THE LEADER’S TOOLKIT: HOW LEADERS CHOOSE THEIR TOOLS OF REPRESSION By Lora DiBlasi A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Political Science—Doctor of Philosophy 2020 THE LEADER’S TOOLKIT: HOW LEADERS CHOOSE THEIR TOOLS OF REPRESSION ABSTRACT By Lora DiBlasi This dissertation analyzes how states use pro-government militias (PGMs) to engage in state- sponsored repression. I aim to better understand the link between PGMs, their leaders, and the civilians they repress. Although most research on state repression assumes that state sanctioned violence will be carried out by the state military, I show that leaders actually tend to dispatch groups with looser ties to the regime, as it provides them greater plausible deniability for resultant abuses. I am interested in understanding why leaders deploy specific types of security agents to engage in repression, as well as how the characteristics of militias influence the forms, targets, and levels of violence in a state. Lastly, I explore how leaders utilize PGMs to prolong their tenure in office and maintain their political power. In order to perform these analyses, I use a novel dataset on PGM characteristics and repressive actions with a global sample from 1989-2007. My dissertation finds evidence that leaders often employ PGMs strategically to maximize their effectiveness and repressive capacities. They also confirm that state-sponsored violence is not always carried out only by official government forces, and leaders can make strategic decisions to maximize their repressive capabilities. In this work, which was recently revised and resubmitted to International Studies Quarterly, I explore the effectiveness of human rights records and naming and shaming, especially by NGOs, to explain the creation of violence security apparatuses that are often used to carry out acts of repression. I demonstrate that leaders are more likely to create and align with PGMs after instances of naming and shaming by the international community. They do so in order to outsource their repression to groups with loose ties to the regime, which they expect to help them to avoid future condemnation from the international community. This raises important questions about how leaders respond to punishment. Copyright by LORA DIBLASI 2020 For Mom and Dad, who made all of this possible. v ACKNOWLEDGEMENTS This dissertation would not have been possible without the help and support of many people. I would like to express my deep gratitude to my committee, Jakana Thomas, Michael Colaresi, Benjamin Appel, and Erica Frantz. They gave me continual and invaluable feedback on the many stages of this project. Their challenging questions and comments pushed me to continually improve and I am forever grateful for their help and advice. My cohort colleagues are some of the brightest people in this discipline. I am grateful to have gone through this experience with them. Throughout my time in graduate school, I was lucky enough to have the continual support and friendship of the brilliant women in Women Hustling in Political Science (WHiPS). Elizabeth Lane, Emma Slonina, Erika Rosebrook, Jamil Scott, and Jessica Schoenherr were my writing partners, cheerleaders, sounding boards, and the most caring group of friends I could ask for. Many others at MSU, notably Caleb Lucas, Lizzie Brannon, Marty Jordan, and Nate Smith made me a better scholar and for that I am incredibly grateful. Also, thank you to the many faculty who opened their doors and offered mentorship, including Christian Houle, Cristina Bodea, Eric Juenke, and Ryan Black. I would also like to thank the many scholars and colleagues who contributed to my knowledge and experiences throughout the years. I am grateful for Sara Mitchell and Elizabeth Menninga encouraging and mentoring me throughout my graduate career. Meeting my Journeys cohort members, especially Jen Spindel, Chelsea Estancona, and Marina Duque, was one of the highlights of my academic career and I am grateful for their friendship. To borrow a sentiment from Abraham Lincoln (and Mike Colaresi), if this dissertation has anything to contribute to the world, it is because I had a family that believed in me and I did not have the heart to let them down. My entire family supported me throughout my PhD, from vi aunts and uncles to cousins and grandparents. While my grandma, Theresa DiBlasi, did not get to see me finish graduate school, I hope this milestone would make her proud. She was a constant source of love and support throughout my life. Her faith in me never wavered and is what inspired me to pursue graduate school. I am extremely grateful for my Aunt Irene, who, without fail, calls me every week to ask if I am a doctor yet. I look forward to finally telling her she can call me “Dr. Lora” now. This long journey would not have been possible without the frequent political discussions, banter, and jokes from my brother Ben. His genuine passion for politics and law has continually inspired me to finish my degree in hopes to keep up with him. I want to end by thanking the people that contributed the most to this project: my parents, Jill Lynch and Frank DiBlasi. My parents instilled an interest in politics in me early in life and wholeheartedly supported my decision to spend a dozen years at Michigan State following my dreams. Their support, love, and encouragement has been unparalleled. They always encouraged my pursuit of education and cheered me on every step of the way (while telling everyone they know about my work). From picking up every time I called, to helping me move multiple times, to traveling across the world with me, they have always been my number one support system. I am eternally grateful for them and their faith in me to finish this dissertation. For all they have done for me, I dedicate this dissertation to them. vii TABLE OF CONTENTS LIST OF TABLES ...........................................................................................................................x LIST OF FIGURES ....................................................................................................................... xi INTRODUCTION ...........................................................................................................................1 BIBLIOGRAPHY ......................................................................................................................5 CHAPTER 1: FROM SHAME TO NEW NAME: HOW NAMING AND SHAMING CREATES PRO-GOVERNMENT MILITIAS ...............................................................................7 Introduction ................................................................................................................................7 Research on Naming and Shaming .........................................................................................11 Human Rights Substitution Literature ..................................................................................17 Pro-government Militias and Plausible Deniability ..............................................................18 Cost-benefit Calculations for Repression ...............................................................................18 Unique Benefits of PGMs .......................................................................................................22 Types of PGMs ........................................................................................................................23 Distance ..................................................................................................................................25 Research Design ........................................................................................................................27 Methods .....................................................................................................................................32 Results........................................................................................................................................32 Naming and Shaming and Repression after PGM Creation ................................................38 Conclusions and Future Work ................................................................................................41 BIBLIOGRAPHY ....................................................................................................................45 CHAPTER 2: THE LEADER’S TOOLKIT: HOW LEADERS CHOOSE THEIR TOOLS OF REPRSSION ..................................................................................................................................51 Introduction ..............................................................................................................................51 Data on State Repression and PGM Violence Literature .....................................................53 Theoretical Framework ...........................................................................................................56 Data ............................................................................................................................................65 Methods .....................................................................................................................................71 Results........................................................................................................................................72 Conclusion .................................................................................................................................88 BIBLIOGRAPHY ....................................................................................................................91 CHAPTER 3: WHO ME?: HOW AND WHEN PRO-GOVERNMENT MILITIAS CAN MAKE REPRESSION LESS EXPENSIVE FOR LEADERS ...................................................................96 Introduction ..............................................................................................................................96 Repression and Leadership Survival ......................................................................................98 Theoretical Framework .........................................................................................................100 Data ..........................................................................................................................................107 Research Design ......................................................................................................................108 Results......................................................................................................................................109 viii Conclusion ...............................................................................................................................116 BIBLIOGRAPHY ..................................................................................................................121 CHAPTER 4: CONCLUSION ....................................................................................................125 ix LIST OF TABLES Table 1: Summary Statistics for Main Dependent and Explanatory Variables .............................32 Table 2: Logistic Regression Models: Examining the Effects of Naming and Shaming on PGM Creation ..........................................................................................................................................34 Table 3: Summary Statistics for Dependent Variables ..................................................................67 Table 4: Distribution of PGM Repression Variable ......................................................................67 Table 5: Distribution of Collaboration Index Variable ..................................................................69 Table 6: Summary Statistics for Independent Variables................................................................70 Table 7: Main Models 1-2: Repression and Violent Attack Dependent Variables........................73 Table 8: Main Models 3-5: Alliances, Acknowledgement, and Training......................................76 Table 9: Main Models 6-8: Funded Salaries, Heavily Armed, and Collaboration Index ..............78 Table 10: Main Models 9-10: Lifespan and Anti-Rebel ................................................................83 Table 11: Main Models – Cox Proportional Hazard Models .......................................................110 Table 12: Combined Model – Cox Proportional Hazard Models ................................................115 x LIST OF FIGURES Figure 1: Number of Instances of Naming and Shaming from Amnesty International .................14 Figure 2: Number of Instances of Naming and Shaming from UNCHR .......................................15 Figure 3: Number of PGMs in Existence over Time .....................................................................16 Figure 4: First Differences of UNCHR and AI Shaming ..............................................................36 Figure 5: Predicted Probability of Violent Attacks by PGM Type ................................................74 Figure 6: Predicted Probability of Violent Attack by Collaboration Index ...................................82 Figure 7: Predicted Probability of Attack by PGM Lifespan ........................................................85 Figure 8: First Differences for Models 2-10 ..................................................................................87 Figure 9: Survival Rate for Informal PGMs ................................................................................111 Figure 10: Survival Rate for Semi-Official PGMs ......................................................................112 Figure 11: Predicted Probability of Relative Hazard with Semi-Official PGMs.........................113 xi INTRODUCTION In July 2020, the United States Treasury imposed sanctions on three senior Chinese officials and a paramilitary organization (Finnegan 2020). The sanctions were issued in response to alleged human rights abuses committed by a pro-government militia (PGM), the Xinjiang Production and Construction Corps (XPCC). The abuses targeted ethnic and religious minorities in western China. The PGM is accused of detaining over 1 million people and repressing Muslim ethnic minorities in the region through re-education and forced labor camps. They also enforced sterilization practices. However, despite these claims, China has denied such abuses. Chinese officials claim this PGM is a counterterror unit and not carrying out abuses against civilians (The Guardian 2020). The case of the XPCC PGM in China is not unique or a new phenomenon. PGMs around the world have been accused of human rights abuses and engaging in acts of violence against civilians (Mitchell, Carey, and Butler 2014). These militias are unique repressive apparatuses, as they are not part of the normal military, but operate in organized units with weapons and still fight or work on behalf of the government. In return, some PGMs receive, pay, weapons, or other benefits (Carey and Mitchell 2017). This relationship between leaders and PGMs is sometimes covert, allowing leaders the ability to deny connections to these groups (Carey and Mitchell 2017). Even if links to these groups can be established, leaders often deny the abuses they carry out, as they did in China. Therefore, it can be a highly beneficial mechanism for leaders. Not only can these groups perform normal security tasks, like their military or police, but their covert relationship and unofficial connections allow them to perform some repressive tasks as well. 1 Over the past decade, attention has turned to better understanding PGMs and their actions. While researchers have begun to explore how PGMs can be used for violence and the implications for civilians, we still do not know exactly what these groups do on behalf of leaders. Furthermore, our knowledge of what characteristics or qualities influence their ability to more repressive is also limited. Better understanding the role of PGMs in state-sponsored repression will help researchers and the international community understand the nature of the PGM- government relationship and what can be done to stymie these abuses. This dissertation explores how leaders around the world have created, used, and possibly benefitted from PGMs, specifically in relation to human rights abuses. Chapter 1 explores the creation of PGMs, which has previously gone underexplored. The reason for the creation of these groups is essential to understanding how and why leaders find them most useful. In this chapter, I analyze the role of naming and shaming as a prompt to pressure leaders into creating alternative security apparatuses, such as PGMs. The international condemnation incentivizes leaders to attempt to avoid future shaming efforts despite their desire to continue their repression. I argue one solution is for leaders to create PGMs. I find that leaders are more likely to create PGMs following acts of shaming by the international community from a variety of sources. The analysis demonstrates that even though the international community is watching, leaders are able to develop new strategies to avoid future punishment. Chapter 2 introduces and uses the Militia Repression Dataset (MRD). This novel data collection effort captures PGM-specific repression. I present data on victims or targets of PGM violence and detailed information about the type and severity of their abuses. In addition, I code PGM characteristics that may influence their capacity to be useful repressive tools for leaders, such as their alliance for formal state forces, like the military, if the government provides heavy 2 artillery or weapons, and if the government has acknowledged the group in any capacity. These features of militias also help explain the nature of the relationship of the PGM and the state. Lastly, Chapter 3 explores whether PGMs are a useful strategy for leaders, specifically in terms of extending their tenure in office. I test if leadership tenure is extended depending on the type of PGMs present in a state. In this chapter, I argue that leaders can derive varying degrees of utility from PGMs depending on their connection to the government. While militias with closer ties can help leaders stay in office longer, the case is not true for more loosely connected PGMs. Interestingly, I find that although leaders may create and utilize these PGMs in order to strengthen their political position and maintain their power, it does not always work. Instead, leaders lose power sooner when they have these loosely connected militias at their disposal. Overall, these results demonstrate that while leaders may choose to create these PGMs and use them for repressive purposes, their strategy does not always work in the long run. While it may benefit leaders to use these loosely connected groups to stymie short-term repression, they risk their power and seat in office if they use them. On the other hand, PGMs that are closely connected to the government are not as repressive and appear to be more helpful for leaders hoping to extend their tenure. These findings present unique challenges for the international community to explore. Evidence suggests leaders create these groups despite the international community’s intention to stop human rights abuses within that state. However, as Chapter 2 shows, these groups are highly repressive apparatuses that are used and deployed for abusing civilians in addition to other security-related tasks. This is a conundrum for the international community, highlighting the unintended consequences of current punishment mechanisms, like naming and shaming. 3 In conclusion, this dissertation aims to contribute to our understanding of the role of PGMs in state-sponsored repression and how leaders use these militias to abuse their own civilians. It addresses concerns of PGM creation, utilization, and utility for leaders in achieving their goals of maintaining their power. While PGMs may be a useful tool for leaders, they have proven to be dangerous security apparatuses for civilians. 4 BIBLIOGRAPHY 5 BIBLIOGRAPHY Carey S and Mitchell N (2017) Progovernment Militias. Annual Review of Political Science 20: 127-147. Finnegan, Conor (2020) US ramps up sanctions over Uighur abuses with penalties on powerful Chinese paramilitary group. ABC News. July 31 https://abcnews.go.com/Politics/us-ramps- sanctions-uighur-abuses-penalties-powerful-chinese/story?id=72105534 The Guardian (2020) US imposes sanctions on senior Chinese officials over Uighur abuses. The Guardian. July 10 https://www.theguardian.com/world/2020/jul/10/us-imposes-sanctions-on- senior-chinese-officials-over-uighur-abuses Mitchell NJ, Carey SC, and Butler CK (2014) The Impact of Pro-Government Militias on Human Rights Violations. International Interactions 40(5): 812-36. 6 CHAPTER 1: FROM SHAME TO NEW NAME: HOW NAMING AND SHAMING CREATES PRO-GOVERNMENT MILITIAS Introduction States and their leaders can derive benefits from repression, including stability and the ability to maintain power. However, repression can be a costly decision for leaders due to the potential of future punishments from both the domestic and international communities (Nordås and Davenport 2013). Therefore, some leaders have incentives to find alternative, less obvious means of engaging in repression, such as dispatching other actors to carry out abuses on their behalf to avoid being punished. In this regard, this article examines how states attempt to use pro-government militias (PGMs) after being named and shamed to skirt future international condemnation. In particular, I analyze how naming and shaming efforts by Amnesty International (AI), a non-governmental organization (NGO), and the United Nations Commission on Human Rights (UNCHR)1, an intergovernmental organization (IGO), incentivizes states to create and then delegate repression to pro-government militias (PGMs). PGMs are non-state forces that can be used to carry out repressive tasks on behalf of governments and their militaries. This strategy of outsourcing repression to maintain control allows leaders to gain not only plausible deniability but also distance from any wrongdoing. Since most leaders seek to avoid naming and shaming, a PGM is a valuable and useful tool for states that would like to use repression to maintain domestic power and order. Nigeria provides an excellent example of this logic. In 1996 Nigeria was named and shamed by Amnesty International a total of 11 times, and the UNCHR chose not to act on allegations of human rights abuses. In 1997 they were shamed only seven times by Amnesty, 1 Known as the United Nations Human Rights Council (UNHRC) as of 2006. 7 but the UNCHR passed a resolution on their human rights record. In 1998 both Amnesty and the UNCHR ramped up their shaming of Nigeria. Amnesty International called out and publicly castigated Nigeria a total of 17 times. The UNCHR again issued a public resolution on Nigeria’s human rights record that year. While Nigeria had experienced some naming and shaming in the past, this spike in international condemnation from multiple sources in 1998 was significantly higher than years prior. In 1999 the Bakassi Boys PGM or the Abia State Vigilante Group emerged in Nigeria. According to Human Rights Watch, the group killed hundreds of civilians through execution, mutilation, and torture and has engaged in unlawful detentions and election violence (Human Rights Watch 2002b). The group’s connection to the state has been documented. The Bakassi Boys received orders directly from the government and had close and regular interaction and communication with government officials. The government has also provided office space, paid their salaries, and equipped them for their missions. Moreover, the militia purportedly announced to their victims, “We are Bakassi Boys. It’s a government order…The government wants you to die” (Human Rights Watch 2002b, 3). Despite these palpable ties to the Bakassi Boys and their repressive actions, the government rejected any connection to the PGM’s human rights abuses in both official statements and news interviews (Human Rights Watch 2002b). Government officials refuted claims of directing and supporting the Bakassi Boys. In July of 2001, for example, when the governor of Abia State, Orji Uzor Kalu, was questioned about alleged involvement with the Bakassi Boys, and the possibility of him utilizing the group for violence ahead of the upcoming 2003 election, he responded, “I don’t think so. I have no hand whatsoever in handling Bakassi. You see, the Bakassi in my state is the most quiet Bakassi. And what I have always said is that 8 nobody will rig election in Abia. We don’t need Bakassi to do that” (Human Rights Watch 2002b, 14). The police commissioner in Anambra state also rejected a connection to the Bakassi Boys, claiming that the group works independently of their police force but does cooperate with them (Human Rights Watch 2002b). The creation of the Bakassi boys in Nigeria presents a puzzle. Why would the government of Nigeria dispatch a new entity to engage in repression following naming and shaming by the international community, specifically IGOs and NGOs? I argue that this anecdote exemplifies a larger pattern. States that have already been named and shamed generally seek to avoid condemnation for their actions. Therefore, leaders turn to alternative security apparatuses, like PGMs, to continue repressing while evading accountability. The Bakassi Boys acted as a scapegoat for the Nigerian government to cast blame upon for human rights abuses, despite the militia being a mechanism of the state (Human Rights Watch 2002b). By distancing themselves from the PGM, the government of Nigeria was able to deny any wrongdoings and potential human rights abuses that may have occurred. This distancing and denial allowed the government to evade punishment, as human rights abuses are not technically committed by the state’s regular forces. This analysis demonstrates that Nigeria’s use of the Bakassi Boys militia is consistent with broader, cross-national trends. Specifically, I examine whether states respond to naming and shaming from NGOs and IGOs by creating PGMs using data on the creation of and links between governments and PGMs across the globe as well as data on naming and shaming from 1986-2000. I argue that governments create and utilize PGMs to avoid international condemnation in the form of naming and shaming, which has previously gone underexplored (Carey and Mitchell 2012). The results show offer support for this argument; PGMs are more 9 likely to be created after a state has been named and shamed by actors in the international community. More specifically, the findings demonstrate that informal PGMs with looser ties to the state are more likely to emerge after instances of international condemnation from both IGOs and NGOs. Semi-official PGMs with closer government ties, on the other hand, are only more likely to be created following instances of naming and shaming from IGOs. This article makes several contributions to the literature. First, existing literature, particularly proponents of naming and shaming, do not consider how naming and shaming may incentivize states to employ alternative mechanisms, specifically PGMs, to continue their repressive behavior while skirting blame. This analysis advances the literature by identifying and this previously unidentified consequence of naming and shaming and explaining why it occurs. Second, little is known about the reasons behind the creation of PGMs, especially beyond their relationship to violence. This analysis also contributes to this body of research by focusing on the impetus for the conception of new PGMs, which has received little attention from existing research. It is essential to understand the creation of entities like PGMs, as they play an important role in politics and repression. By understanding the creation of PGMs, the international community and human rights watch groups can pay particular attention to countries or regions that are indicative of these conditions. In addition, analyzing this previously underexplored strategy of deploying alternative agents of repression can help the international community better understand how governments use alternative agents to carry out repression against civilians. Lastly, this article highlights important implications for policymakers. Most notably, it illuminates a pathway by which state leaders cheat punishment mechanisms by IGOs and NGOs and suggests that the international community should consider alternative punishment and monitoring mechanisms for leaders suspected of human rights abuses. Overall, this article 10 demonstrates that by highlighting the benefits of naming and shaming, scholars miss out on the role that international condemnation can play in exacerbating human rights concerns within states. Research on Naming and Shaming Naming and shaming is a popular and widely used practice to enforce international human rights norms and laws (Hafner-Burton 2008). Various entities, such as the United Nations, Amnesty International, and individual states, “target some nations for particular attention and condemnation in the hope that through such publicity, these governments will be pressured into changing their abusive practices” (Meernik et al. 2012, 234). It is a policy of punishment by publicity designed to inflict “reputational damage on moral grounds” (DeMeritt 2012, 598). A variety of actors can name and shame but is mainly a tool used by NGOs, news media, and IGOs. While specific procedures vary, the basic idea is to first chastise a leader or group of leaders for their repressive acts and violations of human rights norms and laws, and then encourage them to improve their human rights practices. Existing literature debates the effectiveness of such naming and shaming to reprimand states for human rights abuses. There are four main expectations about the relationship between naming and shaming and human rights abuses that appear to garner ample empirical support. First, some scholars expect that human rights abuses may persist despite efforts by the international community to lessen or stop the violations through naming and shaming (Hafner- Burton 2005; Hathaway 2002; Snyder and Vinjamuri 2003–2004). Violators of human rights may choose to simply ignore naming and shaming because they perceive it as cheap talk with no real consequences for their behavior or decision-making. There is ample anecdotal evidence to support this argument, such as Israel’s response to naming and shaming in recent history. The 11 lack of effectiveness of naming and shaming may also arise from the fact that NGOs simply do not have authority over states and therefore lack a mechanism to hold leaders accountable (Hafner-Burton 2008). Second, the argument of suasion suggests that after being named and shamed, state leaders will make genuine efforts or attempts to improve their human rights records (Brysk 2003; Keck and Sikkink 1998; Risse et al. 1999). Here, violators respond positively to naming and shaming. Governments may take the criticisms seriously and reform their ways with passing legislation or holding elections (Human Rights Watch 2002a; Roth 2001; Roth 2004; Williamson 2004). Those that consider naming and shaming a valuable tool, mainly NGOs, believe that the act of naming and shaming will negatively publicize the repressive actions and “shine a spotlight” on the abuses being committed. NGOs view naming and shaming as furthering their own causes, such as crusades against human rights abuses (Hafner-Burton 2008). Third, some argue that naming and shaming alone is insufficient to generate any response from states (Barry et al. 2013; Dietrich and Murdie 2017; Esarey and DeMeritt 2017; Murdie and Peksen 2013; Peterson et al. 2015). Instead, naming and shaming must be coupled with a second mechanism to encourage reform. This indicates that while reform is possible, it is conditional upon additional strategies being employed at the same time. Murdie and Davis (2012) explain that human rights NGOs can help punish human rights abusers, but only if other actors aid them. For example, naming and shaming may be more effective when coupled with pressure from a third-party state, individuals, or organizations. Condemnation may also be effective when a human rights NGO has a domestic presence within a state (Murdie and Davis 2012). Most notably, some scholars demonstrate that naming and shaming can sometimes actually lead to improvements in a state’s human rights practices. This 12 conclusion, however, is met with opposition because of potential unintentional reactions to naming and shaming, such as the spotlight effect (Hafner-Burton 2008). A final argument suggests that naming and shaming can have some unintended consequences. Specifically, this argument claims that naming and shaming may actually incentivize states to ramp up abuses and carry out even more acts of repression than before they were named and shamed (Bob 2005; Conrad and Moore 2010; Escribà-Folch and Wright 2015; Hendrix and Wong 2014; Kuperman 2001). First, these instances of naming and shaming may encourage higher levels of domestic opposition, which may, in turn, threaten the repressive government (Hafner-Burton 2008). States may attempt to quell such opposition and stymie their efforts to compete for power while non-state actors may try to utilize these opportunities to “orchestrate acts of violence large enough to attract the spotlight” (Hafner-Burton 2008, 692). Although this argument highlights one potential mechanism by which naming and shaming can inadvertently increase human rights abuses, the literature has not yet considered other negative externalities that may be caused by naming and shaming, such as the creation of alternative security apparatuses like PGMs. Since 1975, there has been an overall increase in the number of instances of naming and shaming, especially by NGOs (Hafner-Burton 2008). At the same time, an increase in the number of militias created overtime is also evident. Figure 1 displays the frequency of naming and shaming events from Amnesty International while Figure 2 displays the frequency for the UNCHR. Given the heavy reliance on this tool to curb human rights abuses, it is important to understand how effective it is at achieving its desired aims. 13 Figure 1: Number of Instances of Naming and Shaming from Amnesty International 14 Figure 2: Number of Instances of Naming and Shaming from UNCHR Figure 3 displays the number of each type of PGM in existence over time. PGMs are defined as groups that are identified as pro-government or sponsored by the government at either the national or subnational level, are not part of the regular security forces, are armed, and have some level of organization (Carey, Mitchell, and Lowe 2012, 250). These groups are generally created or empowered by the government. 15 Figure 3: Number of PGMs in Existence over Time Overall, naming and shaming from Amnesty International has been increasing since the early 1990s. The number of instances of shaming from the UNCHR has dropped in more recent years; however, the decline is not substantial and appears to have evened out at the end of the sample. Therefore, in general, naming and shaming has increased in more recent decades. However, there is also an increase in the number of PGMs in existence over time, especially informal militias. Together, these figures point to a puzzling trend. If naming and shaming is an effective strategy that is increasing in frequency over time, why do we also see more security organizations delegated for repression over time? One answer may be the utility these groups provide their governments in forestalling future naming and shaming. 16 Although the figures above demonstrate that both naming and shaming and PGM creation have increased over time, scholars have yet to establish a link between naming and shaming and PGMs. However, this connection is important to consider since states choose to outsource abuses, given the high costs of condemnation, to alternative entities like PGMs. Human Rights Substitution Literature To avoid the costs associated with naming and shaming, leaders may turn to alternative forms of repression through a substitution method. Substitution refers to a government’s strategic replacement of “brutal methods of torture with more sophisticated techniques…” (Hafner-Burton and Ron 2009, 379). As Hafner-Burton and Ron explain, when leaders face increasing pressure from human rights watch dogs, they shift toward less visible techniques (also see Ron 1997). There is evidence of leaders substituting when considering signing on to human rights treaties. More specifically, Vreeland (2008) explores why more repressive dictators that practice higher levels of torture are more likely to enter into human rights treaties, specifically the UN Convention Against Torture (CAT), than less repressive dictatorships. In addition, Payne and Abouharb (2016) identify a similar finding with the International Covenant on the Civil and Political Rights (ICCPR). They find leaders that want to repress but also seek to avoid the costs of accountability will be less reliant on extrajudicial killings and instead prioritize forced disappearances. This literature makes clear that dictators do strategically substitute one form of human rights violations for others. This substitution argument can also be applied to naming and shaming. When leaders are punished and scrutinized by the international community, especially human rights watch groups, IGOs, and NGOs, they will attempt to substitute their repressive tactics for those that may engender less scrutiny. The literature has shown this substitution technique at work when states 17 delegate violence to paramilitary groups (Brenner and Campbell 2000). I argue PGMs are another example of this substitution method by leaders. However, instead of substituting one form of human rights abuses for another, states will substitute PGMs for their normal security forces, such as the military. Therefore, states delegate repression to PGMs to decrease their own expected costs. This delegation highlights another example of unintended consequences of naming and shaming by the international community, which has been underexplored. The remainder of this article will focus on this relationship. Pro-government Militias and Plausible Deniability Cost-benefit Calculations for Repression It is assumed that leaders prioritize their political survival (Nordås and Davenport 2013). However, at times, they may face threats to their tenure and power from competitors. As a result, some leaders consider extreme policy actions, namely human rights abuses, to counter or eliminate such threats. Scholars suggest that leaders make cost-benefit calculations regarding the use of repression when faced with dissent from their citizens (Nordås and Davenport 2013). When the benefits of repression to quiet dissent and maintain political office outweigh the costs, some leaders will choose to repress (Krain 2012; Nordås and Davenport 2013; Valentino 2004). The main benefit of repression for leaders is maintaining their hold on power. Governments may choose to utilize repression in various forms, including censorship, political restrictions, terror, and violence. Leaders expect these actions to neutralize political opponents and increase the costs for the opposition such that it is no longer a feasible strategy (Davenport 1995). Such a strategy limits how successful dissent may be and can offer more security for the leadership. The use of repression may also deter future opposition (Pierskalla 2010). Repression can also be risky, as leaders face the possibility of future punishment in 18 response to their actions. Punishments can include sanctions, fewer trade partners, reduced international aid, and reduced foreign investment (Hafner-Burton 2005; Hathaway 2002). In addition, states may face reputational costs (Risse and Ropp 1999). Finally, states may also be named and shamed by the international community in response to repression. These last two consequences are important because leaders generally care about their identity in the international system and want to maintain or improve their status as legitimate regimes (Murdie and Davis 2012). Naming and shaming by the international community is intended to alter the cost-benefit calculation for those trying to decide whether to repress their citizens by increasing the costs for leaders who abuse human rights. By drawing the international community's attention to abuses, naming and shaming shines a spotlight on a state’s repressive actions. Naming and shaming may also serve as a catalyst for future punishments. Therefore, naming and shaming can create a challenge for leaders; they may want to repress to ensure their political survival, but they also want to lower the costs associated with committing human rights abuses (Kirschke 2000; Roessler 2005). Some states are sensitive to the costs of punishments like naming and shaming, as demonstrated by the case of President Daniel arap Moi of Kenya. In 1991 President Moi was named and shamed by the United States and a dozen other aid-donor nations for political repression throughout Kenya. Kenya was mandated to introduce political and economic reforms and improvements in human rights or lose their economic aid within six months (Greenhouse 1991). In response to this naming and shaming coupled with a threat of loss of funding, President Moi embraced political reform within one month. The government lifted a ban on opposition political parties, legalized multipartyism, and eventually held its first multiparty election since 19 1966 (Roessler, 213-4). After being named and shamed, many states will be eager to avoid being chastised publicly again, as was the case in Kenya. However, not all leaders will want to end their spell of repression. An alternative solution is for states to create a separate apparatus to carry out acts of repression on its behalf, such as a PGM. Instead of making genuine efforts to reform their human rights practices in their country, states may instead opt to delegate violence to PGMs to escape the responsibility of subsequent human rights abuses. Since leaders generally do not engage in repression first-hand, they must outsource the task to other groups, such as militias (Gurr 1986). Creating a pro-government militia after being named and shamed allows a state to continue carrying out repressive actions but under a new institution that is separate from their recognized security forces. As Mitchell, Carey, and Butler explain, “The presence of these groups permits governments to shift responsibility for the consequences of the use of violence” (2014, 19). Therefore, PGMs act as a solution to unfavorable costs of repression for leaders, as the cost of repression is potentially diminished for state leaders because they no longer fear scrutiny. The case of the Bakassi Boys in Nigeria demonstrates this point. When PGMs carry out human rights abuses, governments can evade accountability by placing blame on PGMs for acts of repression while still achieving their goals (Carey et al. 2015). Essentially, the PGM presents a scapegoat that can afford state leaders plausible deniability due to the PGM’s weak ties to the government. Therefore, states have institutions to create and employ PGMs after being named and shamed in anticipation of future international scrutiny. As a result, when governments cooperate with or utilize PGMs, human rights will suffer (Mitchell et al. 2014). 20 Although Kenya chose to stop their repressive behavior in response to naming and shaming, not all states will. Other states, such as Iran in the early to mid-1990s, have responded to naming and shaming by shifting the responsibility for repression away from themselves and toward a PGM. In the half-decade leading up to the creation of the Ansar-e Hezbollah PGM in Iran in 1995, the government of Iran was named and shamed a total of 68 times by Amnesty International and was continually the focus of public resolutions from the UNCHR. Amid immense international condemnation, the Ansar-e Hezbollah group was created in 1995 as an informal PGM. Their main targets were unarmed political opposition, government critics, and journalists (Mitchell and Carey 2013). The group has been accused of numerous human rights abuses, including assault and intimidation of writers and intellectuals, disruption of gatherings of critical government policies, and carrying out violent raids on offices of magazines and newspapers with which they disagreed (Amnesty International 1997). Furthermore, there is evidence of corroboration with police in some instances, indicating some support from the government and their normal security forces. For example, in 1999 following a demonstration by several hundred students in response to the government’s closure of a reformist newspaper, police allowed members of the Ansar-e Hezbollah PGM into a student dormitory to attack and detain some of the students (Maloney 2013). They “systematically ransacked student rooms, destroyed property and assaulted students…300 students were wounded, 400 taken into detention and four were killed” (Megally 1999). When students held a rally to protest the initial assault, the PGM again attacked students “with sticks and chains while the police reportedly stood by or joined in the attacks” (Megally 1999). Despite approval and even collusion in multiple instances, the Minister of the Interior, 21 Hojatoleslam Abdolvahed Mousavi-Lari, stated that the assault on campus took place without the ministry’s approval and denied any involvement (Megally 1999). In addition, President Khatami commented that the PGM “instigators should be lawfully prosecuted for what they had done in response to the ‘ugly acts’ of student protests” (Sadeghi 2009). Even Ayatollah Khamenei condemned the PGM to presumably distance the government from the militia by stating, “Entering students’ bedrooms by force is like attacking homes and private residences unlawfully, in the worst cases at night…If, for example, [the students] insulted the leader, we have to be patient, to be reticent, even if they set my picture on fire, or tear it” (Gorgin 2008). Iran provides an example of an attempt to shift the blame for human rights abuses and repression away from themselves and onto groups, like the Ansar-e Hezbollah PGM. Since PGMs are not official security forces, states are often able to deny culpability in the abuses the groups commit, regardless of its tacit approval or support. Unique benefits of PGMs As demonstrated by the case of Ansar-e Hezbollah, PGMs provide several benefits to their respective states that explain why leaders turn to these alternative security apparatuses. First, they reduce the political cost of repression for leadership. By definition, PGMs are outside the state security apparatus. They cannot be officially tied to the state, allowing them to skirt blame for abuses committed by the militia, even if on behalf of the government. These informal ties between PGMs and governments create distance between the two units and affords the government plausible deniability. As Carey, Mitchell, and Lowe explain, these militias can help governments “to evade accountability for strategically useful violence” (2013, 250). Essentially, these militias create space between the regime and the repression they want to be carried out on their behalf because the militia is not a part of their standing security apparatus. Leaders can 22 deny loose links to these groups while still benefitting from their repressive actions. Second, PGMs offer logistical incentives for states (Carey et al. 2015; Staniland 2012). They increase force numbers, lower deployment costs, and can provide informational advantages in their home regions. Most argue that militias can offer strategic benefits to governments in times of conflict that they may not be able to achieve otherwise (Eck 2015). However, PGMs can also be costly for governments. Especially when ties between the group and the state are unofficial, governments may lose their monopoly on violence when they delegate to militias. Carey et al. 2015). This suggests that the strength of the link between the state and the PGM may determine the types of benefits leaders can derive from their use. Types of PGMs The literature generally separates PGMs into two categories: informal and semi-official (Carey et al. 2013). The main distinction between these two classifications is how formally acknowledged the link is between the government and the PGM. According to the Pro- Government Militia Database (PGMD), informal PGMs are described as pro-government, government-backed, or government-allied. They may also be armed or trained by the government. The critical identification characteristic of these groups is that the link to these PGMs is not formally or officially acknowledged. Examples of informal PGMs include death squads and the Young Patriots in Cote d’Ivoire (Carey et al. 2013). The Janjaweed militia in Sudan is another example of a government's use of an informal PGM to carry out human rights abuses. The Janjaweed is a brutal militia that continually carries out violence against the local population. The government of Sudan responded to the issues with rebels in Darfur by creating the group (Human Rights Watch 2005). According to a Human Rights Watch Report, “The Sudanese government’s recruitment and deployment of militia 23 forces, and its strategy of targeting civilians from specific ethnic groups to combat the rebel insurgency resulted in crimes against humanity and war crimes” (Human Rights Watch 2005, 6). The government has not only funded and supported the militia but has also directly encouraged the Janjaweed to carry out human rights abuses. The Human Rights Watch Report cites the attacks on Wadi Saleh, a province in Sudan, as an example of this. In response to the Fur joining the rebellion, the state minister of the interior, Ahmed Haroun, called on the Janjaweed militia to “kill the Fur.” Following this call for action, the Janjaweed army, along with the military, killed members of the ethnic group and looted their property (Human Rights Watch 2005). The Janjaweed militia demonstrates how PGMs can carry out abuses on behalf of the government, and how this creates a rationale for why governments create and support PGMs. Therefore, after funding, support, and direction from the state officials, the militia was able to carry out human rights abuses on behalf of the government. While informal PGMs are created without strict ties to the states, semi-official PGMs have a formally and/or legally acknowledged connection to the government. The PGMD explains that “a semi-official PGM might be sub-ordinate to the regular security forces, but is separate from the regular police and security forces. As such, the link between the PGM and the government is more formal and institutionalized…” (Mitchell and Carey 2013, 10). Examples of semi-official PGMs include the Arrow Boys in Uganda, Saddam’s Lion Cubs in Iraq, and the Revolutionary Guard in Iran. These types of PGMs, while not able to afford the government as much plausible deniability as informal PGMs, are still able to provide some cover for repressive actions. Their close ties to the government make it more challenging to carry out acts of violence on their behalf, and therefore, they may be less able to assist the government. However, as seen in the case of the Bakassi Boys, governments gain plausible deniability with these types of 24 PGMs by denying specific actions of the group and their knowledge of any abuses. Distance Due to the nature of informal militias and their links to their respective states, the distance between the state and the PGM is greater for these groups compared to the distance between semi-official PGMs and states. Informal PGMs are created with loose ties that are difficult to define and identify, allowing the government more plausible deniability than semi- official PGMs (Carey et al. 2013; Rudbeck et al. 2016) 2. Although their connection to the government may be well known within a country, the group has no formalized link to the government or might be denied by the state (Alvarez 2006; Carey and Mitchell 2017). Essentially, it is more difficult to associate informal PGMs with their governments than semi- official PGMs, affording more cover for governments. The relationship between the government and an informal PGM may be entirely denied by a government, unofficial, or simply clandestine. Therefore, their relationship is not always observable by the public or those outside the state (Carey and Mitchell 2017). This makes pinpointing blame and responsibility for human rights abuses much more difficult for the international community. Even if a loose connection is detected by the international community, the leader can exploit their principal-agent relationship 2 PGMs are created for a variety of purposes and can serve multiple purposes to its respective government (Carey and Mitchell 2017). Furthermore, PGMs may change overtime and shift from one purpose to another, depending on the relationship with and the needs of the government at that time. Despite this heterogeneity, recent literature has demonstrated the negative effect of informal PGMs on human rights practices within a state, namely government-sponsored killings, torture, and disappearances (Cohen and Nordås 2015; Mitchell et al. 2014; Stanton 2015). Furthermore, since the distance or “daylight” between the government and informal PGMs is greater than with semi-official PGMs, the risk for civilians is higher when informal militias are employed (Carey and Mitchell 2017). 25 to the leader’s advantage.3 They can simply blame those specific actions of abuse on rogue agents and argue that the group is out of their control (Alvarez 2006; Carey et al. 2015; Carey and Mitchell 2017; Kirschke 2000; Rudbeck et al. 2016). One example of this would be Serbia’s use of the Arkan Tigers, an informal PGM. Officials, such as Slobodan Milošević, were able to avoid international accountability for the group, despite “the flimsy basis for denial” (Carey et al. 2015). When exploring these abuses, the International Tribunal for Former Yugoslavia found evidence that while Serbian officials had connections to this group, “it was not proven beyond reasonable doubt that Stanišić or Simatović planned or ordered the crimes” (UN ICTY 2013). They were then acquitted on all charges. This leads to my first hypothesis: H1: States are likely to create informal PGMs following naming and shaming by the international community. While the literature has distinguished between the two types of militias for various purposes, there is still reason to believe that semi-official militias will also be created following instances of naming and shaming. Semi-official PGMs can also provide some plausible deniability and shelter from the retribution of repressive actions for governments and their 3 In this situation, the state would be the principal, and the PGM would be the self-interested agent that it hires or employs. The state needs to hire or create the PGM to carry out tasks that it cannot do itself, due to lack of skill and opportunity, or in this case, avoid having the normal standing military forces involved. The key to this relationship is delegation of a task, and specifically in the case of PGMS, it is the delegation of repression. In this relationship, the principal is distanced from the agent and is assumed to be the direction-giver and supervisor. They make the rules and delegate the tasks for the agent to carry out. It is important to note that the principal cannot directly observe the actions of the principal at all times. Traditionally, the principal-agent model presents the issue of asymmetrical information possessed by the agent as a disadvantage for the principal. In most cases this issue creates tension between the parties and further problems. However, in the case of PGMs and the governments they represent, it is evident that this distancing of the two parties and the “turning a blind-eye” approach used by governments is actually a benefit for both parties. 26 leaders. This function is mainly because they are a separate apparatus and not formally part of any official military or police entities, providing some distance between the leader and the group. Despite providing somewhat limited plausible deniability for leaders, using semi-official PGMs should still be preferable than using the military for repression. This is because, in addition to providing some plausible deniability, PGMs also require fewer start-up costs. Therefore, some states may be inclined to use them, if only sparingly. Furthermore, the nature of the militia- government link resembles a principal-agent relationship, allowing the government to essentially claim the group cannot be controlled or is a rogue entity (Carey and Mitchell 2017; Hibbs 1973; Tilly 1978). As seen in the case of the Bakassi Boys, governments gain plausible deniability with these types of PGMs by denying specific actions of the group and their knowledge of any abuses. This leads to my second hypothesis: H2: States are likely to create semi-official PGMs following naming and shaming by the international community. In this analysis, I argue that PGMs are, in fact, mechanisms for plausible deniability. Evidence from the cases of Iran and the Bakassi Boys exemplifies exactly how leaders use PGMs to distance themselves from the repression. Furthermore, arguments for the force multiplier effect of PGMs are shaped by the narrative and context of civil war, which ignores the pre- and post-conflict settings, along with times where conflict may fall short of technical qualifiers of civil war. By extending beyond the context of civil wars, we can better understand how PGMs operate and may be used differently from the military. Research Design To test the hypotheses set forth above, I analyze the effects of naming and shaming in two main models. I examine the effects of naming and shaming on the creation of informal 27 PGMs and semi-official PGMs separately. The dependent variables come from the Pro- Government Militia Database (PGMD) by Carey, Mitchell, and Lowe (2012). The dependent variables are dichotomous indicators with a value of one if at least one PGM of a given type was created in a given year and zero otherwise. Almost 3% of the observations are years where informal PGMs were created, and almost 2% are years semi-official PGMs were created.4 The measures only capture the creation of new PGMs in a year according to the PGMD and do not account for existing PGMs. There are 151 informal PGMs, and 59 semi-official PGMs created across 143 countries in the analyses. These data cover a global sample for the period of 1986 through 2000, and the unit of analysis for all models is the country-year. Both main models have 2,711 observations. The two main explanatory variables for this analysis capture two sources of naming and shaming by NGOs and IGOs. I use data from both the United Nations Commission on Human Rights (UNCHR) and Amnesty International (AI) to create two classifications of naming and shaming as naming and shaming can come from a variety of sources. In addition, IGOs and NGOs can both engage in naming and shaming to varying effects. First, for UNCHR naming and shaming, I use data from Lebovic and Voeten to capture a response or action by the UNCHR (2006). The original measure is constructed using five ordered categories, indicating increasingly severe responses by the UNCHR to human rights violations. The first level takes on a value of zero when a state is not named and shamed by the UNCHR. The second level reflects when a state is a target of UNCHR discussion, but the Commission 4 Due to the nature of both dependent variables having an overwhelming majority of observations of zeros, I re-estimate the main models as rare events logistic regressions as a robustness check. The results are consistent with the main models and do not change the findings of the analyses. 28 does not take any action against the state. This can happen when a resolution fails or motion not to consider the resolution passes. The third level indicates that the UNCHR has continued consideration of action under confidential sessions, which means that the unreleased allegations are deemed to have merit. The fourth level demonstrates that the UNCHR issued an advisory or critical statement. This is a milder sanction and comes from the chair of the Commission. Lastly, the fifth level indicates that the UNCHR issued a public resolution on a state’s human rights record, which is the most severe response given by the UNCHR. To distinguish public naming and shaming by the UNCHR from less visible or more temperate efforts by the body, I collapse Lebovic and Voeten’s original measure to construct a three-level ordinal indicator. In this new coding, a value of zero indicates there was no claim against that state for a given year (level one of the original variable). The variable is equal to one when the UNCHR shaming remains at a low level and is not made public (levels two through four of the original variable). Though these represent cases where a state’s human rights record was brought into question, and the IGO found enough merit to internally respond to the accusation, the UNCHR ultimately chose to take no further public actions. Finally, the variable equals two when the UNCHR issues a public resolution on a state’s human rights record (level five of the original variable). This distinction between public and private actions by the UNCHR is important because states are most likely to make alterations to their behavior in response to public chastisement that may spur further penalties. Instances where no public action or where allegations are dismissed are taken are less likely to influence a state’s decision to delegate violence to PGMs.5 5 The results remain consistent when Lebovic and Voeten’s five-point scale is used as an alternative. However, due to space limitations, the full tables for these results are not shown. 29 Second, I use data on non-governmental organization (NGO) human rights reporting from Ron, Ramos, and Rodgers (2005), which provides two measures of shaming from Amnesty International. The first indicator measures the number of Amnesty International’s background reports, which are sent to human rights professionals, United Nations officials, academics, and feature journalists regarding the state’s human rights practices. The second measures the number of Amnesty International press releases, which are generally shorter and aimed at the general public and non-specialized media. By employing the NGO data, I can discern whether states respond equally to naming and shaming efforts by IGOs and NGOs. For my analyses, I use the same coding scheme as the UNCHR variable to construct an ordinal variable of AI shaming severity levels. This new variable is equal to one when AI issues at least one background report on a state. This lower level response is less public and will not be circulated as widely in the international community. The variable is equal to two when AI issues at least one press release on a state, indicating a more public response that will be seen by more of the international community. Lastly, the variable is equal to zero when there is no shaming from AI. I follow the model specification from Carey, Colaresi, Mitchell (2015) to identify covariates that may explain PGM creation. To account for the incentive and ability to avoid domestic accountability with political institutions, I create three dichotomous indicators for regime type using the Polity2 indicator (Marshall et al. 2016). First, I create a dichotomous indicator for strong autocracies, which are coded as countries that score -7 or lower on the Polity2 scale. I then code weak democracies as countries that score between 1 and 6 on the Polity2 scale. Third, I code strong democracies as countries that score 7 or above on the Polity2 scale. The omitted category is weak autocracies. 30 To measure international sensitivity, I use Carey et al.’s measure for aid dependence. To construct this measure, I use the purchasing-price parity adjusted value of aid sent from democracies (those coded at least 7 on the Polity2 scale) to any recipient. I then take the natural log of the sum total of aid received from democracies as a proportion of the recipient’s gross domestic product (GDP) to create the final measure for democratic aid dependence. I also use the distance in kilometers between each country and the nearest democracy as a measure for monitoring costs for the international community. In addition, I include a measure of autocratic aid dependence. In order to capture the effect of state disorder on PGM creation, I use data from the Cross-national Times Series (CNTS) data (Banks 2008) to code dichotomous indicators for the presence of strikes, riots, demonstrations, and guerilla attacks. I control for civil violence and civil war with the Uppsala Conflict Data Programme (UCDP)/Peace Research Institute Oslo (PRIO) Armed Conflict Data (ACD; Gleditsch et al. 2002). Civil violence measures armed conflict above the twenty-five battle-related deaths within one-year threshold and civil war captures more than 1,000 battle-related deaths. As Carey et al. explain, using both indicators of civil conflict allows us to better understand how the varying nature and magnitude of civil conflict can explain PGM creation within the disorder framework. I control for economic development with the log of real GDP per capita from the Penn World Tables. I also control for the logged population from the Correlates of War data (Singer 1987). Lastly, I use Fearon and Latin’s (2003) measure for ethnic fractionalization to measure whether ethnically heterogeneous states are more likely to experience PGM creation. Table 1 displays the summary statistics for all dependent and explanatory variables. 31 Table 1: Summary Statistics for Main Dependent and Explanatory Variables Variable Mean Standard Deviation Variance Minimum Value Maximum Value Informal PGMs (dichotomous) Semi-Official PGMs (dichotomous) Amnesty International shaming (ordinal) UNCHR shaming (ordinal) .071 .257 .066 .061 .239 .057 1.538 .750 .562 .114 .417 .174 0 0 0 0 1 1 2 2 Methods To test my hypotheses, I use two models to analyze each of the two PGM creation dependent variables. Model 1 examines the effects of shaming on informal PGM creation, while Model 2 analyzes the effects of shaming on semi-official PGM creation. Both models use logistic regression as the dependent variables are dichotomous. I cluster standard errors on the country to account for non-independence of observations for the same country over time.6 Lastly, I lag all covariates and control variables by one year to examine the effects of PGM creation in the year following naming and shaming. Results Overall, the results demonstrate support for both hypotheses regarding PGM creation following instances of naming and shaming. Informal PGMs are created after shaming by both the UNCHR and Amnesty International. States also respond to international condemnation from the UNCHR by creating semi-official PGMs but do not create them following shaming from Amnesty 6 Both models produced the same results without clustered standard errors. 32 International. The results of the main models are displayed in Table 2. 33 Table 2: Logistic Regression Models: Examining the Effects of Naming and Shaming on PGM Creation Model 1 Informal PGMs Model 2 Semi-official PGMs .487 *** (.162) .537 ** (.209) .200 (.314) .672 ** (.314) .016 (.595) .139 ** (.063) .133 (.167) -.085 (.052) -.548 ** (.272) .601 * (.308) .467 (.323) .388 (.284) .799 ** (.309) .087 (.457) -.471 ** (.193) .109 (.111) .624 (.483) -4.506 ** (2.029) .211 2711 721.485 827.777 .453 ** (.223) .236 (.264) .287 (.371) .220 (.344) .080 (.862) .028 (.060) .122 (.239) .032 (.036) -.322 (.439) .813 ** (.348) -.160 (.332) .309 (.335) .755 * (.416) .171 (.444) .042 (.245) .309 ** (.126) 1.542 ** (.639) -9.908 *** (2.766) .155 2711 463.991 570.283 Independent Variables UNCHR Shaming (ordinal) AI Shaming (ordinal) Strong Autocracy Weak Democracy Strong Democracy Democratic Aid Dependence Distance to Democracy Autocratic Aid Dependence Strikes Riots Demonstrations Guerilla Attacks Civil Violence Civil War Log GDP Log Population Ethnic Fractionalization Constant Pseudo R2 Number of observations AIC BIC * p < .10, ** p < .05, *** p < .01. Clustered standard errors are in parentheses. 34 Model 1 estimates the effects of NGO and IGO naming and shaming on the creation of informal PGMs. The coefficient for shaming from the UNCHR is positive and significant. Similarly, the coefficient for Amnesty shaming is also positive and significant. Therefore, the findings indicate that informal PGMs with looser ties to the government are created after states are named and shamed by the UNCHR and Amnesty International. These findings demonstrate strong support for Hypothesis 1. Figure 4 displays the first differences from Model 1, which are the changes in the probability of informal PGM creation as the shaming variables increase from their lowest to highest values. The first differences provide information on the substantive impact of the relationship between naming and shaming and informal PGM creation. The graph shows that both sources of shaming have significant effects on the predicted probability of informal PGM creation, with AI shaming having a slightly larger substantive effect. States that experience no shaming from the UNCHR have a 1.13% chance of creating informal PGMs, however, states that experience the highest level of UNCHR shaming with a public resolution have a 2.95% chance of informal PGM creation, indicating a 161% increase in probability. Similarly, states without any shaming from AI have a .55% chance of informal PGM creation, but states with the most public AI shaming of press releases have a 1.58% chance of informal PGM creation, meaning a 187% substantive change. These findings demonstrate substantial impacts of shaming on the likelihood of informal PGM creation. 35 Figure 4 First Differences of UNCHR and AI Shaming Model 2 analyzes the creation of semi-official PGMs. The coefficient for shaming from the UNCHR is positive and significant, indicating that after the UNCHR shames a state, they are more likely to create a semi-official PGM. While the coefficient for AI shaming is also positive, it fails to achieve statistical significance and conventional levels. These findings demonstrate conditional support for Hypothesis 2. Shaming from the UNCHR appears to have a more powerful effect on states than AI’s shaming when considering semi-official PGM creation. While neither source can directly punish a state for a poor human rights record, the UNCHR has the ability to influence member states and increase the likelihood of future punishment. For instance, a public resolution issued by the 36 UNCHR may bring future economic sanctions, military intervention, or reduced amounts of foreign aid for the shamed state. UN member states may be persuaded by the IGOs investigation and findings, and more democratic states may face backlash from their own domestic constituents for supporting repressive regimes. While shaming from Amnesty can bring an issue to light and showcase a state’s poor respect for human rights, it does not appear to induce the same response from states considering semi-official PGM creation. Although Amnesty shaming may constitute preliminary reports that IGOs, like the UNCHR, states, and criminal courts use to begin or further their own investigations, Amnesty is unable to take action on its own based on the findings in its reports. Therefore, Amnesty shaming may be considered less threatening to leaders, especially in the short-term. Furthermore, the nature of semi-official PGMs may also explain this finding. While not official security apparatuses, these militias are more closely tied to states and their leaders than informal PGMs. It would be more difficult for a leader to claim the group is a rogue entity. Thus, the use of such agents may not afford the state much plausible deniability for their actions. Since a lack of plausible deniability may suggest a higher likelihood that abuses will be traced back to the state, leaders may be unwilling to utilize these semi-official militias when the circumstances are not dire. Counterintuitively, the threat of Amnesty punishment may not spur a state to take on added risks, while UNCHR condemnation may. As previously mentioned, Amnesty International can inflict no direct punishment on the states it shames. With UNCHR shaming, however, future condemnation from member-states may carry with it a significant cost inspiring states to adopt riskier behaviors to avoid it. Despite the disadvantages posted by the closer link between semi-official PGMs and the 37 state, there are also clear benefits for leaders contemplating using them. Some states may find a semi-official PGM to be more palatable as they are more easily controlled and may help leaders avoid principle-agent problems. This effectively allows the state to control the level and targets of violence, which may also allow states to effectively deflect attention away from the state. Closer ties may also facilitate synergies with other forces, such as the military, which may bolster their effectiveness and provide leaders with additional oversight. Thus, some states may still find closely linked militias valuable. This may explain the finding that only UNCHR shaming induces the creation of semi-official PGMs. These findings highlight different strategic logics for the use of these militias. The control variables offer a few additional important insights into PGM creation. Weak democracies and increasing levels of democratic aid are associated with informal PGM creation. This finding is consistent with the findings from Carey et al. (2015). In addition, civil violence also increases the likelihood of informal PGM creation. GDP has a negative effect on informal PGM creation, meaning that as states become wealthier, they are less likely to create informal militias. Strikes also have a negative effect on the likelihood of informal PGM creation. Similar to informal PGMs, civil violence is significant for Model 2 exploring the creation of semi- official PGMs. In addition, as the population increases, states are more likely to adopt these militias. Lastly, more heterogeneous states and states experiencing riots are more likely to create semi-official PGMs. Naming and Shaming and Repression after PGM Creation The findings demonstrate that leaders are strategic in their response to naming and shaming. In an attempt to avoid future instances of condemnations, leaders outsource their repression to alternative security apparatuses. However, understanding whether this strategy is actually 38 effective for leaders is essential. Preliminary evidence demonstrates that this strategy can be an effective tool for leaders to avoid future condemnation. Of the states that create a PGM following high levels of Amnesty shaming, meaning press releases issued on their human rights record, only 4.01% of those states were then shamed again the following year by both Amnesty and the UNCHR. On the other hand, states that do not create a PGM following intense Amnesty shaming will be more than twice as likely to be shamed again by both sources in the following year. Of the states that create a PGM following high levels of shaming by the UNCHR, meaning continued consideration or a public resolution, only 10.89% were then shamed again by both sources the following year. Again, states that do not create a militia following intense UNCHR shaming are twice as likely to be shamed again in the future by both sources compared to states that do create these militias. These findings demonstrate that creating PGMs can be a beneficial strategy for leaders hoping to avoid future condemnation as they tend to be shamed less in the future as compared to states with similar levels of shaming that do not create these militias. In addition, evidence also demonstrates that levels of repression will worsen or remain the same after the creation of PGMs in response to naming and shaming. According to the Cingranelli-Richards Physical Integrity Rights Index (CIRI), of the states that create informal PGMs after shaming by Amnesty International, 333 of the 372 cases, or just over 89%, will have the same or worse levels of repression the following year (Cingranelli and Richards 2014). The same is true for 300 of the 320 cases for semi-official PGM creation after Amnesty shaming or just over 93% of the cases. In comparison, just over 78% of states that do not create either type of PGM following shaming by Amnesty will have worse or similar human rights conditions the following year. This indicates that states that do not create militias will be more likely to see 39 improvements in human rights conditions over time. The results for the UNCHR present a similar pattern, but with fewer cases as UNCHR shaming happens less often than Amnesty shaming. Of the countries that create informal PGMs after UNCHR shaming, 24 of the 45 cases, or just over 53% will also have the same or worse human rights conditions. Of the 22 cases for semi-official PGMs, 14 cases, or just over 63%, will have the same or worse human rights conditions after the PGM is created. These same patterns also hold when exploring repression using other measures, such as the Political Terror Scale (PTS), which captures political violence and terror in a state with a 5-level terror scale (Gibney, Cornett, Wood, Haschke, Arnon, Pisano, and Barrett 2019). These findings demonstrate that leaders are able to continue repressing at either the same or more intense levels after the creation of the PGM in response to naming and shaming. The case of the Bakassi Boys presented earlier demonstrates how the decision to utilize PGMs after being named and shamed can be an effective tool to avoid future instances of naming and shaming while still carrying out repression. In Nigeria, shaming reached a maximum value in 1998, from both Amnesty International and the UNCHR. The following year, the PGM was created and the rates at which Nigeria was shamed quickly dropped. In 1999, the government was only shamed by Amnesty at a total of four times, and they were still shamed by the UNCHR. The following year, in 2000, the country was shamed a total of eight times, which is an increase from 1999, but still, less than half the number of instances from before the PGM existed. While Amnesty shaming data does not extend beyond 2000, UNCHR shaming significantly dropped in 2001 and 2002 with no instances of shaming from the Commission. Overall, this demonstrates a significant decrease in shaming following the use of a PGM by the state. The government of Nigeria also continued to repress after the creation of the Bakassi 40 Boys according to the physical integrity rights index from CIRI and the PTS. The CIRI index ranges from 0 to eight, with higher levels indicating more government respect for human rights. The year before the PGM was created, Nigeria’s government respect of human rights was a score of four, indicating some respect for human rights. The score dropped to a value of two the year the PGM was created, then dropped to a value of one the year after the PGM was created, and finally returned to a score of two when the PGM had existed for two years. This demonstrates a significant decrease in government respect for human rights once the PGM was created, which was maintained for several years. Overall, the PTS score for Nigeria remained generally constant throughout this period. There was a slight improvement in human rights conditions the year after the PGM was created, however, it returned to the previous level the following year. Together, these two measures of state repression indicate that government respect for human rights worsened after the creation of the PGM and remained at those lower levels, and overall political terror in the country remained fairly constant after the creation of the PGM. Future work should more thoroughly investigate this relationship to better understand in what ways PGMs can act as a scapegoat for leaders to avoid future international condemnation. Conclusions and Future Work The analyses show that when states are named and shamed by Amnesty International, they are more likely to create only informal PGMs. However, when named and shamed by the UNCHR, states are more likely to create both types of militias. The results generally demonstrate support of the hypotheses, especially Hypothesis 1, that naming and shaming can instigate the creation of informal PGMs. Furthermore, preliminary evidence demonstrates that this strategy of using PGMs to avoid future condemnation does work for leaders and can be beneficial for their 41 international reputation. In addition, this strategy of creating PGMs also allows leaders to continue violence and abusing human rights within their state, demonstrating a common utility of PGMs for governments. Several conclusions can be drawn from these findings. First, naming and shaming is intended to curb human rights abuses by repressive leaders and encourage reform. However, this analysis demonstrates that PGMs are sometimes an unintended consequence of naming and shaming, which highlights a major weakness of this strategy. Instead of improving human rights practices as the naming and shaming actions intend for them to, leaders are able to employ alternative apparatuses to continue carrying out abuses while absolving themselves of blame and the potential costs associated with naming and shaming. This mechanism that is traditionally utilized to improve human rights practices may actually create perverse incentives for leaders to simply be more creative in their attempts to continue abusing their citizens. However, liberal ideas regarding world order depend upon international actions like naming and shaming to effectively reprimand other actors for human rights abuses. If the international community is unable to achieve its goals through means such as naming and shaming, as this analysis indicates, more attention should be given to alternatives. These alternatives may extend beyond simply naming and shaming leaders that create PGMs to also include reductions in foreign aid, economic sanctions, or other forms of punishment. Additionally, it would be beneficial to further explore how states can cheat in order to avoid international attention. In the future, analyzing other forms of international punishment, such as the effects of sanctions or military intervention, may provide more insight into how and when leaders decide to deploy alternative security apparatuses as scapegoats for repression. Second, in addition to identifying and analyzing an additional unexplored consequence of 42 naming and shaming, these findings also provide an explanation for when leaders create PGMs. Although some research explains the relationship between PGMs and violence, previous literature has not yet fully been able to identify what leads to and triggers the creation of both types of militias. A better understanding of the conditions that lead governments to create PGMs can allow the international community, especially IGOs and NGOs, to address this phenomenon. In order to address these concerns and conclusions, it is imperative that the entirety of a state’s repressive apparatuses is reflected in our analyses and in policy decisions. This would include PGMs but also other national and subnational actors that the government may employ for violence. By excluding certain actors or groups, like PGMs, we may underestimate the ability and capacity for certain states to repress. In addition, it is important to better understand not only when these groups influence repression, but in what ways. Some work has begun to explore the link between PGMs, especially informal militias, and human rights abuses; however, the mechanisms by which these groups repress has gone largely underexplored. In terms of policy implications, IGOs and NGOs should be aware of and now anticipate this previously unidentified unintended consequence of naming and shaming by leaders. This study finds that leaders are turning to alternative security apparatuses once already condemned, and therefore trying to cheat to avoid future accountability. However, before naming and shaming states, IGOs and NGOs could proactively monitor human rights conditions in these states, especially those with a known history of militia violence. Monitoring efforts could also be enhanced by further collaboration and information sharing between monitoring organizations and domestic NGOs, as suggested by Murdie and Davis (2012). Second, IGOs and NGOs considering shaming states could call upon specific members of the international community for assistance, notably foreign aid donor states. Previous literature 43 has found that recipients of financial aid from democracies are more likely to create and align with informal militias (Carey, Colaresi, and Mitchell 2015). Therefore, asking donor states to reconsider their type or amount of foreign aid to states known to be engaging with militias could prove an effective tool for stymying PGM creation and collaboration. Although these findings show that states create PGMs after being named and shamed, only preliminary empirical work has been done to examine what happens to a state’s human rights record after they begin employing these militias. Future work might systematically explore the impact that this chain of events has on human rights. Such as, how well and in what ways does this strategy work for leaders and how? How and which types of human rights get worse after PGMs are created in response to naming and shaming? Additionally, it would be interesting to examine how the human rights community responds after PGMs are created. Essentially, do international actors have a learning curve for picking up on things like the use of informal PGMs for repression? Further exploration of state behavior following naming and shaming can shed light on these issues. The implications for this work extend beyond simply concluding that naming and shaming has unintended consequences. We must understand that governments may have alternative apparatuses loosely tied to the state, but nonetheless under their authority and control that can carry out repression on their behalf. These also include actors outside of the official channels. In addition to issues of measurement and conceptual gaps, the larger issue at hand is the fact that current practices to curb human rights abuses, such as naming and shaming, are sometimes failing to discourage the creation of groups like PGMs. The issues of repressive mechanisms, such as PGMs, have largely escaped academic literature and policy agendas until this point; however, the issue now becomes how to address these groups and their leaders. 44 BIBLIOGRAPHY 45 BIBLIOGRAPHY Ahram AI. 2014. “The Role of State-sponsored Militias in Genocide.” Terrorism and Political Violence 26 (3): 488-503. Alvarez, AI. 2006. “Militias and genocide.” War Crimes Genocide Crimes Against Humanity 2: 1-33. Amnesty International. 1997. “Iran: Human Rights Violations against Shi’a Religious Leaders and their Followers.” Index no. MDE 13/018/1997. June 02. Barry CM, Clay KC, and Flynn ME. 2013. “Avoiding the Spotlight: Human Rights Shaming and Foreign Direct Investment.” International Studies Quarterly 57 (3): 532-544. Bob C. 2005. The Marketing of Rebellion: Insurgents, Media and International Activism. Cambridge: Cambridge University Press. Brenner, A and Campbell B, eds. 2000. Death Squads in Global Perspective: Murder with Deniability. New York: St. Martin’s. Brysk A. 1993. “From Above and Below: Social Movements, the International System, and Human Rights in Argentina” Comparative Political Studies 26 (3): 259–285. Byman D and Kreps S. 2010. “Agents of Destruction? Applying Principal-agent Analysis to State-sponsored Terrorism.” International Studies Quarterly 11 (1): 1-18. Carey S, Colaresi M, and Mitchell N. 2015. “Governments' Informal Links to Militias.” Journal of Conflict Resolution 59 (5): 850-876. Carey S and Mitchell N. 2012. Pro-Government Militias Database: Introduction to PGMD.v1. Presented at ENCoRe Meeting, September, Oslo, Norway. Carey S and Mitchell N. 2017. “Progovernment Militias.” Annual Review of Political Science 20: 127-147. Carey S, Mitchell N, and Lowe W. 2012. “States, the Security Sector and the Monopoly of Violence: A New Database on Pro-Government Militias.” Journal of Peace Research 50 (2): 249-58. Cingranelli, David L., David L. Richards, and K. Chad Clay. 2014. "The CIRI Human Rights Dataset." http://www.humanrightsdata.com. Version 2014.04.14. Cohen DK and Nordås R. 2015. “Do States Delegate Shameful Violence to Militias? Patterns of Sexual Violence in Recent Armed Conflicts.” Journal of Conflict Resolution 59 (5): 877-898. 46 Conrad CR and Moore WH. 2010. “What Stops the Torture?” American Journal of Political Science 54 (2): 459-476. Davenport C. 1995. “Multi-Dimensional Threat Perception and State Repression: An Inquiry into Why States Apply Negative Sanctions.” American Journal of Political Science 39 (3): 693- 713. DeMeritt JHR. 2012. “International Organizations and Government Killing: Does Naming and Shaming Save Lives?” International Interactions 38 (5): 597-621. Dietrich S and Murdie A. 2017. “Human Rights Shaming Through INGOs and Foreign Aid Delivery.” Review of International Organizations 12 (1): 95-120. Eck K. 2015. Repression by Proxy: How Military Purges and Insurgency Impact the Delegation of Coercion.” Journal of Conflict Resolution 59 (5): 924-946. Eck K and Hultman L. 2007. “One-sided violence against civilians in war: insights from new fatality data.” Journal of Peace Research 44 (2): 233-246. Esarey J and DeMeritt JHR. 2017. “Political Context and the Consequences of Naming and Shaming for Human Rights Abuse.” International Interactions 43 (4): 589-618. Escribà-Folch A and Wright J. 2015. “Human Rights Prosecutions and Autocratic Survival.” International Organizations 69 (2): 343-373. Gibney, Mark, Linda Cornett, Reed Wood, Peter Haschke, Daniel Arnon, Attilio Pisanò, and Gray Barrett. 2019. The Political Terror Scale 1976-2018. Date Retrieved, from the Political Ter- ror Scale website: http://www.politicalterrorscale.org. Gleditsch NP, Wallensteen P, Eriksson M, Sollenberg M, and Strand H. 2002. “Armed Conflict 1946-2001: A New Dataset.” Journal of Peace Research 39 (5): 651-660. Gorgin I. 2008. “Looking Back at Tehran's 1999 Student Protests.” Radio Free Europe/Radio Liberty. Greenhouse S. 1991. “Aid Donors Insist on Kenya Reforms.” New York Times 27 November. (http://www.nytimes.com/1991/11/27/world/aid-donors-insist-on-kenya-reforms.html). Hafner-Burton EM. 2005. “Trading Human Rights: How Preferential Trade Arrangements Influence Government Repression.” International Organization 59 (3): 593–629. Hafner-Burton EM. 2008. “Sticks and Stones: Naming and Shaming the Human Rights Enforcement Problem.” International Organization 62 (4): 689-716. 47 Hafner-Burton EM and Ron J. 2009. “Seeing Double: Human Rights Impact through Qualitative and Quantitative Eyes.” World Politics 61(2): 360-401. Hathaway OA. 2002. “Do Human Rights Treaties Make a Difference?: Yale Law Journal 111: 1935–2042. Hendrix CS and Wong WH. 2014. “Knowing your audience: How the structure of international relations and organizational choices affect amnesty international’s advocacy.” The Review of International Organizations 9 (1): 29-58. Human Rights Watch. 2002a. “UN: Many Countries Escape Censure. Human Rights Commission Fails to Pass Critical Resolutions on Many Countries” (http:00hrw+org0english0docs020020040190russia3867.htm). Human Rights Watch. 2002b. “The Bakassi Boys: The Legitimization of Murder and Torture” (https://www.hrw.org/reports/2002/nigeria2/nigeria0502.pdf). Human Rights Watch. 2005. “Entrenching Impunity: Government Responsibility for International Crimes in Darfur” (https://www.hrw.org/sites/default/files/reports/darfur1205webwcover.pdf). Keck ME and Sikkink K. 1998. Activists Beyond Borders: Advocacy Networks in International Politics. Ithaca, NY: Cornell University Press. Kirschke L. 2000. “Informal Repression, Zero-Sum Politics and Late Third Wave Transitions.” Journal of Modern African Studies 38 (3): 383-405. Koren, Ore. 2017. “Means to an end: Pro-government militias as a predictive indicator of strategic mass killing.” Conflict Management and Peace Science 34 (5): 461-484. Krain M. 2012. “J’accuse! Does Naming and Shaming Perpetrators Reduce the Severity of Genocides or Politicides?” International Studies Quarterly 56 (3): 574-589. Kuperman AJ. 2001. The Limits of Humanitarian Intervention: Genocide in Rwanda. Washington, DC: Brookings Institution Press. Lebovic JH and Voeten E. 2006. “The Politics of Shame: The Condemnation of Country Human Rights Practices in the UNHCR.” International Studies Quarterly 50 (4): 861–88. Nordås R and Davenport C. 2013 “Fight the Youth: Youth Bulges and State Repression.” American Journal of Political Science 57 (4): 926-940. Rudbeck J, Mukherjee E, Nelson K. 2016. “When autocratic regimes are cheap and play dirty. The transaction costs of repression in South Africa, Kenya, and Egypt.” Comparative Politics 48:147–66. 48 Maloney S. 2013. “Remembering Iran’s Student Protests, Fourteen Years Later.” The Brookings Institution, July 09, 2013. Marshall MG, Gurr TR, and Kaggers K. 2016. “Polity IV Project: Political Regime Characteristics and Transitions, 1800-2015.” Center for Systemic Peace. Megally H. 1999. “Letter to Ayatollah Seyyed Ali Khamenei.” Human Rights Watch, July 12. Meernik J, Aloisi R, Sowell M, and Nichols A. 2012. “The Impact of Human Rights Organizations on Naming and Shaming Campaigns.” The Journal of Conflict Resolution 56 (2): 233-56. Mitchell NJ and Carey SC. 2013. “Pro-Government Militia Database (PGMD) Codebook. V1.1.” (http://www.sowi.uni-manheim.de/militas). Mitchell NJ, Carey SC, and Butler CK. 2014. “The Impact of Pro-Government Militias on Human Rights Violations.” International Interactions 40 (5): 812-36. Murdie AM and Davis DR. 2012. “Shaming and Blaming: Using Events Data to Assess the Impact of Human Rights INGOs.” International Studies Quarterly 56 (1): 1-16. Murdie A and Peksen D. 2013. “The impact of human rights INGOs activities on economic sanctions.” The Review of International Organizations 8 (1): 33-53. Office of the High Commissioner (nd) Mission Statement. The United Nations Human Rights Office of the High Commissioner. Payne, C and Abouharb M. 2016. “The International Covenant on Civil and Political Rights and the strategic shift to forced disappearance.” Journal of Human Rights 15 (2): 1-26. Peterson TM, Murdie AM, and Asal V. 2016. “Human Rights, NGO Shaming and the Exports of Abusive States.” British Journal of Political Science. Epub ahead of print 01 June 2016. DOI: 10.1017/S0007123416000065 Pierskalla JH. 2010. “Protest, Deterrence, and Escalation: The Strategic Calculus of Government Repression.” Journal of Conflict Resolution 54 (1): 117-145. Risse T and Ropp SC. 1999. “International Human Rights Norms and Domestic Change.” In The Power of Human Rights: International Norms and Domestic Change, edited by Risse T, Ropp S, and Sikkink K. Cambridge, UK: Cambridge University Press. Risse T, Ropp SC, and Sikkink K. 1999. The Power of Human Rights: International Norms and Domestic Change. Cambridge: Cambridge University Press. Roessler PG. 2005. “Donor-Induced Democratization and the Privatization of State Violence in Kenya and Rwanda.” Comparative Politics 37(2): 207-27. 49 Ron, J. 1997. “Varying methods of State Violence.” International Organization 51 (2): 275–300. Ron J. 2002. “Territoriality and Plausible Deniability: Serbian Paramilitaries in the Bosnian War. In Death Squads in Global Perspective: Murder with Deniability” edited by Campbell BC and Brenner AD, 286-312. New York: Palgrave McMillan. Ron J, Ramos H, and Rodgers K. 2005. “Transnational Information Politics: NGO Human Rights Reporting, 1986–2000.” International Studies Quarterly 49 (3): 557–88. Roth K. 2001. “Despots Pretending to Spot and Shame Despots.” International Herald Tribune 17 April. (http:00hrw.org0english0docs020010040170global12852.htm). Roth K. 2004. “Defending Economic, Social and Cultural Rights: Practical Issues Faced by an International Human Rights Organization.” Human Rights Quarterly 26 (1): 63–73. Sadeghin S. 2009. “Today is the Anniversary of the 1999 Iran Protest.” The Huffington Post. Snyder J and Vinjamuri L. 2003–2004. “Trials and Errors: Principle and Pragmatism in Strategies of International Justice.” International Security 28 (3): 5-44. Staniland P. 2012. “Between a Rock and a Hard Place: Insurgent Fratricide, Ethnic Defection, and the Rise of Pro-state Paramilitaries.” Journal of Conflict Resolution 56 (1): 16-40. Stanton JA. 2015. “Regulating Militias: Governments, Militias, and Civilian Targeting in Civil War.” Journal of Conflict Resolution 59 (5): 899-923. UN ICTY. 2013. Accessed November 22, 2019. https://www.icty.org/sid/11329. Valentino BA. 2004, Final Solutions. Ithaca, NY: Cornell University Press. Vreeland, J. 2008. “Political Institutions and Human Rights: Why Dictatorships Enter into the United Nations Convention Against Torture.” International Organization 62 (1): 65-101. Williamson RS. 2004. “US Representative to the Commission on Human Rights, Statement on Item 9: March 25.” (http://geneva.usmission.gov/humanrights/2004/statements/ /0325Item9.htm). 50 CHAPTER 2: THE LEADER’S TOOLKIT: HOW LEADERS CHOOSE THEIR TOOLS OF Introduction REPRSSION In December of 1981, a helicopter flew over the city of Cali, Colombia dropping leaflets announcing the creation of a new pro-government militia (PGM), Death to Kidnappers (Muerte a Secuestradores or MAS). The PGM was created by members of the Medellín drug cartel in response to the kidnapping of one of their sisters by the M-19 guerilla forces. This new militia was designed to protect themselves against rebel insurgents and act as a self-defense force. In addition, some locals enjoyed the public protection benefits provided by the PGM, as they too had been victims of the guerilla forces in the region (Carey and Mitchell 2014a). Despite some locals enjoying protection from guerilla groups, Death to Kidnappers was incredibly repressive and well known for its human rights abuses. They charged monthly “war taxes” and “force peasants to patrol under threat of death” according to a Human Rights Watch report (Human Rights Watch 1993). In 1980, there were 92 political killings in Colombia. However, by 1983, less than two years after Death to Kidnappers was established, 240 killings were already linked to the group (Babbitt and Lutz 2009, 38). Although they were a highly repressive and lethal militia, Death to Kidnappers benefited from loose ties and informal connections to police and the military. The formal state security forces supplied the PGM with weapons, ammunition, and even uniforms. The militia took part in joint operations with the army and locals reported witnessing troops accompanied by the militia carrying out extrajudicial killings and destruction of property (Human Rights Watch 1996). In addition, there were government campaigns with wall slogans against “subversion” that supported actions of the militia. This relationship was further exemplified during the search for 51 members of the PGM to hold them accountable for 183 murders committed over a five-year span. Instead of aiding in the commission’s investigation, members of the army and military “actively hampered their search” and even roused the population to attack the members of the commission. This resulted in the commission fleeing for their safety and only being able to securing only one detainee for the investigation (Human Rights Watch 1993). Furthermore, the PGM was rarely punished for their abuses; they faced few convictions by the courts for their killings (Amnesty International 1986). While there was cooperation between the PGM and military, and even acts of violence carried out together, the government and military still denied any formal link and connection to the group and their killings. They had no formal agreement and no connection on paper. They denounced their tactics and publicly called upon the group to stop their human rights abuses in addition to ordering a full investigation on their activities in 1982 (State Department 1982). This practice of denial created distance between the PGM and government in order to deflect blame from the regime for any abuses carried out by the PGM. On the other hand, the government and military of Colombia have been formally tied to another PGM, Soldiers from my Town. However, this PGM was far less repressive and did not target civilians like Death to Kidnappers. They were responsible to the state military structure and provided security for remote regions against rebels and insurgents (Kline 2009, 42). Given the history of militias in the region, civilians worried that this group would also carry out acts of violence, however, they have not been accused of any human rights abuses. While both of these PGMs operated in Colombia, had links to the military, and were self- defense forces targeting insurgents, only the informal militia with looser ties, Death to Kidnappers, was extremely abusive. The semi-official militia with clearer and more visible links 52 to the government, Soldiers from my Town, did not participate in repressive actions and was not accused of human rights abuses. This dichotomy demonstrates that states may choose to employ militias differently and for different purposes. They do so because groups can provide varying levels and modes of utility, and some are more able than others to carry out acts of repression for leaders. More specifically, PGMs with loose ties to the regime will be used for more repressive actions while those that can be traced back to the state more easily will not. This analysis explores this relationship to determine if this pattern applies worldwide. I argue that militias with looser ties to the regime government will repress more than groups with more official ties and formal acknowledgement. This is because groups with looser ties and no acknowledgement can afford the leader more distance from repressive actions they commit. Since leaders cannot be tied as easily to these abuses, they deploy these groups to carry out more brutal acts of violence on their behalf. The findings demonstrate support for the argument and show that groups with more plausible deniability are more repressive. These findings present challenges for the international community in terms of how to address irregular forces that repress civilians and may be used in times of armed conflict. Data on State Repression and PGM Violence Literature Numerous impressive data collection efforts have provided information on state repression around the globe. These datasets have allowed scholars to systematically investigate violence at the state-level in a variety of ways. Over time, several large datasets have become standards measure for capturing state coercion and repressive behaviors. First, three commonly used data sources focus on the variety and extent of violence in an index form. First, the Political Terror Scale (PTS) was among the first quantitative datasets to capture aspects of state respect for human rights (Gibney et al. 2018). This dataset allows 53 scholars to understand and state violations of physical integrity rights. PTS focuses on state-level behavior to provide information on actual violations of physical integrity rights as compared to a focus on general political repression measures (Wood and Gibney 2010). Using the same source material as PTS, CIRI Human Rights Dataset also measures state-sponsored repression in the form of physical integrity rights violations (Cingranelli, et al. 2014). One key departure from PTS is that CIRI disaggregates physical integrity violations into components, including disappearances, killing, torture, and imprisonment (Wood and Gibney 2010). This allows the CIRI dataset to provide a 9-point index to capture the full extent of physical integrity violations within a state. Lastly, the Ill-Treatment and Torture (ITT) Data Collection Project incorporates information on four key concepts, including the incident, perpetrators of the violence, the motivation for the violence, and the judicial response for the violence (Conrad and Moore 2014). Furthermore, the ITT Project uses the individual allegation as the unit of analysis, which also departs from the CIRI and PTS data collections that use a country-year unit of analysis. Other sources, such as the Uppsala Conflict Data Program (UCDP), focus on one-sided state violence (Eck and Hultman 2007; Allansson, et al. 2017). UCDP provides information on organized violence and is the oldest ongoing data collection project for civil war (Eck and Hultman 2007). This data project collects information on conflict dynamics and conflict resolution as compared to indices of state-level repression. UCDP offers information in a variety of forms, including georeferenced event data, yearly datasets, and actor lists. Lastly, others choose to focus on the occurrence of repressive behavior and other repressive activity, such as Freedom House. Through an annual survey and report, Freedom House experts assess countries around the world to evaluate the electoral process, political pluralism and participation, the functioning of government, freedom of expression and of belief, 54 association and organizational rights, the rule of law, and personal autonomy and individual rights (Freedom House 2020). Together, these sources along with many others compromise our understandings of state- sponsored violence. However, these analyses are generally unable to identify the full portfolio of violence for each actor within a state. While some datasets, such as ITT, offer information about the perpetrators of abuse, it is difficult to understand the magnitude and effect of the violence perpetrated by one group or actor over time. Furthermore, while some of these datasets capture violence perpetrated by agents of the state, such as PTS, not all violence by groups like PGMs will be included. This is due to the nature of the relationship between PGMs and their governments. Therefore, while there is certainly a robust offering of datasets on state-level repression, we do not yet have information disaggregated to the perpetrator-level to fully account for the type, severity, and victims of violence within a state. While we do know a great deal about state-sponsored repression given these datasets, we know less about the violence PGMs commit. Much of the literature focusing on PGM violence relies on the principal-agent framework, where states are the principal and PGMs the agent (Carey and Mitchell 2017). This framework allows scholars to explain the moral hazard problem PGMs may present. Essentially, the covert relationship of the PGM and government allows for “rogue” agents to carry out acts of violence on behalf of the leader, sometimes beyond their original intention (Carey and Mitchell 2017). Several case studies and pieces of qualitative work have identified the connection between militias and violence used against civilians (Álvarez 2006; Clayton and Thomson 2014; Kirschke 2000). Similarly, much of the quantitative literature agrees with this central finding and have extended these analyses to provide further insights. In general, scholars find that informal 55 PGMs are used for repression more often than semi-official PGMs. For instance, Mitchell, Carey, and Butler (2014) explore the risk of repression when PGMs are present in a state. They find that informal PGMs make government sponsored killings, torture, and disappearances more likely. However, the same is not true for semi-official PGMs. Focusing on civil wars, both Stanton (2015) and Cohen and Nordås (2015) find that semi-official PGMs are less likely to commit violence compared to informal PGMs. Many, such as Alvarez (2006) argue this is due to the plausible deniability afforded to leaders by these types of militias. Scholarly work on PGM violence has clearly demonstrated that certain types of militias, notably informal, are harmful to civilians and human rights suffer when they are present in a state. However, we do not know how much violence each militia carries out and what other militia characteristics influence violence. This analysis and data collection effort aims to fill this gap in both the data and literature by identifying specific levels, types, and victims or targets of human rights abuses by PGMs around the world and how these groups are used for repression by leaders. Furthermore, it provides more detailed information on the characteristics of these groups that may influence their repressive behaviors. By assessing the means by which leaders deploy PGMs for repression, the international community can work to address these abuses and groups. Theoretical Framework Once leaders or governments decide to use repression to meet their political goals, they must decide how to do so. In most cases, leaders do not carry out repression on their own, but instead must outsource the task to a security apparatus (Gurr 1986). Their repressive strategies and decisions about how to deploy repressive forces will impact potential future punishment for the repression and response from both the domestic and international community. If leaders choose a repressive force that is closely associated with the regime or government, they will have 56 less plausible deniability and potentially face more punishment for their actions. In addition, leaders need to ensure that their forces are loyal and will not present internal threats to their tenure. Therefore, leaders must consider forces that are willing and able to carry out their repressive goals, but ideally choose a security apparatus that will not be closely associated with the regime. This balance of loyalty and covert ties is not easily achieved with traditional security forces such as the military, national police, presidential guard, or gendarmerie. This balancing of priorities can present a conundrum for leaders who wish to repress to achieve their goals but need a unique type of security apparatus to do so on their behalf (Eck 2015). One solution to this issue is for leaders to create or align with alternative security forces, such as PGMs. PGMs offer unique benefits to leaders and can be a useful security force multiplier. Entities such as militias, death squads, paramilitaries, irregular forces, vigilantes, and some private military companies are examples of PGMs. The literature generally considers a PGM to have four characteristics: a group must be sponsored by the government or demonstrate a pro-government position, they must be separate from regular state forces, they must be organized in some capacity, and they must be armed (Carey and Mitchell 2012). However, not all PGMs are created equal. The utility a leader can derive from a PGM may vary depending on certain characteristics of the militia. There are a wide variety of types and extent of connections to the government. Some PGMs may have more covert ties to the leader and only be connected by the provision of weapons and outsourcing of tasks. On the other hand, other groups may be closely linked to the military or police and operate just beyond the bounds of an official security apparatus. This makes them highly visible and easily connected back to the government. 57 Because leaders generally want to avoid blame for their repression, turning to groups with looser links and connections to the regime would be a better strategic move. This looser connection makes it more difficult for both the international and domestic communities to fault the leader for using the group for repression. In addition, the leader can achieve violence beyond what their normal security forces can achieve, as they are very closely tied to the regime (Eck 2015). While semi-official PGMs can still be violent actors, their more obvious connection to the state makes them less useful to deploy for repression if leaders want to avoid accountability for their actions. Therefore, informal PGMs will not only carry out more violence for leaders but they will also provide plausible deniability for their actions. Several informal PGMs demonstrate how they are a better strategic choice for leaders due to their repressive abilities coupled with their distance from the regime. One such PGM is the Beli Orlovi. The Beli Orlovi, or White Eagles, was created in Yugoslavia in 1991 and operated as an informal PGM until 2003. The group largely targeted civilians and aid workers and were more repressive in their earlier years. For instance, in 1993 it was reported that the PGM was responsible for massacres in Bosnia and ransacked offices of media and political opposition figures in Belgrade (Carey and Mitchell 2014b). Overall, this group was fairly useful to the government and helped achieve their political aims. There were also hints of the group being tied to the government, specifically the former President of Serbia, Slobodan Milošević. The same report that indicated the group had been used to kill civilians and ransack opposition offices stated that “Milošević could halt their activities with a wink” (Carey and Mitchell 2014b). Furthermore, additional reports indicated they usually cooperated with the federal army and local territorial defense units, but “carefully retaining an independent command” (Carey and Mitchell 2014b). Therefore, while the group had links to the 58 government and formal security forces controlled by the state, it’s distance and therefore deniability were evident. Despite being a useful repressive tool, the government denied and clearly distanced themselves from the Beli Orlovi. For instance, the political party in charge of the group refuted all allegations that the Beli Orlovi were associated with violent attacks in 1991 (Carey and Mitchell 2014b). In addition, in 1992, the regional committee of the Serbian National Resistance party issued a public statement saying the PGM should “shoulder responsibility for the atmosphere of fear and uncertainty in the town of Bijelo Polje” (Carey and Mitchell 2014b). Out of fear the PGM may turn on the government and implicit them in human rights abuses, the Yugoslav Serb government jailed many members of the White Eagles in 1999 so they could not testify to international tribunals about war crimes (Carey and Mitchell 2014b). This example of the Beli Orlovi PGM in Yugoslavia demonstrates how leaders turn to groups with covert ties to help them achieve their political goals. These groups cannot always be officially tied back to the leader with certainty, which gives leaders the ability to avoid blame for their actions. Because leaders can deny the links to the group and their actions, they will choose to use these informal groups for violence more often than more closely linked militias. Therefore, informal PGMs serve as a useful tool of repression for state leaders. This leads to my first hypothesis: H1: Informal PGMs will be more repressive than semi-official PGMs. In addition to identifying the type of link to the government, other characteristics can provide information about the nature of the connection between PGMs and their respective governments. By exploring additional ways to observe the strength and nature of the relationship between the government and PGM, we can better understand characteristics that may make one 59 type of PGM more repressive than another. Furthermore, analyzing the additional features of these militias affords us the ability to observe the multi-dimensional nature of the relationship between these groups and their sponsoring governments. One such characteristic is an alliance between the PGM and formal security forces, such as the military. Some PGMs perform joint exercises with state forces, cooperate in repressive activities, or simply work as force multipliers. By joining in these security tasks with official security apparatuses, PGMs can benefit the official apparatuses by acting providing force multiplier support to boost troop numbers. Previous literature has also highlighted the logistical and local knowledge that PGMs may offer above and beyond that of the regular military, such as the case of the Arrow Boys in Uganda (Clayton and Thomson 2017). In return, PGMs may be trained by government forces and receive training. Working alongside official forces can provide PGMs with logistical and organizational support, additional and possibly more sophisticated weapons, and legitimacy to missions and security tasks. When PGMs act alongside the official forces, they may also get cover for their abuses. For instance, if a leader wants a particular civilian population repressed, they may deploy the PGM to carry out the abuses and simultaneously deploy their military to control or intimidate any outsiders or observers from intervening to stop the repression. Therefore, when cooperating and aligning with other forces, PGMs will have more repressive capacity and increased plausible deniability for the leader due to confusion of who is responsible for the abuses. The alliance between the Rondas Campesinas and the military in Peru exemplify how this relationship may work and lead to more violence by PGMs. The Rondas Campesinas PGM began in March of 1983 as a grass roots iniative to combat cattle thieves. The alliance between the PGM and the military was documented in several instances. For example, the Washington 60 Post reported in 1985 that the PGM was “encouraged and sometimes ordered by the armed forces” and in 1990 was called the centerpiece of the Peruvian military’s war against the Shinging Path (Carey and Mitchell 2014g). In addition, the ability for the PGM to carry out violence while allied with official forces is also evident. For example, the New York Times reported that “human rights groups assert that the army poorly supervises the militias, allowing them to veer into vigilantism, the settled of old scores and the renewal of land fights” (Carey and Mitchell 2014g). Therefore, governments, like Peru’s, are able to deploy allied forces of PGMs and their formal security forces, like the military, to achieve more repression. Additional PGM characteristics that can demonstrate collaboration with the government or its forces also include acknowledgement by the government, being trained by government forces, receiving a regularly funded salary, and being heavily armed. First, being acknowledged by the government is different from the government connection distinction, as it captures an additional degree of connection to the government. If a leader steps forward and publicly connects themselves to a PGM, they no longer have the ability or as much ability to plausibly deny the relationship in the future. One example of a government leader acknowledging a PGM comes from Tanzania. In 1984, the Minister of Home Affairs, Ndugu Salimin Amour “called upon all members of the traditional defence force, known as Sungu Sungu, to co-operate with other security organs in the country in maintaining security. Amour said that the Government recognized and appreciated the tasks carried out by units of the traditional defence force in ensuring the security of the people and their property” (Carey and Mitchell 2014h). 61 Next, PGMs that train alongside government forces could learn techniques to make them more efficient units. Furthermore, they could develop stronger relations with official government forces that would allow them to cooperate more in the future. Lastly, receiving a regular salary from the government also can demonstrate collaboration with the government. This more overt action of states tying themselves to PGMs with paper trails is a riskier move for leaders. However, it may also help keep PGMs loyal and fighting on their behalf. Similarly, heavily arming PGMs can prove another meaningful collaboration. PGMs that are considered heavily armed may have access to aircrafts, heavily artillery, such as grenade launchers, or an excessive supply of weapons, with a dichotomous indicator. Presumably, a group with access to more weapons will be able to repress more, as it would enhance their capabilities and capacity. Furthermore, leaders that arm PGMs with heavy machinery or provide excessive amounts of artillery demonstrate their trust in the PGM to carry out actions on their behalf. Overall, these covert means of collaborating with governments or their formal security forces may lead to more repression. This leads to the second hypothesis: H2: PGMs that collaborate with the government or state forces will be more repressive than other PGMs. The visibility of a PGM also can impact how a leader utilizes the group for repression. Ideally, most leaders would like to avoid ties back to highly repressive groups and therefore also escape the blame for their abuses. Therefore, using groups that are well-known and more established in a community may be riskier, as the population may be more familiar with their uniforms or members of the militia. On the other hand, lesser known or newer groups may be an attractive option for leaders wishing to use them as a repressive agent. The group may not have 62 the same notoriety and may be less recognizable to observers and the community. This covert nature may allow the leader to carry out more intense or wider spread abuses than they would be able to with a group well-known by the locals. In addition to the Colombia example, this variation in violent activity being dependent upon the longevity of a PGM is demonstrated by the PGMs in Sierra Leone. Two informal PGMs were created in 1999 in Sierra Leone, the West Side Boys and the Bokkie Crew. The West Side Boys only existed for two years, but it was a highly repressive force. A 2000 State Department Report called out the PGM for perpetrating numerous human rights abuses. These abuses included extrajudicial killings, abductions, “deliberate mutilations”, and sexual violence (State Department 2000). On the other than, the Bokkie Crew is still in existence as of 2008. While mentioned in a few news reports in 1999 and 2000 for firing on civilians from helicopters when unsure of their targets, the Bokkie Crew has not been mentioned in any human rights reports (Carey and Mitchell 2014i). Therefore, I expect that the lifespan of a PGM may be an influential characteristic to determine their repressive behaviors. Groups that have been in existence longer will be less repressive, as they do not have the same plausible deniability feature for leaders. However, groups with shorter lifespans may be more useful for the worst abuses. This leads to my third hypothesis: H3: PGMs with shorter lifespans will be more repressive than PGMs with longer lifespans. Lastly, the agenda of the group may be an important indicator. More specifically, groups that espouse an explicit anti-rebel agenda and are dedicated to acting mainly as a self-defense force may act differently compared to groups that are created to repress civilian populations. 63 PGMs that are anti-rebel forces are often times created as self-defense forces by locals from an area that has a security vacuum or is not be sufficiently defended by other state forces. Locals taken an iniative or respond to government requests for “volunteers” to defend their region from insurgencies. Therefore, their attention is not necessarily focused on abusing civilians but instead on preserving their territory and driving out rebel forces. While some civilians may be unintended casualties of the violence, overall it is expected that these groups would be less violent toward non-combatant populations. Furthermore, ties to the community may be stronger in anti-rebel forces. Previous literature has identified the link to society being an important factor in determining the causes and consequences of control of PGMs (Carey and Mitchell 2017). For instance, Stanton (2015) finds that civilians are safer with militias that share an ethnic identity. Therefore, stronger community connections and sometimes being mobilized by the community, anti-rebel PGMs may be less likely to attack their own communities. In addition, they may rely upon information from local sources or resources to aid in their fights against the insurgency (Carey and Mitchell 2017). Several PGMs have demonstrated how having an explicit anti-rebel agenda can lead to fewer instances of violence against civilians. First, the Executive Outcomes PGM in Sierra Leone was an informal PGM from 1995 to 1997. They were awarded a military contract to train the national army to fight an insurgency, indicating an anti-rebel agenda or mission. The militia was never accused of human rights abuses in any news or human rights reports (Carey and Mitchell 2014d). Similarly, the Rhino Defence Force in Uganda was an anti-rebel PGM created using the same model as the Arrow Boys militia to fight the Lord’s Resistance Army insurgency. The group was created in 2003 as a semi-official PGM. There was one instance of violence in 64 2006 committed by a single militia-member killing internally displaced people in a camp in the Lira District in Northern Uganda (Carey and Mitchell 2014f). Lastly, Youth Awareness is an informal PGM in Madagascar. The group was created in 1984 to protect the state from insurgencies and has never been accused of human rights abuses or harming civilians (Carey and Mitchell 2014k). PGMs with explicit anti-rebel missions or agendas like these tend to focus only on protecting the state or borders. Their focus is not necessarily to attack civilians, leading to fewer violent attacks. This leads to the fourth hypothesis: H4: PGMs with anti-rebel agendas will be less repressive than PGMs that do not espouse or demonstrate explicit anti-rebel agendas. Data This analysis introduces a novel dataset, the Militia Repression Dataset (MRD) to capture the targets, severity, and types of repression carried out by PGMs as well as characteristics of groups. Data for MRD was coded using news reports and human rights reports from Human Rights Watch, State Department, and Amnesty International when human rights abuses were tied to specific militias. MRD presents a global sample for all PGMs as identified in the Pro- Government Militia Database (PGMD) from 1989-2007 (Carey, Mitchell, and Lowe 2013)7. The unit of analysis for all models is the PGM-year. There are 2156 observations in each model8. MRD follows the current literature and extant repression datasets to identify a variety of types or categories of repression. Similar to the CIRI Human Rights Data Project (CIRI), MRD identifies four types of repression, including torture, imprisonment or kidnapping, extrajudicial killings, and disappearances with dichotomous indicators (Cingranelli, Richards, and Clay 2014). 7 The Original PGMD is currently being updated to include all PGMs from 1981-2014. Once this update of the PGMD is complete to establish a universe of PGM cases, I plan to update MRD accordingly. 8 Note that the number of observations in the models is less than the total number of observations in the dataset, as observations with missing values for some control variables drop out of the analysis. 65 While these four categories of abuses are useful to classify types of abuses, often times PGMs were documented engaging in low levels of violence with civilians that would not meet the standard definitions of these categories. For instance, PGMs in Indonesia in 1999 were often documented to be abusing civilians with sticks, however, it did not qualify as torture. This low- level violence is still important to capture in this dataset to better understand the variety of abuses PGMs may carry out, as it still signals a lack of respect for human rights. To capture this violence, I also include a dichotomous indicator of any violent action by a PGM in a given year. Therefore, this violent action variable captures both high levels of abuse when any of the four main types of violence are recorded but also low-levels of violence that did not meet those standards. Lastly, MRD also includes a scale of repression to measure the sever that follows other scales of repression in the literature, such as the Political Terror Scale (PTS; Gibney et al. 2018). This scale measures the intensity and severity of abuse committed by the PGM in a single year. Values of 0 indicate no abuse occurred in that given year. A value of 1 indicates a single incident affecting only one person. A value of 2 indicates that a “rare” or “isolated incident” of abuse occurred that affected about a dozen or fewer people. Next, a value of 3 indicates an abuse that is described as happening “occasionally”, “sometimes”, and affected dozens or a significant number of victims. A value of 4 indicates numerous abuses that occurred more regularly. These incidents may be described as “at times” or “several” instances of abuse and affect a hundred or so people. Lastly, a value of 5 indicates the most severe abuses that are considered “widespread” or “systematic” and affect more than about a hundred people. Table 3 shows the summary statistics for these two dependent variables and Table 4 displays the distribution of the repression scale dependent variable. 66 Table 3: Summary Statistics for Dependent Variables Variable Mean Standard Deviation Variance Minimum Value Maximum Value Violent Action .338 .473 .224 Repression scale .756 1.179 1.391 0 0 1 5 Table 4: Distribution of PGM Repression Variable PGM Repression 0 1 2 3 4 5 Total Frequency 1,642 241 315 241 68 7 2,514 Percentage 65.31 09.59 12.53 09.59 02.70 00.28 100 Hypothesis 1 tests the effect of the link or connection between the government and PGM on PGM repression. There are several variables to test this connection. First, the literature generally separates PGMs into two types dependent upon their relationship to the government (Carey and Mitchell 2012). Semi-official PGMs enjoy a closer and much more visible link to the government. On the other hand, informal PGMs have more covert connections to the government that are not formally acknowledged. Therefore, one of the main explanatory variables indicates the type of government link to PGMs. This variable is a binary indicator of informal PGMs from PGMD, where the variable is equal to 1 for groups identified as informal PGMs and 0 for semi- official PGMs. To test hypothesis 2 regarding collaboration with the government and its forces, I use several variables from MRD and the PGMD. First, I use a dichotomous indicator of whether a PGM has been acknowledged by a government or military leader in a public format, such as in a news interview or questioning about their relationship. security of the people and their property” 67 (Carey and Mitchell 2014h). Next, I use a dichotomous indicator of PGM alliance with government forces to test government connection. This alliance variable is equal to 1 when there is evidence of a sustained relationship with a recognized government security force, such as the police or military. One example of this is the Edo State Vigilante group operating in Nigeria starting in 2000 as a semi-official PGM. This group was allied with the police and was originally created to help the police combat crime in the area (Carey and Mitchell 2014c). Third, I use a dichotomous indicator of government training from PGMD. Next, some PGMs are funded and paid regular salaries by their supporting governments. Therefore, I employ a dichotomous indicator if there was evidence of groups being paid regularly by the government or military. Next, I use explore whether a PGM is heavily armed, meaning they have access to aircrafts, heavily artillery, such as grenade launchers, or an excessive supply of weapons, with a dichotomous indicator. Lastly for the second hypothesis, I create an index using five of the above indicators to capture the degree of collaboration the PGM has with the government or its official security apparatuses, such as the military. This variable an additive index created from the dichotomous indicators of government acknowledgement, alliance with government forces, receiving government training, receiving a regularly funded salary, and being heavily armed. Values in this index range from 0, meaning no government connection, to 5 indicating the PGM-year observation is equal to 1 for each indicator. The distribution of this link index variable is displayed in Table 5 below. 68 Table 5: Distribution of Collaboration Index Variable PGM Link 0 1 2 3 4 5 Total Frequency 136 515 1169 561 127 6 2514 Percentage 05.41 20.49 46.50 22.32 5.05 00.24 100 MRD also offers indicators to test the remaining two hypotheses. In order to test the third hypothesis, I use the lifespan of the PGM. This variable ranges from 1-21 years with a mean of 12.931 years. Lastly, I use a dichotomous indicator of an explicit anti-rebel agenda or demonstration of anti-rebel policies to test Hypothesis 4. This variable is equal to 1 when a PGM targets rebel groups or rebellious activities. The West Side Boys in Sierra Leone demonstrate a group that is coded as anti-rebel. Their main purpose was to ally with government forces to fight against insurgencies and the Revolutionary United Front rebels (Carey and Mitchell 2014j). Table 6 displays the summary statistics for all the independent variables. 69 Table 6: Summary Statistics for Independent Variables Variable Informal PGM Acknowledged Allied Trained Funded Salary Heavily Armed Collaboration Index Lifespan Anti-rebel Mean .531 .441 .413 .642 .215 .189 2.018 12.931 .633 Standard Deviation Variance Minimum Value Maximum Value .499 .497 .493 .479 .411 .391 .932 5.888 .482 .249 .247 .243 .230 .169 .153 .868 34.668 .232 0 0 0 0 0 0 0 1 0 1 1 1 1 1 1 5 21 1 In addition to the above variables, I employ a number of control variables also from MRD to help explain why we may observe varying levels of PGM repression. First, I control for whether the group had a strong and organized internal structure. While some PGMs are loosely tied together with no real organizational capacity, others are highly efficient and organized entities with multiple wings and sub-units. Groups with an internal organizational structure may be more repressive due to their increased capacity. In addition, I control for groups tied to drug cartels or the drug trade. Most of these PGMs are in Latin America or the Middle East due to the nature of where drugs are generally harvested. Lastly, I control for the number of PGMs in existence in the state in a given year. If a leader has access to multiple PGMs, they can distribute their repressive needs across multiple groups to lessen their liability with one group. This would mean each group carries out less repression because the responsibility is shared across multiple entities. I also control for several country-level characteristics. First, I include the physical integrity rights index from CIRI to account for state-level repression, which ensures that PGM 70 violence is not simply a proxy for all state-level violence (Cingranelli, Richards, and Clay 2014). Second, I include a count for the number of coup attempts in a given year. If a leader feels their tenure is threatened, they may be more likely to resort to violence to quell the dissent (Nordas and Davenport 2013). In addition, I control for the presence of civil war or civil conflict, the total number of coup attempts (Powell and Thyne 2011), Polity score based on the 21-point scale (Marshall et al. 2016), the log of the gross domestic product (GDP) in current US dollars, and the log of the total population from the World Development Indicators (The World Bank 2016). I also use a binary variable to control for the Cold-War era. Lastly, I include a cubic spline to account for changes over time. Methods Model 1 explores the effect of informal PGMs on the overall level of repression to test the intensity of repression carried out by the different types of PGMs. For this model, I use ordered logistic regression, as the dependent variable is the scale of PGM repression. Model 2 explores the effect of government and PGM ties on the dichotomous indicator of violent actions to test if informal PGMs with looser state ties are more violent and repressive than their counterpart of semi-official PGMs. I explore informal PGMs as the type of relationship between a government and PGM is a binary choice between informal or semi-official. Due to the dichotomous nature of the violent actions variable, I use logistic regression for this model. Models 3-8 explore the effects of the characteristics identified in hypothesis 2 on the likelihood of violent attacks. To test hypothesis 2, Model 3 analyzes the effect of alliances between PGMs and formal security forces on the likelihood of a violent attack. Model 4 explores the role of governments acknowledging PGMs on the chance of violence attacks. In Model 5, I test if government training of PGMs influence their violent attacks. Next, Model 6 analyzes how 71 receiving a regular salary from the government may affect PGMs carrying out violent attacks. Model 7 explores how PGMs that are heavily armed could be more likely to carry out violent attacks. Lastly, Model 8 employs the collaboration index variable to test if having more degrees of collaboration with the government can increase the likelihood of PGMs carrying out violent attacks. To test hypothesis 3, Model 9 explores the effect of the lifespan of the PGM on the likelihood of carrying out a violent attack. Lastly, Model 10 analyzes how having a PGM with an explicit anti-rebel agenda can influence the likelihood of violence by PGMs for hypothesis 4. Models 3-10 use the same dependent variable of violent attack, so I use logistic regression for each analysis. For all models, I cluster standard errors on the PGM to account for non-independence of observations for the same militia over time. In addition, I lag all independent and control variables by one year. Results The results from Models 1 and 2 show support for the first hypothesis and indicate that informal PGMs with looser ties to the state are more likely to commit acts of violence and perpetrate worse human rights abuses than other types of militias. The evidence demonstrates that groups that distance between the government and militia does allow for groups to be more repressive and abusive. The results are displayed in Table 7. 72 Table 7: Main Models 1-2: Repression and Violent Attack Dependent Variables Variables Informal PGM Internal Structure Drug Cartel Ties Number of PGMs Coup Attempts Polity Log GDP Log Population Physical Integrity Rights Civil War Cold War Time 1 Time 2 Time 3 Time 4 Time 5 Number of observations Pseudo R2 AIC BIC Model 1 – Repression Level Model 2 – Violent Attacks 1.072 *** (.252) .707 ** (.280) .912 (.555) -.043 * (.023) .417 ** (.189) .002 (.019) .000 (.000) .000 (.000) .000 (.001) .319 (.195) -.041 (.177) -.069 (.075) .548 (.643) -1.647 (2.612) 1.101 (4.165) .425 (4.070) 2156 .050 4637.298 4750.818 .980 *** (.256) .402 (.253) .811 (.628) -.029 (.021) .431 *** (.159) -.001 (.019) .000 (.000) .000 (.000) .000 (.001) .200 (.192) -.025 (.204) -.064 (.075) .524 (.632) -1.629 (2.616) 1.246 (4.321) .097 (4.357) 2156 .064 2624.200 2715.016 * p < .10, ** p < .05, *** p < .01 Clustered standard errors are in parentheses. 73 The coefficient for informal PGMs in Model 1 is positive and statistically significant. This indicates that informal PGMs are more likely to perpetrate higher levels of abuse than other types of militias. Similarly, in Model 2, exploring the likelihood of violent attacks occurring as the dependent variable, the coefficient is positive and significant. This indicates that informal PGMs are more likely to carry out acts of violence as compared to semi-official groups. These findings demonstrate support for the first hypothesis that informal PGMs are likely to carry out more repression compared to semi-official PGMs. Figure 5 displays the predicted probabilities for Model 2. The predicted probability of an informal PGM carrying out an act of violence is just about .439 or 43.9%, which is significantly higher than the .227 or 22.7% predicted probability for other types of militias. Figure 5: Predicted Probability of Violent Attacks by PGM Type 74 The results from hypothesis 2 exploring how PGM and government collaborations can influence violent attacks are shown in Table 8 for Models 3-5 and Table 9 for Models 6-8. 75 Table 8: Main Models 3-5: Alliances, Acknowledgement, and Training Variables Allied Acknowledgement of PGM Government Training Internal Structure Drug Cartel Ties Number of PGMs Coup Attempts Polity Log GDP Log Population Physical Integrity Rights Civil War Cold War Time 1 Time 2 Time 3 Time 4 Time 5 Constant Number of observations Model 3 - Alliances .358 * (.208) .021 (.227) .839 (.559) -.009 (.022) .525 *** (.152) .001 (.019) .000 (.000) .000 (.000) .001 (.001) .121 (.190) -.022 (.207) -.032 (.073) .393 (.617) -1.300 (2.528) 1.177 (4.131) -.422 (4.158) 61.697 (145.888) 2156 76 Model 4 – Acknowledgement -.514 ** (.242) .304 (.256) .863 (.584) -.016 (.021) .513 *** (.158) .003 (.019) .000 (.000) .000 (.000) .001 (.001) .151 (.189) -.022 (.210) -.019 (.075) .331 (.630) -1.132 (2.576) .931 (4.191) -.038 (4.200) 36.772 (148.828) 2156 Model 5 – Training -.212 (.202) .154 (.325) .875 (.587) -.011 (.021) .497 *** (.152) .004 (.018) .000 (.000) .000 (.000) .001 (.001) .194 (.187) -.045 (.207) -.030 (.073) .362 (.617) -1.156 (2.526) .889 (4.116) -.082 (4.134) 58.301 (146.179) 2156 Table 8 (Cont’d) Pseudo R2 AIC BIC .035 2705.870 2796.686 .040 2692.177 2787.993 .032 2714.059 2804.875 * p < .10, ** p < .05, *** p < .01 Clustered standard errors are in parentheses. 77 Table 9: Main Models 6-8: Funded Salaries, Heavily Armed, and Collaboration Index Variables Model 6 – Funded Salaries Model 7 – Heavily Armed Funded Salaries Heavily Armed Collaboration Index Internal Structure Drug Cartel Ties Number of PGMs Coup Attempts Polity Log GDP Log Population Physical Integrity Rights Civil War Cold War Time 1 Time 2 Time 3 Time 4 Time 5 Constant Number of observations .285 (.255) .075 (.229) .953 (.585) -.014 (.021) .524 *** (.152) .003 (.018) .000 (.000) .000 (.000) .001 (.001) .169 (.189) -.056 (.208) -.037 (.074) .391 (.617) -1.240 (2.526) .974 (4.122) -.125 (4.144) 72.709 (147.079) 2156 .497 * (.285) .062 (.241) .877 (.576) -.010 (.021) .518 *** (.152) .002 (.018) .000 (.000) .000 (.000) .001 (.001) .176 (.186) -.063 (.211) -.031 (.074) .344 (.624) -1.039 (2.555) .672 (4.166) .090 (4.186) 60.058 (147.011) 2156 78 Model 8 – Collaboration Index .195 ** (.099) .063 (.221) .864 (.563) -.013 (.022) .528 *** (.154) .001 (.018) .000 (.000) .000 (.000) .001 (.001) .127 (.183) -.052 (.206) -.031 (.073) .371 (.617) -1.215 (2.540) 1.031 (4.167) -.232 (4.197) 60.781 (144.806) 2156 Table 9 (Cont’d) Pseudo R2 AIC BIC .033 2712.711 2803.527 .036 2704.028 2794.844 .036 2703.221 2794.037 * p < .10, ** p < .05, *** p < .01 Clustered standard errors are in parentheses. Of the six models testing hypothesis 2, four of the main independent variables produced statistically significant coefficients. Overall, these models demonstrate support for hypothesis 2 and show that PGMs that collaborate and work with state forces will be more violent than those that act independently. PGMs working alongside official government forces may be more repressive for several reasons. First, multiple groups working together on security tasks, especially those involving repression, may be overwhelming and confusing for civilians or those observing these events. It may be difficult for observers to ascertain which forces are official state forces and which are PGMs. This may in part be due to some PGMs being outfitted with official looking uniforms, such as the Southern Tribal Militia operating in Afghanistan starting in 2006. Second, when PGMs and state forces perform joint operations, as Death to Kidnappers did in Colombia, PGMs act as the repressing agent while official security forces keep anyone trying to interfere at bay. Essentially, the PGM represses while the official security force, like the military, keeps watch to make sure the PGM can carry out its job. Therefore, this joint operation and alliance allows PGMs to repress and the military to technically avoid blame for the actions. Lastly, PGMs working alongside official apparatuses can sometimes be used as a force multiplier, which may also explain why they are more repressive than groups that work independent from state forces. Since PGMs can perform similar tasks to state forces, they can be a cheap option for leaders to increase their force numbers when considering deploying them for repressive means. 79 Model 3 analyzes how being allied with government forces can influence repression by PGMs. The coefficient a PGM being allied with government forces is positive and statistically significant. Therefore, results for this model indicates that PGMs allied with government forces will lead to significantly more repression than having entirely independent forces. Next, Model 4 exploring PGMs being acknowledged by the government also produces a statistically significant result. The negative coefficient indicates that PGMs with some type of official government acknowledgement will be less repressive. This finding also follows the logic of the main argument that leaders want distance from PGMs and will not tie themselves to highly repressive groups. Therefore, groups that carry out the most violence will not be acknowledged by governments, to avoid blame and responsibility for their actions. Interestingly, Model 5 exploring the effects of government training and Model 6 analyzing the effects of salaries for PGM violence do not produce significant results. This may be because both activities are highly visible to observers and civilians. Funding PGMs is not necessarily a covert activity and may produce a paper trail that leaders want to avoid. If they are tried in court or investigated for these abuses, this evidence could be used against them in the future. While alliances produced a result indicating that PGMs will be more violent when working with state forces, trainings produced a different result. Alliances may be a one-time venture or exercise, whereas training may take place over a longer period of time. This leaves room for civilians or others to observe this training. Furthermore, some military exercises may be public and would be difficult for leaders to deny if questioned. Model 7 examining the role of being heavily armed in PGM violence produces a positive and significant coefficient. This also demonstrates support for hypothesis 2 that increased 80 government collaboration, in the form of supplies and weapons, leads to more violence. Since leaders have access to weapons for their military, police, presidential guards, and other formal units, it would be easy for them to also distribute these same supplies to their PGMs. In addition, leaders may have some plausible deniability and can claim the militia stole the weapons or used them without their permission. This would help leaders avoid full culpability for the violence they commit, while ensuring the group has the ability and capacity to be a violent force. Next, Model 8 uses the collaboration index variable to determine how an increasingly collaborative relationship between a PGM and its government or government forces may impact violence. The coefficient for the index variable is positive and significant indicating that as PGMs have a more collaborative relationship with the state, they will be more repressive. Given that this index is a composition of other independent variables that were statistically significant, this makes sense and again demonstrates support for hypothesis 2. Figure 6 displays the predicted probability of a violent attack being carried out by a PGM by the collaboration index variable. As the value of the index variable increases, indicating a stronger collaboration with the government or its official security forces actors, the predicted probability of violent attacks occurring significantly increases. Each additional measure of collaboration with government actors presents a statistically significant at the .10-level increase in the probability of a PGM being violent in a given year. 81 Figure 6: Predicted Probability of Violent Attack by Collaboration Index Next, Table 10 displays the results for Model 9 testing hypothesis 3 regarding PGM lifespan and Model 10 testing hypothesis 4 regarding anti-rebel PGMs. 82 Table 10: Main Models 9-10: Lifespan and Anti-Rebel Variables Lifespan Anti-Rebel Internal Structure Drug Cartel Ties Number of PGMs Coup Attempts Polity Log GDP Log Population Physical Integrity Rights Civil War Cold War Time 1 Time 2 Time 3 Time 4 Time 5 Constant Number of observations Pseudo R2 AIC BIC Model 9 - Lifespan -.072 *** (.017) .113 (.237) 1.074 * (.557) -.032 (.020) .455 *** (.158) -.004 (.019) .000 (.000) .000 (.000) .001 (.001) .201 (.186) -.022 (.219) .053 (.082) .001 (.653) -.285 (2.620) .385 (4.198) -.047 (4.187) -104.935 (164.195) 2156 .058 2641.347 2732.164 Model 10 – Anti-rebel -.436 * (.236) .176 (.236) .924 (.595) -.014 (.021) .506 *** (.151) .004 (.018) .000 (.000) .000 (.000) .001 (.001) .374 (.205) -.073 (.207) -.033 (.075) .416 (.621) -1.478 (2.538) 1.538 (4.134) -.733 (4.153) 63.929 (148.880) 2156 .036 2702.169 2792.985 * p < .10, ** p < .05, *** p < .01 83 Clustered standard errors are in parentheses. The coefficient for PGM lifespan is negative and significant in Model 9. This means that as PGMs exist longer, they are less likely to carry out acts of violence. This demonstrates support for hypothesis 3 that PGM that have a longer lifespan will be less violent than PGMs that are short-lived. Figure 7 displays the predicted probability of carrying out violent attacks by PGM lifespan for the full range of lifespan values in MRD. As the lifespan of the PGM increases, the chance of carrying out a violent attack against civilians decreases each year. The change in the predicted probability for attacks significantly decreases at the .001-level for each additional year the PGM exists. The predicted probability of carrying out a violent attack for PGMs that only live one year is over 55.3%. However, the likelihood drops by over 60% to just a 21.6% likelihood for groups surviving to the maximum value of 21 years. 84 Figure 7: Predicted Probability of Attack by PGM Lifespan One case that exemplifies the influence of PGM lifespan is Indonesia in 1999. In 1999 there were 20 informal and eight semi-official PGMs in existence. Of those 20 informal PGMs, 14 were created between 1998 and 2000. However, by 2000 there were only 12 informal and five semi-official PGMs, as eight of the informal PGMs created in 1999 only existed for that year. By 2005, the number of informal PGMs dropped by half of the 1999 amount to only 10 and there were only five semi-official PGMs. This demonstrates a high number of PGMs, especially informal PGMs, being created and dissolved within a very short time frame. Many of these short-lived informal PGMs, such as the Ahi, Aitarak, Besi Merah Putih, Dadurus Merah Putih, Laksaur, Peace Force and Defender of Integration, Pro-Integration Fighters PPI, and the Sakunar were accused of abuses human rights in some capacity in 1999. Most of these groups existed for only 1999 or just a few years but were able to carry out acts of 85 repression during this time in Indonesia. Therefore, the case of Indonesia highlights how leaders may decide to create or align with multiple PGMs within a short time frame to achieve their repressive aims and then dissolve or dismantle them shortly after to either avoid blame or the possibility of losing control of the group in the future. Lastly, the results from Model 10 indicate support for hypothesis 4 regarding anti-rebel PGMs being less abusive than PGMs without explicit anti-rebel iniatives. The results demonstrate that having an explicit anti-rebel iniative is correlated with fewer violent attacks, which demonstrates support for hypothesis 4. Many anti-rebel forces begin as a self-defense mechanism. These types of militias have been known to arise in security vacuums where the national military is unable or unwilling to offer protection. Since these militias are focused on solely driving out insurgencies from their homes, they may not choose to target other civilians in the area. This is exemplified by the case of the People’s Militia in Myanmar. The People’s Militia began in 1981 strictly as a defense militia with ties to the military and government. Their main purpose was to protect the civilians and state against insurgencies, especially from the Burma Communist Party (Carey and Mitchell 2014e). Their creation was intended to be the “formation of a people’s militia for the defence of the state” (Carey and Mitchell 2014e). Interestingly, the group has never been accused of human rights violations or any violence against civilians. They have only ever targeted rebels and never targeted civilians, political opposition, or other PGMs in any of their missions. Figure 8 shows the first differences for the main explanatory variables in Models 2-10. The first difference is the change in probability of a violent attack as the variable of interest increases from their lowest value to their highest value in order to provide insight about the 86 substantive relationship between the independent variable and violent attacks for each model. Four of the nine points indicates a significant effect on the predicted probability of a violent attack occurring, meaning the main independent variables in each model have a significant effect on PGMs committing acts of violence. The change for informal PGMs in Model 2, PGM acknowledgement in Model 4, the collaboration index from Model 8, and PGM lifespan in Model 9 are all statistically significant. Conversely, the change for the remaining five models are not significant, meaning their substantive impact on the likelihood of PGMs carrying out violent attacks is not significant. Figure 8: First Differences for Models 2-10 87 The coefficients for most control variables were not statistically significant and did not provide interesting insight to the repressive behaviors of militias. However, the coefficient for coup attempts was positive and significant across all models. This indicates that when leaders are threatened by the possibility of a coup, they are more likely to respond with acts of violence by PGMs. This is an interesting finding that should be further explored in the future to better understand how leaders use PGMs to coup-proof their regimes. Conclusion These analyses presented and utilized new data from MRD to better understand how leaders deploy PGMs for repressive tasks. This novel dataset is the first attempt to capture PGM- specific human rights abuses, which provides subnational information on how security apparatuses are used by leaders. While state-level indicators of repression and human rights abuses have greatly contributed to our understanding of treatment of civilians around the world, advancing our understandings with actor-level data can help solve issues of accountability and prevention of abuses in the future. Overall, these results demonstrate that while leaders use PGMs for repression, their coercive capacity and utility varies across the type of PGM. While informal PGMs with loose state ties have been found to be more repressive, PGMs with more codified and formal relations with the government are less repressive. In addition, militias allied with government forces are more likely to be repressive than those without alliances. I argue this is due to the mutually beneficial relationship between these groups when they cooperate. PGMs that have a longer lifespan are also found to be less repressive. While PGMs that exist longer are not as useful of a tool for repressive as they lessen a leader’s plausible deniability, PGMs that are around for shorter stints are found to be a more useful repressive tool because they are harder to trace and 88 identify. Lastly, PGMs with anti-rebel agendas are less repressive than others. This is possibly because their goal is not to repress civilians but instead to drive out insurgencies. Their targets and intended victims are other armed groups, not necessarily particular civilian populations, so they engage in less repression of civilians overall. Together, these findings demonstrate that leaders are strategic in their choices about how to deploy their repressive forces and may coordinate with or create PGMs to fit their repressive needs. The results are consistent with other findings in the literature that demonstrate that informal PGMs are associated with more state-sponsored violence (Mitchell et al. 2014). However, this analysis extends the argument from exploring levels of state-wide violence to violence of carried out directly by PGMs. By isolating repressive actions carried out by PGMs instead of inferring their patterns and levels of violence from state-level data, we can better understand the ways PGMs interact with civilians through repression and how their behaviors influence patterns of human rights abuses worldwide. This analysis also provides insights for the international community and policy makers regarding how to address state-sponsored violence against civilians. While leaders often have a number of regular security forces at their disposal, these findings indicate that leaders may look beyond the bounds of those regular forces to repress in alternative ways. They do so to avoid blame and punishment from the international community by denying links and ties to the PGMs and their repressive behaviors. This can allow leaders to avoid condemnation for these actions and skirt the blame. This presents a challenge for the international community to address leaders opting to create, engage with, and utilize alternative security forces like PGMs for their repressive benefits. More specifically, monitoring these groups and their behaviors has proven to be difficult due to their irregular nature. This can lead to a lack of information regarding the true 89 levels of abuse within a state. In the future, monitoring and data collection efforts by researchers, practitioners, and policy makers could consider expanding their understanding of state coercive capacities to extend beyond traditional forces. While these analyses provide some insight into how PGMs are used and what characteristics are associated with a higher likelihood of violence, there is still much more work to be done and topics to be explored. Our understandings of PGMs used for violence has improved in recent years, however, there is still room for improvement. MRD may be utilized to help address questions over what types of violence PGMs carry out. For instance, which types of PGMs are more likely to carry out certain types of abuses, such as torture or extrajudicial killings? What PGM characteristics best predict which militias will kidnap? In addition, MRD offers information on the targets of these abuses. Future research could analyze PGM characteristics associated with certain group-targets, such as civilians or the military. Lastly, MRD offers new and different wants to connect leaders to PGMs, such as the acknowledgement, funding, and alliance variables. More in-depth exploration of how leaders or governments are linked to these PGMs and the respective outcomes on violence would be a fruitful future research agenda. In conclusion, PGMs have been shown to be a unique repressive tool for regimes, especially informal militias. They can extend a leader’s ability to repress beyond the bounds of regular forces while acting as a scapegoat for blame for the abuses. The informal links to PGMs allows leaders to deny accusations from domestic and political actors regarding their behaviors. Therefore, PGMs are a strategic tool for leaders that would like to repress. However, these groups present new challenges to the international community and those tasked with addressing human rights abuses. 90 BIBLIOGRAPHY 91 BIBLIOGRAPHY Allansson, Marie, Erik Melander & Lotta Themnér. 2017. “Organized violence, 1989-2017.” Journal of Peace Research 54(4). Amnesty International .1986. “Colombia Annual Report.” Arjona, Ana .2017. “Civilian Cooperation and Non-Cooperation with Non-State Armed Groups: The Centrality of Obedience and Resistance” Small Wars & Insurgencies 28(4): 755-778. Babbitt, Eileen F. and Ellen L. Lutz. 2009. Human Rights & Conflict Resolution in Context: Colombia, Sierra Leone, & Northern Ireland. Syracuse, NY: Syracuse University Press. Carey S and Mitchell N. 2012. “Pro-Government Militias Database: Introduction to PGMD.v1.” Presented at ENCoRe Meeting, September, Oslo, Norway. Carey S and Mitchell N. 2014a. “Pro-Government Militias: Death to Kidnappers (Muerte a Secuestradores) (Colombia)”. https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/387/ Carey S and Mitchell N. 2014b. “Pro-Government Militias: Documentation for Beli Orlovi (White Eagles).” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/385/evidence/ Carey S and Mitchell N. 2014c. “Pro-Government Militias: Documentation for Edo State Vigilante.” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/409/evidence/ Carey S and Mitchell N. 2014d. “Pro-Government Militias: Documentation for Executive Outcomes.” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/70/evidence/ Carey S and Mitchell N. 2014e. “Pro-Government Militias: Documentation for People’s Militia.” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/132/evidence/ Carey S and Mitchell N. 2014f. “Pro-Government Militias: Documentation Rhino (Amuka) Defence Force.” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/200/evidence/ Carey S and Mitchell N. 2014g. “Pro-Government Militias: Documentation for Rondas Campesinas.” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/301/evidence/ Carey S and Mitchell N. 2014h. “Pro-Government Militias: Documentation for Sungu Sungu.” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/12/evidence/ 92 Carey S and Mitchell N. 2014i. “Pro-Government Militias: Documentation for The Bokkie Crew.” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/76/evidence/ Carey S and Mitchell N. 2014j. “Pro-Government Militias: Documentation for West Side Boys.” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/86/evidence/ Carey S and Mitchell N. 2014k. “Pro-Government Militias: Documentation for Youth Awareness.” https://militiasdb.sowi.uni-mannheim.de/militias-public/pgag/9/evidence/ Carey S and Mitchell N. 2017. “Progovernment Militias.” Annual Review of Political Science 20: 127-147. Carey S, Mitchell N, and Lowe W. 2013. “States, the Security Sector and the Monopoly of Violence: A New Database on Pro-Government Militias.” Journal of Peace Research 50(2): 249-58. Cingranelli, David L., David L. Richards, and K. Chad Clay. 2014. "The CIRI Human Rights Dataset." http://www.humanrightsdata.com. Version 2014.04.14. Clayton, Govinda and Andrew Thomson. 2017. “Civilianizing Civil Conflict: Civilian Defense Militias and the Logic of violence in Intra-State Conflict.” International Studies Quarterly 60(3) 1-12. Conrad, Courtenay R., Jillienne Haglund, and Will H. Moore. 2014. “Introducing the Ill- Treatment and Torture (ITT) specific allegation data.” Journal of Peace Research 51(3): 429- 438. Davenport, Christian. 1995. “Multi-Dimensional Threat Perception and State Repression: An Inquiry into Why States Apply Negative Sanctions.” American Journal of Political Science 39(3): 693-713. Davenport, Christian. 2007. “State Repression and Political Order.” Annual Review of Political Science 10: 1-23. Davis, David and Michael Ward. 1990. “They Dance Alone: Deaths and the Disappeared in Contemporary Chile.” Journal of Conflict Resolution 34: 499-75. Duvall, Raymond and Michael Shamir. 1983. “Governance by Terror.” In The Politics of Terrorism, ed. Raymond Duvall and Michael Stohl. 3rd edition. New York: Marcel Dekker. Easton, David. 1975. “A Re-Assessment of the Concept of Political Support.” British Journal of Political Science 5: 435-57. Eck K. 2015. “Repression by Proxy: How Military Purges and Insurgency Impact the Delegation of Coercion.” Journal of Conflict Resolution 59(5): 924-946. 93 Eck K and Hultman L. 2007. “One-sided violence against civilians in war: insights from new fatality data.” Journal of Peace Research 44(2): 233-246. Freedom House. 2020. “An annual study of political rights and civil liberties.” https://freedomhouse.org/report/freedom-world Gibney, Mark, Linda Cornett, Reed Wood, Peter Haschke, Daniel Arnon, and Attilio Pisanò. 2018. “The Political Terror Scale 1976-2017.” http://www.politicalterrorscale.org. Goldstein, Robert J. 1978. Political Repression in Modern America: From 1870 to the Present. Boston: Schenckman/G.K. Hall. Gupta, Dipak, Hanrder Singh, and Tom Sprague. 1993. “Government Coercion of Dissidents: Deterrence or Provocation.” Journal of Conflict Resolution 37: 301-39. Gurr, Ted. 1986. “The Political Origins of State Violence and Terror: A Theoretical Analysis.” In Governmental Violence and Repression: An Agenda for Research, ed. Michael Stohl and George Lopez, eds. New York: Greenwood Press. Hafner-Burton EM. 2005. “Trading Human Rights: How Preferential Trade Arrangements Influence Government Repression.” International Organization 59(3): 593–629. Hathaway OA. 2002. “Do Human Rights Treaties Make a Difference?” Yale Law Journal 111: 1935–2042. Hibbs, Douglas. 1973. Mass Political Violence. New York: Wiley. Human Rights Watch. 1993. “Human Rights Watch Report: Colombia” (https://www.hrw.org/reports/1993/WR93/Amw-02.htm). Human Rights Watch. 1996. “Colombia’s Killer Networks: The Military – Paramilitary Partnership and the United States.” (https://www.hrw.org/legacy/reports/1996/killertoc.htm). Kline, Harvey F. 2009. Show Teeth to the Dragons: State-Building by Colombian President Álvaro Uribe Vélez, 2002-2006. Tuscaloosa, Alabama: The University of Alabama Press. Krain M. 2012. “J’accuse! Does Naming and Shaming Perpetrators Reduce the Severity of Genocides or Politicides?” International Studies Quarterly 56(3): 574-589. Nordås, Ragnhild and Christian Davenport. 2013. “Fight the Youth: Youth Bulges and State Repression.” American Journal of Political Science 57(4): 926-940. Marshall MG, Gurr TR, and Kaggers K. 2016. “Polity IV Project: Political Regime Characteristics and Transitions, 1800-2015.” Center for Systemic Peace. 94 Mitchell NJ and Carey SC. 2013. “Pro-Government Militia Database (PGMD) Codebook. V1.1.” (http://www.sowi.uni-manheim.de/militas). Mitchell NJ, Carey SC, and Butler CK. 2014. “The Impact of Pro-Government Militias on Human Rights Violations.” International Interactions 40(5): 812-36. Pierskalla JH. 2010. “Protest, Deterrence, and Escalation: The Strategic Calculus of Government Repression.” Journal of Conflict Resolution 54(1): 117-145. Powell, Jonathan & Clayton Thyne. 2011. “Global Instances of Coups from 1950- Present.” Journal of Peace Research 48(2):249-259. Risse T and Ropp SC. 1999. “International Human Rights Norms and Domestic Change.” In The Power of Human Rights: International Norms and Domestic Change, edited by Risse T, Ropp S, and Sikkink K. Cambridge, UK: Cambridge University Press. Staniland P. 2012. “Between a Rock and a Hard Place: Insurgent Fratricide, Ethnic Defection, and the Rise of Pro-state Paramilitaries.” Journal of Conflict Resolution 56(1): 16-40. Stanton, Jessica. 2015. “Regulating militias: governments, militias, and civilian targeting in civil war.” Journal of Conflict Resolution 59: 899-923. State Department. 1982. “Annual Report on Colombia.” State Department. 2000. “Country Reports on Human Rights Practices: Sierra Leone.” Walker, Stephen and Ivy Lang. 1988. “The Garrison State Syndrome in the Third World: Research Note.” Journal of Political and Military Sociology 16: 105-16. Wood, Reed and Mark Gibney. 2010. “The Political Terror Scale (PTS): A Re-introduction and a Comparison to CIRI.” Human Rights Quarterly 32: 367-400. The World Bank. 2016. “World Databank Indicators.” 95 CHAPTER 3: WHO ME?: HOW AND WHEN PRO-GOVERNMENT MILITIAS CAN MAKE REPRESSION LESS EXPENSIVE FOR LEADERS Introduction The Arrow Boys of Uganda were created as a defense force in 2003 to fight back against Joseph Kony’s Lord’s Resistance Army’s (LRA) insurgency. The national military left a void of protection in the Teso region, so the Arrow Boys, whose title came from their primary weapon of choice, organized to protect their villages. Armed with government supplied weapons, the Arrow Boys operated as a pro-government militia (PGM) in the region. Their strategic advantages over the LRA included knowledge of local terrain and the fact that the group consisted of many former soldiers from Uganda’s People’s Army (UPA) allowed the Arrow Boys to lead a successful campaign against the LRA. Acknowledging the military’s shortcomings in combating the LRA, President Museveni recognized the efforts of the Arrow Boys against the insurgents. He supplied the militia with an additional 7,000 rifles and formally integrated the militia into the army. Eventually, the Arrow Boys were able to drive out the LRA and provide President Museveni with stability in the region (Taylor 2017). The Arrow Boys were originally designed to act as a village defense force and to protect local civilians. However, the group has been accused of numerous human rights violations against the very same civilians they were tasked with protecting (Human Rights Watch 2006b). According to a Human Rights Watch Report, the Arrow Boys have been accused of “lawlessness and terrorizing the local population” (Human Rights Watch 2006b). Despite their “abusive behavior toward civilians”, the PGM was selected and trained to assist with the 2006 presidential election in Uganda (Human Rights Watch 2006b). The election was the first multi-party election held in Uganda in two decades. The Electoral Commission in 96 charge of arranging the election was inadequately prepared and faced numerous complaints from voters. In order to provide sufficient police to guard polling stations, the Commission turned to the PGM for assistance as “election constables” (Human Rights Watch 2006b). In addition, the militia was commanded by the NRM-O parliamentary candidate, State Minister for Health, Mike Mukukla, at the time of the election. Just prior to the election, Human Rights Watch issued a letter asking election observers to address issues of “intimidation and violence by security agents against opposition supporters; and the selective use of prosecutions to hinder and intimidate opposition leader” (Human Rights Watch 2006a). There were 58 documented cases of intimidation and violence by security agents against the opposition (Human Rights Watch 2006c). Despite accusations of vote rigging, violence, intimidation, and violations of free assembly, expression, and association, President Museveni declared victory in his re-election campaign. The Arrow Boys supported Museveni’s regime multiple times, from defending the Teso region from insurgency by the LRA to providing security support at the polls just a few years later in a re-election bid. These instances have both directly and indirectly assisted Museveni’s attempts to stay in power despite opposition and some uprising. Is this a unique case of a PGM prolonging a leader’s tenure? Or can PGMs alter a leader’s length of time in office? In this manuscript, I ask, how do PGMs impact a leader’s length of time in office. Using the Archigos dataset (Goemans, Gleditsch, and Chiozza 2009) on executive leaders and the Pro-Government Militia Database (Carey, Mitchell, and Lowe 2013), also known as PGMD, I explore these questions. I argue that PGMs can significantly influence a leader’s chances of survival. However, the direction of the impact varies depending on the type of PGM in question. While semi-official 97 PGMs with stronger ties to the leader may help a leader stay in power longer, PGMs with less institutionalized relationships with leaders, known as informal PGMs, can drastically increase the likelihood that the leader will be ousted from power. The reason for this difference lies in the amount of control a leader has over the militia. When leaders are unable to effectively control and monitor the PGM, as is the case with informal PGMs, they will be less likely to effectively utilize them for their own gains. However, groups that can be closely monitored and wielded at the leader’s discretion will be more effective to help them maintain their power and possibly extend their tenure. The results find support for the argument and demonstrate the varying effect of PGMs. While leaders with semi-official PGMs are 33.1% less likely to be ousted than leaders without semi-official PGMs, leaders with informal PGMs are about 20% more likely to be removed from power than those without informal PGMs. Therefore, leaders can derive utility from one type of PGM but can actually be harmed by the other riskier type. In conclusion, PGMs can significantly alter the length of time a leader is in power. Repression and Leadership Survival Literature on leadership survival has identified several factors that may impact a leader’s length of time in office. One such factor is the role repression. Scholars have explored the role of repression in lengthening leadership tenure, however, there is very little empirical evidence to suggest that repression actually helps leaders stay in power longer (Davenport 2007). Earlier work focuses on repression as a tool to remain in power (Arendt 1962). Scholars have concentrated on the coercive capabilities and capacities of regimes as a survival strategy. Later work shifted to explore the trade-offs of various survival strategies for dictators once they are already in power (Tullock 1987). As Escribà-Folch and Wright (2010) explain, “the focus 98 turned from repression to buying loyalty and the combination of these two strategies.” A more recent study by Escribà-Folch (2013) uses two-stage estimation methods and finds that repression increases the likelihood of a leader’s survival and an increased probability of exist is associated with increases in repression levels. Relatedly, other scholars have focused more on regime type and its relationship to a leader’s ability to stay in power based on the assumption that repression is a basic tool of survival for authoritarian leaders. One main conclusion in this body of work is that repression levels are lower in democratic systems (Poe and Tate 1994). Others have pushed back on this notion and suggested the presence of a threshold effect, meaning democracy does not have a negative effect on repression until democracy scores reach a certain level (Davenport and Armstrong 2004). Lastly, others have argued that the relationship between regime type and repression is curvilinear, indicating a “more Murdie in the middle” theory (Fein 1995). Other work has explored the role of elite supporters. In order to remain in power, authoritarian leaders rely on loyalty and support (Wintrobe 1998). Retaining the political support of their winning coalition is essential for authoritarian regime survival. Leaders are able to do this through patronage and policy concessions. Bueno de Mesquita, Morrow, Siverson, and Smith (1999) find that a leader’s tenure can be shortened by a large winning coalition but lengthened by a large selectorate. However, other scholars have instead focused on the threats dictators face and the institutions they can use to appease them instead of the size of the coalition (Gandhi and Przeworski 2007). Despite these notable contributions, recent literature has overlooked the connection between militias and their effect on leadership tenure. By examining the link between leaders and their alternative security apparatuses, I seek to explain whether or not these apparatuses 99 actually work to their leader’s benefit as intended or if they can have negative consequences. If leaders can be directly helped by utilizing militias, they may be incentivized to create more groups and carry out more acts of repression that are more difficult to punish them for. However, if these militias do not behave the way leaders may intend for them to, this may highlight an inefficiency in leadership logic and demonstrate the inadequacy of alternative security apparatuses. Theoretical Framework Leaders of states may face many challenges to their tenure, including coups, dissent from civilians, and even international pressure for change. When faced with these challenges from inside and outside their state, leaders must evaluate how to best respond to such threats to maintain their tenure. In order to address many of these concerns, leaders may consider a variety of repressive tools to stymie dissent and combat insurgencies. Usually, options for leaders would include their military, a national police force, gendarmerie, a presidential guard, or other similar units. However, not all agents of repression are equally useful for leaders in these circumstances. In fact, while some types may provide short term relief in terms of large-scale repression, they actually harm a leader’s ability to maintain their power. Ideally, leaders would like to reap the benefits of repression while mitigating or even possibly avoiding the costs associated with it; essentially, they would like to alter the cost-benefit calculation. One way to do so is to use alternative security apparatuses outside of the normal military. These alternative security apparatuses would carry out repressive acts on behalf of the leader while also acting as a scape goat for the leader when faced with the threat of punishment. Therefore, these groups would allow the leaders to reap the benefits of repression to maintain their tenure and discount the cost of repression by avoiding some accountability. 100 These groups provide numerous benefits to the leader, including offering logistical incentives. They increase force numbers, lower deployment costs, and provide informational advantages in their home regions, as was the case of the Arrow Boys. However, when engaging some groups with unofficial ties, the government may also lose their monopoly on violence within the state (Carey, Colaresi, and Mitchell 2015). One type of alternative security apparatus that a government can utilize in this way is a PGM. PGMs are defined as armed, organized, and pro-government groups that are separate from the regular security forces (Carey, Mitchell, and Lowe 2013). One way to classify the types of PGMs is by their link to the government. There are two main categories of PGMs when considering their ties to the government, informal PGMs and semi-official PGMs. The main distinction between these two classifications is how formally acknowledged the link is between the government and the PGM (Carey and Mitchell 2017). According to the Pro-Government Militia Database (PGMD), informal PGMs are described as pro-government, government- backed, or government-allied. They may also be armed or trained by the government. The key identification characteristic of these groups is that the link to these PGMs is not formally or officially acknowledged, although their connection to the government might be widely known (Carey and Mitchell 2017). Examples of informal PGMs include death squads, the Janjaweed militia in Sudan, and the Shabiha militias in Syria (Carey et al. 2013). On the other hand, semi-official PGMs have a formally and/or legally acknowledged connection to the government. The government may establish the group by law and militia members may be compensated for their service (Carey and Mitchell 2017). The PGMD explains that, “a semi-official PGM might be sub-ordinate to the regular security forces but is separate from the regular police and security forces. As such, the link between the PGM and the 101 government is more formal and institutionalized…” (Mitchell and Carey 2013, 10). Examples of semi-official PGMs include the Bakassi Boys in Nigeria, Revolutionary Guard in Iran, and the Home Guard in India. Recent literature has demonstrated that PGMs carry out acts of repression and human rights abuses for the government. As a result, when governments cooperate with or utilize PGMs, human rights will suffer (Mitchell, Carey, and Butler 2014). The distance between the government’s official security forces and the PGM is beneficial for the government and strategic. Even if the government creates and financially supports or equips the militia, state leaders can often deny culpability with respect to the abuses the group commits, as PGMs are not official security forces. Due to the nature of their relationship, governments are able to utilize the distance between them and the militia to deflect blame and responsibility for the repression. Leaders are able to exploit the issues of monitoring and control that naturally arise in the principal-agent framework by claiming they are not in control of the militia and their actions. However, the traditional principal-agent framework is complicated in the context of militias, as the leader faces a moral hazard issue. According to Carey and Mitchell (2017), “a key claim in the literature is that militias provide deniability…but some types of militias deflect blame better than others” (134). Although leaders are able to shift responsibility onto the “rogue” agents, they are limited in their capacity to do so with informal PGMs (Carey and Mitchell 2017). They are able to shift responsibility onto semi-official groups more easily than informal groups due to the degree of control. Since governments have a minimal degree of control over informal militias, they are less able to deny the violence the group perpetrates (Carey and Mitchell 2017). 102 This distancing and denial are exemplified by the case of the informal Janjaweed militia in Sudan. The Janjaweed are a notably brutal repressive force utilized by former Sudanese President Omar al-Bashir in the Darfur region. The militia consisted of a mix of camel herders and bandits to fight insurgencies and other threats to Bashir. This militia was a favorite repressive tool of Bashir’s, often noted as the “spoiled child” of the Sudanese regime. In fact, this militia was often times better paid and equipped than the military. There is evidence of Bashir congratulating militia members on the front lines of battles and rewarding them for a job well done with medals and praise. Conversely, a commander of militia, Mohammed Hamdan has implicated the government in the PGM’s activities. In a public news conference he stated, “All the hardware we have – where did we get it from?...Do you think we magic[k]ed it out of thin air? It belongs to the government.” Furthermore, Hamdan even showed his militia identity card to reporters (El Bagir 2008). However, publicly, Bashir denounced the militia and maintained his distance. Bashir strongly pushed back on the international community’s concerns stating, “genocide like the International Criminal Court claims is non-existent” (Al-bazzaz 2008). He denied links to the group and even criticized them calling them “bandits and thieves”. Bashir also denied providing arms and supplies for the group. The international community faced challenges with holding Bashir accountable for the abuses. Bashir restricted access to the country for reporting and investigating claims of genocide, limiting the international community’s ability to monitor and observe the violence. Therefore, the international community had little ability to act and respond to concerns of the militia’s abuses and Bashir could not be punished accordingly (Human Rights Watch 2004). As demonstrated by the Janjaweed, informal PGMs carry a high risk for leaders. This is 103 because governments have less control over informal PGMs and are therefore less able to direct their violence. This lack of control is result of the covert relationship shared between the PGM and the leadership. Informal PGMs are loosely connected to the government by nature, and they operate at a distance. They may be more likely to have private goals and create control problems (Carey and Mitchell 2017). While acting on the ground, they may be more likely to deviate from the government’s directive. The less control a leader has, the less likely that leader will be able to effectively wield that group to achieve its political aims. Essentially, these PGMs may become rogue entities and act on private goals with little ability for the government to reign them in and reestablish control. In addition, informal PGMs are more repressive than semi-official PGMs (Carey and Mitchell 2017). Due to their highly repressive nature that is difficult to control, leaders appear to turn to these groups for short term repressive needs. For instance, Indonesia in 1999 experienced significant internal conflict. There was a total of 20 informal PGMs present that year to engage in violence on behalf of the government. However, many of these PGMs were dismantled or dissolved that same year. In 2000 Indonesia had only 12 informal PGMs, which indicates a significant decrease from the previous year’s 20 and only two informal PGMs in 2001 after the violence significantly tapered off. Therefore, leaders face a tradeoff when considering using an informal PGM. These PGMs are riskier as they are more difficult to control due to their covert relationship, however, they offer an opportunity to carry out more violence. However, the levels of violence may be too extreme or used against the wrong populations and actually harm a leader. The leader could be blamed for the violence by the civilians or even the international community. In addition, they could be chastised for their inability to control the militia. Therefore, leaders are less likely to 104 benefit from the use of informal PGMs and may in turn be harmed by their rogue actions. This leads to my first hypothesis: H1: The presence of informal pro-government militias is likely to decrease a leader’s chance of survival. On the other hand, semi-official militias, such as the Arrow Boys in Uganda, are more closely controlled and can be directed more easily by the government as they often times work alongside recognized government security forces. More formal links to these groups lessens goal variance, increases transparency, decreases the information gap between the principal government and agent militia, and creates a degree of accountability (Carey and Mitchell 2017). Therefore, semi-official militias may be less likely to attract opportunist and extremist recruits. Instead, recruits may value their position and the compensation that comes with it. This allows leaders to direct the militia and have more control their repressive actions, ensuring they are benefiting the leader. These militias present the opposite tradeoff for leaders compared to informal militias. Semi-official PGMs offer leaders a less risky option for repressive tasks. While semi-official PGMs still repress and can help leaders achieve their goals, their more overt relationship to the regime makes it more difficult for leaders to distance themselves from the violence. They are not as able to deny the violence or links to the group, and therefore may choose to have them only repress when absolutely necessary. Therefore, the increased control of these groups allows leaders to utilize them for violence, but at lower levels and not as often as informal PGMs. This strategy allows for more effective repression and helps leaders stay in power longer. Jerry Rawlings led Ghana from 1981 through 2001, including a military junta until 1992. Throughout his tenure, which was one of the longest leader tenures in modern history, Ghana 105 had eight semi-official PGMs, including the Association of Committees for the Defence of the Revolution, Ghana Private Road Transport Union, Committee for the Defense of the Revolution, Civil Defense Organization, Progressive Voluntary Organization, SEMACO, Patriots Club of Ghana, and Battalion 64. While five of the militias were created between 1982 and 1984, three were created in the 1990s. The country has no record of any informal PGMs. Several militias were called the “organs of the revolution” and were considered key features to Rawling’s presidential bid and success (Abduli 1992). For instance, the Association of Committees for the Defence of the Revolution fired on opposition demonstrators in Accra protesting the high cost of living and unemployment in the state in 1995 to help stymie uprising and dissent. At least two civilians were killed and 17 wounded in the shootings (Canada: Immigration and Refugee Board of Canada 1996). The group was generally used as thugs against anti-government protestors and government critics. In 1999 the group raided the Supreme Court building and ransacked the offices and molested personnel (Carey and Mitchell 2014). The following year the PGM was also tied to preventing voters from voting on election day in the Ashanti Region to “frustrate voters” (Carey and Mitchell 2014). These militias were easily controlled by Rawlings, as they were fairly public. However, they still carried out less egregious forms of human rights abuses on his behalf to directly benefit his tenure. The use of militias to push back on dissenting protestors, interfere with elections, and carry out general security tasks on behalf of Rawlings demonstrably contributed to his lengthy tenure and ability to stay in office longer than he may have without the group, especially after the presidential election. Therefore, it is clear how the use of a semi-official PGM allowed Rawlings control over the group’s actions to limit their repression and to maximize their utility in his pursuit of extending his tenure. 106 This leads to my second hypothesis: H2: The presence of semi-official pro-government militias is likely to increase a leader’s chance of survival. Data In order to test the hypotheses set forth above, I utilize the Archigos (Goemans, et al. 2009) dataset on leadership survival to define time and the sample of executive leaders. The unit of analysis is the leader-year, and I observe 236 leaders across 78 countries. The dependent variable is the length of time a leader remains in power until their failure, or removal from office. Leaders that leave office due to natural death are censored. In addition, exits that are coded as “unknown” are also censored. I use a global sample from 1989-2007 and each model has 1,074 observations. The independent variables for the presence of pro-government militias come from the Pro-Government Militia Database (PGMD) by Carey, Mitchell, and Lowe (2013). The PGMD provides information on militia types and characteristics. For each type of militia, informal and semi-official, I use the number of PGMs in the country at a given time. For informal PGMs, these values range from 0 to 20. Semi-official PGMs are not as frequent as informal PGMs, with values ranging from 0 to eight. The highest values for both types of PGMs are from Indonesia in 1999 when the country was experiencing significant civil conflict. In addition, I employ a number of control variables for all models. First, I control for the repression PGMs are responsible for within the state. This variable captures the highest level of violence carried out by a PGM in that given year. This ordinal variable is from the Militia Repression Database (MRD) with values ranging from 0 indicating no repression to five, indicating widespread abuses. In addition, I control for the presence of at least one anti-rebel 107 PGM within the state. Many PGMs are created as defense forces to drive out insurgencies, which could presumably greatly influence the stability of the regime and leadership. I also use a dichotomous indicator from MRD to capture PGMs being used for election violence within a state. Previous literature has demonstrated that militias can be a useful tool for leaders at election polling places, as was the case with the Arrow Boys. I also control for several country-level characteristics. I include the Polity 2 IV score based on the 21-point scale (Marshall et al. 2016), presence of civil war according to the UCDP/PRIO Armed Conflict Dataset (Gleditsch, Wallensteen, Sollenberg, and Strand 2002), the presence of interstate war, the log of the gross domestic product (GDP), and the log of the total population from the World Development Indicators (The World Bank 2016). Lastly, I use a dichotomous variable to control for the Cold War era. Research Design In order to analyze the effect of PGMs on leader survival, I employ a survival analysis, often also referred to as an event history or duration model. More specifically, I use a Cox proportional hazards model. This model specification is most appropriate for this analysis for several reasons. First, a survival model estimates a hazard rate, which is the natural rate for the ending of some event or process (Bueno de Mesquita and Siverson 1995). Second, a survival model allows me to include cases that may be excluded or treated improperly with other model specifications. For instance, the survival model allows the censoring of observations when the leader dies through natural causes while in office, indicating the end of their tenure. Lastly, no assumptions are made regarding the nature or the shape of the hazard function. The Cox regression results are interpretable as hazard of risk ratios. Each hazard ratio represents the factor change in the risk of the leader losing power as the independent or control 108 variable increases one unit, holding all else constant. Hazard ratios above 1 indicate that the variable of interest increases leader’s chance of being removed from power. Conversely, a hazard ratio less than 1 for an independent variable indicates that variable decreases a leader’s likelihood of being ousted. For instance, a hazard ratio of 2.0 indicates the likelihood that the leader is removed from power doubles as the variable increases. A hazard ratio of 0.5 indicates that the likelihood of a leader being removed from power drops by half. Results Table 11 displays the main results for Model 1, analyzing only informal PGMs, and Model 2, analyzing only semi-official PGMs. Overall, the results provide support for both hypotheses. The presence of informal PGMs are likely to decrease a leader’s chance of survival, while semi-official PGMs have the opposite effect and are more likely to prolong a leader’s time in office. In Model 1, which analyzes only informal PGMs, there is a hazard ratio of 1.200. This indicates that leaders with informal PGMs are 20% more likely to be removed from power than those without informal PGMs. Therefore, leaders with informal PGMs have a lower likelihood of survival than leaders without informal PGMs. This finding demonstrates support for Hypothesis 1 that informal PGMs decrease a leader’s chance of survival. 109 Table 11: Main Models - Cox Proportional Hazard Models Model 1 (Informal) Hazard Ratio 1.200 0.900 1.004 Independent Variables PGMs PGM Repression PGM Election Violence Anti-Rebel PGM Log Population Log GDP Civil War Interstate War Polity Cold War N Number of Failures 167 Log-Likelihood 0.596 1.000 1.000 2.404 .000 1.134 .999 1074 SE p-Value .110 .082 .002 .258 .000 .000 .693 - .034 .280 .047 ** .245 .027 ** .231 .085 * .055 * .002 *** - .000 *** .998 Model 2 (Semi-Official) Hazard Ratio 0.669 0.981 1.005 .694 1.000 1.000 2.517 .000 1.156 .950 1074 167 -895.624 SE p-Value .107 .084 .002 .284 .000 .000 .711 - .034 .262 .012 ** .826 .016 ** .373 .670 .056 * .001 *** - .000 *** .853 -896.947 *p<.10, **p<0.05, ***p<0.01, two-tailed tests. Figure 9 displays the survivor function for leaders in Model 1. The dashed line represents leaders with three informal PGMs and the solid black line represents all other leaders. Multiple informal PGMs is fairly common and 95% of leaders have up to three PGMs, capturing most of the data in the analysis. The survival rate for leaders with at least one PGM is significantly lower than leaders without informal PGMs and declines at a faster rate. 110 Figure 9: Survival Rate for Informal PGMs Model 2 analyzes only semi-official PGMs, which are more closely tied to the government. The hazard ratio for the PGM variable is .669. This indicates that leaders with semi- official PGMs are 33.1% less likely to be ousted than leaders without semi-official PGMs. Therefore, leaders with at least one semi-official PGM have a higher likelihood of survival and leaders without them. This finding provides support for Hypothesis 2. Figure 10 displays the survivor function for Model 2. Like Figure 1, the dashed line represents leaders with PGMs, but this time two semi-official PGMs. The black line represents leaders without any semi-official PGMs. Unlike Figure 1, the dashed line representing PGMs is now higher than the line representing leaders without PGMs. The survival function for leaders 111 with semi-official PGMs is higher than leaders without semi-official PGMs, and the difference becomes more significant as the leader stays in power longer. Figure 10: Survival Rate for Semi-Official PGMs In addition, Figure 11 displays the predicted probability of the hazard function for leaders with semi-official PGMs. The predicted hazard rate when leaders have no semi-official PGMs is .809, indicating a 19.1% chance of success. However, when leaders have one semi-official PGM, chances of survival more than double to 45.9%. Leaders with two semi-official PGMs have an even lower risk of failing, with a 63.8% chance of survival. Once leaders have three semi-official PGMs at their disposal, their risk of failure decreases even further with a 75.8% survival rate. 112 Each additional semi-official PGM significantly reduces a leader’s risk of failure, indicating that they are a highly valuable tool for regimes wishing to extend their tenure and insulate their power. This finding again lends support to the second hypothesis. Figure 11: Predicted Probability of Relative Hazard with Semi-Official PGMs Model 3 includes both types of PGMs in one model, leaving only leaders without any type of PGM as the omitted category. The results for Model 3 are displayed in Table 12. These findings are consistent with Models 1 and 2. The hazard ratio for informal PGMs is again positive and above 1, indicating an increased hazard ratio for leaders with informal PGMs 113 compared to leaders without any PGMs. Conversely, the hazard ratio for semi-official PGMs is .654, indicating that leaders with semi-official PGMs have a higher likelihood of survival than leaders without any PGMs. These findings demonstrate support for both hypotheses. 114 Table 12: Combined Model - Cox Proportional Hazard Models Model 3 (Both Types) Independent Variables Hazard Ratio SE p-Value Informal PGMs Semi-Official PGMs PGM Repression PGM Election Violence Anti-Rebel PGM Log Population Log GDP Civil War Interstate War Polity Cold War N Number of Failures 1.204 0.654 0.929 1.005 0.619 1.000 1.000 2.362 .000 1.147 1.013 1074 167 Log-Likelihood -893.833 .116 .109 .084 .002 .264 .000 .000 .683 - .034 .284 .054 * .011 ** .419 .018 ** .260 .748 .117 .003 *** - .000 .964 *p<.10, **p<0.05, ***p<0.01, two-tailed tests. Across all three models, the control variables are consistent and stable. PGM repression fails to achieve significance in all three main models. This may be due to the fact that high levels of violence call too much attention to the militias or leaders are held responsible for losing the monopoly of violence within a state. However, having at least one PGM that carries out election violence appears to have a small effect on leader survival. The hazard ratio for election violence varies between 1.004 and 1.005 for all three models, indicating a slight risk of using this tactic 115 for leaders. Interestingly, this tactic does not appear to lower the hazard ratio for leaders, meaning it is most likely not the most efficient use of PGMs. On the other hand, having at least one PGM with an explicit anti-rebel agenda also interestingly did not help a leader to stay in power longer. While the hazard ratio was below 1 for each model, it failed to achieve statistical significance. This may be because these forces are used to drive out small-scale insurgencies in remote regions, like the Arrow Boys. It is possible that these insurgencies would not threaten the regime and cause it to fail, but instead would be a nuisance. If a leader can deploy a PGM to handle the smaller insurgency by just supplying weapons, it would save the cost of deploying an army unit and allow the leader to use the military branches for other larger insurgency concerns. The hazard ratios for the log of the population and the log of GDP are consistently 1.000 and fail to achieve significance in all models. The dichotomous indicator for the presence of a civil war is consistently positive and above 1. This means that having a civil war greatly increases a leader’s risk of being removed from power. Due to the rare occurrence of interstate war, this variable fails to produce any meaningful results. The Polity hazard ratio is just above 1 and always statistically significant at the .01-level, meaning as a state becomes more democratic with an increasingly Polity score, leaders face an increased risk of being removed from power. This is mostly likely due to the regular office turnover feature in democratic systems. Lastly, the dichotomous indicator for Cold War is just above 1 and insignificant for all models, indicating that there is no significant difference in risk for leaders in the Cold War era. Conclusion The effect of PGMs on leadership tenure has gone underexplored in recent literature. Previous work on PGMs has suggested that leaders choose to employ PGMs for acts of repression on their behalf. Ideally, this repression would help keep the leader in office longer and 116 prevent the opposition from forming a realistic threat to the regime. However, these analyses have demonstrated that this may not always be the case. The findings suggest that PGMs with closer connections to the government are better able to help a leader stay in power longer. This may be due to the control the leader can exercise over the group and their repressive behavior. In addition, because these groups are generally not as repressive, they may not draw as much attention and scrutiny for the leader. Conversely, their informal counterparts, groups with looser ties to the government, have the opposite effect and can actually harm a leader’s chances of staying in power longer. These groups may go rogue and carry out abuses that do not necessarily benefit the leader and may be harder to oversee. Therefore, I argue the utility of a PGM for a leader largely is related to the control the leader can exercise over the militia to ensure their maximum efficiency. The results from all three models demonstrate suggest for this argument. These findings suggest that although leaders may create and utilize PGMs in order to strengthen their political position and maintain their power, it does not always work. Instead, by using more informal PGMs, leaders actually increase their risk of being tossed out of power. This means that leaders attempting to make repression less costly by outsourcing to PGMs are not always able to do so. In fact, by utilizing some alternative security apparatuses to make repression cheaper, they actually face much higher costs in the long run, including being ousted from power. Overall, informal PGMs appear to be a risky tool for leaders wishing to extend their tenure. On the other hand, leaders may be able to stay in power longer by using the more closely controlled and monitored semi-official PGMs. This points to a conundrum for the international community to grapple with. If leaders are able to artificially prop themselves up using semi- 117 official PGMs to stay in power longer, civilians and political rights may be suffering. Therefore, the results suggest that the international community should turn its attention toward leaders with semi-official PGMs under their control. Since these groups have been found to help a leader stay in power longer by means of repression, new approaches to punishing leaders for utilizing such groups should be considered. In addition, incentivizing leaders to not only dismantle existing PGMs but also to stop the creation of new militias would be advantageous for the civilians in these states that may be adversely affected by their repression and ability to keep human rights abuses in power longer. 118 BIBLIOGRAPHY 121 BIBLIOGRAPHY Abduli, David. 1992. “Rawlings ‘Wins’ Ghana’s Presidential Elections: Establishing a New Constitutional Order.” Africa Today 39(4): 66-71. Al-bazzaz, Noor. 2008. “Sudan’s president denies involvement in Darfur atrocities.” The Guardian. August 20. Arendt, Hannah (1962) The Origins of Totalitarianism. New York, NY: Meridian. Bueno de Mesquita, Bruce and Randolph M. Siverson. 1995. “War and the Survival of Political Leaders: a comparative study of regime types and political accountability.” American Political Science Review 89(4): 841-855. Bueno de Mesquita, Bruce, James D. Morrow, Randolph M. Siverson, and Alastair Smith. 1999. “Policy Failure and Political Survival: The Contribution of Political Institutions.” The Journal of Conflict Resolution 43(2): 147-161. Canada: Immigration and Refugee Board of Canada. 1996. “Ghana: Information on whether charges were laid in connection with the anti-VAT demonstration in May 1995, and whether persons are still sought by the authorities.” UNHCR. Carey, Sabine, and Neil Mitchell. 2014. “Documentation for Association of Committees for the Defence of the Revolution.” Pro-Government Militias. Carey, Sabine, and Neil Mitchell. 2017. “Progovernment Militias.” Annual Review of Political Science 20: 127-147. Carey S, Colaresi M, and Mitchell N. 2015. “Governments' Informal Links to Militias.” Journal of Conflict Resolution 59(5): 850-876. Carey S, Mitchell N, and Lowe W. 2013. “States, the Security Sector and the Monopoly of Violence: A New Database on Pro-Government Militias.” Journal of Peace Research 50(2): 249-58. Colaresi, Michael. 2004. “When Doves Cry: International Rivalry, Unreciprocated Cooperation, and Leadership Turnover.” American Journal of Political Science 48(3): 555-570. Davenport Christian. 1995. “Multi-Dimensional Threat Perception and State Repression: An Inquiry into Why States Apply Negative Sanctions.” American Journal of Political Science 39(3): 693-713. Davenport, Christian. 2007. “State Repression and Political Order.” Annual Review of Political Science 10: 1-23. 122 Davenport, Christian and David A. Armstrong II. 2004. “Democracy and the Violation of Human Rights: A Statistical Analysis from 1976 to 1996.” American Journal of Political Science 48(3): 538-554. El Bagir, Nima. 2008. “We were armed by Sudan, say Darfur Killers.” Telegraph March 12. Escribà-Folch, Abel. 2013. “Repression, political threats, and survival under autocracy.” International Political Science Review 34(5): 543-560. Escribà-Folch, Abel and Joseph Wright. 2010. “Dealing with Tyranny: International Sanctions and the Survival of Authoritarian Rulers.” International Studies Quarterly 54: 335-359. Fein, Helen. 1995. “More Murder in the Middle: Life-Integrity violations and Democracy in the World, 1987.” Human Rights Quarterly 17(1): 170-191. Gandhi, Jennifer and Adam Przeworski. 2006. “Cooperation, Cooptation, and Rebellion under Dictatorships.” Economics and Politics 18(1): 1-26. Gleditsch NP, Wallensteen P, Eriksson M, Sollenberg M, and Strand H. 2002. “Armed Conflict 1946-2001: A New Dataset.” Journal of Peace Research 39(5): 651-660. Hathaway OA. 2002. “Do Human Rights Treaties Make a Difference?” Yale Law Journal 111: 1935–2042. Human Rights Watch. ND. “Uganda Elections 2016.” (https://www.hrw.org/tag/uganda- elections-2016). Human Rights Watch. 2004. “Darfur Destroyed: Ethnic Cleansing by Government and Militia Forces in Western Sudan.” (https://www.hrw.org/report/2004/05/06/darfur-destroyed/ethnic- cleansing-government-and-militia-forces-western-sudan). Human Rights Watch. 2006a. “HRW Letter to Uganda’s Election Observers.” (http://pantheon.hrw.org/legacy/english/docs/2006/02/17/uganda12689.htm). Human Rights Watch. 2006b. “In Hope and Fear: Uganda’s Presidential and Parliamentary Polls. No. 1” (https://www.hrw.org/legacy/backgrounder/africa/uganda0206/uganda0206.pdf). Human Rights Watch. 2006c. “Uganda: Monitors Must Investigate Threats to Free Elections.” (https://www.hrw.org/news/2006/02/16/uganda-monitors-must-investigate-threats-free-elections) Marshall MG, Gurr TR, and Kaggers K. 2016. “Polity IV Project: Political Regime Characteristics and Transitions, 1800-2015.” Center for Systemic Peace. Mitchell NJ and Carey SC. 2013. “Pro-Government Militia Database (PGMD) Codebook. V1.1.” (http://www.sowi.uni-manheim.de/militas). 123 Mitchell NJ, Carey SC, and Butler CK. 2014. “The Impact of Pro-Government Militias on Human Rights Violations.” International Interactions 40(5): 812-36. Murdie AM and Davis DR. 2012. “Shaming and Blaming: Using Events Data to Assess the Impact of Human Rights INGOs.” International Studies Quarterly 56(1): 1-16. Pierskalla JH. 2010. “Protest, Deterrence, and Escalation: The Strategic Calculus of Government Repression.” Journal of Conflict Resolution 54(1): 117-145. Poe, Steven C. and C. Neal Tate. 1994. “Repression of the Human Right to Personal Integrity in the 1980s: A Global Analysis.” American Political Science Review 88(4): 853-872. Risse T and Ropp SC. 1999. “International Human Rights Norms and Domestic Change.” In The Power of Human Rights: International Norms and Domestic Change, edited by Risse T, Ropp S, and Sikkink K. Cambridge, UK: Cambridge University Press. Taylor, Magnus. 2017. “On the Trail of Uganda’s Arrow Boys.” International Crisis Group 16 March (https://www.crisisgroup.org/africa/trail-ugandas-arrow-boys). Tullock, Gordon. 1987. Autocracy. Boston, MA: Kluwer Academic Publishers. Wintrobe, Ronald. 1998. The Political Economy of Dictatorship. New York: Cambridge University Press. The World Bank. 2016. “World Databank Indicators.” 124 CHAPTER 4: CONCLUSION This dissertation has explored the role of PGMs in state-sponsored repression. I aim to better understand how, when, and in what ways leaders use these militias for abuses. Using a novel dataset, I argue that PGMs are highly useful for leaders intending to harm their own civilians and provide leaders with political cover for these actions. PGMs have been proven as repressive mechanisms used by leaders, which presents several challenges for the international community. These findings demonstrate that current punishment mechanisms for abuses, such as naming and shaming, are producing unintended consequences. The pressure from the international community incentivizes leaders to create and use alternative mechanisms instead of changing their behavior. While the intentions of naming and shaming may be to curb human rights abuses, evidence demonstrates this may not always be the case. One possible solution is to increase monitoring of human rights practices on the ground with attention to the creation or utilization of groups like PGMs. Future research on this topic might also explore other ways leaders are punished by the international community. For instance, how do leaders respond to sanctions, such as those issued by the United States against China for its use of a PGM? Furthermore, does the method or rationale for creation influence the types and severity of abuses that PGMs commit? In addition, further exploring factors and characteristics that can explain PGM repression would contribute to our understanding of this type of repression. Lastly, additional data collection efforts should continue to focus on the role of PGMs and how they may be integrated or collaborate with other actors. 125 In conclusion, this dissertation has demonstrated the important role of PGMs in our understanding of state-sponsored repression. Leaders have continually turned to these groups to carry out human rights abuses on their behalf while enjoying the covert relationship. Denial of the abuses has so far proven somewhat effective for some leaders to at least slow the international community’s response. On the other hand, PGMs have also proven to not always help leaders in the long run. While we continue to learn more about PGMs and their violent actions, their human rights abuses and role in political violence remains clear and significant. 126