Rebel Legitimacy and Wartime Sexual Violence. Katherine Sawyer, Ph.D. Candidate University of Maryland

Transcription

1 Rebel Legitimacy and Wartime Sexual Violence Katherine Sawyer, Ph.D. Candidate University of Maryland Kathleen Cunningham, Associate Professor University of Maryland & Peace Research Institute Oslo Kanisha Bond, Assistant Professor University of Maryland Abstract: Rebel organizations vary dramatically in how they are perceived in the international community. These perceptions are often deeply rooted in questions about group legitimacy. In this paper, we connect the concept to both the internal organization and conflict behavior of rebel organizations. First, we contend that leader selection mechanisms are an important vector of group legitimacy: specifically, quasi-democratic leader selection practices are positively correlated with perceptions of legitimacy from states, rebel group members, and potential group supporters. Further, we argue that an important implication of this is that perceived legitimacy can constrain rebels against civilian abuse during war. Using original data on rebel leadership selection practices in conjunction with data on civilian victimization from the Sexual Violence in Armed Conflict (SVAC) project, we show that rebel groups who choose leaders by election are less likely to engage in systematic acts of sexual violence in wartime. Draft October 2016: Comments welcome, do not cite without permission. 1

2 Wartime sexual violence and rape are pervasive, pernicious, and by some accounts, possibly unpreventable. 1 Moreover, there is a misconception that all rebels, as the anti-authoritarians, are the main perpetrators of wartime rape. We know that rebels in failed states are often providers of public goods (Mampilly 2011), such as policing, taxation, and even education, functioning in the place of the state. However, this civil service can go hand and hand with abuse of power, as was the case with a faction of the Sudanese People s Liberation Movement, SPLM-A, with reports of massive rebel perpetrated civilian violence and rape. 2 Rebel leaders in groups such as the Revolutionary United Front in Sierra Leone systematically engaged in sexual violence and encouraged their recruits to as well, generating a sort of sadistic comradery and cohesion amongst the rank-and-file members (Cohen 2013). But, rebel groups can also serve as a haven for women, where they purposely join for personal protection and increased political agency. Kurdish women fighters describe their experience in the Kurdish Democratic Party of Iran (KDPI) as an opportunity for equality, battling a state that they feel is repressive to both their identity as Kurdish people and as women. 3 During the group s seven decade fight in Iran, there have been no reports of rebel perpetrated sexual violence. 4 What then explains this discrepancy? Why do some rebels commit acts of sexual violence while 1 See Cohen, Green, and Wood 2013, and Wood 2009 for a discussion of the conventional wisdom that wartime rape is inevitable during conflict. 2 U.S. State Department Annual Report, Senk, Marta. 'Everyone is a Peshmerga': The Kurdish men and women who are fighting Iran. 27/6/2016. Accessed online 05/10/2016: 4 Based on annual reports from Amnesty International, the Human Rights Watch, and the U.S. State Department. 2

3 others do not? And, are other, more positive tools available to rebels to generate group cohesion while establishing legitimacy both locally and internationally? Early research into wartime sexual violence would suggest that all conflicts incur some level of rape and sexual abuse of women (Benard 1994), while others found that some conflicts do not incur cases of sexual violence or rape at all (Bourke 2007, Wood 2008). Recent scholarly and policyoriented work has emphasized the need to understand under what conditions actors chose to commit acts of sexual violence and predict when civilians, and particularly members of minority and disempowered groups, will be targets. Beyond the direct societal and individual consequences of sexual victimization, this particular form of violence also has important negative downstream consequences for post-conflict peace and reconstruction. Conflicts plagued by acts of sexual violence are both harder to resolve and when resolved, are more likely to see conflict recurrence than conflicts not characterized by sexual violence (see Plümper and Neumayer 2006). Wartime sexual violence varies substantially by perpetrator as well (Cohen 2013). It has been used as a tool of coercion and compellence by both state and rebel forces (Cohen, Green, and Wood 2013, Wood 2009, Wood 2008), and as such has been condemned by the global community as a crime against humanity (ICC Rome Statute, 1998, Article 7 (1) (g), pp. 8-9). Yet, not all actors, be they state or rebel forces, engage in sexual violence. Both within and across conflicts, there is particularly wide variation in its use by rebel organizations. In explaining this variation, some scholars tend to focus on ideological constraints: for example, the general restraint against sexual violence exhibited by insurgents in the Salvadoran civil war is often attributed to the groups Marxist foundations (Cohen 2013, Leiby 2009). Others focus on institutional characteristics: for example, the systematic rape and sexual enslavement of Yazidi women and girls by ISIS in Iraq and Syria is 3

4 said to be used as a tool of compensation for its fighters. 5 Yet, there is still little comparative analysis of those organizations that avoid sexual violence alongside those that do. If sexual violence is not ubiquitous, what explains why only some rebel organizations engage in sexual violence? We argue that variation in rebel actors use of sexual violence is directly related to perceptions of organizational legitimacy. The concept of rebel legitimacy combines insights about the structural and ideological bases of group behavior, particularly by focusing on the formal channels through which the preferences of leaders, rank-and-file, and popular supporters are generated and aligned within organizations, and then signaled to external actors. 6 Specifically, we expect when rebel groups use leader selection mechanisms that prioritize consensus-building, constrain leaders against extreme battle tactics, and minimize principal-agent problems they benefit from higher levels of legitimacy, which in turn reduces the likelihood that the group will engage in such an extreme form of civilian abuse. Using original data on rebel leader ascension processes coupled with the Sexual Violence in Armed Conflict (SVAC) dataset (Cohen and Nordås 2014), we find empirical support for our theory that rebels are both less likely to engage in and be targets of sexual violence when they have a history of quasi-democratically elected leaders. We are able to show who these actors are that refrain from sexual violence and how that is paradigmatic of a larger group structure that divides rebels by their credibility as political actors both domestically and in the international community. Below, we first outline the progress that has been made to date researching the causes and effects of sexual violence in conflict. Next, we turn to our theory of rebel group legitimacy and 5 Iraq: ISIS Escapees Describe Systematic Rape Accessed 06/05/16: 6 Here we follow on arguments by Paul Staniland (2014) and others that emphasize the importance of the multitude of possible links between rebel leaders, members, and civilians for group behavior. 4

5 discuss how this impacts the likelihood of sexual violence perpetration by rebel actors. We then discuss the data and the findings from our empirical analyses. We use a series of statistical tests to identify broad patterns in the relationship between rebel legitimacy, leader selection methods, and sexual violence perpetration, as well as process tracing to provide a more nuanced exploration of rebel legitimacy and sexual violence through the examples of the Kurdish Democratic Party of Iran and Sudan People s Liberation Movement - North. We conclude with a discussion of implications for the study of sexual violence and rebellion movements. Causes and Sources of Sexual Violence Until recently, there has been little systematic study of sexual violence in civil war. Conflict literatures focusing on the costs of war primarily examined killings, either civilian or combatant, and battle deaths still remains a common indicator of conflict severity (Kalyvas 2006, Valentino, Huth and Balch-Lindsay 2004, Weinstein 2007). However, that has changed in the last few years with large-n empirical work on variation in the determinants of rape and sexual violence (Butler and Jones 2014, Cohen 2013, Cohen and Nordås 2014). As a form of civilian abuse, wartime rape and other forms of sexual violence are common tools available to perpetrators for asserting dominance over their victims, and sometimes even their supporters. This motivation can be especially pronounced in the wartime context, as struggles over the hearts, minds and resources of unaffiliated individuals are central, and resource mobilization is critical - particularly for the most power-disadvantaged (Lichbach 1995). At first blush, it appears that arguments about the motivations or effects of wartime rape and sexual violence are not qualitatively different from existing arguments on motivations for other forms of civilian targeting, including indiscriminate killings and suicide bombing (Bloom 1999). 5

6 However, emerging research shows that the effects of sexual violence during war are particularly insidious, due mostly to its ability to promulgate lasting psychological as well as physical harm on targets. For instance, in a sample of 68 women victims of rape during the mid-1990s war between Croatia and Bosnia and Herzegovina, both direct victims of and witnesses to wartime rape and torture have been found to suffer as a result deep immediate and long-term consequences on social and interpersonal functioning, including post-traumatic stress disorder, depressiveness, suicidal ideation, social phobias and sexual dysfunction (Lončar, et al. 2006, pp ). Especially in communities where manifestations of individual sexuality are considered part of the public interest, the shaming and blaming of men and women rape victims often constitutes a form of psychological terrorism that can be inter-generational as well as highly personal (Bastick, Grimm, and Kunz 2007, Sivakumuran 2007). Extant research has put forth a variety of underlying explanations for why some combatants might choose to use sexual violence to levy such costs. On one hand, much normative and feminist research in this area tends to focus on the role that society-level gender attitudes and relations play in encouraging a tolerance for, if not outright acceptance of, deeply personal and particularly sexualized violence during conflict (Hagen and Yohani 2010, DeLargy 2013, Benard 1994, Green 2006, Kaufman and Williams 2010, Seifert 1996). On the other hand, many positivist treatments have focused on more proximate causes both at the society-level, such as work by Hansen (2001) or Koo (2002) on the role of gender inequality on such outcomes, and at the individual-level, including a focus on greed and opportunism (Goldstein 2001, Mueller 2000, Weinstein 2007) or ethnic prejudice (Bloom 1999, Fearon 2006, Horowitz 1985). Even within the positivist camp however, there is meaningful debate about whether the use of sexual violence is a manifestation of personal opportunism exercised under low levels of organizational control or a strategy of civilian victimization that is deliberately chosen and encouraged by group leadership. 6

7 Arguments in favor of the opportunism and lawlessness hypotheses include work by Jeremy Weinstein, who argues that rebels are more likely to engage in sexual violence when the rank-and-file are primarily motivated by material resource gains and thus more willing to forcibly extract needed goods and/or support from the local population (2007). Christopher Mullins, similarly argues that winner-take-all conflicts such as genocidal wars and other ethnically-charged conflicts are most likely to include wartime rape, particularly as a central technique (2009, p. 15). 7 Butler et al. (2007) argue that sexual violence coming from pro-government forces against non-combatants is most often is committed by out-of-control state agents. Their argument underscores the point that while sexual violence can produce some benefits for the perpetrators, the divergence in preferences on combat strategies that it highlights between principals (group leaders) and agents (rank-and-file) also jeopardizes the organization itself by encouraging insubordination and murky authority structures. The authors conclude specifically that weak accountability mechanisms, low economic wealth/salaries, and a lack of democratic decision-making encourage the principal-agent problems that encourage sexual violence during war. 8 In contrast to these arguments, Dara Cohen (2013) finds that the use of wartime rape by rebel organizations against noncombatants is not driven by opportunism among perpetrators, nor is it more likely to occur during ethnic conflict, cases of genocide, or in countries with higher degrees of gender inequality. Instead Cohen shows that wartime rape often emerges as a strategy for encouraging combatant socialization, or enabling intra-organizational cohesion and strengthening 7 MacKinnon (1994) and Sharlach (2000, p. 90), among others, refer to the use of rape in genocidal wars as specifically, genocidal rape. 88 This is a problem because leaders are keenly aware of the possibility that they can, in effect, abuse their potential supporters into shifting support away from the rebels and towards its opponents, as they seek protection and refuge from victimization. 7

8 bonds of loyalty among perpetrators (2013, p ). 9 Relatedly, the sanctioning of sexual violence as a tool for both recruiting new members and retaining existing ones has also been used to some effect as a mechanism of internal control. For instance, along with other forms of social pressure and entrapments, the forced perpetration of rape, sexualized violence, and other abuses can help maintain high levels of compliance amongst forced recruits (Eck 2014). We argue that organizational legitimacy influences and reflects the likely strength of not only principal-agent problems within a rebel group but also the relationship between organizational insiders and the broader populations with which they interact. These relationships underscore degrees of both internal and external legitimacy. We focus specifically on the institutional bases of rebel group legitimacy, and particularly on the incorporation of quasi-democratic practice in leader selection processes. We argue that this institutional base portends important information about the preferences of both leaders and the rank-and-file: the use of some democratic process in leader selection reflects not only a certain respect for consensus-building from the leader s perspective but also a desire for constrained yet representative leadership from the soldiers perspective. Rebel Legitimacy and Leader Selection All rebel and nonstate actors are not equal; neither in the eyes of the state, the international community, nor the people the rebels often claim to represent. While some rebel movements emerge from the populations to challenge illegitimate governments, others emerge as essentially opportunistic brigands. Empirically, it appears that some rebel actors are viewed as more legitimate than others. Historically, the concept of political legitimacy has reflected a normative belief within a 9 This is not necessarily going to work if victims of rape within rebel organizations themselves. However, militaries have significant problems with rape among service members as well, and we have little research on the effects of this on cohesion among members of the armed forces. 8

9 population that the existing institutionally defined rules ought to be followed (Alagappa 1995, Beetham 1991). Weber (1947) defined legitimate power as the probability that commands will be obeyed by a specific group of people without mass coercion. Rebel legitimacy may come from many sources, and be perceived or identified in a number of ways. Internally and externally however, estimates of organizational legitimacy are often tightly tied to expectations about the predictability and stability of authority. 10 Extant research underscores this point: rebels with more coherent group structures are seen as more credible bargaining partners, leading to increased negotiations and successful war termination (Cunningham and Sawyer 2016, Staniland 2014). For the rebel organization, internally legitimate leaders enjoy minimal principal-agent difficulties: leader preferences are likely to be highly congruent with followers preferences, and so leader commands are likely to be followed as a matter of course. Externally legitimate leaders are able to secure support and participation from the broader population without the need for highly coercive action. In many ways, the adoption of democratic practice in the selection of leaders can work to normalize the organization of authority within a given rebel group. It can send a strong signal to outsiders that the group has internal legitimacy, as the electoral selection process can underscore the leader s claim to a popular mandate within the rebel group. 11 This effect is similar to the ways in which elections are used to create perceived mandates in both democratic and non-democratic state settings (Collier 2009, Gandhi and Lust-Okar 2009, Geddes 2005). Such mandates constitute a 10 Extant research underscores this point: rebels with more coherent group structures are seen as more credible bargaining partners, leading to increased negotiations and successful war termination (Cunningham and Sawyer 2016, Staniland 2014). 11 Rebel groups that gain legitimacy through elections exist in stark contrast to predatory rebel groups that rely primarily on coercion to maintain power (Thies 2010). 9

10 source of power for rebel leaders as they pursue different policies and actions, such as negotiating over the settlement of a dispute. At minimum, the staging of elections for rebel group leadership highlights the rank-and-file membership s interest in collective decision-making. Rebel legitimacy can also be indexed to the depth of social cohesion as it radiates throughout the group s mobilization potential. Elections may also suggest popular mandate among their local constituencies, providing rebel leaders with the legitimacy necessary for pursuing policies with small chances of insubordination or reckless divergence. For example, in the 1984 election of the Somali National Movement leader, civilians were described as being able to regain control of the leadership in electing Ahmad Mahammad Mahamuud Silanyo. 12 Furthermore, the non-negligible degree of respect for the leader s authority communicated by the intentional and public selection of an individual may also suggest reasonable congruence between the preferences of the leader and the rank-and-file. 13 For instance, groups such as RENAMO in Mozambique have used democratization in order to bolster the movement s legitimacy in the eyes of international donors and domestic constituents. 14 Perceptions of legitimacy from outside of the organization can also influence the ability of rebels to gain and maintain external support from third party states and other actors. Rebel groups that eventually take power in some capacity (whether that is at the center of the state or in a break- 12 UCA Dynamic Analysis of Dispute Management (DADM) Project. Retrieved January 12, present/ 13 The extent of popular support conferred likely depends on the size of the selectorate and the openness of the process, however. 14 This also allows the movement to become politically relevant despite electoral defeats and decreases the risks to leaders of participating in the political settlement. 10

11 away territory) will rely on both explicit and tacit support from the international community to be recognized as the government. 15 The breakaway rebels in the Eastern Ukraine appear to be thinking along these lines. The election on November 2, 2014 is seen as being primarily an exercise in generating legitimacy. 16 From outside the group, democratic leader selection processes can work to signal that preferences are consistent enough throughout the group to warrant serious attention from international actors promoting negotiated settlement, as the rebel leadership in such cases will be more credible in its promises to uphold agreement terms (Cunningham and Sawyer 2016). Conversely, rebels that fail to gain international recognition can face severe consequences. A key debate in the United States since the start of the Syrian civil war has been over which rebel faction to support. Syrian opposition factions have worked to demonstrate their legitimacy, and this has been a contested process among rebel factions. The Syrian National Coalition was recognized as legitimate by a number of external actors, but has struggled to maintain its place as the primary representative of the opposition. 17 Legitimacy and Rebel Use of Sexual Violence 15 See Coggins (2011) on the critical role of great power recognition for new states. 16 Quoting a candidate in the election, the New York Times reports: The overriding goal of the election, Mr. Sivokonenko said, was to boost the legitimacy of the separatists state. Andrew E. Kramer Rebel-Backed Elections to Cement Status Quo in Ukraine. The New York Times Nov 2, France on Nov. 13 became the first Western country to recognise the Syrian National Coalition as the sole legitimate representative of the Syrian people. Keesing s Record of World Events Volume 58, November, 2012 Syria, Page Nov New opposition coalition. 11

12 We argue that the internal and external legitimacy that is conferred upon rebel organizations by their adoption of democratic processes and norms is likely to have a dampening effect on the likelihood that they will use sexual violence against non-combatants during civil conflicts. 18 For the rebel command, the challenge is to balance its use of coercion in dealing with civilians with its need for reliable civilian support (Mampilly 2011, p. 55). 19 Rebel forces are typically more reliant on the local population for provisions and support than state military forces, and thus, risk alienating a primary means to combat the state by engaging in sexual violence against their own potential participants and support network. Returning to the intra-organizational foundations of sexual violence as described by Cohen (2013) and Butler, et al. (2007), it is reasonable to expect that the use of violence to encourage investment in the group should be discouraged when rebel leaders are elected. It also follows that principal-agent problems should be minimized as the authority of elected leaders is more likely to be respected (or enforceable). This suggests a high level of top-down control over group behavior, and lessens the chances that sexual violence will emerge as a consequence of internal lawlessness. To the extent that rebel leaders understand that the effects of sexual violence can easily discourage collective action among the unaffiliated, we should expect the rank-and-file members to hew more closely to the leader s preference for protecting, rather than abusing, non-combatants We treat this as a constrained strategic decision by the group. This is not to suggest that there are no principal-agent problems that could lead to sexual violence as Butler et al. (2007) suggest. 19 A distinction can be made from the rebel s perspective between the use of sexual violence as a recruitment tool and as a combat tactic (Cohen 2013; Butler et. al 2007); however, from the perspective of civilians the effect in terms of building local population support is arguably the same. 20 It is not uncommon for leaders of guerrilla organizations or other non-state groups that rely on broad-based civilian support to use rhetoric that elevates the protection of civilians and other 12

13 Rebel groups with greater internal legitimacy stemming from leader elections are also more likely to display good bottom-up control as well, as the rank-and-file demonstrate a commitment to consensus-based decision-making. If group members already display such concern for collective decision-making, participation in the collective act of democratic leader selection may also eliminate the need for sexual victimization as a tool for internal bonding. 21 Thus, the internal legitimacy conferred onto leaders by the rank-and-file through the election process reduces the chances that sexual violence will be needed to prevent or solve intra-organizational principal-agent problems. Normatively, if the election of rebel group leaders genuinely reflects a commitment to the democratic values of moderation and consensus-building, these groups should be less likely to support such methods of extreme violence than groups based on personalist regimes and cults of personality. Those groups with elected leaders have a broader commitment to democratic values are more likely to respect human rights and legal constraints, whether for normative or strategic reasons. Such groups are often more reliant on international support, some of which have strategically chosen to adopt quasi-democratic processes, in order to secure international aid (Gleditsch et al. 2011, Manning 2004). 22 vulnerable populations. For examples, consider the Revolutionary Laws codified by the EZLN in Mexico, or the 1977 Political Programme devised by the EPLF in Eritrea. 21 This argument is similar to others made about the sources of effective in-group policing, in which high levels of horizontal accountability within a given group facilitate the sort of monitoring and enforcement that can credibly constrain group members against acting out of the collective s interest (Fearon and Laitin 1996). 13

14 Further, we also expect that externally legitimate rebel groups (i.e. those that present a positive freedom fighter public face to domestic and international audiences) will have less utility for sexual violence as a tool for encouraging recruitment and/or discouraging solidarity with the government. For these groups, abusing the civilian population with sexual violence can eliminate the positive effects that come from being perceived as legitimate by external actors as well. In fact, much like states in the international system, these rebel groups may face real local and international audience costs for engaging in sexual violence against non-combatants. With respect to the pool of potential recruits and supporters, rebel leader elections can portend reduced chances of sexual violence through potentially reducing the number of unaffiliated individuals that may (need to) be coerced into participation. If members are encouraged to join in order to enjoy the selective benefit of a say in the group s direction and activities, selection of the leader through elections and the legitimacy that they confer to the group may reduce the expected efficiency of sexual violence as a method of recruitment. The successful use of electoral selection in a rebel group can act as a cohesion generating process. Extant research demonstrates that political participation and quality of life are interwoven (Andrews, Jilke, and Van de Walle 2014, Bartlett and Popovski 2013, Berger-Schmitt 2000, Phillips 2006). Political participation in the election of local rebel leaders, particularly when it entails the inclusive participation of members of the community, increases social cohesion inasmuch as it is representative of the interests of those involved in the electoral process and improves the quality of life of the local population. Elections for the local rebels in the Eastern Ukraine are demonstrative of this. The political centralization and economic inequality in the Ukraine that has continued for the last twenty years (Bartlett and Popovski 2013) has exacerbated the distrust of politicians amongst Ukrainians; lending support to more localized governance where community participation in the electoral process has bolstered the Eastern rebels. 14

15 Finally, to the extent that the adoption of democratic norms and processes internally may confer credibility in negotiating with the state and securing support from international actors, rebel groups should be reticent to engage in behaviors that seem to undermine their respect for other seemingly-democratic norms among states, such as the protection of civilians during war and a minimizing of collateral damage. Thus, we expect that, all else being equal, rebel groups with greater legitimacy should be less likely to engage in sexual violence. H1: Rebels with greater legitimacy are less likely to commit acts of sexual violence in civil conflict. Data and Empirical Analyses We use original data on rebel leadership and organizational structure in conjunction with data on sexual violence from the Sexual Violence in Armed Conflict (SVAC) dataset to examine the relationship between rebel legitimacy and rebel and government perpetrated acts of sexual violence (Cohen and Nordås 2014). The dataset used in this analysis builds on the Uppsala Conflict Data Project and rebel groups are identified using UCDP s Dyadic Dataset which employs a 25 battledeath threshold. 23 We used the UCDP Dyadic Dataset to identify all rebel groups active between 1989 and We include dyad years prior to 1989 if the war was still active in our study period. 23 Uppsala Conflict Data Program (Date of retrieval: 13/06/30 14/11/07) UCDP Conflict Encyclopedia: Uppsala University. Rebel groups are engaged in armed conflict with the state: An armed conflict is a contested incompatibility which concerns government and/or territory where the use of armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths. 15

16 Measuring rebel legitimacy Legitimacy itself is multifaceted and difficult to measure outright. Using this list of rebel groups, we collected data on the history of the leadership of the group over the length of the group s existence. Leadership data was collected through secondary academic sources, as well as news sources found in LexisNexis Academic and Keesing s Record of World Events (formerly Keesing s Contemporary Archives). The process was conducted chronologically from the founding year of the group forward, so that the founding leader(s) was/were identified, the year of their subsequent descent from power was noted (and any information on the process by which the leader left his leadership position). Information on the newly ascendant leader was recorded, including the date of his ascension 24 and the method of ascension. Based on the case histories and summaries of the leadership in these groups, we then categorized ascension into seven, mutually exclusive types. Between 1989 and 2011, there were 133 civil wars around the globe. Including multi-party and two-party civil wars, 287 rebel groups have fought their state. Among these rebel groups, we have identified the leader of the group about 95% of the time. 25 For the 373 leaders we have identified in our study, 26 each is classified in one of the following categories: 24 No female rebel group leaders were identified in the data during the period of study. 25 For 83 of 1652 dyad-years, we were unable to identify leadership. These were primarily in rebel groups in Asia and there is little temporal variation in the missing data (all instances occurring before 1976). See appendix for details of which cases have unidentified leaders. 26 There are 370 unique leaders in the dataset, but for our purposes we are interested in the process of selection in each rebel group. Ernest Wamba dia Wamba, Idriss Deby and Yasser Arafat each led two distinct rebel groups. Ernest Wamba dia Wamba led the RCD and RCD-ML after its split from 16

17 Elected Founder Inherited/Non-democratic successor Splintered off of existing rebel groups Merged existing rebel groups Selected by a third party Selected by cadre of rebel officers Our primary independent variable, rebel elections, is coded as one if the rebel group has a history of elected leaders, specifically one or more elected leader(s) in the group s tenure. 27 Sexual Violence in Armed Conflict Data The dataset captures six dimensions of sexual violence: prevalence, perpetrators, victims, forms, location, and timing (Cohen and Nordås 2014). We employ their definition (and measures) of sexual violence which was originally outlined by the International Criminal Court (ICC) as any of the following: 1) rape; 2) sexual slavery; 3) forced prostitution; 4) forced pregnancy; 5) forced RCD. Idriss Deby led both the Mouvement Patriotique du Salut (MPS) and the Revolutionary Forces of April 1st; Yasser Arafat led both Fatah and the PNA. 27 Weighting this measure by the length of the group s tenure or ratio of number of elected to nonelected leaders does not impact the sign or significance of the effect of rebel elections on the probability of committing acts of sexual violence by either the rebel or state forces. See Appendix Table 4. 17

18 sterilization/abortion (ICC, 2000). 28 Cohen and Nordås (2014) followed Wood s (2009) inclusion of sexual mutilation, and sexual torture, as do we. 29 We use information on both the perpetrators and victims of sexual violence from this dataset in the analyses below. Dummy variables code whether the state engages in sexual violence in a given year, specifically engagement in at least 25 or more acts of sexual violence in a given year. 30 Perpetrators are defined as being organized armed actor groups here but do not code individuallevel perpetrators of sexual violence or domestic violence. We use their data to create dichotomous variables indicating whether the perpetrator was a state or rebel armed actor and the target of the violence. Controls We incorporate a number of state and conflict controls, as well as measures capturing competing arguments for engagement (or the lack-thereof) in sexual violence derived from the extant literature. Since our argument expects that legitimate rebel groups may be more likely to receive international support, we include measures of material external support and intervention to 28 See Appendix for a list of complete definitions from the ICC, 2000: International Criminal Court, Elements of Crimes, U.N. Doc. PCNICC/2000/1/Add.2 (2000) cited in Cohen and Nordås (2014). 29 Cohen and Nordås (2014) note that their definition of sexual violence does not include what other researchers have deemed acts of sexual humiliation including forced undressing and insults of a sexual nature (Leiby 2009). 30 Cohen and Nordås (2014) code three prevalence levels of sexual violence (with 0 being no acts of sexual violence reported in a given year): isolated (1-25 victims of sexual violence reported in a year); numerous ( victims of sexual violence in a year); massive (1,000 or more acts of sexual violence in a year). See Cohen and Nordås (2014) for further description of the coding scheme. 18

19 our models. The two dichotomous measures of external support are coded from the UCDP External Support Dataset, and measure support provided to the rebels and support provided to the state from states, rebels, or other international actors external to the civil conflict. 31 In addition, we control for negotiations between the rebels and the state in the previous year (dichotomous variable). This information is coded from the UCDP conflict profiles to identify when and with whom negotiations occurred during these civil wars. 32 As these are not private negotiations, and often include an international monitor/third party, we would expect that the increased public attention would diminish acts of sexual violence in the subsequent year. With respect to competing arguments, we incorporated a dichotomous measure of territorial control from the Expanded Armed Conflict Data (EACD) v dataset as the extant literature argues that victimization is more likely in areas of contested control (Kalyvas 2006). As warfare based on ethnic cleavages has been argued to be a precipitant of conflict-wide rape (Cohen 2013), we include a measure of ethnic fractionalization (Fearon 2003). We also control for a number of state characteristics that are commonly included in civil conflict studies such as the size of the state s population (natural log transformed), 34 and GDP per 31 See Pettersson, Therése, Pillars of Strength External Support to Warring Parties, in Therése Pettersson & Lotta Themnér eds., States in Armed Conflict Research Report 94. Uppsala: Universitetstryckeriet. Support can be provided as: financial support, troop support, weapons support, material/logistic support, military/intelligence support, training/expertise, and/or access to territory, as well as other and unknown support. 32 Negotiations were coded from the field: Negotiations in dyad in the UCDP Conflict Encyclopedia. Accessed: 09/01/01-14/11/ See Salehyan, Gleditsch, and Cunningham (2011) for further information. 34 World Bank. World Development Indicators. 19

20 capita (natural log transformed) We also include a measure of armed strength, measured as the size of the state and rebel troops (natural log transformed), coded from the UCDP Conflict Encyclopedia Rebel use of Sexual Violence: Empirical Results Table 1 below displays the results of a logit model predicting rebel perpetrated acts of sexual violence in a given year. We find strong support for hypothesis 1 with our measure of elected leaders in rebel groups. A group having a history of quasi-democratically elected leaders consistently decreases the probability of the group committing acts of sexual violence. 39 Using Hanmer and Kalkhan s (2013) observed values approach, we find that, on average, having a history of rebel elections decreases the probability of a rebel group committing sexual violence by approximately 35 Data from World Bank. World Development Indicators. 36 In Appendix Table 6, we include a measure of gender equality: fertility rate measured as total births/per woman by country year (World Bank. World Development Indicators). Including this measure does not change the substantive (or statistical) effect of rebel elections on the use of sexual violence. 37 Uppsala Conflict Data Program (Date of retrieval: 09/01/01-14/11/07) UCDP Conflict Encyclopedia: Uppsala University. 38 Using the ratio of rebel to state troop forces does not change the results of the models. See Appendix Table As an alternative measure of election history, we use ratio of elected leaders where the numerator is the total number of elected leaders in the group s tenure over the total number of leader changes in the group s history. The results are substantively and statistically similar. See Appendix Table 4. 20

21 10.76 percentage points holding all other variables to their observed values. 40 Substantively, these models indicate that having a history of elected leaders has the largest negative effect on the probability a group will commit acts of sexual violence. 40 This difference is statistically significant at the 95% confidence level and is calculated using the variables in Model 3 of Table 1. 21

22 Table 1. Logit Model - Rebel Acts of Sexual Violence (1) (2) (3) Rebel Sexual Rebel Sexual Rebel Sexual VARIABLES Violence Violence Violence History of rebel elections ** ** ** (0.490) (0.538) (0.598) Territorial control 0.640*** (0.211) (0.241) (0.280) ln(troop size - rebel) 0.247*** 0.257*** 0.250*** (0.068) (0.076) (0.087) ln(troop size - state) *** *** (0.060) (0.064) (0.119) Secondary support - rebel ** ** (0.285) (0.332) Secondary support - state ** * (0.268) (0.292) Independent intervention ** (0.393) (0.419) Democracy (0.502) Ethnic fractionalization 2.515*** (0.647) ln(population) *** (0.163) ln(gdppc) *** (0.247) Constant ** *** (0.833) (0.867) (2.332) Observations 1, Reporting logistic coefficients. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 22

23 Both secondary support to the rebels and independent intervention decrease the probability that the rebel group will engage in sexual violence. This supports our argument that groups with international audiences are less likely to engage in activities that are seen as illegitimate or inhumane and stand to threaten foreign aid. In particular, external support to the rebels decreases their probability of engaging in sexual violence by 5.76 percentage points on average and intervention in the conflict decreases the probability of the rebel group committing acts of sexual violence by 8.79 percentage points on average, holding all else to their observed values. On the other hand, these models suggest that increasing rebel troops makes acts of sexual violence more likely, even while controlling for state troop size. In the Appendix, Table 5, we use a logged ratio of rebel to state troops and find a substantively (and significantly) similar result lending some support to the literature that suggests that relative power may increase violence including sexual violence. On average, increasing rebel troop size from the first to third quartile 41 increases the probability of rebel acts of sexual violence by 2.10 percentage points. Three of the state-level control variables have impacts on the probability of rebels committing acts of sexual violence. In support of Bloom (1999) and Plumper and Neumayer (2006), we find that ethnic cleavages play a pivotal role in predicting rebel acts of sexual violence. The more ethnically fragmented the country, the more likely rebels are to engage in sexual violence. This suggests that ethnic divisions can play a role in acts of sexual violence. Ethnic fractionalization has the largest positive effect on the probability that the rebels engage in sexual violence: increasing the level of ethnic fractionalization from the first to the third quartile increases the probability on average by approximately percentage points, holding all other variables to their observed values. 42 More populous and richer states are associated with a decreased chance of sexual violence 41 This is a change from (first quartile) to (third quartile). 42 This is a change from (first quartile) to (third quartile). The maximum value is 1. 23

24 by rebel groups operating in their territory. Increasing the state population from the first to the third quartile decreases the probability of sexual violence by rebel groups on average by 4.72 percentage points, holding the other variables to their observed values. 43 Similarly, increasing GDP per capita (natural log) from the first to the third quartile decreases the probability of sexual violence by rebel groups by on average 9.48 percentage points, holding all else to observed values. 44 Figure 1. Marginal Effects: Probability of Rebel Acts of Sexual Violence 43 This is a change from (first quartile) to (third quartile). 44 This is a change from (first quartile) to (third quartile). These first differences are also calculated using Hanmer and Kalkan s observed values approach (2013) using the variables in Model 3 of Table 1. The results are statistically significant at the 95% confidence level. 24

25 Causal Inference Although the previous statistical analyses demonstrate that groups with a history of elected leaders are negatively correlated with acts of wartime sexual violence, does the electoral process confer legitimacy that constrains groups from engaging in sexual violence? We address this issue of causality in two ways: first, quantitatively, we adopt a statistical causal inference model designed to isolate the effect of elections on sexual violence. As it is possible that there are other processes by which groups may foster legitimacy and decrease the observed likelihood of committing wartime acts of sexual violence, we use a causal modeling method, in particular Wooldridge s regression adjusted estimator, or treatment effects estimator, to parse out the specific effect of rebel elections on committing acts of sexual violence (Cattaneo 2010, Sloczynski and Wooldridge 2014, Wooldridge 2007, Wooldridge 2010). Our second approach is to use causal process tracing to outline the effects of electoral processes on legitimacy building and the rebel use of sexual violence. Treatment Effects Estimators As we acknowledge that the process by which groups assign power is not randomly assigned, and likely the processes by which groups determine leaders is related to their use of wartime tactics including sexual violence, we account for this statistically by using treatment effects estimators to measure the effect of rebel elections (our treatment ) on the group s use of sexual violence (outcome). Treatment effects estimators are designed to simulate experimental outcomes where we estimate the unconditional means of outcomes by treatment levels (Sloczynski and Wooldridge 2014). In both experiments and observational data, we can only observe the outcome for each subject conditional on the treatment the subject receives. The difference being that for experiments, this assignment process is random. Here we model the process by which the treatment is assigned, 25

26 i.e., the process by which groups assign leadership elections or otherwise and how it in turns impacts the group s use of sexual violence. 45 Below we report the Average Treatment Effect (ATE), the mean difference in the use of sexual violence between groups with electorally ascended leaders and those groups whose leaders come to power through non-democratic processes. The results reveal that rebel groups with elected leaders are significantly less likely to have members committing acts of sexual violence than those that do not: the mean predicted probability of acts of sexual violence decreases by approximately 6.20% and is statistically significant at greater than a 95% confidence level. 46 Table 2. Treatment Effects Estimator Average Treatment Effect of Elections on Rebel Sexual Violence Rebel Sexual Violence Coefficient Robust Std. Errors z P> z [95% Conf. Interval] ATE Election Tenure (1 vs 0) < Potential Outcome Means < In Table 3, we report the regression adjusted coefficients on the covariates used in the model to specify both the outcome, sexual violence, and the treatment assignment process, or how groups 45 Although this method is referred to as causal inference modeling, the ability to infer causal effects and the quality of the results are highly dependent on the model specification of the assignment process, and should not be interpreted as the definitive causal effect. 46 The results above use a Regression Adjustment (RA) estimator specifically to model the nonrandom treatment assignment; the results are robust, however, across several causal model specifications including the inverse probability weighting (IPW) estimator, the IPW with regression adjustment estimator, and the augmented IPW (AIPW) estimator, which adds bias-correction term to the IPW estimator to account for misspecification in the treatment model (see Table 3). 26

27 elect leaders. Specifically, we use an Augmented Inverse Probability Weighted (AIPW) estimator which combines the double-robust properties of the IPW estimator, meaning that the estimates of the effects should be consistent if either the treatment or outcome models are misspecified (Wooldridge 2010). In addition, the form of this model includes a bias-correction term, which adjusts for issues in the specification of the treatment model, important as this is the process by which we statistically account for the nonrandom assignment of rebel group leaders either electorally or otherwise. Model 1 displays the regression adjusted coefficients for the untreated condition, or for the case where leaders are not elected into power, and Model 2, the adjusted coefficients for the case where rebel leaders are elected into power. In comparing the two conditions -- with (1) and without elected leaders (2) -- many of the former statistically significant effects on sexual violence disappear when accounting for the electoral process. Using the conflict in Bosnia-Herzegovina between the years as an illustrative example, conservative estimates of the reported rape and sexual violence totaled 20,000 women (with an early report by the government estimating 50,000 women), approximately 6,666 women were victims of sexual violence per conflict year. 47 Decreasing the probability of wartime rape in a given year by 6.20% would mean, on average, 413 less women a year would have undergone these acts of sexual violence. Descriptively, only three rebel groups with elected leaders were reported by Amnesty International, the Human Rights Watch, or the U.S. State Department to have committed acts of sexual violence against women between the years 1989 and 2009, in comparison to 44 rebel groups whose leaders came to power through some other process. Taken together, these results 47 The Parliamentary Assembly of the Council of Europe estimated that 20,000 women were subjected to rape and other forms of sexual violence during the years ,000/3 ~ 6,666. 6,666*0.062 = See: 27