this regard surrounds the government s implementation of agreements and protocols previously signed with presently-active rebels. For example, in November 2008, the CNDP demanded the government in the Democratic Republic of Congo implement a peace deal they signed in January of that year. Additionally, new rebel groups emerge to induce their governments to comply with deals made with other groups. In 2007, the MNJ was fighting the government in Niger in an attempt to force them to implement a peace accord they signed in 1995 with another rebel group, the Organization of Armed Resistance/ Coordination of Armed Resistance (ORA/CRA). The third most frequent type of demand falls in the other category. Demands coded other were stated aims that did not fall neatly into any of the other categories in the typology. A separate category was coded when the principle of a type of demand was distinctly different from all other existing categories, was theoretically interesting or when a demand appeared recurrently across groups. In order to make the typology effective, the number of categories coded needed to be limited. Although I attempted to group as many demands as possible in a single category, some demands proved difficult to fit conceptually into a pre-existing category. For example, despite it being observed across a few rebel groups, the demand for the closure of detention centers and refugee and regroupment camps as demanded by the AIS in Algeria, and the CNDD-FDD and Palipehutu-FNL in Burundi, was coded as other. A similar demand also coded in this category was the return of refugees. While coding a category for refugee related concerns could certainly be 51

61 reasonable, there was not a great mass of these demands to immediately warrant its own category. 4 Most other demands aggregated into the other category were quite disparate and were more difficult to group. Some of these demands included calls by the GIA in Algeria for the closure of the French Embassy, the death of all foreigners and most surprisingly, the conversion of French President Jacques Chirac to Islam. Other aims coded in the other category were demands like UNITA s calls for bodies of dead rebels to be returned to the organization, Al-Shabaab s demands for the banning of Khat, Boko Haram s calls for the jettisoning of western behavioral patterns in Nigeria, rebel demands for televised debates, the end to propaganda and name-calling as well as press blackouts. Demands such as calls for national reconciliation and unity, calls for the emancipation of women and demands for international condemnation of atrocities occurring under the guise of war were also coded in this category. While these are conceptually important, they are also different from most other types of demands in that they were symbolic in nature and perhaps, less tangible. <Insert Figure 3.2 Here> Other very important political claims, were those made for power-sharing (6.5%), political autonomy (4.3%), military restructuring (4.3%) and religious rights (4.2%). Demands for religious rights may be considered in the same vein as demands about ethnicity. Of the 108 rebel groups in the data, only 12 groups demanded concessions about religious rights. This is evident from Figure 3.2. Figure 3.2, shows the distribution of demands by rebel group, while Figure 3.1 shows the distribution of all demands, allowing for repeated demands. The demands on religious rights tended to fall dichotomously into two 4 A failure to include an exhaustive list of demand types has no effect on the results of the empirical models. The typology simply makes it easier to draw comparisons across types of demands. Having too few categories only makes it a little more difficult to see patterns in the distribution of demands as there is this large aggregated Other category. 52

62 main types: demands for the repeal of Sharia law and the establishment of Sharia law. More accurately, these demands can be classified as either for an Islamic state or for separation of state and religion. The installation of an Islamic government and strict state adherence to Sharia law or Islamic courts were professed as aims for the AIS and GIA in Algeria, Ahlul Sunnah Jamaa and Boko Haram in Nigeria, ADF in Uganda, MPCI in Cote d Ivoire and Al- Shabaab in Somalia. The EPRDF in Ethiopia and the SLM, JEM, SPLM, and NDA in Sudan all demanded the separation of state from religion. The EPRDF was the only group not seeking to overturn Islamic law in that country. Figures 3.1 and 3.2 give a clear picture of the distribution of concessions demanded by rebel groups in this sample. The next three figures exhibit what governments actually give in response to these demands. Based on the four-point ordinal scale capturing the extent to which governments concede to rebel demands, Figures respectively display the types of moderate, substantial and maximal concessions governments tend to make. Figure 3.3 shows governments are more likely to offer moderate concessions on demands for power-sharing than any other type of demand. Power-sharing tends to be one of the most frequent type of concessions overall, as governments are also most likely to make substantial concessions (Figure 3.4) and are likely to make a large number of maximal concessions (Figure 3.5) on power-sharing demands. Governments are also very likely to make moderate concessions on religious rights, but are not nearly as likely to make substantial or maximal concessions on these demands. This suggests that while governments are willing to compromise on issues of religion, the extent to which they are willing to compromise is very limited. Although governments are not extremely likely to make moderate concessions on issues related to the change in government or government policy, governments are willing to 53

63 make larger compromises on these issues as this is the second most frequent type of substantial concession and the most frequent type of maximal concession. While governments do not often compromise on the extreme versions of these demands, like full control over the government, they are likely to make concessions on the more moderate calls for changes in government and government policy. Surprisingly, governments are highly likely to alter the status quo by redistributing power through power-sharing, modifying the structure of the government, and altering the balance of power within the military. Further research should focus on determining whether the type of concessions offered rebels in civil war can be explained by the characteristics of rebels, states, or the conflict. It is likely that certain types of government structures are more permissive of compromises on the distribution of power than are others. While patrimonial governments, for example, have a lot to lose by altering the status quo, they are likely to lose more when they are forced to relinquish power entirely. On the other hand, while power-holders in highly structured institutions are less disadvantaged when they are forced to share government or military power, alterations are likely to be more difficult as institutionalized bureaucratic structures often prevent states from making substantial changes easily. Which groups receive negotiations and concessions is likely systematic. In the next chapter, I examine group-level features that make government concessions more probable. Certain features of a rebel organization are likely to affect the probability the group will receive concessions from the government, with negotiation being a main concession. Rebel strength and viability are especially likely to be key factors in determining whether rebels and governments even enter into negotiations. When rebels are strong or can levy heavy costs on their opponents, governments are more likely to seek peaceful settlements (DeRouen and 54

64 Sobek 2004). Other factors very likely to affect governments decisions to make offers are likely to be those allowing rebels to impose extreme costs on governments. Rebel strategies of terror, terrorism and crime are likely to affect a government s willingness to concede. Rebel credibility is also likely to impact whether governments decide to negotiate with rebels. If groups are likely to be punished for changing or misrepresenting their aims, their bargaining behavior is likely to be more consistent than groups with no accountability. Assuming governments are actually able to assess rebel credibility, they should be more willing to negotiate with groups they believe to be credible as these groups are more likely to stick to bargains they make. Negotiating solely with credible groups should partially alleviate commitment problems as governments can be more confident rebels will stick to bargains they strike. When conflicts are long or costs borne by society are particularly high, conflicts are likely to be most ripe for settlement. As the costs of conflict mount, both sides should be more willing to pursue a peaceful settlement to their dispute. DEMAND-LEVEL DATA CODEBOOK In this codebook I list and describe the variables I code from the data I collected for this project. I also include tables elaborating on my coding rules. Dyad ID- The dyad id is taken from the UCDP Dyadic Armed Conflict dataset version It is a unique identifier for each dyad in conflict. This ID variable marks the rebel-state dyad to which the demands and concessions data correspond. Conflict ID- This identification variable was taken from the UCDP Dyadic Armed Conflict dataset version and marks the conflict the dyad is involved in. This variable is 55

65 different from the dyad id as some conflicts contain multiple actors. (ex. SLM/A, JEM, NRF, and SLM/A-MM were all part of the same conflict with the government of Sudan) Side A- This variable identifies the state in the conflict dyad. Side A, in almost every case, is the actor expected to respond with concessions to rebel demands. This variable is taken from the UCDP Dyadic Armed Conflict dataset version GWNOA- This variable is the Gleditsch and Ward state code for Side A. It is taken from the UCDP Dyadic Armed Conflict dataset version Side B- Side B is the rebel group in the dyad. This variable identifies the group expected to issue demands in a conflict. This variable is taken from the UCDP Dyadic Armed Conflict dataset version Month- Month is the month in which the demand or concession was made. Year- Year is the year in which the demand or concession was made. Period- The period variable is a count of months conflict within the dyad has persisted. This variable is a unique identifier for each time period the dyad is in conflict. Demand Type- The demand type variable records the type of demand that rebels issued in that month. There are 32 disparate demand types. The number corresponding to each type of demand is recorded below in the right hand column followed by the name of the category and a description of the category. Rebels can demand multiple demands at a time. Demand Detail- This cell contains information about the detail of the rebel demand. The exact content of the demand was recorded as precisely as possible in this variable. Political Demand- In attempting to ascertain which demands might be more important to rebel groups, I created this variable to capture whether demands were political in nature. This variable was coded dichotomously, with political demand taking on a value of one. Demands were not coded as political if they could be considered exclusive to the conflict situation. If the same demand might be made by a group not engaged in violence, it was coded as a political demand. For example, security-related demands were generally not considered political nor were demands regarding the withdrawal of foreign forces as in most cases, the security situation would not be dire, nor would intervening forces be present if not for the violent conflict. The end to civilian targeted violence and the release of political prisoners could be considered political goals, as instances of violence against civilians and the detaining of political prisoners might occur outside the context of war. Specific demands were not coded as political, but the categories were coded either political or apolitical. Whether a category is considered political is listed in the last column of the table above. Previous Demand- Whether a specific demand was previously voiced is recorded here. This variable takes on the value of one if the same demand was voiced in a prior period. 56

66 Concessions- This measure of concessions is an ordinal variable that captures the level of concessions governments offer rebels in a month. Concessions are coded when a government addresses, at least in part, something rebels demand in that month. Although a delay between rebel demands and government offers is conceivable, concessions are only coded within the month that rebels demand them. If the issue is still important, rebels are likely to repeat their demand again until they believe it has been addressed. Thus, there is likely to be great overlap between concessions and demands. In some months, governments make multiple offers. In these cases, I code the last concession governments offer on a demand in a month. The coding rubric is below in Table 3.2: <Insert Table 3.2 Here> Action Against Foreign Government- This dichotomous variable takes on the value of one, when the demand is targeted at a foreign government. This includes demands made for ransoms. Also coded in this variable are demands where consequences related to the failure to offer concessions is directed toward a foreign government s interests or the interests of a foreign business. DYAD-MONTH-LEVEL DATA CODEBOOK Dyad ID- The dyad id is taken from the UCDP Dyadic Armed Conflict dataset version It is a unique identifier for each dyad in conflict. This ID variable marks the rebel-state dyad to which the demands and concessions data correspond. Conflict ID- This identification variable was taken from the UCDP Dyadic Armed Conflict dataset version and marks the conflict in which the dyad is involved. This variable is different from the dyad id as some conflicts contain multiple actors. (ex. SLM/A, JEM, NRF, and SLM/A-MM were all part of the same conflict with the government of Sudan) Side A- This variable identifies the state in the conflict dyad. Side A, in almost every case, is the actor expected to respond with concessions to rebel demands. This variable is taken from the UCDP Dyadic Armed Conflict dataset version GWNOA- This variable is the Gleditsch and Ward state code for Side A. It is taken from the UCDP Dyadic Armed Conflict dataset version Side B- Side B is the rebel group in the dyad. This variable identifies the group expected to issue demands in a conflict. This variable is taken from the UCDP Dyadic Armed Conflict dataset version Month- Month is the month in which the demand or concession was made. Year- Year is the year in which the demand or concession was made. 57

67 Period- The period variable is a count of months that conflict within the dyad has persisted. This variable is a unique identifier for each time period the dyad is in conflict. Main Group- This dichotomous variable was coded one when the rebel group (side b) in the dyad is the main group in the conflict that year. In most cases, there is only one rebel group considered the main group in a conflict, but in some cases multiple rebel groups are coded a main group in a conflict. If reports indicated the government considered the group a main group, this variable was coded one. This variable was also coded one if the group inflicted the greatest number of fatalities on the government (side a). In cases where these two forms of information conflict, multiple groups were coded main groups in that conflict. Multiple groups could also be considered a main group if there were multiple groups that inflicted roughly the same number of fatalities on the government in that year. Negotiations- Negotiations is a dichotomous variable taking the value of one when the government and the rebel group are engaged in dialogue in that month. This variable is coded without respect to whether the negotiations were formal or were supported by a third party. If the rebels and government held a meeting with the intention of talking about the demands and the road to peace, this variable was coded one. The belligerents also have to have been the main interlocutors at the meeting and not just both in attendance. Also coded one, are dyad months where there was no longer a formal government but the rebels and the main authority in the country (former rebel group, mediators) meet to discuss the return to peace. Agreement- Agreement takes on a value of one when the rebel group in the dyad and the government sign a formal agreement. Both parties must sign the agreement in order for it to be considered. Agreements can be political or apolitical. They may be formal cease-fires that contain no political provisions or they may be peace accords that are more specific about the peace process. There may be formal concessions on rebel demands, but this variable is not only restricted to such agreements. Agreement Detail- This cell provides details for the agreement recorded in the Agreement variable. In most cases, it contains the name, date and main provisions of the agreement. Political Agreement- This variable is a subset of the Agreement variable. Political Agreement takes on a value of one only when the provisions of the agreement are political in nature. Thus, agreements that are only concerned with security arrangements are excluded. Number of Demands- This variable is a count variable that records the number of demands rebels made in that month (period). Number of Political Demands- Using the definition of political demand I proffer above, I generate a count variable recording the number of political demands rebels made in that month. Number of Concessions (Strict)- Using the definition of concessions offered above, I code a strict measure of concessions by counting the number of substantial and maximal concessions governments offer rebels in a month. 58

68 Number of Concessions (Weak)- I code a weak measure of government concessions by counting the number of moderate, substantial and maximal concessions governments offer rebels in a month. Percentage of Concessions (Strict)- I calculate this strict measure of government concessions by dividing the Number of Concessions (Strict) variable by the Number of Demands variable and multiplying it by 100 to generate a variable that represents a strong interpretation of the percentage of rebel demands governments make concessions to in a month. Percentage of Concessions (Weak)- I calculate this weaker measure of government concessions by dividing the Number of Concessions (Weak) variable by the Number of Demands variable and multiplying it by 100 to generate a variable that represents a weak interpretation of the percentage of rebel demands governments make concessions to in a month. Number of Political Concessions (Strict)- This count variable captures the number of substantial and maximal concessions governments offer rebels on their political demands within a month. This variable is a subset of the Number of Concessions (Strict) that excludes concessions made on non-political rebel demands. Number of Political Concessions (Weak)- This count variable captures the number of substantial, moderate and maximal concessions governments offer rebels on their political demands within a month. This variable is a subset of the Number of Concessions (Strict) that excludes concessions made on non-political rebel demands. Percentage of Political Concessions (Strict)- I calculate this weaker measure of government concessions by dividing the Number of Political Concessions (Strict) by the Number of Political Demands and then by multiplying the quotient by 100 to generate a variable that represents a strict interpretation of the percentage of rebel demands governments make concessions to in a month. Percentage of Political Concessions (Weak)- I calculate this weaker measure of government concessions by dividing the Number of Political Concessions (Weak) by the Number of Political Demands and then by multiplying the quotient by 100 to generate a variable that represents a weak interpretation of the percentage of rebel demands governments make concessions to in a month. Additional Concessions- This dichotomous variable is coded when the government offers concessions that do not directly address any demand rebels made in that month. This variable would be coded one if a government responded with a concession to a rebel demand made in the month prior. Additional Concessions Detail- This cell provides details for the concession variable recorded in the Additional Concessions variable. The exact content of the concessions was recorded as precisely as possible in this variable. 59

69 Chapter 4: Rewarding Bad Behavior: How Governments Respond to Rebel Group Tactics in Civil Wars Introduction: Putting the Horse Before the Cart A fair amount of recent scholarly work focuses on explaining why civil wars are so unlikely to have peaceful resolutions (Zartman 1993; Stedman 1997; Walter 1997, 2002, 2009; Fearon 2007; Fearon and Laitin 2007; Toft 2006; Cunningham 2011). There has been very little focus, however, on whether systematic features of the parties that make up bargaining dyads explain why the prospects of peace are so poor. Very little work systematically examines which groups governments extend negotiations and subsequent concessions to. Surely this is be important. 1 If dyads engaged in civil wars are frequently composed of at least one party lacking the desire or the ability to commit to peace, it is reasonable to expect a bleak forecast for the probability of peaceful conflict resolution. To understand fully how civil wars end, it is important to consider the characteristics of the parties fighting them. Specifically, if we are interested in determining whether conflicts will be resolved peacefully, we should determine whether the dyad is a reasonable candidate for peaceful resolution. We cannot realistically expect to witness the peaceful termination of a conflict if the parties involved have not made even the smallest of steps toward peace. More precisely, belligerents should not be expected to agree to end their conflicts before they have even agreed to negotiate with one another. Rather, negotiations themselves are a major concession. 1 Cunningham (2011) and Walter (2006) examine whether self-determination movements are likely to attract concessions from governments. At the end of this chapter, I discuss Cunningham (2011) at length and contrast her findings with mine. 60

70 Most research focusing on conflict resolution in civil war proverbially puts the cart before the horse because it ignores the stage preceding agreements and termination, that is, it ignores whether there are negotiations in the first place. Paying attention to the negotiation stage in bargaining is important. Uncovering which groups governments are more disposed to enter into negotiations with is an important question on its own, but is also important because who gets to negotiate necessarily affects which groups are candidates for agreements. Negotiations serve at least two purposes: to formally recognize groups and their demands, and to actively discuss parties conditions for peace. Groups often use violence as a means to gain formal recognition from the state (Crenshaw 1981:386). If groups use violence as a means of publicity, they should be unlikely to stop fighting if governments deny them the attention they seek. Thus, if governments disregard negotiations, groups should be less likely to agree to settlements. Moreover, groups should be reticent to sign agreements with governments before their demands are properly discussed, as accords reached outside of negotiations likely only reflect government s preferences. 2 Groups that do not attain negotiations should be significantly less likely to enter into formal agreements while groups that do negotiate should be much more inclined. Whether all groups are equally likely to experience negotiations has not been tested. Thus, whether a dyad attempts negotiations is likely related to the composition of that dyad, or the characteristics of the belligerents. The characteristics of rebel groups should explain a great deal of the variation in which dyads negotiate. While it is certainly important to examine whether some features make some states more or less likely to negotiate with their oppositions, the 2 Since rebels announce their demands publicly, it is conceivable that the governments can unilaterally create an agreement that rebels would be happy to accept. This is unlikely, however, because if governments do not think highly enough of rebels to initiate dialogue with them they probably do not think enough of them to offer generous settlements. 61

71 characteristics of the rebels likely explain just as much, if not more. In order for a dyad to engage in dialogue, governments must make the first move in extending negotiations. 3 As it is most states first inclination to reject compromise with their opposition, we can assume that most governments will not do so unless they are compelled. Something must change the government s obdurate posture to one that is more conciliatory. Features of a conflict, like an increase in costs or a hurting stalemate, may explain a change in the government s posture (Zartman 1993). In a related fashion, features of an opponent might produce specific conditions, like a steep increase in costs, that explain a government s re-orientation toward compromise. In this chapter, I examine the characteristics of rebels that might convince governments to extend negotiations. I argue that characteristics related to rebels ability to hurt governments and to frustrate governments ability to hurt them back are likely to produce a more conciliatory government posture. Governments will be more inclined to cooperate when rebels are able to impose extreme costs. They should also be likely to cooperate as their ability to reciprocate costs declines. Basically stated, when forced to, governments will negotiate. As I am interested in examining conditions that make governments more disposed to compromise, I also examine whether features related to rebels power to hurt affect states decisions to offer more concessions to rebels. The empirical results demonstrate that governments are indeed more likely to compromise when rebels possess a power to impose significant costs on them. 3 Although rebels can demand negotiations, governments must ultimately offer them before they are initiated. Similarly, even after governments extend negotiations rebels must agree for them to begin. However, I characterize the decision to negotiate as the government s decision here. As each rebel group has aims they seek to accomplish and most groups can only accomplish their aims through negotiated settlements, most rebel groups will prefer negotiations. While some rebel groups do, very few will reject negotiations if they are offered. Most governments, on the other hand do initially reject calls for negotiations as they typically prefer not engage in dialogue with any group. Thus it is reasonable to consider the decision to open negotiations as a decision made by the government. 62

72 It is incredibly important to determine whether there is a systematic component that explains when governments decide to compromise or who they decide to compromise with. There are a few cases of agreements without negotiations, but it is nevertheless, a highly unlikely trajectory for the road to peace. It suggests there is a systematic element to the process that generates data on formal agreements. If whether they sign agreements is conditional on some unobservable features related to whether belligerents reach negotiations, there is likely some selection bias in the estimates of extant studies. Selection bias can lead to incorrect inferences about the effect of covariates on the dependent variable. Thus, determining whether there is a systematic component to negotiations and accounting for it in models examining agreements is prudent. In this chapter, I find that governments are overwhelmingly likely to offer concessions to violent groups that behave badly. Specifically, I uncover that governments are likely to open dialogue with and offer concessions to groups that employ extreme tactics, like terrorism, banditry, and crime in civil war. More moderate groups are systematically excluded from negotiations and concessions. I demonstrate that governments make peace overtures to groups that use extreme tactics, like terrorists, and groups with strong incentives to see the continuation of conflict, like bandits and criminals, in order to stop the pain caused by the tactics these groups employ. As groups using terrorism are often thought to be radical extremists or groups with aversions to compromise on their extreme goals, it should be much more difficult to come to peaceful settlements when dyads contain one of these groups. Even if these dyads are able to come to agreements, governments should be much less likely to honor these bargains as they are incredibly costly to implement. Groups that find conflict profitable should also be likely to renege on bargains that are struck because for them, peace is less lucrative. That governments appear to reward these types of groups suggests they are likely selecting as bargaining partners groups with whom there are already significant commitment problems at the outset. So, in addition to selection bias that might be 63

73 pervasive in existing models on agreements, there might be unaccounted selection bias in models examining the durability of peace after settlements as well. First, rewarding with negotiations and concessions groups that employ extreme tactics is counterproductive to any peace process. By offering concessions to extremely violent but not to more moderate groups, governments appear to reward the means by which groups seek to accomplish their ends. Such policies likely encourage other groups to use extreme strategies as a means to accomplish their aims as well (Toros 2008). Governments unintentionally promote terrorism or extreme violence as a legitimate and acceptable means to gain concessions by offering concessions and negotiations to these groups exclusively. Second, groups that use extreme tactics like terrorism are expected also to possess extreme goals. Crenshaw (2008:25) argues that terrorist groups are often organizations seeking radical changes to the status quo. She (1981:383) argues that terrorism is usually the work of extreme factions within a broader movement. For settlements to actually placate these groups, groups extremely dissatisfied with the status quo, agreements must contain exorbitant concessions. The concessions these groups require for settlement are likely to be those that governments are unwilling or unable to implement. Even if belligerents are able to come to an agreement in the short run, neither side is likely to be able to commit in the long run. Thus, negotiations with extreme groups should lead to incredibly fragile peace, if any at all. This chapter attempts to put the horse back before the cart by first examining what groups are most likely to attract concessions from governments. In the next section, I advance an argument that explains why governments should be more likely to offer negotiations and concessions to organizations that engage in bad behavior. I apply Slantchev s (2003) power to hurt argument to civil wars to elucidate the connection between rebel tactical choices and government concessions. I test this argument and find that, contrary to conventional wisdom, governments are overwhelmingly 64

74 likely to cooperate with groups that use extreme and unpopular tactics, including terrorism and banditry. In the following chapter, I use these analyses as the first stage in a selection model to determine whether accounting for governments tendencies to negotiate and offer concessions to certain types of groups actually affects the relationship between concessions and agreements. Why Not Negotiate With The Terrorists? Most governments claim they do not negotiate with terrorists. While western democracies are most renowned for this expression, African governments like Liberia s also know that it's the policy of most democratic governments in the world not to negotiate with terrorists. 4 Yet, we witness quite often that governments, including those in the United States, Britain and Spain, do in fact negotiate with groups they label terrorists. The Burundian government, for example, has participated in negotiations with the FNL, a rebel group they routinely dismiss as bandits. That governments claim never to negotiate with terrorists but actually do so is remarkable and important. The fact that governments frequently offer negotiations and concessions to groups that use abhorrent tactics to execute their conflicts is likely damning for the entire enterprise of peaceful conflict resolution. Scholars are concerned that offering negotiations to terrorists is especially inappropriate and dangerous because granting negotiations confers a measure of legitimacy on groups (Neumann 2007; Toros 2008). Negotiations are not only likely to legitimize the groups, they are expected to legitimize the strategies, tactics and means they use. Scholars have voiced concern that negotiating with groups that choose terrorism as a strategy likely marginalizes groups that choose to seek change peacefully, or at a minimum, according to the laws of war. 4 Liberia; Liberian Government Rejects Calls for Ceasefire Africa News May 21,

75 Negotiations are likely to incentivize such behavior and negatively influence groups that have previously shied away from such tactics (Wilkinson 2001; Neumann 2007; Toros 2008). Essentially, negotiating with groups employing extreme tactics glorifies or rewards bad behavior. Negotiating with groups that use illegitimate means to achieve their political goals is destructive to the legitimacy of the governments that initiate dialogue. Negotiating with such groups is also destructive to the prospects of peace with these particular groups and with other groups attempting to enact changes in the status quo. As such, peace should be less likely to result from negotiations with illegitimate groups or groups that behave badly. The problem that these negotiations pose is likely to be negligible if negotiations with these groups are infrequent. However, if governments negotiate with terrorist groups often, it is likely to have a sizeable and important impact on peace and stability in countries experiencing internal conflict. Negotiations with extremist groups are likely to result in the adoption of extremist tactics by a greater number of other groups. There is sufficient reason to believe that governments offer negotiations and concessions to groups that can be deemed extremist quite frequently. A reasonable explanation for a group s tactical choice is that they believe it will be effective in helping them achieve some end. In civil war, this end is often political concessions or alterations in the status quo. Terrorism is but one tactic in a dossier of repertoires that violent groups can choose from to achieve these goals (Crenshaw 2008:24). For a variety of reasons, groups make intentional choices to employ one set of tactics over others but in civil war it is likely that groups choose terrorism from all other available strategies because they believe it will be the most effective strategy for garnering concessions from the government. 5 But is it? That many governments, like those of Sudan, Niger, Mali and Burundi, have 5 Groups often claim to select terrorism as a last resort to achieve their goals. They may select terrorism when they feel that they have exhausted all other options. 66

76 all declared they do not negotiate with terrorists seems to suggest that terrorism might not be an effective strategy. However, that many of those same governments have negotiated with the very groups they labeled terrorists, criminals, and bandits (the FNL, SLM/A, and ATNMC), might cause one to rethink the effectiveness of the strategy and decide that it likely is. At least anecdotally, it appears governments are overwhelmingly likely to backpedal on their vows not to negotiate with unsavory groups. This may be due to governments insistence on branding any armed opposition groups terrorist groups, armed bandits, or criminal gangs. Whether these groups actually terrorize, or engage in banditry or criminal activity is usually an unrelated consideration. Sambanis (2008:177) argues that regimes label or mislabel the activities of their armed opposition as banditry or terrorism to downplay the political threat they pose. Toros (2008) suggests, similarly, that governments name groups in this way to ensure that their opposition is delegitimized, stigmatized and isolated. Governments attempt to make their opposition groups pariahs in order to lessen their threat and to ensure they do not garner support that will allow them to constitute an even greater threat. Governments are highly likely to regard any opposition group as terrorist or criminal, regardless of their actual tactics. As a matter of policy, they also claim not to negotiate with these types of groups. As such, governments can hoose to never negotiate with any group offering violent resistance, or prove themselves false by opening dialogue with groups with whom they said they would not negotiate. Since civil wars are less likely to end in victories by either side (DeRouen & Sobek 2004), these choices leave governments with the options of letting civil wars continue indefinitely or backtracking on their word never to negotiate with the terrorists. Most governments are unlikely to have the ability to sustain internal conflict indefinitely, and are thus 67

77 likely to be forced to initiate dialogue or create some opening for a peaceful settlement. This means that we should expect to see at least some negotiating with terrorists. If we limit the designation of terrorist, criminals, or bandits to only those groups that actually use terrorism, banditry, and crime as a strategy, we should still expect to see that despite the declarations of states, governments are likely to negotiate with objectionable groups. Governments should only be expected to pay attention when they are forced to. Thus it is reasonable to interpret groups using terrorism as a strategy to gain government attention. Crenshaw (2008: 29) argues that terrorism can be an effective tool because by attracting attention it makes the claims of the resistance a salient issue in the public mind. The government can reject but not ignore an opposition s demands. The very public, overt nature of terrorism makes governments pay attention, and its painful nature makes governments seek compromise. Terrorism is particularly painful for governments to endure because it disrupts the rule of law. Unavoidable attacks against civilians humiliate the government, challenge its legitimacy, and ostensibly make the state ungovernable (Crenshaw 2008:29, 1981:386; Hultman 2007, 2009). Hultman (2009) proffers that rebels strategies of terrorism can inflict severe costs on governments, especially when the civilian targets form a core government constituency. Crenshaw (1981: 386) and Kydd and Walter (2006) argue that groups utilize terrorism to execute wars of attrition against their governments. Hultman (2009:823) suggests that by destroying the government s ability to maintain control, and by proving that the state is unable to provide security, a rebel group can force the government to back down and offer them concessions. As terrorism is so painful for governments enduring the attacks, groups that use it should be particularly successful in forcing government cooperation. 68

78 Terrorism, Terror and the Power to Hurt Slantchev (2003) argues that the realization of the power to hurt one s opponent increases the probability that belligerents will pursue a peaceful settlement as it opens up a range of bargains that would not otherwise exist. States are expected to accept less-than-favorable settlements when they recognize that their ability to hurt their opponent has diminished and/or when they realize their opponent has the ability to inflict severe pain on them. Slantchev (2003:128) writes, the diminished, or eliminated, capacity to hurt the enemy is a major reason to terminate war and seek a negotiated settlement. He further argues that since the power to hurt is a kind of bargaining power, the denial of such power undermines the bargaining position of the opponent (Slantchev 2003:128). In the context of civil war this means that when a government is forced to recognize they no longer possess the capacity to hurt their opponent but their opponent can inflict unbearable costs on them, they will be forced to pursue a peaceful settlement. Translating this logic from the inter- to the intrastate level, it follows that an asymmetry in the power to hurt that favors rebels should lead to governments seeking compromise (or vice versa). Groups that use terrorism as a strategy in civil war have the ability irreparably to hurt governments. By targeting civilians, they are able to call the governments legitimacy into question by demonstrating first that the government is not the sole proprietor of force within the state and further that the government may be illegitimate because it is unable to protect its civilians (Hultman 2007, 2009). When governments demonstrate that they are unable to protect civilians from violence, civilians are more likely to seek protection from rebels (Kalyvas 2004). This is important, because in civil wars the ability to protect civilians often coincides with control over territory and maintaining or capturing territory is a measure of battle success. Thus, for rebels terrorism becomes a means to undermine the government s control of the state (Crenshaw 1981: 386). 69

79 Under the power to hurt argument, groups using terrorism should be especially likely to gain concessions as the strategy they pursue limits or eliminates the government s ability to hurt them back. Like insurgent or guerrilla strategies, the quintessence of terrorism is the lack of direct engagement. If governments do not have the ability to engage, they lack the ability to respond effectively, and also lack the power to hurt. Thus, when rebels use terrorism as a strategy we can expect governments to offer negotiations and attempt to locate peaceful settlements. Similarly, we expect rebels to possess the power to hurt governments when they are strong and when they use other strategies that terrorize, like kidnapping or forcible recruitment. These strategies should make governments more likely to offer peaceful settlements. This argument produces the following expectations: H1a: In civil war, governments should be more likely to offer negotiations to groups using terrorism as a strategy. H1b: In civil war, governments should be more likely to offer concessions to groups using terrorism as a strategy. H2a: In civil war, governments should be more likely to offer negotiations to groups using kidnapping or forcible recruitment strategies. H2b: In civil war, governments should be more likely to offer concessions to groups using kidnapping or forcible recruitment strategies. Especially in Africa, natural resources are important for the functioning of the state. Many African governments rely on the export of natural resources, or primary commodities, as a main means of raising revenue. Governments are often able to fund their wars through the sale of future contracts to foreign investors (Ross 2004). Governments are able to use the sale of natural resources 70

80 to hire mercenary groups to defend their regimes against rebels. If governments lose access to these resources, they also lose access to a mechanism to resist the coercion of rebel groups. If resources help governments remain viable parties in war, the loss of these resources should diminish their ability to inflict pain. As they are less able to inflict pain on their opponent, they should be more likely to agree to a peaceful settlement of the conflict even if they know they are likely to get a raw deal. Thus, H3a: In civil war, governments should be more likely to offer negotiations to groups extracting valuable resources including, oil, gems or drugs. H3b: In civil war, governments should be more likely to offer concessions to groups extracting valuable resources including, oil, gems or drugs. The loss of the power to hurt explains why governments might be more likely to negotiate with the type of group they specifically claim they are unwilling to negotiate with. Despite governments vehemence to the opposite, I expect them to be more likely to negotiate with groups that have the power to hurt them. This means that while governments like that of al-assad in Syria argue no political dialogue or political activity can succeed while there are armed terrorists groups operating, we know that political dialogue will succeed when the government needs it to. I test the power to hurt argument in civil wars. Specifically, I am interested in determining whether governments are more likely to engage in dialogue with groups when they use extreme and painful tactics, like terrorism, banditry, and crime. I, then, examine whether governments are as likely to offer groups greater concessions during peace processes when they use such tactics. Using the UCDP Dyadic Armed Conflict Dataset (ACD) on African rebel groups from , I find that governments are, in fact, more likely to dialogue with groups that can demonstrate the 71

81 power to hurt them. Moreover, governments are more likely to offer meaningful concessions to groups demands when they use some of these strategies. In the next section, I discuss the research methods and data I use to test this argument. Subsequently, I discuss the results from logistic regression models examining whether governments hold formal talks with rebel groups and negative binomial models examining the number of concessions governments offer rebels on their demands. I conclude the paper by explaining why it is a big deal that governments negotiate with these groups more frequently and follow up with expectations of how this affects the prospects for long-term peace in civil war, broadly. Research Methods Independent Variables: To test whether governments are more likely to make peace overtures to organizations that have the power to hurt, I identify three types of groups possessing the power to hurt: terrorists, kidnappers, and criminals/bandits. By their very nature, terrorist groups should possess the power to hurt. I use the Global Terrorism Database (GTD) to identify African groups within the ACD that employ terrorism as a strategy. The GTD dataset records individual instances of terrorist attacks globally. The GTD records attacks by subnational actors intended to coerce a large audience and/or attain broader social, religious, political, or economic goals. Types of incidents coded in the dataset include assassinations, armed assaults, bombings, attacks on facilities, hijackings, and kidnappings. These data include both successful and unsuccessful attempts at domestic and international attacks. Since I am interested in rebels ability to inflict pain on the domestic government they are fighting, I exclude cases of transnational terrorism and include only successful domestic attacks. Therefore, I only include observations where the target and the government are of the same nationality. The independent variable is a count of successful domestic terror attacks executed by rebel groups in a 72

82 month. The number of attacks in a month ranges between 0 and 49. Kidnapping and abductions are included in the GTD dataset only when the actions are intended to coerce the government into concessions. However, it is likely important to consider kidnappings or abductions outside of those intended to produce concessions. Specifically, rebels kidnapping and abducting children to enlist them in their armies should be considered a form of terror. Because data on abductions are difficult to locate, I use the recruitment of child soldiers as a proxy. I code whether a group was reported as using children under the age of 18 from the Child Soldiers Global Report Like other forms of terror, I expect abducting children to have a positive effect on negotiations and concessions offered to rebels groups. Extracting natural resources, specifically gemstones, drugs, and oil, can be considered a criminal or quasi-criminal activity, or a form of banditry. By occupying or controlling areas with natural resources, groups prevent governments from benefiting from these goods. If governments cannot use these resources to execute their wars with rebels, rebels are able to reduce government s power to hurt. Rebels are further able to hurt the government by potentially crippling the economy. Such a move would hurt the government by generating new civilian grievances and creating new challengers to the state s legitimacy. I use data from Lujala et al. (2009) to identify whether there were extractable resources in the conflict region. I include three measures of natural resources: gems, drugs, and oil as each of these might have a different impact on the likelihood of government cooperation. As control variables, I include relative rebel strength, explicit rebel support, and main group. Unequivocally, strong rebels are much more likely to have the power to hurt than weaker rebel groups. I use the rebel strength variable from the Cunningham et al. (2009) Non-State Actor Dataset (NSA) to examine the relative power between belligerents in conflict. This is an ordinal 73

83 variable that captures whether rebels are much weaker, weaker, at parity with or stronger than the government. From this dataset, I also include whether rebels have explicit outside support. If rebel groups have supporters assuring their viability, they are much more likely to have the power to hurt. As groups may inflict a great number of costs on the government even when they are not necessarily the strongest or most capable militarily, I also code whether a group was the main group inflicting casualties on the government in a month. When groups inflict the greatest number of casualties in a conflict, they can be thought to have greater power to hurt, relative to the other groups in a conflict. This allows for the possibility that governments evaluation of a rebels power to hurt is not objective, but it is relative to that possessed by other groups. In this way, governments may choose to cooperate with a weak group if they are able to inflict more pain than can other groups. I coded this variable from the UCDP database listing the number of casualties each group caused in a year. If a group caused the greatest number of deaths in a conflict, they were coded as the main group in the conflict for that year. Multiple groups were coded a main group if they inflicted roughly the same number of casualties. I include a measure of battle-related deaths to control for the intensity of the conflict. I use the best estimate of battle deaths included in the dyadic UCDP Battle-Related Deaths Dataset. These data include all casualties related to fighting between the warring parties, including civilians killed in the cross-fire. It is likely that certain types of governments are more sensitive to costs than others. In particular, high war costs are likely to be more damaging to democratic governments, as they are expected to be more accountable to their constituents (Kydd and Walter 2006; Valentino et al. 2010). Additionally, democratic governments should have a decreased capacity to retaliate against attacks as they have greater constraints on their ability to reciprocate. Here, using painful tactics should be more effective at producing concessions as they must restrain their power to hurt. Kydd and Walter (2006) argue that terrorism is most likely to be used as a strategy of attrition against 74

84 democracies as these states are more restrained in their power to retaliate against attacks. Thus, I expect democratic governments to be more likely to adopt conciliatory postures when faced with steep costs than are autocratic governments. To test the effect that regime type has on negotiations and concessions. I include a regime type measure from the Polity IV dataset. I use the Polity2 variable, which ranges from -10 to 10 in the Polity data and -8 and 8 in the sample. I also include a variable measuring the total number of demands rebels made in that month. The number of demands variable ranges from 0 to 21. I include a measure of time, ln(time), in the fully specified models. I coded the number of months the conflict episode had been ongoing and I took its natural log. I also account for the number of conflict episodes the dyad has experienced. The greatest number of episodes in the sample was 6. The variables representing the number of months in the conflict episode and the number of episodes in the conflict measure how protracted the conflict within the dyad is. This is necessary because protracted conflicts may be riper for settlement. If both sides have sustained significant costs over time, they may be better able and more willing to attempt to locate bargains preferable to continued conflict. Moreover, after a significant period of time, learning is likely to have taken place and both sides should be more likely to estimate accurately future costs as well their probability of victory. To determine whether governments are any more likely to offer concessions to credible groups, or grops with support bases, I use data from the ACD2EPR dataset (Wucherpfennig et al. 2012). From this dataset, I include a dichotomous measure, Ethnic Claim, that assumes a value of one when a group makes a claim or demand on behalf of any ethnic group in that state. These ethnic groups need not be co-ethnics with the rebels in the group representing them. 75

85 Dependent Variables: Because I am interested in evaluating two different phenomena, the incidence of negotiations and the number of concessions offered, I code two dependent variables. I code whether governments engaged rebels in negotiations in a given month. This variable is a dichotomous measure that takes on the value of one when negotiations take place in a month and is zero otherwise. The second dependent variable assesses the volume of concessions governments offer rebels in a given month. Concessions are coded when a government addresses, at least in part, something rebels demand in that month. The measure of concessions, I include in the count model captures the number of maximal and substantial concessions rebels receive on their demands in a month. To address the concern that rebels using painful tactics are more likely to gain only meaningless concessions, I include a dependent variable that examines the number of maximal and substantial political concessions rebels gain in Model 8 of Table 4.2. This variable differs from the main independent variable as it only includes concessions and demands that are political as opposed to those that are tactical or procedural (a change in venue of negotiations for example). I also include leads of both dependent variables which amounts to lags of all of the independent variables in the models. Thus in Table 4.2, for example, Model 5 examines the effect that the independent variables in month t have on the number of concessions in the following month (t+1). I include measures that account for the effect of the independent variables in month t on the dependent variables in t+2 and t+3 as well. The unit of analysis for all of the models is the dyad-month. For each month a dyad is actively fighting, there is one observation. I use logistic regressions to examine the effect of rebels ability to hurt the government on negotiations because negotiations is a dichotomous dependent variable. The results of these analyses are reported in Table 4.1. Table 4.2 displays the results of 76

86 Negative Binomial Regressions. I use Negative Binomial Regression because the dependent variable here is a count of the number of concessions governments offer rebels. I employ a Negative Binomial model because I expect that the underlying rate of events (λ) within each period is not constant as assumed by the Poisson model. Specifically, I expect there to be over-dispersion in the data as the occurrence of one event is likely to lead to another. In these data, events are concessions. It is reasonable to suspect that after the government offers one concession, the probability that they offer another increases. Therefore, I expect that the rate of events until we see at least one concession is significantly different from the rate after we have observed at least one concession. This violates a main assumption of the Poisson model which the Negative Binomial model relaxes. 6 From the non-zero α parameter in the models (which indicates over-dispersion), it is clear that the Negative Binomial Model is the correct model for these data. 7 In the analyses examining the number of concessions offered I cluster the errors on the conflict, as concessions may be offered to multiple groups in that conflict when they are given at all. There may be other dynamics of the conflict that explain the rate of concessions. I reran these analyses without the clusters and very few of the results differed from those reported in Table A main assumption of the Poisson Model is that the probability of an event is independent of all other events in that period. 7 Because I cluster on the conflict, Stata does not compute a likelihood ratio (LR) test. The LR test evaluates the null-hypothesis that α is equal to zero. A non-zero α suggests that a Poisson model is not the appropriate model. The α in all of the models are in fact non-zero, but whether this is a statistically significant test-score cannot be verified without a formal statistical test (ex. LR or Wald test) or a manual test that checks whether the distribution of the errors of the Poisson model are statistically different from λ. When I exclude the clusters, the likelihood ratio test of α is significant, suggesting we can reject the null hypothesis that α=0. 77

87 Results <Insert Table 4.1 Here> Table 4.1 demonstrates that the power to hurt argument applies to civil wars. I include several models to demonstrate the results I garner are not produced by missing data as some variables have quite a bit of data missing (natural resources, rebel strength and explicit support). I introduce variables with missing data one group at a time to demonstrate that the main results are robust to the variables and number of observations included. All the models in Table 4.1 demonstrate that groups using terrorism as a strategy are highly likely to be invited to negotiations with the government. While I include multiple models, particular attention should be given to Models 4 through 7 as these are fully-specified models. While terrorism is a significant predictor of negotiations, using terror or kidnapping civilians is less likely to induce government conciliation. Arguably, the kidnapping of non-strategic children in rebel zones of control should have less of an effect on a government s legitimacy and should, therefore, be much less hurtful. Alternatively, abducting children may only serve to delegitimize rebel groups using these tactics. Therefore, civilians should be less supportive and governments less conciliatory. Model 4 also demonstrates that groups extracting resources or fighting in an area with gems and drugs are more likely to be included in negotiations. Such a strategy likely reduces the government s power to hurt rebels and increases rebel viability which increases these groups power to hurt. As a state s power to hurt decreases, or its opponent s power to hurt increases, governments will be more likely to seek peaceful settlements. While groups that extract drugs are likely to gain negotiations, it is dubious that states are as likely to profit from drugs as from oil or gems. It is possible that the drugs variable is largely capturing rebel viability. I do not believe this undermines my argument as viable groups are likely also more capable and wield the capacity to hurt 78

88 governments in more conventional ways. While it is significant in all other models, Model 4 demonstrates that rebels extracting oil or fighting in the proximity of oil fields, does not assure them negotiations. Model 4 also illustrates that when rebels are either strong or the main group engaging in violent conflict with the government, they are more likely to be selected for negotiations than groups that are weaker or otherwise are peripheral to a conflict. Rebel relative strength is a direct measure representing how strong rebels are vis a vis their state opponent. If a group is stronger than their opponent, they no doubt yield the power to hurt. If states are more likely to offer concessions to groups that are able to exact costs on them, they should surely offer concessions to stronger rebels. Whether a group is a main group is an indirect measure of rebel strength measuring whether fighting between a dyad is more intense than that between other dyads in the conflict. Even if rebels are not numerous or as strong as the state, they may still be able to inflict more pain than other groups fighting the government. What matters more than how many rebels belong to a group is how much trouble they are able to cause. The significant coefficient on the main group variable demonstrates that governments are, in fact, likely to offer concessions to the group that inflicts the most costs on them. The results of Model 4 also demonstrate that groups with explicit outside support are not more likely to gain negotiations. However, some models (2, 3, 7) do suggest explicit outside support does improve rebels prospect of achieving recognition at the bargaining table. Dyads including a more democratic government and those involved in protracted conflicts are likely to negotiate. Dyads experiencing a greater number of episodes, however, are less likely to be involved in negotiations. This result may indicate that groups experiencing more conflict episodes are involved in an enduring intrastate rivalry (DeRouen and Bercovitch 2008), and are thus less likely 79

89 to attempt negotiations under any circumstances. Also unlikely to engage in talks are dyads involved in very intense conflicts. Models 5 through 7 examine the effect of rebels wielding the power to hurt in a given month on the probability of negotiations in the following period. It is possible that the effects of rebel strategies are not readily apparent. Governments may need time to recognize how damaging rebel s strategies are, and may need more than a few weeks to reassess the utility of settlement and reformulate their policies regarding how best to deal with rebels. Thus, governments should be just as likely or perhaps more likely to hold negotiations in future periods when rebels inflict significant costs on them as they are in the present period. The results demonstrate this. They show that groups that use terrorism as a strategy, exploit natural resources, are stronger and inflict the greatest number of casualties in a conflict, are significantly more likely to be involved in subsequent negotiations than are groups that do not have a significant, asymmetrical power to hurt. Perhaps one of the most interesting and telling results is that rebels groups that garner the support of a significant portion of the population are significantly less likely to be included in negotiations with the government. This finding holds across all of the models where the variable is included. The theory in this dissertation suggests groups with ethnic support bases have mechanisms that can help generate credible commitments. These groups should be more likely to uphold agreements and stop fighting when they gain concessions on their demands as they are bound by their constituencies. If governments are systematically less likely to include these groups in talks, it may explain why civil wars are so unlikely to end in negotiated settlements. It is reasonable to expect conflicts to be less likely to end peacefully when at least one party in a negotiating dyad is unlikely to honor their agreements. The results of Table 4.1 demonstrate governments are more likely to offer negotiations to groups that are, in theory, less credible. Groups engaging in banditry and crime, or 80

90 groups exploiting natural resources and kidnapping civilians to increase their ranks, do not have the credibility mechanism that groups relying on an ethnic support base have. Despite this, governments are significantly more likely to offer negotiations to (some types of) non-credible groups over groups with the ability to generate audience costs, or more credible groups. <Insert Figure 4.1 Here> Figure 1 shows predicted probabilities of the effect of different independent variables on the probability of negotiations. Figure A shows the effect that terrorism in time t has on the probability of negotiations. When rebels do not use terrorism as a strategy, the likelihood of being offered negotiations is about 10 percent. There is a 50 percent chance of negotiations when rebels use around 27 terror attacks in a month. If rebels use 49 terror attacks, the probability of negotiations is incredibly high. The confidence bands get larger around the larger values of terror attacks because there are far fewer observations where rebels launch a huge number of successful terror attacks against the government than those where rebels do not launch any successful attacks at all. 8 Figure B demonstrates a very similar trend. This figure depicts the effect of terrorist attacks on the likelihood of negotiations in the following month. When rebels launch no successful terror attacks, the likelihood they will participate in negotiations is about 10 percent. If they launch roughly 25 successful attacks, the chance of negotiations has increased to a little greater than 60 percent. At 35 attacks, they should expect to be invited to negotiations with a probability of around 85 percent. Figure C displays the effect of rebel strength on the probability of negotiations. When rebels are weaker than the government, they can expect to gain negotiations at t+1 with a probability of about At parity, the probability that a dyad will negotiate in the following month jumps to 8 The mean number of terror attacks in the sample is In only 14 percent of the observed months was there at least one terror attack 81

91 about 18 percent, and when rebels are stronger than the government the probability increases to about a 27 percent. Finally, Figure D exhibits a clear positive relationship between the length of time an episode has been ongoing and the probability of negotiations in the subsequent month. At about 4 years of sustained violence the probability of negotiation in the following month is about 15 percent. The probability of negotiations doubles when the conflict has been ongoing for 12 years. <Insert Table 4.2 Here> Table 4.2 examines how rebel characteristics affect the number of concessions governments offer. The full model, Model 4, demonstrates that groups using more terrorism in civil war are likely to gain a significantly greater number of concessions on their demands. Here, violence leads to more concessions. Very few studies have systematically examined which rebel groups gain concessions in civil war. Cunningham s (2011) is one such study that does assess the effect of group-level factors on whether the government offers concessions. She examines whether violent and non-violent selfdetermination movements are more likely to gain concessions when they are divided or unified, and finds that when movements contain multiple divided factions, they are more likely to gain concessions as governments employ divide and conquer strategies. Cunningham (2011) examines the effect of violence on whether movements gain concessions and finds that groups using violence are only more likely to gain concessions in the year that the civil war begins. Other measures of violence, including whether there is conflict in the previous year and whether violence is ever used, do not appear to explain government concessions. This diverges from my results in that I find more violence prompts government concessions. While the initial onslaught of violence seems to predict concessions in Cunningham s models, I expect sustained violence to be responsible for inducing concessions. Concessions should be more likely after costs have mounted. The divergence between our findings may come from the differences in our samples. Cunningham examines whether 82

92 violence explains concessions to self-determination movements including groups using violent and non-violent strategies while I examine only groups engaging in armed conflict. Cunningham surveys whether using violent strategies helps groups attain concessions. In contrast, I examine how the level of violence among violent groups affects the probability of concessions, and I find that governments are significantly more likely to make concessions when violent groups adopt more extreme violent strategies. In these models, whether rebels operate in areas rich with gemstones and oil does not make government concessions any more or less likely than cases where groups do not exploit resources. This result suggests that while these groups may be more likely to be more involved in negotiations, these negotiations do not always lead to a greater number of concessions. On the contrary, if groups extract drugs they are likely to gain significantly fewer concessions. Model 4 suggest that if a rebel group inflicts more costs on the government than any other group they are no more likely to receive concessions. Whether a group is strong or whether they have outside support also does not affect whether they gain concessions. Democratic governments are not more likely to provide concessions. Only in Model 2 does recruiting children affect the probability of concessions. In all other models, whether groups recruit children in a war has no effect on the number of concessions rebels will gain. Again, it is likely that governments are just less vulnerable to this type of rebel tactic. Alternatively, since the variable is measured to include any recruitment of children, whether forceful or voluntary, it is possible that it is not precise enough to distinguish between groups that practice routine abduction and groups that begin legitimately recruiting soldiers at 17 and 18 years old. A better measure might produce different results. Models 5, 6, and 7, show the use of terrorism always increases the number of concessions rebels can be expected to gain in civil war, even up to three months after the attacks. The models 83

93 with the lagged independent variables demonstrate that rebel strength enhances the prospect that rebels will gain greater concessions on their demands in later months. Explicit backing acts in a similar way. Again, operating in areas with drugs decreases the number of concessions rebels can gain in negotiations. Three months after rebels have been operating in oil fields, they are still more likely to gain concessions. While rebels that exact the greatest costs on governments gain more concessions a month or two after, this effect wanes. Being the main group engaged in violence with the state has no effect three months later. Abrahms (2012) also examines whether terrorism is an effective strategy. Our results differ in that he finds terrorism to be ineffective at garnering concessions from the state. A significant reason for the divergence in our conclusions is that we examine completely different types of cases. Abrahms (2012:371) intentionally excludes groups involved in protracted civil wars of attrition as he believes they are conceptually distinct phenomena from terrorist campaigns. As I am interested in uncovering how rebel groups tactics contribute to their likelihood of gaining concessions in civil wars, I examine only those groups involved in conflicts with their government. While I only examine this subset of groups, I acknowledge that terrorism might also be an effective strategy outside of civil war. 9 That we come to different findings may be useful as it can give us insight into the effectiveness of terrorism in different contexts. Terrorism is likely to prompt a government to offer concessions in a civil war when rebels couple the tactic with other painful tactics. Abrahms study suggests terrorism outside of the context of war does not work in the same way. By only examining groups using terrorism as their predominant strategy, his study likely demonstrates that terrorism on 9 Violent, revolutionary groups oscillate between strategies. In addition to or instead of terrorism, groups may strategically choose to engage their opponents directly, or may opt to use guerrilla strategies only targeting military targets. Groups may also opt to use non-violent strategies at times. As terrorism is a strategy that can be chosen from alternatives, any group using terrorism as a strategy can be considered a terrorist. 84

94 its own is not painful enough to bring a government to its knees. It might be profitable to combine both types of groups into a single analysis to determine whether groups relying on a single strategy are as successful in achieving their goals as groups that employ a variety of tactics. Rebel Claims Together Tables 4.1 and 4.2 demonstrate that although rebels with support bases are significantly less likely to be included in negotiations, they are more likely to be granted concessions when are invited to the bargaining table. In each model displayed in Table 4.2, with the exception of Model 8, the results show that rebels that make claims on behalf of a support base are likely to gain more concessions when they are included in negotiations. Model 8, however, shows that these concessions tend not to be political in nature. Governments are likely to concede to groups that are bound by their constituencies, but they are only likely to offer concessions on these rebels apolitical demands. In the next two chapters I examine whether offering concessions to rebel groups with support bases affects the probability that the dyad will come to an agreement and terminate their wars. In the final chapter of the dissertation, Chapter 7, I discuss why it is relevant that governments tend not negotiate with credible rebels. I also discuss why it is important that groups with support bases are more likely to gain concessions on their demands when they are included in negotiations. Discussion: Overall, the analyses support the notion that governments are more conciliatory to groups that make war more painful for them. Governments might believe that negotiating with these groups is akin to putting a bandage on a wound to stop the bleeding, but instead they are in effect putting a Band-Aid on a leaky pipe. Rather than selecting bargaining partners based on whether they have legitimate claims or whether they are likely to uphold agreements, governments choose to offer olive branches 85

95 to groups that are found to have the power to hurt them. Governments forgo peaceful settlements until conflicts become unbearable, only when they must sue for peace are they likely to begin explicit bargaining. 10 Of course, this places the government at a huge disadvantage in bargaining. If it is clear that their opponents unilaterally possess the power to inflict damage, governments will be forced into suboptimal agreements each time. When bargaining with extremists, suboptimal agreements often mean the government is forced to offer concessions they cannot commit to, even in the short term. As extremism tends to be less amenable to compromise, extremists are likely to renege on any agreement not containing absolute concessions. Thus, neither party in the bargaining dyad has the capacity to commit to an agreement. Governments may create commitment problems in civil wars by self-selecting into bargaining situations where both parties are overwhelmingly likely to renege on any bargain that is struck, ostensibly stacking the deck in favor of peace failure. Extant research on conflict resolution fails to consider that there might be selection bias in the data generating process unaccounted for in models examining whether agreements lead to lasting peace. However, it is important to account for this bias: specifically scholars should account for which groups are even invited to participate in the peace process before they examine whether agreements are struck and whether these agreements hold. I do this in the chapter that follows. 10 This differs from a hurting stalemate because both parties are not equally hurting. Only the government is. 86

96 Chapter 5: The Effect of Concessions on Negotiated Settlements In the previous chapter I examined factors that cause governments to offer rebel groups concessions. The findings show that when groups are able to inflict significant costs that the government is unable to reciprocate, states are more likely to respond by offering negotiations and concessions. Among the most significant factors that determine whether belligerents engage in negotiation is whether rebels use terrorism as a strategy, cause a significant number of battle related casualties, extract natural resources and are strong relative to the government. The use of terrorism explains a great deal about which groups gain concessions in civil wars. Whether belligerents enter into negotiations and whether governments make offers of concessions is incredibly important but does not completely explain whether an agreement is signed. Governments may make offers to certain types of groups when they are weak, but it is not obvious that those same groups would be inclined to accept concessions. It is very likely that the types of rebel groups governments feel compelled to offer concessions to, are those groups with the ability to evade settlement and continue fighting even after generous offers. For the same reasons governments decide to offer concessions to these groups, because they are strong and have the ability to remain viable, groups are able to continue war even after the government indicates an initial willingness to compromise. As strong groups might have incentives to hold out until victory and groups that maintain their viability through the extraction of natural resources may find conflict lucrative, these types of groups should be less likely to settle. Table 5.1 from the previous chapter demonstrates that groups with support bases are actually less likely to be included in negotiations and less likely to gain concessions on their demands. This is true despite the fact that groups with support bases should be more likely 87

97 to sign agreements and lay down their weapons when their demands are met, at least theoretically. In Chapter 2, I argued that rebel groups drawing a significant portion of their support from a civilian base should have more credible demands as their ability to change or deviate from their aims will be constrained by their support base. 1 Groups are likely to face punishments from their support bases when they change or abandon their aims before attaining meaningful government concessions on them. As punishments severely hamper an organization s ability to function, groups with support bases should be particularly cautious about upsetting their supporters. Thus, groups relying on civilian support should be more likely to sign agreements and stop fighting only once their aims are addressed. Conversely, if these groups do not gain concessions on their demands they should be more likely to continue to fight. Groups that do not rely on support bases, specifically groups that finance their operations by the extraction of natural resources or maintain a steady stream of soldiers through forceful recruitment, should not have this credibility mechanism and should not be any more likely to agree to settle their conflicts peacefully when they are awarded concessions on their demands. In this chapter, I test whether these expectations are true. In particular, I test the first hypothesis from Chapter 2: Rebel groups with civilian support bases will be more likely to sign formal agreements when governments make meaningful concessions on their demands. In the following section, I discuss previous literature examining the determinants of agreements. I then describe the data and methods used to test the argument about how 1 Here, credible demands connote demands rebels maintain until they are met. Credible demands are important as they have explicit meaning in negotiations. 88

98 concessions impact peaceful settlement. I conclude this chapter with a discussion of the results and a substantive interpretation of their importance. Agreements in Civil War Many scholars examine the effect of peace agreements on conflict resolution. Fortna (2004), among others, argues that the content of an agreement matters for peace. She argues that the terms of agreements should make the benefits of settling outweigh the costs and benefits of fighting. Therefore, if agreements provide clauses that make settlement more attractive, belligerents should be expected to prefer peace over further conflict. Mattes and Savun (2009) similarly believe the content of agreements is important. They suggest provisions in agreements are important when they address commitment problems. As agreements containing provisions that ensure high costs for reneging decrease belligerent s fear of being taken advantage of, they should alleviate commitment problems and support more durable peace. Glassmyer and Sambanis (2008) find that providing for military integration in an agreement promotes self-enforcement. Agreements containing military integration should be less prone to fail due to commitment problems. By mandating the integration of rebels into a state army, agreements are able to lessen time inconsistency problems. Belligerents signing such agreements should feel more secure committing to peace. Hartzell and Hoddie (2003) examine the effect of agreements containing power-sharing provisions. Like military integration, power-sharing agreements are expected to address security concerns as the provisions provided therein create self-enforcing agreements. Power-sharing agreements are expected to have this effect as they ensure rebels, in particular, will be allowed to participate 89

99 in future decision making processes. Hartzell and Hoddie (2003) also suggest peace will be more durable when there are more provisions specified in agreements. While studies like Glassmyer and Sambanis (2008), Hartzell and Hoddie (2003) and Mattes and Savun (2009) highlight the importance of examining the contents of agreements, none focus on whether the provisions written into agreements address the underlying grievances of rebels and their support bases. Military integration and power-sharing should undoubtedly have an effect on whether agreements are signed but mainly when they address rebel demands. The existing literature focuses on whether agreements make for durable peace but does not provide much insight into how belligerents get to those agreements. Specifically, literature on agreements does not explain which dyads are most likely to sign or implement such agreements, nor does it explain the conditions under which agreements are signed. Very little literature explicitly theorizes about factors that lead to formal agreements once negotiations have begun. An exception can be found in the spoiler literature that provides general explanations for the breakdown of conflict at any stage. The presence of spoilers, then, is expected to explain which conflicts are least likely to end in peace agreements. Cunningham (2006, 2011) provides one version of a spoiler argument. Stedman (1997) and Greenhill and Major (2007) provide other accounts of how spoiler dynamics affect agreements. Cunningham (2006, 2011) argues neither information problems, a lack of trust, a desire to profit from conflict nor issue indivisibility are sufficient explanations for the settlement of conflict. Instead, he argues the multiparty context of most civil wars provides a sufficient explanation for the duration and intractability of settlements. In multiparty conflicts, spoilers seeking to disrupt the peace are among the greatest inhibiters of 90

100 agreements. Parties have incentives to hold out to be the last signer expecting to pull the bargain closer to their ideal points. As each party has a similar incentive, the probability that conflict will end decreases with each additional party. Factors that deter, address, or overcome spoilers, should also be useful in explaining where peace can be established. Stedman (1997) offers several ways to address spoiler problems in civil war. He suggests inducements to help manage peace spoilers. 2 This strategy amounts to meeting spoilers demands in order to establish peace. This is congruent with the solution I propose, however, with a modification. In order to be effective, negotiated peace settlements should address the demands of all rebel groups in a conflict, not only those that are successful at spoiling peace. Greenhill and Major (2006), who re-examine the spoiler argument proffered by Stedman, also suggest that solutions should not only focus on successful spoilers of the peace, but that would-be spoilers, or those with the potential to disrupt peace agreements, should also be pre-empted. Offering concessions that address rebel groups main demands should have a deterrent effect on peace spoiling immediately and in the long-term. Providing concessions that satisfy the civilian population, specifically the rebels latent or active support base, should prevent conflicts that might erupt in later years over the same issues. Regan (2002:58) suggests that alleviating poverty through external economic inducements should inhibit rebel mobilization as poverty tends to be a predictor of participation in rebellion. Thus, by removing incentives to participate in war, governments 2 Stedman argues that alternative strategies for managing spoilers are to declare rebel s demands illegitimate or to suggest that peace would move forward with or without their participation. While these may be sufficient strategies for getting an agreement signed, it is not clear why this would lead to successful peace. 91

101 can reduce the probability of further conflict. Concessions that target civilians grievances should have a similar effect on peace. The provision of concessions that alleviate the qualms of a subset of the population should disincentivize participation and undercut the rebels support base. Thus, concessions should increase the probability that rebels will agree to settle. Concessions should have the greatest effect on groups that mobilize resources from civilian support bases. Research Design In this chapter, I examine whether concessions to rebel demands make groups more likely to sign formal agreements. The unit of analysis is the dyad-month. One observation is included for each month the dyad is actively engaged in conflict. Thus, there are a total of 3413 observations in these data. The dependent variable is a formal agreement. Agreement, takes on a value of one when a formal agreement is signed by both belligerents in a dyad. This variable includes both formal cease-fires and political agreements. Informal arrangements, or agreements that were not signed by both belligerents, are not coded as agreements. These data were coded from a variety of sources, primarily the United States Institutes of Peace (USIP) Margarita S. Studemeister Peace Agreements Digital Collection, the University of Ulster s International Conflict Research Institute (Incore) Peace Agreements database, and based on news reports from Lexis Nexis Academic. There are 99 cases of a formal agreement coded in the dependent variable. In Tables 5.1 and 5.2, I include eight variations of the main explanatory variable capturing the scope of government concessions to rebel group demands. I include strong measures of concessions that include only substantial and maximal concessions on rebel demands in a month. Number of Concessions (Strong) represents the number of substantial 92

102 and maximal concessions governments offer rebels on their demands. Percentage of Concessions (Strong) is the ratio of how many maximal and substantial concessions the government offers over the number of demands rebels made in that month. I also include weak measures of both of these aforementioned variables, Number of Concessions (Weak) and Percentage of Concessions (Weak). These variables include moderate, maximal and substantial concessions. I also include four measures that only include political concessions. These are Percentage of Political Concessions (Strong), Number of Political Concessions (Strong), Percentage of Political Concessions (Weak), Number of Political Concessions (Weak). These variables are coded in the same way as the other variables, but they exclude all demands and concessions that are not political in nature. I examine this many varieties of my measure of concessions to demonstrate the robustness of my explanatory variable. After demonstrating that the independent variable is robust in Tables 5.1 and 5.2, I use the Percentage of Concessions (Strong) in all analyses in Table 5.3 (excluding Model 4). To test the effect that rebel support bases have on the likelihood that rebels will formally agree to a peaceful settlement, I include, Ethnic Claim,. I also include an interaction term that multiples the effect of Ethnic Claim by the Percentage of Concessions (Strong) to test the propensity of groups with support bases to sign agreements when they gain more concessions on their demands. While I expect the majority of rebel groups to settle when their demands are met, all groups will not pursue peace when they receive what they demand in negotiations. Groups that do not rely on support bases for survival should be less likely to settle even given 93

103 concessions. 3 Organizations with access to natural resources and a steady stream of recruits are likely to be self-sufficient and should be less likely to rely on support bases. Since rebel groups with access to extractable resources raise their own revenues, they are less likely to rely on civilians for financial backing and are, therefore, less likely to be responsive to civilians demands. Groups recruiting forcibly are similarly less likely to consider any supporters preferences for settlement. Such self-sufficient groups should lack mechanisms forcing them to commit to settlements, as there are no audiences to inflict costs for failing to follow thru on their stated demands. As nothing prevents these groups from engaging in cheap talk or failing to commit to agreements when governments make important concessions, self-sustaining groups should not be as likely as groups with support bases to sign agreements when they gain concessions on their demands. I use the data from Lujala et al. (2009) to code whether there are extractable resources, or Gems, Oil and Drugs in the conflict region and child recruitment, as a proxy for forcible recruitment. I code whether there are explicit negotiations between the dyad in a month. Negotiations do not ensure concessions to groups, but agreements are more likely when belligerents are already engaging in dialogue. In most cases, offering negotiations to a rebel group signifies the softening of a government s bargaining position and a (relative) receptiveness to compromise. As is evident in Chapter 4, when governments are downtrodden and unable or unwilling to bear further costs in war, they are most likely to offer negotiations and concessions. As negotiations constitute a significant concession to 3 Groups that neither have support bases nor are self-sustaining are highly unlikely to survive and are also unlikely to appear in the data. Even if they were to survive, concessions to these groups should induce settlement as only groups that find conflict lucrative in some sense should want to continue war. With few exceptions, groups finding conflict lucrative will only be those extracting resources. 94

104 rebels, it should be an indicator of conflict ripeness where conflicts with negotiations should be more prone to peaceful settlements. In these analyses, negotiations may also account for the number of demands and concessions that are made during a period of explicit bargaining. During negotiations, demands, offers, and counteroffers are made frequently. Thus a disproportionate number of demands and concessions should be expected to be made during talks. This directly affects the calculation of the percentages in the concessions variable. Because it simultaneously affects the dependent and independent variables, negotiations can be a potential confounding variable and should, therefore, be included in the analyses. Existing literature has tended to rely on the balance of capabilities and commitment problems as predominant explanations for the success or failure of settlements (Walter 1997; Werner 1999). Walter (1997) argues that security concerns prevent belligerents from agreeing to settlements even when they contain the most favorable conditions. When belligerents fear their opponents will take advantage of their prostrate positions, they will be least likely actually to terminate their conflicts. Even if rebels were to gain all they demanded through bargaining, they should still be unlikely to lay down their weapons if they do not believe the environment is permissive for settlement. Third parties, however, can alleviate this security concern. To capture whether there is a security environment favorable to settlement, or a credible guarantor present, I code whether there is a United Nations Peacekeeping mission present in the country in each month. Specifically, I code the number of United Nations military troops, military police, and military observers in the conflict region. I also code whether a major power is among the troop contributors on a mission. This information is available on the United Nations Peacekeeping website. The UN records 95

105 fine-grained statistics on each of their missions that allows the values of these variables to vary by month. This is useful as the level of analysis for the study is the dyad-month. Although information on African Union and ECOWAS missions are likely to be more appropriate for Africa, reliable and complete statistics on these missions are quite difficult to find. Werner (1999) and Werner and Yuen (2005) suggest that conflicts tend to reignite when settlements did not mirror the distribution of power between the belligerents or when shifts in the distribution of power occur. Thus, the distribution of power between the belligerents should explain a lot about patterns of settlement. I use the rebel strength variable from the Cunningham et al. (2009) NSA dataset to examine the relative power between the belligerents in conflict. This is an ordinal variable that captures whether rebels are much weaker, weaker, at parity with or stronger than the government. From the NSA data, I also include Explicit Support as outside support can alter the balance of capabilities in the rebel s favor. While concerns about relative capabilities and the distribution of power among belligerents is surely important, it should not explain quite as much about patterns of conflict settlement as whether the underlying issues in conflict are addressed. Although security concerns are going to be incredibly important for determining whether belligerents put down their weapons, whether they actually agree to a formal settlement should be more dependent upon the terms of an agreement than the security conditions. However, I still include variables that might impact whether rebels believe the security environment to be permissible for peace. The state s use of violence against civilians should certainly influence rebel s attitude toward settlement. Rebel groups may be more inclined to agree to settle when civilians are 96

106 forced to bear the brunt of conflict. Groups that are connected to civilians being targeted by indiscriminate violence may be especially likely to seek settlements as high levels of violence may lead their support base to become war-weary and less inclined to continue supporting violent means of dissent. Lyall (2009) argues that governments often use indiscriminate violence to suppress insurgent attacks. He further suggests that as government attacks on civilians rise, rebels support bases will pressure groups into ending the conflict as the violence associated with it endangers them without offering any apparent benefits. Governments are expected to intentionally employ violence to coax a civilian base into pressuring rebels to stop their attacks. If Lyall s account is accurate, government violence should increase the likelihood of settlements. Conversely, Kalyvas (2004) argues that this type of government violence is inefficient as indiscriminate attacks drive civilians into the rebel s arms. This is especially so when rebels have some level of control over territory. Here, indiscriminate attacks aid rebel mobilization, therefore exacerbating conflict. Although both arguments are plausible, I find more reasonable the expectation that rebels will become more inclined to settle as their support base disproportionately bears the cost of war. Many of the variables that impact whether the government offers negotiations and concessions are characteristics that should approximate a general inclination toward peace and should be relevant in examining the determinants of peace agreements. Therefore, I include a variety of variables explained in greater detail in the previous chapter, including Main Group, ln(time), ln(deaths), Number of Conflict Episodes. To test the effect of concessions to rebel group demands, I use probit models for all the models in this chapter with the exception of Model 5 in Table 5.3, where I use a bivariate probit model. In Chapter 4, I suggested it would be important to account for whom 97

107 governments offer negotiations to when examining whether both sides sign agreements. If belligerents have not taken the most preliminary steps in the peace process, they should be less likely to reach peace then those dyads that have been taking gradual, consistent steps towards peace. I expect these two processes are interrelated as the decision to negotiate and the decision to settle a dispute peacefully are often made together. 4 I use a bivariate probit model to account for the relationship between these two phenomena. I eschew a Heckman selection model because the Heckman model assumes that the outcome stage only happens after the selection stage has occurred. Although dyads should not, theoretically, enter into agreements without first beginning negotiations, it is still possible that they will. 5 Like the Heckman model, the bivariate probit allows the errors of the two models to be correlated, but without assuming that one event must precede the other. A significant ρ in this model indicates that the two models are in fact related. Results I use probit models to test the effect of offering concessions to rebel s demands on the probability belligerents will sign an agreement in a month. Table 5.1 presents very parsimonious models that examine the effect of the main independent variables only controlling for the number of demands and whether there were concessions offered in addition to those demanded by rebels in that month. Each model in this table employs a different operationalization of concessions to rebels demands. Model 1 examines the effect of a greater percentage of strong (maximal and substantial) concessions on the likelihood of an agreement. The second model looks at the number of strong concessions governments 4 Just because belligerents decide that they will pursue peace it does not mean that peace will be established. There are many steps to be undertaken after the initial decision to pursue peace. 5 While there are a few occurrences of agreements without negotiations in the data, it is quite infrequent. 98

108 offer rebels in a month. The next two models, Models 3 and 4, test the effect of the weak measures of concessions that include maximal, substantial and moderate concessions to rebel demands. Models 5 through 8, examine the effect of only political concessions on the probability of an agreement. Model 5 and 6 examine the strong measures of political concessions while the Model 7 and 8 examine the weak measures of political concessions. Generally, this table shows a strong positive effect of concessions on the probability of an agreement. As expected, when governments offer greater concessions on rebel demands, belligerents are more likely to locate acceptable settlements. The relationship is robust as all eight models demonstrate a similar significant positive effect. There is not a large difference between the models where all rebel demands are considered and those that only consider the subset of political demands. Comparing Model 1 and Model 5 in Table 5.1, the effect of concessions on the probability of an agreement is only slightly stronger when all concessions are included in the main independent variable. <Insert Table 5.1 Here> From Table 5.1, it is evident that offering additional concessions to rebels has a large positive effect on the probability of an agreement. When governments appear generous or are able to sweeten a deal with extra concessions, dyads are more likely to agree to settle their dispute through peaceful means. An alternative explanation for the salience of this variable is that it might capture the effect of governments offering concessions to demands made in previous months. As the main independent variables only account for concessions made in the same month as rebel groups demand them, instances where governments provide concessions in following months are excluded. Additional Concessions, then, would 99

109 include any of the concessions granted at time t that address rebels demands made before time t. It should also capture any demands missed when coding the main concessions variables. In Table 5.1, the number of demands rebels make in a month is also positive and significant. Reasonably, the more demands rebels make in a month, the more concessions they are likely to attract as groups asking for more are likely to receive more. It is important that this variable is included in the model as it accounts for the effect making a demand (as opposed to not making one) has on the probability of an agreement. I include this variable to moderate any exaggerated effect demand-making might have on the dependent variable. Specifically, conflicts where rebels make some demand in a month should be closer to settlement than conflicts where rebels are not asking for anything. While most groups make a demand each month, not all groups do. As such, it is not reasonable to expect governments to offer concessions to rebels that do not make any demands. 6 It is not only prudent to account for whether rebels make a demand at all, it is also important to measure the volume of demands made as the number of demands can explain quite a bit about the state of an explicit bargaining process and therefore the likelihood of a peace settlement. Dyads engaged in serious bargaining might be exchanging position papers, modifying bargaining positions, and revising their demands and offers more frequently than groups that are not engaged in real negotiations. The rapidity and volume of demands, then, might indicate a mature negotiating process and a conflict closer to settlement. <Insert Table 5.2 here> 6 It is not reasonable to exclude groups not making any demands from the analysis as groups can potentially receive concessions without first making any demands. 100

110 The relationships found in Table 5.1 are consistent even after other covariates are included in the model. This is evident in Table 5.2. I include both Table 5.1 and 5.2 to demonstrate the results I garner are not due to missing data. Between the two tables, 1000 observations are lost mainly due to missing data on the natural resources data. An additional 400 observations drop out between Table 5.2 and 5.3, when additional covariates are added. Despite this, the substantive effects of the main independent variables are unchanged. In Table 5.2, I include measures accounting for rebels support bases and whether rebels have the means to sustain themselves without civilian support. This table also contains an interaction term capturing the effect of the percentage of concessions on rebel demands when rebels have an ethnic support base. The results in Table 5.2 confirm hypothesis 1. When rebels have ethnic support bases and when they gain greater concessions on their demands, they are more likely to sign agreements. Rebels are likely to sign agreements when governments offer greater concessions on their demands, even when they do not rely on civilian support bases. The negative sign on Ethnic Claim demonstrates that when groups have ethnic support bases and are offered no concessions on their demands, they less likely to agree to settle their demands peacefully. However, this effect is insignificant across all the models, suggesting there is no effect of an ethnic support base, when there are no concessions. Whether groups can sustain themselves without a support base tends to have no significant effect on the probability of settlement. Whether groups extract oil or drugs or forcibly recruit, all negatively impact the probability of agreements, but these measures are statistically insignificant. If a group extracts gems, however, the probability of a settlement does increases by a lot. Tables 5.1 and 5.2, clearly support hypothesis 1; offering greater 101

111 concessions to rebel groups increases the prospects of peace. The anticipated negative effect of the measures symbolizing self-sustaining rebels, however, is not supported. <Insert Table 5.3 about here> In Table 5.3, I run four additional probit models examining the effect of government concessions on agreements, adding more contextual variables. For example, I add measures that capture whether dyads are involved in negotiations in that month or in the previous one. I add variables that suggest the difficulty of the bargaining situation- the number of rebel groups, the length of the current conflict episode and the number of conflict episodes. I include variables to proxy whether a government suffered recent costs which might prompt it to relax its obdurate bargaining posture and pursue conciliation- terrorism, battle related deaths, Polity, and whether a group was the main group inflicting costs on the government in that month. Finally, I control for factors found in previous studies to be important for settlement, namely power distribution (relative strength and explicit support), whether there are third party guarantors and the level of government coercion being levied against civilians. Model 1 in Table 5.3 represents a fully specified model. Model 2 is a variation of the first model but contains a one-month lag on the negotiation covariate. Model 4, employs one-month lags on all independent variables, with negotiations being twice lagged. Model 5 is a bivariate probit model controlling for the interdependence in the equations on agreement and negotiations. Overall, the results demonstrate that as governments offer a greater percentage of concessions to rebels demands, conflicts are more likely to be settled through agreement. In every model, with the exception of Model 3, a greater percentage of concessions offered to rebels with support bases is associated with an increase in the likelihood of an agreement. 102

112 The more concessions governments offer to rebels, even those without support bases, the more likely dyads are to find acceptable agreements. The effect of rebels having support bases without having gained any concessions in that month is clearest in Model 2 and Model 5. Shifting from Model 1 to Model 2, the negative effect of having just an ethnic claim reaches significance, marginally, at the.10 level. This can be attributed to the lagged negotiation variable in Model 2. Controlling for whether a dyad was engaged in bargaining in the previous period, or had some bargaining history, makes the effect of the main independent variables more pronounced. In addition to revealing a statistically significant effect of the conditional effect of having an ethnic support base, the magnitude of the coefficient representing the percent of concessions offered rebels becomes stronger and confidence in that point estimate increases. The interaction effect also increases in magnitude moving from Model 1 to Model 2. Model 3, replicates Model 2, but adds lags to all of the independent variables. This model demonstrates the effect of all the independent variables in the previous month (t-1) on the probability of an agreement this month (t). Most notable from this model is the change in signs on the Ethnic Claim and the interaction variables. Model 3 indicates that groups with support bases are less likely to sign agreements when they gained more concessions on their demands in the previous period. This is counterintuitive and unexpected. The conditional effect of an ethnic support base is insignificant, but positive. When the effects of the independent variables on an agreement in the following month are considered, the number of UN troops no longer has a significant effect, but UN observers negatively impacts the probability of an agreement. 103

113 In Model 4, I examine the effect that offering political concessions has on agreements. 7 Aside from the concessions variable and its interaction with Ethnic Claim, Model 4 is specified identically to Model 2. This allows for the effects of covariates in these two models to be directly compared. The results are not drastically different. Whether governments address all types of rebel demands or just those that are political in nature, greater concessions still increase the probability of conflicts being settled through negotiated agreements, especially when rebels rely on support drawn from the population. The results of the bivariate probit model are presented in Model 5 of Table 5.3. The two columns in this model represent the two outcome variables of interest. When accounting for the interdependent nature of negotiations and agreements, the relationship between the main independent variables and the dependent variable of interest becomes clearer. Model 2 and the agreement equation in Model 5 are specified identically and can be directly compared. As compared to Model 2, the negative relationship of Ethnic Claim in Model 5 is stronger and more significant. Thus, after accounting for dyads decision to negotiate in time t, the effect of having a support base and gaining no concessions decreases the probability that a dyad will end their conflict with a peace agreement. 8 This is important as the argument set forth in Chapter 2 posits rebel groups with support bases should be less likely to stop fighting until their demands are met. The interaction term between the Percentage of Concessions and Ethnic Claim is also stronger in the bivariate probit model displayed in Model 5. As rebels with support bases gain more concessions on their demands they are much more likely to agree to settle their disputes peacefully. Finally, the positive 7 The interaction term reflects the difference in the concessions variable. 8 I do not test for the effect of negotiations at time t in the bivariate probit model because both models must be specified identically (for more information about the assumptions and restrictions of the bivariate probit model see Sartori (2003)).Since the first dependent variable in Model 5 is negotiations in time t, it was not possible to use negotiations in time t as a covariate in either model. I use negotiations in the previous period instead. 104

114 effect of concessions when rebels do not make any demands on behalf of a civilian group is stronger in the bivariate probit model suggesting that generally, rebel groups should be more likely to sign agreements when they get what they want out of negotiations. The bivariate model is useful as it shows a stronger impact of the main independent variables on the probability agreements than did the probit models that do not account for whether the dyad decides to negotiate. Furthermore, the model suggests that the decision to negotiate and the decision to end a conflict peacefully are related decisions. As in Table 5.2, the effect of the variables capturing whether groups are self-sustaining are largely insignificant in Table 5.3. Only exploiting gems has an effect on the probability of a settlement and that effect is opposite to what I posited. Groups that extract diamonds, for example, are much more likely to sign agreements than groups that do not. While unexpected, the finding is perhaps not unreasonable. It is possible that groups extracting this kind of natural resource are more likely to be accommodated in negotiations. 9 Governments may be able to offer more attractive concessions such as those allowing rebel groups or their leaders to continue to maintain control over these resources even after conflict. An example of such an offer is that extended to Foday Sankoh by the government of Sierra Leone. As mentioned previously, the terms appointed him chairman over strategic resources allowing him control over the diamond industry. The finding is likely showing that governments are able induce settlements with rebels when they can assure the benefits of a settlement will outweigh the benefits of continued conflict. 9 Groups extracting gems are much more likely to be offered negotiations and are marginally more likely to be offered concessions than groups that do not extract gems. In Chapter 4, Table 5.2 shows groups extracting natural resources are much more likely to gain negotiations while Model 3 from Table 5.2 supports the notion that they gain more concessions. 105

115 Extant literature on civil war settlement suggests the power distribution between belligerents should explain when settlements are most likely to occur. In the last chapter I demonstrated that when rebels are stronger and can impose more costs, governments are more likely to adopt conciliatory postures and offer negotiations and concessions. This relationship is evident in the negotiation stage in Model 5. Governments are more likely to adopt conciliatory postures when rebels are able to inflict significant costs. Thus, it should be no surprise that agreements are also more likely to be reached when rebels are strong. When rebel groups have explicit outside supporters, agreements are also more likely as an intervention on behalf of rebels is likely to shift the balance of capabilities in favor of a rebel victory. Balch-Lindsay et al. (2008) demonstrate when there is an intervention on the side of the rebels the hazard of conflict terminating in opposition victory or a negotiated settlement increases. As the prospects of a government victory decline, states should be more likely to attempt to pursue negotiated settlements. The results here demonstrate a similar pattern; belligerents are more likely to pursue agreements when rebels have outside support. Overall, having credible guarantors, or the UN present in a conflict, increases the prospects of a peace settlement. However, this does not diminish the importance of the findings regarding the effect of providing concessions on the probability of an agreement. That government violence against civilians increases the probability of an agreement expresses support for the idea that making war painful for a group s civilian base increases the probability that they will pressure the group they are supporting into settling. The finding that a greater number of rebel groups increases the probability of a settlement is puzzling. Cunningham (2006) argues and finds that multiparty conflicts should be more difficult to settle as the inclusion of more groups with diverse preferences makes it more 106

116 difficult to locate a bargain each party prefers to the continuation of violence. The findings in this chapter differ from this account as they demonstrate an increased probability of a bargain when there are more rebel groups. The dissimilarity might be accounted for by considering the difference in the variables we employ to test this effect. The operationalization of Cunningham s veto player variable is much more nuanced than that in the present analysis. He separates original groups from splinter groups and considers a veto player only a group with both differing preferences and an ability to continue war unilaterally. I only include the number of groups as an independent variable. Substantive Effects Figure 5.1 graphically displays the effect government concessions have on the predicted probability of agreements. This figure shows the difference in the probabilities of agreements when rebels with support bases gain concessions from governments and when rebels without support bases gain concessions. The dashed line represents the predicted probability of an agreement when the government offers concessions and when rebels make claims on behalf of an ethnic group in the state. The shaded region is the confidence intervals around the estimates. The solid line is the predicted probability of an agreement by the percentage of concessions governments offer when rebels do not make any ethnic claims. The spiked lines are confidence intervals around these estimates. It is clear from this figure, there is only a miniscule positive effect of concessions to rebel demands when rebels have no ethnic support bases. When rebels do have support bases, the probability of an agreement is much larger and positive. For example, at zero concessions to rebel demands, the probability of an agreement is very close to zero. However, at 30 percent concessions there is a 25 percent chance of an agreement. At 50 percent concessions, the probability of 107

117 an agreement is about It is necessary to note there is only one observation where the government offered substantial concessions on more than 50 percent of the demands made by rebels with support bases. As such, this observation is incredibly influential and necessarily affects greatly the trend observed. This is evidenced by the shape and size of the confidence intervals, specifically the upper bound. Conclusion <Insert Figure 5.1 Here> The analyses in this chapter test the first hypothesis advanced in the theory chapter of this dissertation. I examine whether government concession offers bring conflicts closer to settlements. The results show they do. When governments offer more concessions on rebel demands dyads are more likely to agree to the peaceful settlement of their dispute. The positive effect of concessions is particularly evident when rebels rely on civilian support bases. When these groups gain no concessions, they are considerably less likely to sign agreements. In the next chapter, the second hypothesis is tested. This chapter will examine whether conflicts are likely to terminate when concessions are offered to rebels. 108

118 Chapter 6: Stopping When They Get Enough: The Effect of Concessions to Rebel Demands on Civil War Termination Literature While scholars have approached the topic of civil war termination from a variety of angles, the predominant explanation for civil war termination, or lack thereof, is the commitment problem. Fearon (2004), for example, argues that fluctuations in government capabilities are responsible for the difficulty in settling intrastate conflicts. Governments are motivated to offer great concessions to end conflicts when they are weak but have incentives to renege on these offers once they become strong again. Given these incentives exist, rebels will have difficulty accepting such offers. While less common in recent years, information problems have also been noted as explanations for the persistence of civil wars. Wagner (1993:258), for example, states wars will end when the combatants have convergent expectations about the outcome of further fighting, regardless of the goals that led them into war in the first place. Here, civil wars are expected to endure until a sufficient amount of learning has occurred. Once learning has taken place, belligerents will come to an agreement on the cessation of hostilities in order to avoid the further costs of war. As Wagner s comment suggests belligerents will settle despite having gotten their demands met, it may be assumed he severely discounts any effect accomplishing war aims has on settlement. I do not rule out existing arguments as viable explanations for termination; information and commitment problems certainly abound in civil war and they most certainly have their respective effects on termination. Instead, I offer an alternative and additional explanation for termination overlooked in the present literature. I argue that demands matter. If rebel demands are not 109

119 addressed, belligerents should be less likely to stop fighting, regardless of the information available to them. Especially when rebel groups act on the behalf of a civilian support base, they should be less likely to stop fighting until they meet the goals they set out to attain. Zartman (1993:29) suggests an insurgent organization may suffer defeat and its leaders may be killed, but the group remains and its sources of grievance still remain potentially alive. The struggle is not over the existence of the group, but over the particular deprivations it suffers and the effectiveness of the mechanisms for handling them. Convergent expectations may impact belligerents perceptions of whether rebels will win or lose, but should not determine when a war ends. Whether rebels goals are met, however, should explain termination. If concessions tackle the root causes of rebellion, rebels will have very little incentive to continue fighting and more importantly, civilians will have very few reasons to continue supporting violence. Barring concessions of this kind, wars will continue. Even if rebels are defeated, conflicts are likely to recur over the same issues, as the conditions that supported rebellion in first place do not cease to exist. Research Design In this chapter I test the second hypothesis offered in Chapter 2: conflicts will be much more likely to terminate when governments provide meaningful concessions to rebel groups with civilian support bases. Providing substantial concessions to rebels demands is likely to satiate rebel support bases such that supporting rebellion is no longer reasonable or profitable. Thus, as the volume of government concessions increases, the probability of termination of the violence associated with a civil war should also increase. While most groups have incentives to stop fighting once their demands are met, this effect should be most pronounced for groups mobilizing resources from a civilian support base. Conversely, not offering concessions to groups with support bases should lead to a very low probability of conflict termination. 110

120 Below, I discuss the results of three different types of analyses conducted to determine the impact giving concessions to groups with support bases has on conflict termination. Tables 6.1 and 6.2 employ probit models to determine the predictors of termination. I examine the results of bivariate probit models in Table 6.3 to determine whether accounting for previous stages of conflict resolution (negotiations and formal agreements) actually affects the inferences we make about conflict termination. The final sets of analyses are duration models displayed in Table 6.4. These models examine the affect that addressing rebel demands with concessions has on the time until hostilities cease. As this chapter relies on most of the covariates discussed in the chapter on agreements, I only discuss variables not included in Chapter 5. An interaction term combining the concessions variable and the Ethnic Claim variable perfectly predicts termination, and thus could not be used in the analyses presented in this chapter. In other words, there are no cases in these data where a termination occurs and the interaction between the percentage of government concessions the Ethnic Claim variable is zero. Although perfect prediction between the dependent variable and the interaction term strongly supports my hypothesis, it is a problem because it does not allow me to statistically estimate a relationship. Thus, the operationalization of ethnic support base has changed. The variable used in previous chapters, Ethnic Claim was coded 1 if a rebel group made claims against the government on behalf of any ethnic group within a state. Instead, I use Ethnic Recruitment which is coded 1 if a rebel group actively recruits from a local ethnic group. Although, Ethnic Claim is the best variable to test the mechanism I suggest in the theory section, Ethnic Recruitment should be a sufficient substitute. Rebels that make claims on behalf of civilian groups should certainly be beholden to the preferences of those civilians as the demands they make are directly related to the support the garner. In other words, civilian support is completely predicated on rebels making acceptable demands on their behalf. Rebel groups recruiting their troops from an ethnic group should also be considerate of that 111

121 ethnic group s preferences. When rebels do not recruit forcefully, they attract civilian support with the demands they espouse. In most cases, the claims they make will help them recruit. Therefore, Ethnic Recruitment is likely a derivative of Ethnic Claim and should be an adequate, albeit inferior, substitute theoretically. The correlation coefficient between these two variables is 0.62 and is significant at the 0.01 significance level, indicating they are in fact statistically related. As these variables are both conceptually and statistically related, I am satisfied they are sufficient substitutes. To be sure, I reanalyzed the models displayed in Chapter 5 using Ethnic Recruitment in lieu of Ethnic Claim and found the relationships between the variables of interest to be very similar 1 Even after replacing the Ethnic Claim variable with Ethnic Recruitment, there is still some difficulty estimating the models examining civil war terminations. One of the main explanatory variables is dropped from the probit models as a perfect prediction is reported between termination and the interaction term (between the percent of concessions offered on rebel demands and a support base) 2. To discern a relationship between the dependent and independent variables, I include lagged versions of the dependent variable and report those analyses in the first three tables. It is probable that these estimation problems are the result of insufficient variation in the data. If this is so, the issues encountered here might be ameliorated by addressing the problem of missing data. Since a lot of observations are list-wise deleted in models containing many covariates, a large number of the terminations in the sample are excluded from the analyses. In the full sample, there 1 I replicated the fully specified model examining the effect of offering concessions to groups with support bases on the signing of formal agreements and while the directions of all of the variables are the same across models, the Ethnic Claim variable seems to be a marginally better measure than the Ethnic Recruitment variable. The relationships among the main covariates (concessions, ethnic support base, and the interaction between the two) and the explanatory variable are more significant (significance level shifts from 0.01 to 0.05 and 0.05 to 0.10) and slightly larger in magnitude in the models using Ethnic Claim. This trend suggests that the use of Ethnic Recruitment will likely yield results very similar to but less significant than the actual effect of support bases on conflict resolution. 2 The perfect prediction issue was much more acute with the Ethnic Claim variable. Even lagged values of this variable were reported to perfectly predict termination. On the other hand, an interactive relationship between concessions and the Ethnic Recruitment variable could be estimated when lagged versions of the constitutive terms of the interaction was included. 112

122 are only 124 terminations (in 3148 observations). After list-wise deletion, there are only 64 cases of episode termination (in 1908 observations). In addition to excluding nearly half of the events the models are trying to predict, 1240 observations are dropped from the analyses by list-wise deletion of observations with missing values. Between losing cases of successful termination and losing so much variation in the independent variables, there might not be enough information for the models to be properly estimated. In the next chapter, I discuss some of the data issues I confront in this chapter in greater detail. I also suggest ways that I will fix these issues in future iterations of this project. The dependent variable used in this chapter is a binary measure representing the termination of a conflict episode. These data come from the UCDP Dyadic Termination Dataset (Kreutz 2010). A termination is recorded when there is at least one year where the three criteria UCDP use to code conflict activity are not met. That is, a conclusion to a conflict episode would be recorded when fewer than 25 battle-related deaths are documented, the parties cease to be organized, or their preferences for an outcome cease to be incompatible. In the cases where UCDP is unsure of the month in which the episode was terminated, they code an end in December of that year. I retain this coding. Doing so will likely only bias the results against my hypothesis. As I expect terminations to follow large numbers of concessions, terminations occurring in December after concessions were offered in June, for example, would not provide support for my hypotheses even if this termination is a direct result of the concessions given earlier in that year. Thus, I expect any inaccuracies in the termination variable to lead to the estimation of weaker relationships between concessions and termination. 3 3 That is unless the default termination coding always covaries with an offer of a great number of concessions, which is unlikely. 113

123 Results Table 6.1 shows the first set of analyses on termination using probit regressions. There are four models presented in Table 6.1. Model 1 and 2 are base models only including the main covariates of interest. Model 1 employs a one-month lag on the independent variable while Model 2 employs two-month lags. 4 These lags correspond to termination in the following month (t+1) and two months later (t+2). Models 3 and 4 include many potential confounding variables as well as many variables frequently employed in termination research as controls. Models 3 and 4 employ one- and two-month lags, respectively. The results presented in Model 1 of the first table demonstrate that termination is more likely the month after governments offer rebels more concessions on their demands. However, this relationship only applies to groups without support bases. Groups with support bases are not any more likely to stop fighting the month after they have been offered meaningful concessions. This relationship is evident in Model 3 as well. Model 2, however, shows that for both groups with and without support bases, a greater percentage of concessions on rebels demands increases the probability that belligerents will lay down their weapons two months after those concessions are given. In Model 4, the positive conditional effect of concessions is no longer significant, but the interaction term still indicates groups with support bases are more likely to stop fighting two months after they have gotten what they wanted from their governments. <Insert Table 6.1 Here> Table 6.2 displays the results of bivariate probit models examining episode termination as well as the prior stage of negotiation or agreement. In addition, these models use a termination dependent variable aggregated across time. I believe operationalizing the dependent variable this way 4 As previously noted, models could not be estimated without lags. 114

124 is justifiable as it is hard to believe belligerents will be able to stop fighting as soon as they come to an agreement or as soon as concessions are offered. In addition to the security concerns highlighted by Walter (1999, 2009), there may be a number of factors impeding immediate peace. Actors engaging in spoiling behaviors may use additional violence to attempt to disrupt peace immediately following an agreement. Furthermore, fighting might continue as members of rebel groups remain in the bush without knowledge that the war was declared over. Coordinating demobilization and disarmament may also take a couple of months. This creates the opportunity for armed, undisciplined, yet-to-be cantoned troops to fall back into fighting. In some severe cases, the practical considerations related to ending fighting can be a cause of further conflict for far longer than just a couple of months. Consider an example. In July 2003 the Burundian rebel groups Palipehutu-FNL and CNDD-FDD, were still fighting with government troops as they demanded the establishment of cantonment sites that would meet the basic survival needs of members of the organizations, including food and health care. Although the parties had agreed to end hostilities seven months prior, in December 2002, the rebels would not stop fighting until their logistical concerns were dealt with in a serious manner. 5 In situations like this, it is reasonable to expect that instantaneous peace will be unlikely If the details outlining how belligerents should stop fighting are not established or executed in a way that prevents further fighting, we shouldn t be surprised if fighting continues. This should be true even when belligerents are able to come to agreements regarding the fate of the status quo. Given the difficulty belligerents face in the immediate wake of a settlement, looking at whether conflicts end a few months after concessions are offered is quite reasonable. This is my 5 Burundi: Rebel soldiers demand "better conditions" in cantonment sites. BBC Sumary of World Broadcasts. July

125 strategy in measuring termination for the analyses reported in Table 6.2. This table shows the results of a probit regression where the dependent variable is coded 1 if an episode terminates within three months. 6 More precisely, the dependent variable includes conflict terminations that occur in either time t, t+1 or t+2. I also add lagged independent variables going back to time t-2 to see how additional time affects the probability of peace after concessions. For example, a lag at t-1 shows the effect of concessions in March on the probability of a termination occurring in either April, May or June while a lag at t-2 shows the effect of concessions in March on the likelihood that conflict will terminate in May, June or July. This table reveals a positive and statistically significant effect of concessions at time t on the probability of a termination within three months of t. This result is conditional on there being no ethnic recruitment or civilian support base. The conditional effect of Ethnic Recruitment is negative and statistically significant. Therefore, rebels recruiting from an ethnic support base are significantly less likely to stop fighting when they gain no concessions from their government. The interaction between the two of these terms is positive but insignificant. This indicates there is no effect of concessions at time t on the probability of settlement within three months for groups with support bases. <Insert Table 6.2 Here> The lagged independent variables in these models echo the results in Table 6.1. A onemonth lag on the independent variables produces the exact relationship hypothesized. A greater number of concessions to groups without support bases increases the probability of termination within three months after the concession is given. When a group with a support base gains greater 6 I also ran analyses using dependent variables that examine terminations within two and four months of time t. The results did not change drastically. Instead of the interaction term being insignificant at time t, the interaction term is negative and significant when examining termination within two months. 116

126 concessions on their demands, they are significantly more likely to stop fighting in the three months that follow. However, if a group with a support base gains no concessions, they are significantly less likely to stop fighting in the next three months. The concessions variable lagged two months (to t-2) does not have any effect on the probability that a conflict will terminate within two months when rebels do not recruit from an ethnic support base. If governments offer concessions in March, groups without support bases are no more likely to terminate their conflicts between May and July. Oddly, groups with support bases that gain no concessions in March are much more likely to settle between May and July. Directly related to the hypothesis I am examining in this chapter, the results show that as groups relying on civilian supporters gain greater concessions on their demands termination in the next few months becomes more likely. Groups that gain concessions in March are much more likely to cease fighting by July. <Insert Table 6.3 Here> Table 6.3 employs bivariate probit models to examine the relationship between concessions and termination. As one can imagine, the processes of negotiations, agreements and termination are interrelated. When a government makes the decision to negotiate with rebels they likely also decide they will offer concessions to end their wars. Similarly, the decisions to sign agreements and cease hostilities are also likely related, as most well meaning dyads would not bother to ink an agreement until they believe their wars can be ended peacefully. Thus, it makes sense to employ a model that can address the interrelated nature of these processes. To this end, I use the bivariate probit model that examines the correlation between the two models errors. Since the models seem to perform better when they include independent variables lagged at t-2, I only display models with these lags. The results on both of these models show that once accounting for the stage of conflict resolution 117

127 preceding it, a greater number of concessions two months prior still increases the probability of termination. The interaction term reveals once negotiations or agreements are accounted for, offering more concessions to rebels with support bases increases the probability that that conflicts will terminate two months later. There is no effect of having a support base without concessions on the probability of termination in the next two months. <Insert Table 6.3 Here> Although ρ in both models is statistically significant, the results change only slightly compared to those from the probit model presented in Table 6.1, Model 2. The only result that varies meaningfully across the three models is the significance of the variable representing the percentage of concessions at t-2. In the bivariate probit model, this variable attains a higher level of statistical significance. While the ethnic recruitment variable also changes, the variable is statistically insignificant and is therefore a negligible change. That there are only minor changes across these models suggests the use of the bivariate probit model is more interesting theoretically but not more interesting statistically. While a statistically significant ρ coefficient suggests that there is value added to using the bivariate probit, it can also result from model misspecification. I cannot discount this as a possible explanation as including a measure of the percentage of concessions in a model examining whether belligerents engage in negotiations makes very little sense substantively. The final analyses included in this chapter are Cox proportional hazards regressions presented in Table 6.4. These models are important because unlike the probit and bivariate probit models, the duration model demonstrates the effect covariates have on the time until termination. The dependent variable in Model 1 examines the effect of the covariates on the probability that a conflict will terminate in time t. Model 2 examines termination two months after (t+2). The 118

128 dependent variable in Model 3 accounts for terminations that occur within three months, either in time t, time t+1 or time t+2. <Insert Table 6.4 Here> The results here support the conclusions from the other models. Model 1 suggests offering greater concessions to groups that do not have support bases increases the probability that a conflict episode will end in the next period. Overall, the model supports the idea that offering a greater percent of concessions to rebels with civilian support increases the probability of settlement in that month. 7 Model 2, shows this trend more clearly opposite. Although offering concessions to more credible groups does not make peace more likely immediately, it does increase the hazard of termination in subsequent months. Model 3 also shows that offering a greater percentage of concessions to groups with civilian support increases the probability that a conflict will end in the following three months. Substantive Effects Figure 6.1 demonstrates quite clearly that hypothesis 2 is supported. I used Model 2 in Table 6.1 to graphically depict predicted probabilities from this model. The resulting figure shows the different effects concessions have on the probability of an episode termination when rebels recruit from civilian bases and when they do not. The trends show that while offering concessions to rebels without support bases increases the probability of an episode termination, the effect of concessions is much more pronounced when rebels do have civilian support bases. Concessions to rebels 7 Although there is a negative sign on the interaction term in Model 1, when the effect of the interaction term is considered in conjunction with the effects of the constitutive terms, the overall of concessions on termination is positive. Thus, concessions to rebels generally increase the likelihood of episode termination. 119

129 generally should be effective at producing successful termination, but should be especially effective when rebels have constituencies that can bind them to agreements. <Insert Figure 6.1 Here> Conclusion This chapter examined whether rebels stop fighting when governments grant them enough concessions on their demands. The analyses offer support for my second hypothesis: Groups with civilian support bases are more likely to stop fighting when they gain greater concessions on their demands. The results suggest that while concessions have an immediate effect on the probability of termination, the effect becomes clearer with time. The findings from this chapter and Chapter 5 demonstrate that rebels are not only more likely to sign agreements when governments concede to their demands, they are also more likely to lay down their weapons when they are granted significant concessions. This effect is especially pronounced when examining rebels with support bases. The results support the notions that rebel civilian support bases force rebels to hold out until they gain concessions but they also prevent them from fighting after they have gained concessions on their demands. Even if rebels with support bases wanted to continue fighting, once they have gained concessions that satisfy their support base, conflict would become unsustainable. Although fighting without the cooperation of a support base may be possible for a month or two, rebels are likely to run out of the resources necessary to continue conflict and will be forced to abandon fighting shortly after. This likely explains the stronger effect of concessions to rebels with support bases over time. 120

130 Chapter 7: Conclusion Throughout this dissertation I have challenged the assumption frequently made in conflict literature that rebel groups are inherently non-credible because they have incentives to misrepresent privately held information. I argue, instead, that some rebel groups are able to reveal credible information about their resolve and their reservation points for settlement when they issue public demands. Most rebel organizations act as agents on behalf of a group of civilians that can be considered their support base. When rebel groups publically announce their political programs, they gain support from those sympathetic to their causes. Supporters remain devoted, however, only so long as rebels continue to espouse the aims that originally attracted their support. The rebel support base withdraws financial and tactical support if rebel groups abandon or drastically change their aims, such that they are no longer congruent with the preferences of their supporters. Thus, when rebels publicly profess to fight for some aim, they tie their hands ensuring they will fight until that aim is met. Rebel groups also face being alienated by their support base if they fail to accept reasonable settlements governments offer. When concessions are obtained, rebels should stop fighting as supporters have no reason to remain supportive and most groups cannot sustain rebellion without a support base. Thus, meaningful government concessions to rebel group demands should be associated with conflict settlement, specifically agreements and termination. 121

131 Results In Chapters 5 and 6, I conducted analyses examining how government concessions to rebel demands affect the probability belligerents will sign agreements and terminate their wars. I find that belligerents are indeed more likely to sign formal agreements when government offers address more of rebels demands. Rebels are also more likely to stop fighting when they are offered greater concessions. When groups rely on civilian support to execute their wars, they are only more likely to stop fighting when they gain greater concessions on their demands. When these groups gain no concessions they are significantly less likely to agree to settlements or to lay down their weapons. The results of these two chapters largely support the expectations developed in the theoretical discussion. While the main results are supported, there is significantly less support for the notion that selfsustaining groups are not as likely to stop fighting when their demands are met. The results show that while rebels having a support base profoundly affects the resolution of conflict, there is no effect of rebels being self-sufficient. Chapter 4, however, demonstrates the importance of rebel group self-sufficiency as groups extracting gems, drugs and oil are much more likely to be invited to participate in negotiations. There is weak support for the notion that groups extracting gems gain more concessions, but no support for the idea that groups extracting oil gain more concessions. Groups that extract drugs are much less likely to gain concessions even after being more likely to engage in negotiations with their governments. By far, groups using terrorism are most likely to be invited to participate in negotiations and to gain concessions on their demands. Chapter 4 demonstrates that groups with the power to inflict severe costs on governments are much more likely to be compensated with negotiations or concessions. 122

132 Perhaps the most astonishing finding is that groups mobilizing resources from a significant subset of the civilian population are least likely to be offered negotiations but more likely to be offered concessions. This is counterintuitive. When examining together the results from Chapters 4, 5 and 6, it is clear that although groups with support bases are actually more likely to stop fighting when they are offered what they demanded in settlements, they are less likely to be offered negotiations. Groups that use terrorism as a strategy and groups forcibly recruiting children, however, are significantly less likely to terminate their wars, yet are more likely to gain negotiations and concessions. 1 The results show that although governments are more conciliatory toward rebel groups that inflict severe costs on them, these groups are not any more likely (sometimes significantly less likely) to settle their wars as they do not have mechanisms to ensure they will honor bargains. Given these results, it is no wonder internal conflicts last as long as they do. In contrast, groups with support bases agree to settle their conflicts sooner so offering concessions to this type of group should ultimately lead to shorter conflicts. The results in Chapter 4 reveal, however, that governments are more preoccupied negotiating with groups they feel they have to bargain with as a matter of survival. That groups operating with support bases tend to be weaker and less likely to use painful tactics, like terrorism, is clear from cross-tabulations. Nearly 82% of rebels with support bases are weaker than governments and not even one is stronger. Less than 1.5% of groups using terrorism were those with civilian support bases. Thus, if governments only bargain with groups that have the power to hurt and rebels with support bases are less likely to wield this power, it is 1 I attempted to but could not run models with interactions on the covariates indicating that rebels are selfsufficient, thus I cannot make the claim that these types of groups do not stop fighting when their demands are met. 123

133 understandable why they are so much less likely to be invited to participate in negotiations. While more than 13% of groups negotiating are groups using terrorism, only 2% of groups with claims participate in these negotiations. This difference is statistically significant. There may be at least two explanations for governments failures to bargain with groups more likely to uphold agreements. The first explanation involves governments being shortsighted. While they may recognize that rebels with support bases are more credible, governments may care more about the pain they are suffering today and discount the future. Thus, they decide to bargain with groups using costly tactics in order to stop the pain they cause in the present. Fearon s (2004) commitment problem is applicable here. He suggests commitment problems arise because governments offer concessions when they are weakest. They promise extraordinary concessions to convince rebels to end the costly violence. Governments propose large concessions knowing they will renege on these promises when they recuperate from conflict. Governments, then, do not care whether rebels will uphold their bargains as they intend to renege on any agreement they sign themselves. In such a scenario, peace is not necessarily spoiled by rebels but by governments who cannot commit to implement their costly concessions. An alternate explanation for these empirical findings is that governments do not distinguish between types of groups even though they can and they should. I begin the theory chapter with the observation that both scholars and governments alike believe rebels are generally non-credible. Just as the extant bargaining literature fails to, governments may also fail to distinguish between groups that can misrepresent freely and those that pay costs for doing so. If governments believe all rebels act on incentives to misrepresent information equally, then it is not surprising when governments do not notice that rebel groups with 124

134 support bases are more credible than groups without them. Beginning to elucidate these relationships is a contribution of my dissertation. The work begun here demonstrates the effect of offering concessions to certain types of groups. It demonstrates the necessity of differentiating between rebel groups types and it establishes that rebel credibility is variable. If governments were to pay attention to the characteristics of the rebels they fight, they would find that settling with groups that were more credible would lead to more durable peace. Offering concessions to rebel groups with support bases appear to alleviate commitment problems due to rebel incentives to renege. As this would not diminish government incentives to renege on bargains, time-consistent custodians of peace are still necessary to ensure governments uphold deals they strike. To better understand why governments tend not to negotiate with more credible groups, it might be useful to examine more features of the state in analyses focusing on which rebels gain concessions and negotiations. That groups with support bases are more likely to gain concessions, demonstrates that although most governments do not, some governments do negotiate with groups with support bases. It is possible that some types of states care more about whether rebels are credible and might do a better job at determining how credible rebels are. Perhaps these governments are the more likely to offer negotiations and concessions to credible groups. Other governments may be more sensitive to costs. These states may care less about rebel credibility and more about rebels ability to impose costs. States of this kind should be more likely to offer negotiations and concessions to groups with the power to hurt. There may be significantly more states of the latter type than the former. Therefore, further exploring the distribution of states with the aforementioned characteristics may help explain why credible groups tend to be excluded from negotiations, 125

135 yet are more likely to be afforded concessions when they are included. This will be the subject of future work. Future Work Although there has been a groundswell of recent literature examining peace accords as an explanation for peace durability in civil war, there have been no empirical studies examining whether governments followed through with the terms of these settlements and how this might affect conflict renewal. Until now, research in this area has focused on whether the security environment was permissive for disarmament or whether settlements were commensurate with the distribution of power among belligerents (Walter 1997; Werner 1999). Scholarship has neither focused on how grievances related to onset are addressed in settlements, nor whether concessions on these grievances are actually delivered to aggrieved parties in civil war. These factors, however, are likely to explain why conflicts are so prone to recur after they have been initially settled. The causal link between the implementation of accords and peace durability has yet to be tested in any systematic way. In a subsequent related project, I will examine whether governments efforts to honor terms written into peace agreements affects the probability that conflicts restart after initial settlement. I will also address how the failure to implement deals affects the probability that new groups will emerge to seek settlements promised in old conflicts. Governments frequently make promises of concessions to rebels. These offers may influence groups decisions to sign agreements and even compel them to call off wars, as we have seen here, but if governments fail actually to deliver concessions promised in peace deals, the peace will not endure. That in 2007 the MNJ was fighting for the implementation of a 1995 peace accord signed between the government of Niger and the OAR/CRA, and in 126

136 November 2008, CNDP was demanding the implementation of a peace deal signed with the DRC government in January of that year, reveals that governments do make promises without implementing them, that old groups continue to fight for concessions promised them in previous accords, and that new groups emerge to secure concessions previously promised by unfaithful governments. I plan to address quantitatively the extent to which the aforementioned statements are true. Using expanded data from my dissertation on offers of government concessions, I will gather further information on whether and how governments deliver concessions they promise in peace accords. I will bolster the statistical analyses with case studies to elucidate the processes I argue. Sudan provides an ideal case in this context, as there have been two recent conflicts with two very different outcomes. Within and across each case, there have been a number of rebel groups with varying war aims that have culminated in a number of agreements between the rebels and the government. The amount and degree of the concessions offered have also varied across the cases, as has the extent to which concessions have been implemented. The conflict in southern Sudan represents a successful case. Among the main aims of the Sudan People s Liberation Movement/ Army (SPLM/A) were the selfdetermination/independence of the South and the separation of religion from the state. Despite the protracted nature of the conflict and the fact that the SPLM/A demanded indivisible aims, the government eventually offered and implemented substantial concessions and the conflict ended. In the 2005 peace accord, the government of Sudan offered a referendum for self-determination for the South as well as substantial concessions 127

137 on the separation of state from religion. In 2011, South Sudan held the referendum, voted for and received independence from Sudan. The outcome of the conflict in the Darfur region in western Sudan offers a stark contrast. While numerous deals have been signed between different rebel groups and the government of Sudan, there has been no real settlement to the conflict in Darfur. After rebels signed two 2004 deals providing for the neutralization of the Janjaweed militia and the right for refugees to return home (respectively), a 2006 peace agreement that called for a referendum on Darfur s status and a 2011 peace deal which promised a number of political concessions, including a truth and reconciliation commission, the conflict still persists. Although one rebel group signed the most recent agreement, the more prominent rebel groups, the JEM and the SLM/A, continue to hold out. They argue that peace has not been established because the government has failed to implement the terms of the previous agreements. These groups argue that conflict will continue until the government makes good on previous promises. The preliminary evidence from Sudan suggests that the implementation of agreements might explain the durability of peace after agreements. In the southern Sudan case, the provisions of the accord were eventually implemented, and the peace seems to have held. In western Sudan, the government has made numerous offers of concessions and has failed to follow through with any of them. I anticipate that the government s unwillingness to commit to settlements in Darfur is both responsible for the continuation of violence and the emergence of new violent groups there. A detailed analysis of the conflicts in Sudan would enhance a cross-national quantitative study. 128

138 Data Issues and Solutions In addition to the extension of this project outlined above, I plan to do significant work improving the analyses displayed in this dissertation. A profitable next step would be to collect data from other regions and extend the timeframe of the study backward. Although these data are cross-national, they only record the characteristics of conflicts in one region. Having data from countries outside of Africa would show that the results I garner in this study are not due to some peculiarity of African wars. As the argument I present is general and not specific to African conflicts, I expect a broader dataset to yield the same results. While expanding the dataset would allow for more general results, a more immediate and pressing concern is to find a solution to the missing data issues here. There are many technical approaches to fixing problems of data missingness. The most common method is multiple imputation, which fills in missing values based on Monte Carlo simulations. By treating missing values as extreme cases of measurement error, Blackwell, Honaker and King (2011) create a tool that simultaneously corrects for measurement error and missingness. Multiple overimputation, an extension of multiple imputation, uses a two-step procedure that involves relatively minimal assumptions, allows for random measurement error in any number or combination of variables or cell values in a data set (Blackman et al. 2011:1). The first step corrects for measurement error while the second step allows any statistical model to be estimated as if the data were not missing or incorrectly measured at all. Correcting for measurement with such a model is essential as measurement error can prejudice estimates. Bias affects the relationship between the inaccurately measured covariate and the dependent variable. As such bias is not reduced by simply adding more data, adding 129

139 data for more countries as I propose above would not improve the statistical tests. As random measurement error attenuates the effect of covariates, correcting for this would likely only increase the significance of the variables in the model while not correcting for the measurement error in one covariate could alter the effect of all the others in the model. Conclusion When I embarked on this project I wanted to know whether rebel group demands had an effect on conflict settlement. Specifically, I wanted to determine whether addressing rebel demands with concessions positively impacted conflict resolution in civil wars. As much extant literature casts doubt on the credibility of rebel group demands, conventional wisdom would suggest that acknowledging rebel group demands would not change the likelihood of conflict settlement. On the contrary, I found that addressing rebel group demands does matter. When governments address rebels grievances, conflicts are much more likely to be resolved peacefully. While government concessions to rebel demands really affects the probability of agreements and termination, these findings do not suggest other factors do not matter. In fact, this dissertation demonstrates that rebels strength and ability to inflict costs on an opponent impacts settlement greatly as strong groups imposing great costs are more likely to be met with government cooperation. Moreover, this study discusses the influence of rebel support bases on rebel group behavior. 2 I find that rebel groups that rely on civilian support 2 Other studies also discuss the importance of rebel support bases, Weinstein (2007), for example finds that the type of support base rebels use effects the level of violence they use. 130

140 bases to mobilize resources for conflict are much more likely to settle when they are offered what they demand. These groups tend be more credible and more committed than groups that do not rely on civilian support. Finally, this project highlights the importance of considering rebel group characteristics when trying to understand conflict dynamics. Rebel groups have varying levels of credibility that affect the probability of settlement and the duration of conflict. Examining the notion of rebel credibility, particularly empirically, is a profitable area of study. 131

141 References: Abrahms, Max The Political Effectiveness of Terrorism Revisited. Comparative Political Studies 45(3): Ayres, William Strategies, Capabilities, and Demands: Explaining Outcomes in Violent Intrastate Nationalist Conflicts. International Interactions 27(1): Blackwell, Matthew, James Honaker, and Gary King. Multiple Overimputation: A Unified Approach to Measurement Error and Missing Data. Working Paper, copy at Blimes, Randall The Indirect Effect of Ethnic Heterogeneity on the Likelihood of Civil War Onset. Journal of Conflict Resolution 50(4): Coalition to Stop the Use of Child Soldiers Child Soldiers Global Report 2008 accessed from child-soldiers.org Collier, Paul and Anke Hoeffler On the Duration of Civil War. Journal of Peace Research 41(3): Crenshaw, Martha The Causes of Terrorism. Comparative Politics 13(4): Crenshaw, Martha Theories of Terrorism: Instrumental and Organizational Approaches in Inside Terrorist Organizations, ed. David Rapoport, London: Frank Cass. Crenshaw, Martha The Logic of Terrorism: Terrorist Behavior as a Product of Strategic Choice in Terrorism in Perspective, Sue Mahan and Pamela Griset,eds. Thousand Oaks, CA: Sage Publications Inc. Cunningham, David Veto Players and Civil War Duration. American Journal of Political Science 50(4): Cunningham, David Barriers to Peace in Civil War. New York: Cambridge University Press. 132

142 Cunningham, David, Kristian Skrede Gleditsch, and Idean Salehyan It Takes Two: A Dyadic Analysis of Civil War Duration and Outcome, Journal of Conflict Resolution 53 (4): Cunningham, Kathleen Gallagher Divide and Conquer or Divide and Concede: How Do States Respond to Internally Divided Separatists? American Political Science Review 105(2): DeRouen, Karl and Jacob Bercovitch Enduring Internal Rivalries: A New Framework for the Study of Civil War. Journal of Peace Research 45(1): DeRouen, Karl and David Sobek The Dynamics of Civil War Duration and Outcome. Journal of Peace Research 41(3): Ellingsen, Tanja Colorful Community or Ethnic Witches Brew? Multiethnicity and Domestic Conflict During and After the Cold War. Journal of Conflict Resolution 44(2): Farrell, Joseph and Robert Gibbons Cheap Talk with Two Audiences. The American Economic Review 79(5): Fearon, James D Rationalist Explanations for War. International Organization 49(3): Fearon, James D Why Do Some Civil Wars Last So Much Longer Than Others? Journal of Peace Research 41(3): Fearon, James and David Laitin Ethnicity, Insurgency and Civil War. American Political Science Review 97: Filson, Darren and Suzanne Werner A Bargaining Model of War and Peace: Anticipating the Onset, Duration, and Outcome of War. American Journal of Political Science 46(4):

143 Fortna, Virginia Page Does Peacekeeping Keep Peace?: International Intervention and the Duration of Peace After Civil War. International Studies Quarterly 48(2): Glassmyer, Katherine and Nicholas Sambanis Rebel-Military Integration and Civil War Termination. Journal of Peace Research 45(3): Greenhill, Kelly and Solomon Major The Perils of Profiling: Civil War Spoilers and the Collapse of Intrastate Peace Accords. International Security 31(3):7-40. Guisinger, Alexandra and Alastair Smith Honest Threats: The Interaction of Reputation and Political Institutions in International Crisis. Journal of Conflict Resolution 46: Gupta, Devashree Radical Flank Effects: The Effect of Radical-Moderate Splits in Regional Nationalist Movements. Paper Presented at the Conference of Europeanists, Chicago, March 14-16, Haines, Herbert Black Radicalization and the Funding of Civil Rights: Social Problems 32(1): Harbom, Lotta, Erik Melander and Peter Wallensteen Dyadic Dimensions of Armed Conflict, Journal of Peace Research 45(5): Hartzell, Caroline, and Matthew Hoddie Institutionalizing Peace: Power Sharing and Post-Civil War Conflict Management. American Journal of Political Science 47 (2): Hultman, Lisa Battle Losses and Rebel Violence: Raising Costs for Fighting. Terrorism and Political Violence 19: Hultman, Lisa The Power to Hurt in Civil War: The Strategic Aim of RENAMO Violence. Journal of Southern African Studies 35(4):

144 Humphreys, Macartan and Jeremy Weinstein Who Fights?: The Determinants of Participation in Civil War. American Journal of Political Science 52(2): Jenne, Erin A Bargaining Theory of Minority Demands: Explaining the Dog that Did not Bite in 1990s Yugoslavia. International Studies Quarterly 48(4): Kalyvas, Stathis New and Old Civil Wars: A Valid Distinction? World Politics 54(1): Kalyvas, Stathis The Paradox of Terrorism in Civil War. Journal of Ethics 8(1): Kalyvas, Stathis and Matthew Kocher How Free Is Free Riding in Civil Wars? World Politics 59(2): Kreutz, Joakim How and When Armed Conflicts End: Introducing the UCDP Conflict Termination Dataset. Journal of Peace Research 47(2): Kydd, Andrew and Barbara Walter The Strategies of Terrorism. International Security 31(1): Lujala, Päivi; Halvard Buhaug & Scott Gates, Geography, Rebel Capability, and the Duration of Civil Conflict, Journal of Conflict Resolution 53(4): Lyall, Jason Does Indiscriminate Violence Incite Insurgent Attacks? Journal of Conflict Resolution 53(3): Marshall, Monty and Keith Jaggers Polity IV Project: Political Regime Characteristics and Transitions, Mattes, Michaela and Burcu Savun Fostering Peace After Civil War: Commitment Problems and Agreement Design. International Studies Quarterly 53: McAdam, Doug Political Process and the Development of Black Insurgency, 135

145 Chicago: The University of Chicago Press. McAdam, Doug The Decline of the Civil Rights Movement in Waves of Protest: Social Movements Since the Sixties, eds. Jo Freeman &Victoria Johnson, Oxford, UK: Rowman & Littlefield Publishers, Inc. Morrow, James D A Continuous- Outcome Expected Utility Theory of War. Journal of Conflict Resolution 29(3): Mukherjee, Bumba Why Political Power-Sharing Agreements Lead to Enduring Peaceful Resolution of Some Civil Wars, But Not Others? International Studies Quarterly 50: National Consortium for the Study of Terrorism and Responses to Terrorism (START) Global Terrorism Database [Data file]. Retrieved from Neumann, Peter Negotiating With Terrorists. Foreign Affairs 86(1): Powell, Robert Bargaining and Learning While Fighting. American Journal of Political Science 48(2): Ramsay, Kristopher Politics at the Water s Edge: Crisis Bargaining and Electoral Competition. Journal of Conflict Resolution 48(4): Regan, Patrick Third-Party Interventions and the Duration of Intrastate Conflicts. Journal of Conflict Resolution 46(1): Reiter, Dan Exploring the Bargaining Model of War. Perspectives on Politics, 1(1): Ross, Michael How Do Natural Resources Influence Civil War?: Evidence from Thirteen Cases. International Organization 58(1): Rothchild, Donald and Caroline Hartzell. The Peace Process in the Sudan, in Stopping the Killing: How Civil Wars End, Roy Licklidder ed. New York: New York University Press:

146 268. Sambanis, Nicholas Partition as a Solution to Ethnic War: An Empirical Critique of the Theoretical Literature. World Politics 52(4): Sambanis, Nicholas Do Ethnic and Nonethnic Civil Wars Have the Same Causes? Journal of Conflict Resolution 45(3): Sambanis, Nicholas Terrorism and Civil War In Terrorism, Economic Development, and Political Openness, Philip Keefer and Norman Loayza eds. New York: Cambridge University Press Sartori, Anna An Estimator for Some Binary Outcome Selection Models Without Exclusion Restrictions Political Analysis 11(2): Schelling, Thomas An Essay on Bargaining. The American Economic Review 46(3): Shultz, Kenneth Domestic Opposition and Signaling in International Crisis. American Political Science Review 94: Shultz, Kenneth Democracy and Coercive Diplomacy. New York: Cambridge University Press. Smith, Alistair and Allan Stam. Bargaining and the Nature of War. Journal of Conflict Resolution 48(6): Siqueira, Kevin Political and Militant Wings Within Dissident Movements and Organizations. Journal of Conflict Resolution 49(2): Slantchev, Branislav The Principle of Convergence in Wartime Negotiations. American Political Science Review 97(4):

147 Slantchev, Branislav The Power to Hurt: Costly Conflict with Completely Informed States. American Political Science Review 97(1): Smith, Alastair International Crises, and domestic Politics. American Political Science Review 92: Stedman, Stephen Spoiler Problems in Peace Processes. International Security 22(2):5-53. Toft, Monica Duffy Issue Indivisibility and Time Horizons as Rationalist Explanations for War. Security Studies 15 (1): Toros, Harmonie We Don t Negotiate With Terrorists: Legitimacy and Complexity in Terrorist Conflicts. Security Dialogue 38(4): Uppsala Conflict Data Program (Date of retrieval: 2012/03) UCDP Conflict Encyclopedia: Uppsala University. UCDP Battle-Related Deaths Dataset v , Uppsala Conflict Data Program, Uppsala University Valentino, Benjamin, Paul Huth and Sarah Croco Bear Any Burden? How Democracies Minimize the Costs of War. Journal of Politics 72(2): Wagner, Harrison The Causes of Peace in Stopping the Killing: How Civil Wars End, Roy Licklidder ed. New York: New York University Press: Wagner, Harrison Bargaining and War. American Journal of Political Science 44(3): Walter, Barbara F The Critical Barrier to Civil War Settlement. International Organization 51 (3): Walter, Barbara F Designing Transitions from Civil War: Demobilization, Democratization, and Commitments to Peace. International Security 24 (1):

148 Walter, Barbara F Does Conflict Beget Conflict? Explaining Recurring Civil War. Journal of Peace Research 41(3): Walter, Barbara F Information, Uncertainty and the Decision to Secede. International Organization 60(1): Walter, Barbara F Bargaining Failures and Civil War. Annual Review of Political Science 12: Watt, Nigel Burundi: Biography of a Small Country. London: Hurst Publishers. Weinstein, Jeremy Resources and the Information Problem in Rebel Recruitment. Journal of Conflict Resolution 49(4): Werner, Suzanne The Precarious Nature of Peace: Resolving the Issues, Enforcing the Settlement, and Renegotiating the Terms. American Journal of Political Science 43 (July): Werner, Suzanne and Amy Yuen Making and Keeping Peace. International Organization 59(2): Wilkinson, Paul Terrorism Versus Democracy: The Liberal State Response. London: Routledge. Wucherpfennig, J., N. Metternich, L.-E. Cederman, and K. S. Gleditsch (2012). Ethnicity, the state and the duration of civil war. World Politics 64 (1), pp Zartman, William I. The Unfinished Agenda in Stopping the Killing: How Civil Wars End, Roy Licklidder ed. New York: New York University Press. 139

149 Appendix A: Figures and Tables for Chapter 3 Table 3.1: List of Demands Demand Number Demand Type Demand Detail Political Demand? 1 Independence/ Secession/ Irredentism If rebels demand independence, secession or irredentism, their demands were coded in this category. Demands were coded in this category when rebels claim sovereignty over a particular territory, or when they demand to be reconnected to a nation with which they were previously a part. 2 Self-determination A demand was coded in this category if a rebel group wants independence or some regional autonomy, but wants the outcome to be decided by the inhabitants of the territory through vote or referendum. While this is related to the previous category of independence/secession/ irredentism, it is coded separately since the main demand is for the right for the people on the territory to decide their own future. 3 Political Autonomy 4 Change/ Restructuring in Government The group demands some form of autonomy to control the political landscape of their territory. The group may be explicit about the type of institutional arrangement that would grant them autonomy (ex. confederation, federalism) or may just demand political or regional autonomy. Categorizing this type of demand is sometimes complicated because of its relationship with other types of demands. For example, when the desired institutional arrangement for autonomy is different than the current institutional set-up in the state the demand could have also been coded as a change in government. The variable is only coded as a demand regarding autonomy, however, when the purpose of the change in government is to allow a territory or a group more control over its governance. Power-sharing arrangements are also coded separately, although most forms of autonomy can be considered a form of power-sharing. Change in government is a far-reaching category. It is coded when rebels demand that some aspect of the government change. Rebels may demand that the structure of the government changes from a presidential to a parliamentary system or from one-party rule to multiparty rule, or may simply demand democratization (without explicitly asking for elections). Rebels may alternatively ask for a new government or for the resignation of the leader. 5 Power-sharing Demands were coded in this category when rebels demanded more control in the constitution of the government. In most cases, rebels demanded for their own group to be integrated into the government, but in other cases simply demanded a more inclusive government. Specifically, rebels may call for ethnic equilibrium in government or a better distribution of governments. This category sometime overlaps with the change in government category as calls for a transitional government or a government of national unity could have been coded as either. I code these calls as demands for power-sharing, when the rebel group explicitly requests they take part in a new transition government. yes yes yes yes yes 140

150 6 Elections Demands in this category are related to the execution of elections. Most of these demands are explicit calls for elections. Sometimes rebels demand the postponement of elections or more detail on scheduled elections. 7 Change in Election Procedure 8 Constitution Change/Reform 9 Political Freedoms/Civil Liberties 10 Recognition as a Political Party This category was coded when rebels requested a change in the structure of current or impending elections. This category includes demands for recounting votes, changing rules and review of election outcomes. If rebels demand changes in the current constitution, the demands are coded in this category. This category includes calls for specific amendments to a state s current constitution, general constitutional reform, reversion to a previous constitution, or an entirely new constitution. If rebels demanded greater political rights, including the right for opposition parties to participate, media freedoms, the lifting of a state of emergency that may curtail basic rights, the demands were coded in this category. This category is distinct from the change in government category because the rebels are not explicitly asking to change the structure or participants in government, but they are asking for greater freedoms while the current government persists. Demands were coded in this category when rebels explicitly demanded that their group be allowed to participate in political life as a legitimate political party. 11 Religious Rights Demands were coded in this category when rebels made explicit claims related to the right to freely practice, or be governed by the laws of their religion. The variable can almost completely be bifurcated into two categories, demands for religious secularization and demands for religious fundamentalism. Groups may demand that the state instill and adhere to religious laws, or that the state be completely separated from any type of religion. 12 Cultural Autonomy 13 Economic Development/ Reform 14 Resource Rent Distribution Cultural autonomy was coded in cases where rebels wanted their culture to be respected in the country where they live. Demands for language rights, the respect and recognition for various identities in a state, and the right to develop an ethnic identity were coded in this category. Demands were recorded as pertaining to economic development when rebels demand that the government develop a particular region. Building or rebuilding of the infrastructure, better services, economic recovery and economic rights in an area are all examples of demands that would fit in this category. Demands for a more advanced economic system in the state were also included here. Demands coded in the resource rent distribution category are those explicitly calling for better allocation of a country or region s economic resources. This includes demands for the fair distribution of revenues from the country s natural resources like hydrocarbon, minerals or demands for a just division of wealth. yes yes yes yes yes yes yes yes yes 141

151 15 Military Restructuring/ Integration Demands were coded in this category if they concerned changes in the nation s military. This category includes changes in the military leadership as well as changes in the constitution of the military institution. Calls for rebels to be integrated into the army are included here. 16 Cease-Fire Explicit calls for the cessation of hostilities were coded in this category. 17 Military Withdrawal/ Pull-out 18 End of Foreign Occupation/ Intervention/ Influence Demands in this category all concern the military withdrawal, cantonment or confinement of government troops (Side-A) or troops supported by the government, including police and some paramilitary groups. When rebels demanded the removal of foreign forces, including peacekeepers, foreign interveners or foreign diplomatic influence the demands were coded here. This category also includes calls for the closure of military bases or embassies, the removal of diplomats from their posts and the abrogation of treaties with foreign states. 19 Security Related Demands were coded as having to do with security when they concern the security of the intra- and post-war environment for either combatants or civilians in the conflict zone. Specifically, these are demands that might influence belligerents ability or willingness to adhere to settlements. This variable may include demands for the freedom of movement for combatants, and the disbandment of other hostile forces in the country. 20 End to Civilian Targeted Violence 21 Justice for Atrocities 22 Release of Political Prisoners/ Militants Demands were coded in this category when rebels demanded explicitly that hostility towards civilians stop. These demands are similar to the demands in the security-related category, but the insecurity must be targeted at civilians in order to be included here. In other words, violence affecting civilians must not simply be a side-effect of war, but must be intentionally directed at noncombatants. Demands were coded in this category when rebels demand that perpetrators of war crimes are held accountable for their actions. This includes rebel calls for the firing of public officials, trials of military personnel and political leaders, calls for truth and reconciliation commissions, or creation and appointment of investigative commissions. Demands were included in this grouping when rebels demanded the government release political prisoners, militants, or rebel supporters or leaders. 23 Amnesty This category represents demands for amnesty and pardon for political prisoners or exclusively for rebels. yes yes yes yes yes yes yes yes no 24 Military Intervention 25 Humanitarian Assistance Calls for peacekeeping or military interventions were coded here. Calls for humanitarian assistance were coded in this category. Specifically, when rebels demanded the free movement of aid workers, improvement in the humanitarian situation, or the return of humanitarian aid, the demands were coded here. no no 142

152 26 International Ceasefire Observers/ Election Monitors 27 Change in Structure of Peace Process 28 Personalistic Benefits 29 Food/Supplies/ Medicine Demands were coded in this category when rebels demanded that independent observers monitor a specific process related to the return to peace. If rebels requested international supervision of a ceasefire or elections, the demand was coded here. This category is coded when rebels desire changes in the structure of the peace process. Specifically, this category corresponds to demands for changes in mediators or guarantors, the venue for negotiations, and the agenda of peace talks. Demands for negotiations were also included in this category. Demands were coded as requesting personalistic benefits or sidepayments when concessions do not provide any benefit to anyone except the person receiving them. While conceptually distinct, practically, this category overlaps with some of the other categories, including the Other category and the food supplies and medicines category. Demands were coded in this category when rebels demanded food, supplies, or medicine. This category could be considered a subset of the personalistic benefit category or can be considered separately. This category is made distinct because the calls for food, supplies and medicine may benefit a large range of people and may allow rebels to advance their cause. The demands in the personalistic benefits category are not intended to advance any political aim, but are designed to provide side-payments to individuals on the rebel side. 30 Ransom Demands were coded in this category when rebels held someone hostage in exchange for concessions. This category could include money or may be more political in nature. These demands may fit into any of the other categories, but were coded here because they have a different potential for being responded to than other types of demands. 31 Coup This category was coded when the main aim of the rebel group was to overthrow the government. This is infrequently verbalized by the rebel group before their aim is achieved. This category was coded to capture the aims of groups that had very short-term goals and, thus, may not have publically announced any demands. 32 Other The other category encompasses all demands that did not neatly fit in any of the other categories. no yes no no yes yes yes/no 143

153 Table 3.2: Coding Rubric Value Description of Values on Concessions Variable 0 Rejected/no response 1 Minimal concession 2 Moderate concession 3 Substantial concession 4 Maximal concession 144

154 Figure 3.1: Frequency of Rebel Demands (Repeated Demands) 145

155 Figure 3.2: Frequency of Rebel Demands by Group 146

156 Figure 3.3: Moderate Government Concessions by Demand Type 147

157 Figure 3.4: Substantial Government Concessions by Demand Type 148

158 Figure 3.5: Maximal Government Concessions by Demand Type 149

159 Appendix B: Statistical Tables from Chapter 4 Table 4.1: Logit Regressions of the Effect of Group Characteristics on Civil War Negotiations Variable Model 1 Model 2 Model 3 Model4 Model 5 Model 6 Model 7 T t t t t t+1 t+2 t+3 # Terrorist Attacks 0.085*** 0.105*** 0.097*** 0.093*** 0.115*** 0.114*** 0.111*** (0.027) (0.034) (0.036) (0.036) (0.039) (0.039) (0.038) Child Recruitment 0.324*** 0.340** (0.116) (0.140) (0.187) (0.232) (0.228) (0.226) (0.224) Gems 0.421*** 0.422** 0.398** 0.366* 0.336* (0.168) (0.195) (0.193) (0.197) (0.198) Drugs ** 0.591* 0.522* (0.303) (0.350) (0.345) (0.348) (0.351) Oil 0.582*** * 0.278* 0.298* (0.152) (0.185) (0.181) (0.182) (0.182) Relative Rebel Strength 0.749*** 0.770*** 0.500*** 0.500*** 0.598*** 0.633*** (0.085) (0.096) (0.112) (0.110) (0.111) (0.112) Explicit Support 0.613*** 0.350*** ** (0.128) (0.144) (0.177) (0.173) (0.175) (0.175) Main Group 0.238* ** 0.565*** 0.467** 0.386* (0.137) (0.162) (0.180) (0.229) (0.225) (0.225) (0.223) Ethnic Claim -1.10** * * * (0.455) (0.417) (0.426) (0.436) ln(time) 0.266*** 0.291*** 0.307*** 0.266*** (0.084) (0.084) (0.086) (0.086) ln(deaths) *** ** (0.065) (0.063) (0.063) (0.062) Polity 0.106*** 0.119*** 0.108*** 0.101*** (0.026) (0.026) (0.024) (0.027) Number of Demands 0.425*** 0.304*** 0.227*** 0.174*** (0.036) (0.032) (0.030) (0.030) Number of Conflict Episodes ** *** ** * (0.0529) (0.161) (0.160) (0.163) Constant (0.122) (0.227) (0.279) (0.529) (0.533) (0.551) (0.557) N Logit models presented with robust standard errors clustered on conflict; Statistical significance denoted by ***=p.01, **= p.05, * = p

160 Table 4.2: Negative Binomial Regressions of the Effect of Group Characteristics on the Number of Government Concessions Variable Model 1 Model 2 Model 3 Model4 Model 5 Model6 Model 7 Model 8 T t t t t t+1 t+2 t+3 t # Terrorist Attacks 0.082*** 0.045*** 0.044*** 0.049*** 0.056*** *** 0.086** (0.027) (0.010) (0.012) (0.015) (0.016) (0.071) (0.012) (0.039) Child Recruitment 0.436* (0.263) (0.277) (0.294) (0.334) (0.362) (0.402) (0.765) Gems 0.437* (0.247) (0.245) (0.371) (0.384) (0.432) (0.807) Drugs -1.74*** -1.39*** -1.51*** -1.95*** -1.97*** 1.26 (0.417) (0.360) (0.483) (0.426) (0.447) (1.07) Oil * -1.10** (0.236) (0.374) (0.439) (0.452) (0.454) (0.567) Relative Rebel Strength * 0.796*** 1.00*** (0.264) (0.274) (0.315) (0.291) (0.253) (0.341) Explicit Support * 0.773* (.322) (0.347) (0.322) (0.369) (0.475) (0.448) Main Group * 0.487* 0.777** 0.664* *** (0.225) (0.223) (0.290) (0.362) (0.316) (0.371) (0.667) Ethnic Claim 0.720** *** 1.10** -1.24* (0.355) (0.403) (0.370) (0.494) (0.702) Number of Demands 0.537*** 0.494*** 0.513*** 0.330*** 0.236*** 0.191*** 0.832*** (0.054) (0.071) (0.073) (0.056) (0.063) (0.046) (0.315) ln(time) ** (0.122) (0.127) (0.132) (0.160) (0.153) ln(deaths) (0.122) (0.098) (0.092) (0.106) (0.160) Polity (0.064) (0.091) (0.088) (0.083) (0.086) Number of Conflict Episodes ** *** -1.02*** ** -1.24* (0.169) (0.345) (0.356) (0.436) (0.733) Constant (0.259) (0.215) (0.609) (0.775) (1.18) (1.19) (1.52) (1.57) α N Logit models presented with robust standard errors clustered on conflict; Statistical significance denoted by ***=p.01, **= p.05, * = p.10. This dependent variable examines only the number of political concessions. All non-political concessions excluded. 151

161 Figure 4.1: Predicted Probabilities of Negotiations by Terrorism, Rebel Strength and Time 152

162 Table 4.3: Rebel Groups Coded as Using Terrorism in the Global Terrorism Database (GTD) State Name Niger Ethiopia Uganda Algeria Somalia Sierra Leone Algeria Nigeria Congo (Kinshasa) Chad Niger Ethiopia Angola Djibouti Uganda Mali Somalia Algeria Sudan Sierra Leone Liberia Uganda Senegal Niger Mozambique Congo (Kinshasa) Burundi Sudan Burundi Sudan Angola Congo (Brazzaville) Ethiopia Ethiopia Burundi Sierra Leone Rwanda Rebel Group Name (GTD) Air and Azawak Liberation Front (FLAA) Al-Ittihaad al-islami (AIAI) Allied Democratic Forces (ADF) Al-Qa`ida Islamic Maghreb (AQIM)/ Salafist Group for Preaching and Fighting (GSPC) Al-Shabaab Armed Forces Revolutionary Council (AFRC) Armed Islamic Group (GIA) Boko Haram Bunda Dia Kongo (BDK) Committee for a National Drive for Peace and Democracy (CSNDP) Democratic Front for Renewal (FDR) Eritrean People s Liberation Front (EPLF) Front for the Liberation of the Enclave of Cabinda (FLEC) Front for the Restoration of Unity and Democracy (FRUD) Holy Spirit Movement Islamic Arab Front of Azawad (FIAA) Islamic Courts Union (ICU) Islamic Salvation Front (FIS) Justice and Equality Movement (JEM) Kamajor Hunters Liberians United for Reconciliation and Democracy (LURD) Lord's Resistance Army (LRA) Movement of Democratic Forces of Casamance (MFDC) Movement of Niger People for Justice (MNJ) Mozambique National Resistance Movement (MNR) National Congress for the Defense of the People (CNDP) National Council for Defense of Democracy (CNDD) National Democratic Alliance of Sudan (NDA) National Liberation Front (FNL) (Burundi) National Redemption Front (NRF) National Union for the Total Independence of Angola (UNITA) Ninjas Ogaden National Liberation Front (ONLF) Oromo Liberation Front (OLF) Party for the Liberation of the Hutu People (PALIPEHUTU-FNL) Revolutionary United Front (RUF) Rwanda Patriotic Front (RPF) 153

163 Niger Somalia Sudan Angola Uganda Chad Chad Uganda Saharan Revolutionary Armed Front (FARS) Somali Patriotic Movement (SPM) Sudan People's Liberation Army (SPLM/A) The Front for the Liberation of the Cabinda Enclave Renewed (FLEC-R) Uganda People's Army (UPA) Union of Forces for Democracy and Development (UFDD) United Front for Democratic Change (FUCD) West Nile Bank Front (WNBF) 154

164 Appendix C: Statistical Tables for Chapter 5 Table 5.1: Probit Regressions of the Effect of Government Concessions on the Probability of Agreements (Parsimonious Models) Variable Model 1 Model 2 Model 3 Model4 Model 5 Model 6 Model 7 Model 8 % Concessions (Strong) 0.017*** (0.002) # Concessions (Strong) 0.421*** (0.121) %Concessions (Weak) 0.015*** (0.002) #Concessions (Weak) 0.396*** (0.111) % Pol Concessions (Strong) 0.013*** (0.003) #Pol Concessions (Strong) 0.019* (0.012) %Pol Concessions (Weak) 0.013*** (0.003) #Pol Concessions (Weak) 0.022* (0.013) Additional Concessions 0.795*** 0.823*** 0.775*** 0.794*** 0.842*** 0.952*** 0.824*** 0.950*** (0.226) (0.220) (0.219) (0.226) (0.219) (0.200) (0.217) (0.200) Number of Demands 0.138*** 0.103*** 0.129*** 0.078*** 0.140*** 0.157*** 0.132*** 0.155*** (0.022) (0.027) (0.023) (0.088) (0.022) (0.025) (0.022) (0.025) Constant (0.089) (0.088) (0.090) (0.088) (0.091) (0.087) (0.093) (0.087) N Probit models presented with robust standard errors clustered on conflict; Statistical significance denoted by ***=p.01, **= p.05, * = p

165 Table 5.2: Probit Regressions of the Effect of Government Concessions on the Probability of Agreements (with Support Bases and Many Controls) Variable Model 1 Model 2 Model 3 Model4 Model 5 Model 6 Model 7 Model 8 % Concessions (Strong) 0.014*** (0.002) # Concessions (Strong) 0.464*** (0.123) %Concessions (Weak) 0.013*** (0.002) #Concessions (Weak) 0.441*** (0.122) % Pol Concessions (Strong) 0.012*** (0.002) #Pol Concessions (Strong) 0.019* (0.012) %Pol Concessions (Weak) 0.013*** (0.003) #Pol Concessions (Weak) 0.016* (0.009) Ethnic Claim (0.189) (0.188) (0.189) (0.200) (0.178) (0.180) (0.172) (0.183) Ethnic Claim*Concessions 0.035*** 0.168*** 0.024*** *** 0.268*** 0.015*** 0.242*** (0.011) (0.137) (0.007) (0.124) (0.004) (0.087) (0.005) (0.077) Drugs (0.291) (0.284) (0.275) (0.281) (0.263) (0.291) (0.264) (0.291) Gems 0.432*** 0.419*** 0.436*** 0.420*** 0.444*** 0.440*** 0.441*** 0.439*** (0.149) (0.146) (0.150) (0.149) (0.145) (0.147) (0.149) (0.148) Oil (0.181) (0.167) (0.184) (0.180) (0.178) (0.189) (0.179) (0.190) Child Recruitment (0.197) (0.201) (0.207) (0.226) (0.194) (0.194) (0.194) (0.194) Additional Concessions 0.634** 0.662** 0.604** 0.648** 0.650** 0.795*** 0.638** 0.801*** (0.315) (0.309) (0.309) (0.315) (0.309) (0.273) (0.310) (0.273) Number of Demands 0.130*** 0.097*** 0.122*** 0.070** 0.131*** 0.146*** 0.124*** 0.144*** (0.030) (0.029) (0.031) (0.029) (0.030) (0.033) (0.029) (0.034) Constant (0.172) (0.161) (0.174) (0.165) (0.169) (0.177) (0.170) (0.177) N Probit models presented with robust standard errors clustered on conflict; Statistical significance denoted by ***=p.01, **= p.05, * = p

166 Table 5.3: Probit and Bivariate Probit Regressions of the Effect of Government Concessions on the Probability of Agreements Variable Model 1 Model 2 Model 3 Model 4 Model 5 T t t t+1 t Neg. Agreement % Concessions (Strong) 0.006* 0.011** 0.005* 0.018*** 0.012*** (0.0003) (0.003) (0.003) (0.004) (0.003) % Pol Concessions (Strong) 0.010*** (0.003) Ethnic Claim * *** ** (0.282) (0.244) (0.207) (0.211) (0.179) (0.244) Ethnic Claim*Concessions 0.068*** 0.072*** ** 0.025*** 0.589*** 0.076*** (0.019) (0.018) (0.005) (0.009) (0.100) (0.018) Drugs *** (0.270) (0.259) (0.256) (0.229) (0.127) (0.247) Gems 0.338* 0.327** 0.396*** 0.352** 0.280*** 0.359** (0.194) (0.149) (0.147) (0.144) (0.089) (0.165) Oil (0.209) (0.178) (0.156) (0.180) (0.126) (0.185) Child Recruitment (0.226) (0.208) (0.187) (2.01) (0.108) (0.207) Negotiations 1.233*** 0.477*** (0.029) (0.172) Negotiations (t-1) 0.296** 0.330*** 0.091*** 0.300** (0.140) (0.131) (0.127) (0.149) Additional Concessions 0.767** 0.795** 0.365* 0.777** ** (0.361) (0.353) (0.172) (0.357) (0.240) (0.018) Number of Demands *** 0.069*** 0.106*** 0.218*** 0.102*** (0.036) (0.032) (0.024) (0.033) (0.027) (0.034) Main Group ** (0.121) (0.154) (0.170) (0.153) (0.179) (0.151) Relative Rebel Strength 0.338** 0.378*** 0.357** 0.386*** 0.247*** 0.403*** (0.155) (0.153) (0.151) (0.155) (0.089) (0.155) Explicit Support 0.282*** 0.234*** 0.303*** 0.289*** *** (0.107) (0.105) (0.101) (0.104) (0.148) (0.113) # UN Military Troops *** *** 4.67e ** -8.98e ( ) ( ) (.00001) ( ) (9.59e-06) ( )*** 157

167 # UN Observers ** * (0.0005) (0.0005) (0.0006) (0.0006) (0.0004) (0.0004) #UN Police 0.003*** 0.003*** ** ** (0.001) (0.001) (0.0009) (0.001) (0.0008) (0.001) #Major Power -1.18*** *** * 0.547*** *** (0.318) (0.243) (0.292) (0.320) (0.159) (0.244) Polity 0.044* 0.052** 0.058*** 0.055** *** 0.051** (0.026) (0.025) (0.020) (0.025) (0.015) (0.025) # Terrorist Attacks *** (0.045) (0.037) (0.017) (0.039) (0.019) (0.034) Government Violence 3.34e-06*** 2.30e-06*** 2.60e e-06*** -2.20e-06*** 2.25e-06*** (8.79e-07) (7.82e-07) (7.62e-07) (8.60e-07) (4.26e-07) (7.49e-07) Number Rebel Groups 0.238*** 0.234*** 0.138*** 0.213*** 0.160* 0.257*** (0.079) (0.068) (0.045) (0.066) (0.091) (0.060) ln(time) *** (0.084) (0.087) (0.079) (0.084) (0.057) (0.085) ln(deaths) ** ** ** ** * (0.035) (0.037) (0.041) (0.037) (0.042) (0.037) Number of Conflict Episodes * (0.075) (0.073) (0.056) (0.068) (0.051) (0.063) Constant (0.357) (0.439) (0.379) (0.406) (0.435) (0.424) N ρ=0.639 χ 2 =28.17*** Probit & Bivariate Probit models presented with robust standard errors clustered on conflict Statistical significance denoted by ***=p.01, **= p.05, * = p

168 Figure 5.1 Predicted Probability of Agreement by Ethnic Claim and Percentage of Concessions 159

169 Appendix D: Statistical Tables for Chapter 6 Table 6.1: Probit Regressions of the Effect of Government Concessions on Conflict Termination Variable Model 1 Model 2 Model 3 Model 4 % Concessions (Strong) t *** 0.010*** (0.0003) (0.004) % Concessions (Strong) t * (0.003) (0.006) Lagged Ethnic Recruitment (0.189) (0.295) (0.166) (0.169) Lagged Ethnic Recruitment*Concessions *** ** (0.005) (0.003) (0.004) (0.008) Drugs (0.512) (0.542) Gems (0.162) (0.149) Oil (0.160) (0.163) Child Recruitment *** *** (0.260) (0.241) Negotiations 0.432** 0.483*** (0.188) (0.183) Additional Concessions (0.261) (0.183) Number of Political Demands (0.050) (0.047) Relative Rebel Strength (0.093) (0.086) Explicit Support (0.200) (0.200) # UN Military Troops -1.35e e-06 (.00003) ( ) # UN Observers (0.0005) (0.0006) #UN Police 0.002* 0.002* (0.001) (0.001) #Major Power * * (0.376) (0.411) Polity 0.048** 0.058*** (0.021) (0.021) # Terrorist Attacks *** ** (0.158) (0.165) Government Violence 2.49e-06*** 3.04e-06*** (6.72e-07) (6.53e-07) Number Rebel Groups *** (0.092) (0.097) ln(time) (0.062) (0.067) ln(deaths) * * (0.041) (0.043) Number of Conflict Episodes (0.076) (0.079) Constant (0.072) (0.073) (0.478) (0.500) N Probit models presented with robust standard errors clustered on conflict Statistical significance denoted by ***=p.01, **= p.05, * = p

170 Table 6.2: Probit Regressions of the Effect of Government Concessions on Termination (Lagged Dependent Variable) Variable Model 1 Termination w/in 3 months % Concessions (Strong) t 0.009*** (0.003) % Concessions (Strong) t *** (0.02) % Concessions (Strong) t (0.003) Lagged Ethnic Recruitment t -3.86*** (0.239) Lagged Ethnic Recruitment t *** (0.398) Lagged Ethnic Recruitment t *** (0.214) Lagged Ethnic Recruitment*Concessions t (0.003) Lagged Ethnic Recruitment*Concessions t * (0.003) Lagged Ethnic Recruitment*Concessions t ** (0.004) Constant (0.09) N 3148 Probit model presented with robust standard errors clustered on conflict Statistical significance denoted by ***=p.01, **= p.05, * = p

171 Table 6.3: Bivariate Probit Regressions of the Effect of Government Concessions on Negotiations, Agreements & Termination Model 1 Model 2 Variable Negotiations Termination Agreement Termination % Concessions (Strong) t *** 0.006** 0.012*** 0.006** (0.003) (0.003) (0.004) (0.003) Lagged Ethnic Recruitment t (0.276) (0.306) (0.227) (0.312) Lagged Ethnic Recruitment*Concessions t *** 0.015*** *** 0.015*** (0.003) (0.003) (0.004) (0.003) Constant (0.135) (0.073) (0.104) N ** 11.68*** Bivariate Probit models presented with robust standard errors clustered on conflict Statistical significance denoted by ***=p.01, **= p.05, * = p

172 Table 6.4: Cox Regressions of the Effect of Government Concessions on Time until Termination Model Model 1 Model 2 Model 3 Termination w/in Termination w/in Termination w/in T One Month (t) Two Months (t+2) Three Months(t+3) % Concessions (Strong) 0.015*** ** (0.005) (0.465) (0.007) Ethnic Recruitment 0.588** 0.016* (0.289) (0.01) (0.39) Ethnic Recruitment*Concessions *** 0.015*** 0.018** (0.010) (0.006) (0.009) N Coefficients from Cox Proportional Hazards models presented with robust standard errors clustered on conflict Statistical significance denoted by ***=p.01, **= p.05, * = p

173 Figure 6.1 Predicted Probability of Episode Termination by Ethnic Claim and Percentage of Concessions 164