Burundi’s rebel groups and the stigmatisation of wartime sexual violence

ABSTRACT Despite assumptions about the ubiquity of wartime sexual violence, some armed actors work hard to generate negative views of rape and other abuses. This article qualitatively explores rebel group stigmatisation (and stigma) of wartime sexual violence and prohibitive normative practices. Regularly discussed with reference to the shaming of victims or survivors, stigmatisation is nonetheless utilised here as a concept for understanding how sexual coercion is “made” deviant and consequential for potential perpetrators. Two rebel groups from Burundi’s civil war (1994–2008), CNDD-FDD (National Council for the Defence of Democracy-Forces for the Defence of Democracy) and FNL (Palipehutu-Forces for National Liberation), are compared. The FNL stigmatised rape and sexual assault, and the nature and quality of its practices shaped negative social norms surrounding rape. The article’s main contribution is to demonstrate the need to deepen and widen the evidence base on the prevention of wartime sexual violence.


Introduction
The rebel groups The CNDD-FDD The CNDD-FDD launched its armed campaign in 1994, in response to the October 1993 assassination of Melchior Ndadaye, the first democratically elected Hutu president. This event precipitated the country's descent into full-scale civil conflict. The rebellion underwent internal splits and divisions, one of which resulted in Pierre Nkurunziza's assumption as chairman in 2001. Between 8,000 and 12,000 armed soldiers were part of CNDD-FDD (Nindorera, 2012). Less than 5% were women, who were not represented in the CNDD-FDD high command (Nindorera, 2012). It relied on support from a range of sources (Dilworth, 2006). It taxed the Burundi population in areas it controlled. Arms, funds, travel documents and safety were provided, to some extent, by governments in neighbouring countries. The CNDD-FDD used persuasion as well as the threat of force to secure civilian support. It had a reputation of forcibly recruiting young men, and was alleged to have abducted local administrators to demonstrate the government's inefficacy (Human Rights Watch, 2003).
When given the opportunity, CNDD-FFD refused to participate in the international negotiations that led to the 2000 Arusha Peace and Reconciliation Agreement that was signed in Arusha, Tanzania (and is often referred to as the Arusha agreement or accord). It and FNL objected to the settlement's power-sharing stipulations: quotas ensuring an ethnic equilibrium between the majority Hutu and minority Tutsi groups. Further negotiations and fighting continued. Eventually, in November 2003, CNDD-FDD agreed to a settlement. In 2005, Burundi held elections for municipalities, parliament, the senate and the presidency, with CNDD-FDD claiming the majority. Nkurunziza was elected president. However, FNL persisted in demands for constitutional changes to exclude power-sharing. The armed conflict continued between the CNDD-FDD-led government and FNL.

The FNL
The FNL emerged from Burundi's 1972 genocide, and espoused an uncompromising ethnic ideology against the Tutsi minority (Watt, 2008). The rebellion was estimated as a force of 2,000-3,000 fighters (IRIN News Service, 1999). An unknown, but observable, minority within the movement were women. The group was most active in eastern Burundi and along the border with DRC (then Zaire). The FNL members were required to take up Christian, mainly Protestant, evangelical beliefs. They held prayer sessions regularly, including before and after offensive operations. Among other strategies, FNL raised funds for operations among villagers and residents of Bujumbura, and recruited child soldiers.
There were reports of civilian abuse, e.g. the massacre at Gatumba (Human Rights Watch, 2004;Watt, 2008).
As noted earlier, following the 2005 elections, FNL continued its military campaign against the CNDD-FDD-led government. Initially, Agathon Rwasa, the main leader of the rebellion, resisted international pressures to negotiate. Ultimately, FNL signed a comprehensive agreement with CNDD-FDD in September 2006, but engaged in some low intensity fighting until 2008.

Contribution
This article aims to contribute to understanding sexual violence behaviour, and its prevention, in the context of civil war. It complements and relates to empirical research on restraint (see Hoover Green, 2016Green, , 2017Green, , 2018; International Committee of the Red Cross, 2018; Wood, 2009Wood, , 2014Wood, , 2018. I expand upon research pioneered by Wood (2009Wood ( , 2014Wood ( , 2018Wood & Toppelberg, 2017) that underscores the role of within-group institutional processes in the production of wartime rape. The article contributes, at a granular level, to the conceptualisation of the "content" of the institutional practices that result in shame and derision for sexual coercion. It explores two cases from one conflict and social context, illuminating how actors with similar aims and backgrounds can make different choices, leading to different norms and preferences for or against sexual abuse and violence. The main research question was: How do non-state armed groups stigmatise sexual violence?

Definitions
Wartime sexual violence 3 is defined as rape, sexual torture, abuse, forced prostitution or acts of a sexual nature based upon coercion by a party to an armed conflict. The scope of the study is limited to violence that is not organised, ordered or explicitly sanctioned. Wood (2018) would characterise this type of sexual violence as a practice, as a "pattern of violence that is not authorised or ordered but tolerated by officers, and one that is often driven by social dynamics among soldiers, not just by individuals acting on private preferences" (Wood & Toppelberg, 2017, p. 621). It is not part of the formal or authorised repertoire of the armed group (see Hoover Green, 2016Green, , 2017Green, , 2018. It is not utilised for operational or tactical reasons, or instrumentalised to threaten, punish or control an enemy group, members of the rebellion or the general civilian population. The individual or group of perpetrators may commit abuses indiscriminately or opportunistically. Based on the Uppsala Conflict Data Program (UCDP), I define non-state armed groups (UCDP, 2009) as organised opposition actors or organisations that use force or advance organised violence to contest the incompatibility in an armed conflict. The scope of this research is limited to formally organised groups, with a clear name, organisation and political identity.
Theoretical framework I follow the theoretical insights from Wood's (2009Wood's ( , 2014Wood's ( , 2016 work on different types of wartime rape and Hoover Green's (2016Green's ( , 2017Green's ( , 2018 explication of leadership, political education and training in the narrowing and widening of repertoires of sexual violence. An essential premise is that armed group leaders must, as Weinstein (2007, p. 135) states, "create and rely on an internalised set of professional norms" in order to control the way that their organisation achieves results. Militaries and rebel groups alike may attempt to limit the extent to which their members engage in abuse of civilians (Bell, 2016;Thaler, 2012;Weinstein, 2007). It is not enough to focus on the outcome of victory on the battlefield, commanders strive to avert defections and to ensure that their subordinates carry out orders in tandem with the leadership's strategic objectives. Social norms and how these are transmitted can contribute to this process of behavioural control, and cumulatively, to organisational preferences for acceptable acts of physical aggression, or conversely, their prevention (see for example Bandura, 1991Bandura, , 1978Checkel, 2017;Cohen, 2013Cohen, , 2016Cohen, , 2017Hoover Green, 2016Humphreys & Weinstein, 2006;Tilly, 1993;Weinstein, 2007;Wood, 2009Wood, , 2014Wood, , 2018. Of concern to this article are the processes that generate such norms and preferences, and how they are reinforced.

The stigmatisation (and stigma) of wartime sexual violence
The main outcome of interest is the stigmatisation (and stigma) of wartime sexual violence, marking coercive sexual conduct as negative, perhaps even shameful, unworthy or deviant. I suggest that stigma is not a foregone conclusion of formal edicts or acts, but also an outcome arising from informal processes that "make" sexual violence consequential and status-reducing. Most likely, stigmatisation is dynamic, fluid, and difficult to observe systematically. It is surely subjective to a given social group (in this article the rebel group), and difficult to extract from norms that are built-up, reinforced, reformed and rebuilt, sometimes anew. Nonetheless, the starting point of this article is that it should be possible to identify and observe stigmatisation and stigma. For instance, members of an armed group might fear being exposed by their compatriots for using force or violence in their sexual relations. Group members might pressure one another to avoid any association with rape. Coercion of sexual partners could be deemed inferior, low-status behaviour.
From a theoretical perspective, stigmatisation is chosen for its utility in understanding conflict actors' sexual violence norms and preferences and its potential relevance to insights about the prevention of abuses. First, the concept is normally discussed in relation to the poor treatment and shaming of victims and survivors of sexual violence. Developing it with regard to perpetration is unusual. Initially, the relevance of this concept for the prevention of wartime predation arose from anecdotal observations of sanctioned sexual conduct in some peacetime contexts, and masculine (sometimes contradictory) ideals of "romantic" or "consensual" behaviour that denigrated sexual aggression. In such a paradigm, rapists or sexual predators are depicted as sexually and socially inferior. Theoretically, this type of stigmatisation could also apply to perpetrators of rape in war contexts. Perhaps, stigma could contribute to explaining how some armed groups inculcate preferences for consensual sex or sexual abstinence. Second, exploration of stigmatisation in armed conflict could provide additional information about the processes that contribute to variation in wartime sexual violence. While individual sexual behaviour ultimately fluctuates and differs depending on background and situational factors, at the level of a group, such as the rebel organisation, norms and preferences are shaped socially. In turn, norms and preferences feed into the individual's situation. Understanding stigmatisation could enhance our overall comprehension of the factors that contribute to wartime sexual violence, as well as its restraint. Stigmatisation allows us to develop theoretical insight earlier in the causal chain, and to identify proximate conditions of prevention.

Prohibitive normative practices
The main constitutive elements of stigmatisation, resulting in stigma, are prohibitive normative practices, defined as disciplinary measures and procedures. If stigmatisation is the outcome (dependent variable), the answer to how it arises is found in these measures and procedures and their unfurling. Narrowly, prohibitive normative practices are evident in disciplinary rulemaking, rule-dissemination and rule-enforcement from the top-down; and more broadly, they include social interactions across and throughout the organisation. Most armed actors have some rules to restrain individuals. These rules may be managed explicitly through training, political education and disciplinary practices (Hoover Green, 2016. In this article, I seek to illustrate how an armed actor may carry out such practices. I propose that how rebel groups forbid sexual violence (rule-making), communicate this prohibition (rule-dissemination) and follow through both from commanders, and across the organisation (rule-enforcement), are the main components of stigmatisation. Without some aspect of these practicesfor instance a costly rule against rapethere could be no stigmatisation or stigma. If an actor does not work hard to maintain and enforce a prohibition and stops at rule-making, intense socialisation must transmit and entrench the code. Hoover Green (2016) indicates that such implicit approaches are also relevant, since they valorise specific types of violence, and imbue fighters with identities that narrow the acceptable repertoire of violence.

Design
The qualitative research design is based on case study methodology, an approach belonging to "building block studies" (George & Bennett, 2005, p. 76), and that is particularly relevant for developing theoretical insights about new concepts. The unit of analysis is the non-state armed group. My case selection of CNDD-FDD and FNL is based on John Stuart Mill's method of difference, whereby the researcher explores cases that vary in the outcome of interest, which in this article is the stigmatisation of sexual violence. The cases should be comparable, and I have hazarded that this is so, given that they recruited from the same population and had the same motivations for entering into armed conflict. In both groups, most members, regardless of rank or geographic origin, spoke the same language and had the same religious background, mainly Catholicism, or evangelical Protestantism. They also experienced the same gender and social norms at the outset of the war. Their leaders had similar socio-political backgrounds. Still, there is one critical difference. The insurgencies differed in terms of ideological commitment to a "one man-one vote" end to the civil war. During the period under study, CNDD-FDD became a ruling party through its acceptance of a power-sharing arrangement and the principle of ethnic quotas, and it acquired control over the state and its use of force. Nonetheless, the main insights of this study are from the time of rebel activity and orientation.
The data in this study were gathered using different collection methods. First, secondary sources formed the basis for empirical material about patterns of sexual violence of the armed groups, and for triangulating and contextualising the findings from qualitative fieldwork. Second, the study's main findings are based on qualitative analysis of focus groups and semi-structured interviews.

Data and procedure I: Approximating armed group patterns of sexual violence
The focus of this article is the stigmatisation of sexual violence. Presumably, stigmatisation will result in lower prevalence or incidence of rape and other abuses. Nevertheless, this causal relationship is difficult to prove, and is beyond the scope of this article. Sexual violence can have multiple causes. Even so, it was useful to establish an approximation of the patterns of abuse and assault that may be attributable to CNDD-FDD and FNL. Thus, data used for this article include information from a review of sexual violence events reliably attributed to armed groups from the period of 1998-2011 that was undertaken for an earlier study that examined armed group impunity for sexual violence (Muvumba Sellström, 2015b).
A systematic search of media and journal articles was conducted, using the Dow Jones Factiva Database; and reports by Amnesty International, Human Rights Watch and the US State Department. These latter three are sources used in the Sexual Violence in Armed Conflict Dataset (SVAC) (Cohen & Nordås, 2014) which assesses prevalence of wartime sexual violence. My review focused on events whereas SVAC assessed overall patterns, based on the descriptions in these same sources. The search command in Dow Jones was a variation of the conflict party name followed by: AND "sex*" OR "sexual violence" OR "rape*" OR "raping" OR "tortur*" OR "mutil*" for the government of Burundi, CNDD-FDD and FNL. In order to be attributed to CNDD-FDD or FNL, each specific sexual violence event had to be clearly identified as perpetrated by the actor, and not only identified as nonstate groups or "soldiers" or "rebels". Finally, reports had to include other types of details, such as timing (with a particular month, day or period identified) or location. This information was supplemented by insights from background questions about sexual violence, gathered during the qualitative fieldwork.
Overall, qualitative material proved useful for triangulation with the secondary sources on a number of factors of the study, but mainly was utilised to assess, confirm and specify differences between the armed groups in relation to stigmatisation (and stigma).

Participants
Nineteen focus groups (see Appendix I) were conducted in Burundi in November and December 2011 and in January 2013. A total of 74 individuals (10 women, 64 men) participated as respondents in the focus groups, the majority were ex-combatants of CNDD-FDD or FNL. The average age was 31 years. Eighty-eight percent of the participants described themselves as Hutu. Thirty-two participants had been fighters in CNDD-FDD between 1993 and 2006. Of these, 31% (10 out of 32) were female and 69% (22 out of 32) were male. Importantly, four of the women from CNDD-FDD left the group to join FNL and demobilised in 2008. Most of the ex-combatants from CNDD-FDD had served under Nkurunziza. Forty-four participants in the study had been fighters in FNL between 1994 and 2009. Of these, 14% (6 out of 44) were female and 86% (38 out of 44) were male. All of the FNL ex-combatants served under Agathon Rwasa. Seven of the focus groups were mixed in terms of gender. Two groups had only one woman in the group. In addition, I conducted 20 (19 individual) separate interviews with 6 former CNDD-FDD soldiers, 7 ex-FNL combatants, and 7 local journalists, experts and civil society members.
The participants in the focus groups were selected through purposive snowball sampling, based on their affiliation to a particular armed group. They were chosen for their knowledge, experience and service. Often, the participants in a focus group knew one another. Selection was voluntary and sampling was not designed for the purposes of probability or representativeness. I assembled several homogenous focus groups for both CNDD-FDD and FNL. I began my fieldwork with assistance from an interlocutor at a civil society organisation that worked with ex-combatants. This associationthat wishes to remain anonymousalso provided translators and, in some instances, a place to conduct focus group sessions. Through it and its contacts, different respondents were suggested. This assistance was essential, since I could not independently identify and verify an authentic, considerable and diverse set of ex-combatants. In turn, participants recommended others, including people from, or living in different parts of Burundi. Since CNDD-FDD and FNL had varied loyalties and wartime activity in different geographic strongholds, focus groups took place in four parts of the country, namely Bujumbura, Rumonge, Gitega and Bujumbura-Rural. Finally, in order to enrich the comparison of information and viewpoints, a few focus groups included ex-combatants from other armed organisations or were heterogeneously mixed between CNDD-FDD and FNL.

Focus group format
The focus group format was used for learning about stigmatisation and stigma of sexual violence and prohibitive normative work of CNDD-FDD and FNL, the central focus of this article. Focus groups were used to draw out the nature of social interaction and to identify composite ideas, memories and perceptions at group, not individual, level. This form of data collection seeks out responses that reflect the social relations among participants and between them, and allows these different power dynamics to shape the outcome of each discussion (see Aryton, 2018;Frith, 2000;Söderstrom, 2011). This approach has proven useful for sex research (Frith, 2000), and the study of sensitive issues around conflict (Söderstrom, 2011). The social setting, rhetorical phrasing, identity and demeanour of all those participating has a bearing on the outcome of focus group discussions. Agreement and discord, or dialogue among the participants can contribute to deeper understanding of group perceptions (see Aryton, 2018;Frith, 2000;Söderstrom, 2011). Such interactions reveal dominant hierarchies both in terms of identities and attitudes, as well as social preferencesa main objective of the research design.

Focus group procedure
I conducted all of the focus groups myself. Each focus group followed the same pre-determined order of discussion and set of questions. Focus groups (and semi-structured interviews) were not recorded because of the sensitive nature of the topic(s) (i.e. politics, the civil war, justice, identity, family life, sex and sexuality (at the time)). Recording could deter participation and hinder frank exchange, particularly if a respondent was concerned about his or her future security and social and political status. Instead, detailed, precise notes were taken during the focus group/interview, written up, coded, anonymised and stored in a password protected file.
The same set of questions, in the same order, were interpreted by a translator from English to Kirundi. Participants took turns speaking, and waited until interpretation was completed, before providing further responses. The pacing also fostered a measured tempo and communication style. Probing, clarifying and validating questions were asked to gauge the level of consensus or disagreement within the group. The study covered 35 questions, not all of which dealt with rules against sexual violence. Translators were prepped ahead of time by me, and trained in the methodology, approach and pacing. Most sessions lasted two to three hours.

Informed consent procedure, anonymity and confidentiality
The procedure for each focus group as well as the individual interviews included an overview of the purpose of the project, a brief introduction to my background and a verbal process for acquiring informed, voluntary consent. Respondents were assured of anonymity and confidentiality. I stressed that I did not wantfor ethical reasonsto learn about specific sexual violence events, but instead, to know more about life in the armed group and the way that sexual violence was addressed. During a few, all-male focus groups, actual events of rape were discussed, although the identities of victims, perpetrators and precise dates of these events were not invoked.

Analytical approach
I reviewed the empirical material from the focus groups, interviews and the secondary materials (also used in determining the sexual violence patterns of the rebel groups), triangulating different sources and evidence using a series of indicator questions. These questions were asked of each case and then in a cross-case comparison.
First, I assessed the material for how the rebel groups described the consequences for wartime sexual violence. The analytical steps included answering the following questions: . Did members describe their organisation as willing to mete out consequences for sexual violence? . How did the armed group view cohort perpetrators of sexual violence? . What would be the status of a perpetrator, within the group? Second, to understand the prohibitive normative practices, I explored the data for insight about the rule-making, rule-dissemination and rule-enforcement against sexual violence: . Did the armed group have a code prohibiting sexual violence? . Was the rule clear? . Could members recall it? . Was it written down? . How was it disseminated? . Were there any respondents that did not remember the prohibition? . Did all the participants, or most, recall the rule and its enforcement? . Was ít costly to commit sexual violence? If so, how? . Did the armed group alter the rule within its units or in particular times and circumstances? . Did armed group members think it applied to leaders as well as foot soldiers?
Triangulation among the various sources of the study entailed balancing between the credibility, interests and reliability of respondents and various sources of information. Focus group material was assessed for consistency across different ranks, experience, gender, age, affiliation and for how these differences could influence the inputs made in each session.
Subsequently, I interpreted the evidence from the focus groups through the theoretical framework, and utilised the characteristics of clarity, costliness, depth and constancy to organise and assess the quality of rule-making, rule-dissemination and rule-enforcement. My emphasis was on the relevance of this prohibitive work for enacting stigmatisation and resulting in stigma.

Armed group patterns of wartime sexual violence
Respondents used the term for sexual violence in the local language, Karundi, gufata konguvu/ k'nguvu (translated as forced sex), to describe rape, sexual coercion and abuse and assault associated with the threat of force or violence (Muvumba Sellström, 2015a). This formulation mirrors the definition of wartime sexual violence used in this article, since it incorporates an array of acts well beyond single perpetrator rapeincluding forced marriageand resonates with the meaning of coercive sex within both the research and policy communities.
Based on the review of sexual violence events, during 1998-2011, out of 56 reports, only two events were attributed to FNL and three to CNDD-FDD in 2003 and 2004, respectively. Going further, in the period after CNDD-FDD's election (2005)(2006)(2007)(2008)(2009)(2010)(2011), reports of the Burundi government's sexual violence acts were more numerous than that of FNL. Since the two groups continued to be in conflict until 2008, this difference is notable. The government was linked to 16 reports of events during 2005-2011 while FNL was attributed to three.
The focus groups and interview respondents also referred to sexual violence, rape and other assaults perpetrated by FNL and CNDD-FDD members alike. Significantly, the overwhelming pattern they attested to was that FNL was less likely than CNDD-FDD to be associated with sexual harms, coercion or aggression. The view that FNL committed less sexual violence was supported through interviews with experts (e.g. Author interview no.10 and Author interview no.16 from 2013) who were not aligned with FNL. Further, all the focus groupsregardless of their heterogeneity or homogeneity, their mixed status in terms of gender or armed group affiliationclaimed that FNL was much less likely to order or permit sexual violence, while CNDD-FDD was viewed by insiders and outsiders as perpetrating wartime rape. This is particularly important as ex-combatants from the state military or CNDD-FDD were attesting to the FNL's (their rival) better sexual conduct. In addition, there is no evidence that FNL used or allowed sexual violence in its assaults on civilians, such as the August 2004 attack on a refugee transit centre in Gatumba. Moreover, a subsequent version of the FNL later operated in Uvira, South Kivu, DRC, and its fighters were not reported to commit sexual violence as part of their operations (Van Acker, personal communication, November 18, 2014). Overall, this implies restraint of sexual violence among FNL cohorts, even after the group's original members were demobilised and its leadership changed (Author interview no.18, 2013).

Stigmatisation (and stigma) of wartime sexual violence
The stigmatisation of wartime sexual violence differed between CNDD-FDD and FNL. Of the two, CNDD-FDD former fighters were prone to describe sexual violence in war (and in peace) with euphemistic language, as a normal everyday occurrence, arising from romance or even, love. CNDD-FDD members described sexual violence as part of human nature, with men finding an opportunity during warfare in order to "enjoy sex" (Author interview no.4, 2011). In contrast, the FNL fighters described rape negatively, as disruptive to the social order and operations (Author interview no.5, 2011). Their image of an ideal soldier was of heroic sacrifice, including in terms of sexual conduct. This was a way to demonstrate allegiance to the armed organisation.
CNDD-FDD cohorts tended to dismiss incidents of sexual violence as harmless. In contrast, FNL members uniformly described rape as consequential, problematic, harmful and, to some extent, status-reducing. Among ex-CNDD-FDD, the moral meaning of wartime sexual violence was situational. Negative connotations applied to perpetrators who had violated young children in one instance, or if a victim was related to a commander in the rebellion. For former FNL fighters, all coercive acts were unwelcome and threatening to the group and the population. To be clear, some members reported the occurrence of consensual as well as coercive sex within the FNL, however, they distinguished between sexual or romantic partnerships that were forbidden, and unsanctioned aggression for sexual gratification. The ex-FNL respondents did not provide caveats or conditions in the denigration of all wartime sexual violence. The FNL and CNDD-FDD came from the same society, and shared similar backgrounds and experiences. Yet, all former or current FNL members, regardless of the composition of the focus group they participated in, their prior deployment, time in the conflict or experiences and status after the war, depicted rape as illicit.

Clarity
The CNDD-FDD armed rebellion did not have a clear rule against sexual violence. Participants would explain that sexual violence was banned. However, the rule was not articulated with precision across the different focus groups. Indeed, they often provided contradictory depictions. They described the actual rule and the punishment for sexual violence differently: Many ex-combatants reported that there was no written code of conduct, giving evidence of unclear rule-making. Further, they did not present systematic information about the investigation and prosecution of sexual violence perpetrators.

Costliness
The statements made indicated that fighters within the CNDD-FDD would not have been able to predict the penalty for sexual predation, abuse or assaults since its costliness varied. Focus groups explained that the consequence for sexual violence could entail corporal punishment, deprivation, killing or detention. A former combatant from Gitega noted that: Punishment depended on the mistake you had done [my emphasis] … you could spend time in the trench, this was a punishment. It depended on the region you were in [my emphasis], and then you would make an agreement to the victim (woman) and if you were caught they made an investigation, if it is found you used her, you could be killed, if it was found that it was a genuine agreement between you and the victim you would be beaten or put in jail (trench, cave) held by ropes and/or the prisoner spent time in isolation [with limited food and water]. (G16, 2013: P1) Moreover, members of CNDD-FDD could, to some extent, mitigate the costliness of the penalty. Commanders could intervene to alter or influence investigations, thereby circumventing rule-enforcement. Suspected perpetrators could use their power, leverage resources to influence their commander's judgment, change testimony and/or enter into informal, customary (and highly coercive) marriages with victims.
Depth I asked whether punishment for sexual violence, despite the variation in its costliness, would have been applicable to commanders. The ex-combatants provided examples of the kinds of cases that would warrant punishment of a commander, for example, if the child of a CNDD-FDD leader was attacked and molested. Mainly, ex-combatants agreed that rules regarding sexual violence were either not applicable to commanders or were not enforced uniformly, particularly in the theatre of the war, in the bush or at the frontline: Who will make a judgment against him? A chief was a chief, you must accept such rules. It was [also] very difficult because we were in the bush. If the civilians reported it, he [the perpetrator] could be beaten or killed … . ( G19, 2013) CNDD-FDD-only focus groups also suggested that leaders' invulnerability to the consequences varied depending on the unit they served in, and their operational base. In summary, CNDD-FDD's enforcement was variable, with some commanders enjoying impunity for sexual aggression.

Constancy
A further topic for the focus groups was whether the CNDD-FDD changed its approach to sexual violence over time. Former CNDD-FDD combatants pointed out that prohibitions became more definitive as the insurgency negotiated peace. First, during the period before 2002, sexual violence was punished "rarely" (G19, 2013), and "not a lot of people were punished" (G16, 2013). Then, CNDD-FDD became more vigilant in the transitional stages after the 2003 ceasefire agreement with the then transitional government of Burundi. Several respondents agreed that " … these rules came after the ceasefire. They came from the government and the rebel army". (G19, 2013). Although the death penalty was never universally a consequence for sexual violence in the armed group, for some, " … the [CNDD-FDD] force became bigger and things became more disciplined. [And,] things were the most disciplined at demobilisation" (G16, 2013).

Clarity
The FNL ex-combatants agreed that its code of conduct demanded abstinence from sex, alcohol and smoking. Unvaryingly, the FNL participants recalled that sexual violence was forbidden and punished severely, Punishment was death for gufata ku nguvu [forced sex] (G10, 2013) It happened that FNL members raped as did members of the national army. But the difference was the punishment. If you were caught [in FNL] you would be killed (G4, 2011: P3) All focus groups reported that the FNL's response to a sexual act that was associated with force or violence was execution. Consensual liaisons, while forbidden, were punished severely if the offender was a lower-ranking soldier (G3, 2011; G9, G12, G14, G17, 2013).
The movement's main prohibitive work revolved around its code of conduct; leaders' documentation of the code and those who violated it; and the established penalty. Leaders and commanders repeated the prohibition in other formal and informal interactions, and dissemination occurred throughout the rebellion's units. This information was corroborated by all ex-FNL as well as former combatants from other armed organisations in this study. Overall, respondents agreed that the FNL rule was written down. Recruits were instructed in the code at induction and during the early stages of their training. The rule was disseminated through the organisational hierarchy and chain of command, verbally as well as through documentationthe commander's notebook. Commanders used notebooks for recording activities, violations and various sentences. Hence, focus group respondents noted that: There was a rule book and it was interpreted to determine the punishment. (G8, 2013) Yes, there was a well-written code that existed. It was hand written by chiefs in their own books. (G9, 2013) FNL rules were written in a small notebook when they joined. We were taught the rules. (G13, 2013) You were told the rule when you joined the group. (G12, 2013) This information illustrates the codification of the armed group's rule. Given the historic pattern of forced marriage in Burundi, the effort and investment in FNL's explicit prohibition against sexual violence is noteworthy.

Costliness
Turning to the high cost of the penalty for sexual violence in FNL, respondents described incidents of stoning perpetrators to death, or the use of capital punishment by blunt instrument: You would be killed, and death was the punishment. A hoe was used to beat perpetrators to death and they were killed. (G12, 2013: P4) You would be killed and they used a hoe, it was a less expensive instrument and they used it to kill the guilty. (G13, 2013) Members, particularly of a low rank, suggested that individuals who could not abide by the rule were unlikely to remain in the group and desertion was a genuine concern. Again, respondents provided similar information: It [sexual violence] happened very little since the law was so harsh, most likely [a] strategy of perpetrators who were caught was to try to escape, to run away. (G12, 2013) The general sentiment among FNL participants in the focus groups was that sexual violence incidents were uncommon, because of the costliness of rape. Sexual aggression and coercion, therefore, were consequential.

Depth
Variation in the depth of the FNL prohibition is evident in terms of who was likely to be punished with death, as opposed to a less costly punishment. Overall, the respondents concurred that the edict forbidding sexual violence applied across the rebellion and up and down the chain of command. Participants claimed that they knew of perpetrators who were commanders and who had been killed, noting that "some sergeants were killed for sexual violations" (G8, 2013: P2). However, some respondents suggested that there were offenders who were not addressed appropriately or were exempted from execution. Those exempted were likely commanders or foot-soldiers with special access to the higher levels of leadership. Still, most of the respondents agreed that even under exceptional circumstances, the rapist was demoted, detained and punished in some manner.

Constancy
The prohibition of sexual violence and rule dissemination of the code remained the same throughout the civil war, with the death penalty the main consequence. In contrast, there were some adjustments after the final ceasefire. Allowances were made, and the central command agreed that sexual violence would not be punished by execution (Author interview no.18, 2013). This amounted to a shift in rulemaking, with implications for rule-enforcement and rule-dissemination.
After the May 2008 ceasefire agreement between FNL and the Burundi government, the regular members awaited demobilisation in temporary processing centres. Leadership was more diffuse, with power and authority over the members dwindling. The rebel command also sought to swell its ranks in advance of integration with the national military. It sought to reduce defections and desertions. The logic of demobilisation was fundamentally different from wartime, and the FNL cohort engaged more widely in consensual sexual partnerships: After the 2008 ceasefire there were unplanned pregnancies on the increase among the combatants. Also, in the rassemblement camps there was more sexual violence reported, and perpetrators were beaten, not killed. (G8, 2013:P2) To be clear, FNL cohorts distinguished between the rule against any sexual activity, which changed after the 2008 ceasefire, and the prohibition against sexual violence, which did not alter. Notwithstanding, the FNL leadership altered the punishment for rape post-settlement. However, the focus group participants did not seem to hold a different view of the consequential nature of sexual violence, even in the aftermath of the conflict.

Discussion
The aim of this article was to contribute to understanding sexual violence behaviour, and its prevention, in the context of civil war. The main outcome of interest was the stigmatisation (and stigma) of wartime sexual violence. The main constitutive elements of stigmatisation were theorised as prohibitive normative practices, or disciplinary measures and procedures. I proposed that prohibitive normative practices lead to stigma. I conducted a qualitative study of two armed groups that fought in Burundi's civil war -CNDD-FDD and FNL. The design and data collection entailed gathering insights from former combatants of these rebellions and an analysis of the rule-making, rule-dissemination and rule-enforcement within each armed organisation. The results highlighted the disciplinary dimensions of the group's normative work, and less so the types of social interactions that went on. Some of these are explored more in this discussion, since they arose as additional observations in the qualitative material.

Prohibitive normative practices varied between the groups
The ex-combatants of CNDD-FDD cited different types of penalties. There was variation in enforcement within the armed group. Its rule against sexual violence was transmitted in an ad hoc manner. In contrast, FNL had a fixed written code of conduct and a culture of documentation. The ex-combatants of FNL recalled rule-dissemination at the start of their indoctrination and training. The death penalty as a consequence for sexual violence was known uniformly across FNL units and cohorts. All ex-FNL members could recite the rationale and content of the rule. Documentation seemed important in conveying to members the importance of the norm against sexual violence. Recording the rule and violations, by the commanders, professionalised the rule. It was made a part of their responsibilities. While there were key differences between the two groups, both rebel practices granted greater leniency to higher-ranking commanders. Of the two, however, FNL's leaders were more likely to experience some kind of punishment for sexual violence.
In each armed group, the rule was transmitted along organisational structures and through social interactions. In CNDD-FDD, rule-dissemination was limited. Its fighters never mentioned social pressure around whether one could or could not commit rape. Whereas, ex-FNL respondents described the rule in social terms and it was socialised horizontally through shared religious values. The FNL cohorts denounced one another for sexual violence. In additional observations from the field, ex-members of the FNL recalled that improper behaviour was exposed in collective confessional settings, usually during group prayers that were part of everyday life and social interaction. The leadership of the group used its religious ideals and values, its "ideology", to justify these forced confessions (Author interview no.18, 2013). FNL members could police one another and expressed a specific ideal of soldiering, with sexual violence as deviant or status-reducing and sexual aggression depicted as the "behaviour of animals" (G8, 2013). In some additional observations, the abhorrence of rape was evident among FNL ex-combatants, who suggested that the leaders who tried to break the rule were punished more. The rebellion's leadership killed a commander after treating him "so badly", since he should have been a good role model and sexual violence was detrimental to "the culture of FNL" (G17, 2013). Furthermore, rape "brought bad luck" (G17, 2013).
Conversely, ex-CNDD-FDD members reported some permissive association between some types of coercive sex and soldieringwith a preponderance of euphemism and an absence of shame for rape or breaches in the prohibition of sexual violence. Its members did not describe rape as harmful to life in the rebellion, nor was there evidence of an organisational moralism, an idealised notion of good conduct. In contrast, ex-FNL fighters idealised sexually pure soldiers, and sexual violence was against the group's own interests. Soldiering in FNL included contradictory motifs: soldiers were meant to be violent, but also prayerful and spiritual. Control and restraint were characteristics of an ideal male soldier in the insurgency. Fighters had to be vigilant and to adhere to the operational plan: "if you go to fight, you go straight to the fight, for there was to be no deviation from the plan" (G14, 2013).
These conclusions concord with existing research on restraint and its origins in political education and training (Hoover Green, 2016, the processes of the socialisation of sexual violence behaviour (Hoover Green, 2017) and the socialisation of violence more broadly (Checkel, 2017). The FNL was able to stigmatise sexual violence, at least in part, based on rule-making, enforcement and dissemination. Its clear and constant approach of clearly denouncing sexual violence, informing its members formally and informally, and routinely and regularly meting out punishment, were socialised across units and up and down the chain of command.

Limitations
Limitations in the study arise from the sources of data specifically, and from the overall research design generally. First, one local civil society organisation was relied upon to provide translators and to identify respondents. This could have affected participant selection, created a bias in the data and ultimately influenced the results. Also, respondents may have felt compelled to provide responses that resonated with the organisation, particularly if the translators were recognised as associated with the organisation and its views.
Second, self-reporting bias and social desirability bias most likely affected the validity of the findings (Breen, 2006;Tourangeau & Yan, 2007). The topics covered in fieldwork were sensitive, with implications for the participants' self-perception politically, socially and morally; as ex-combatants, activists, serving soldiers, local experts or as men or women with sexual, family and social histories of their own. They were unlikely to recall the past accurately and completely. Their responses would have been affected by the time passed since the civil war, as well as their individual preferences, attitudes and values. Social desirability bias, in particular, is likely to have led to omission, exaggeration or misrepresentation. Participants may have wanted to protect their self-perceptions, or to please me in terms of my status as an "outsider", both as a woman of East African origin and a European-based academic researcher.
At the same time, social desirability also existed among the respondents. The focus groups simulated social hierarchies within each armed group. During the homogenous focus groups (both for CNDD-FDD-only or FNL-only members), ex-combatants confirmed or validated different perspectives. Higher status members, such as former lower-to-mid-level commanders or individuals who were more educated, appeared to shape some of the discussions. I interpreted these occurrences as evidence of dominant hierarchies and identities. This was useful and also problematic. The focus groups were part of a research design that aimed to gather insights about group-level social norms and to incorporate the dominant attitudes of each armed group (Söderstrom, 2011). Yet, in doing so, marginal but valid perceptions about the social consequences of sexual violence or the pattern of abuse (for example) may have been silenced or diminished. I relied upon triangulation with other sources and comparison with other focus groups to verify important trends, such as patterns of abuse, rule-dissemination or the role of commanders. Nonetheless, the likelihood that respondents were influenced by one another is a major limitation.
Third, the methodological choice not to record the focus groups/interviews and to rely instead on written notes places a major burden on the researcher. Some information could have been misunderstood or missed. The lack of a recording makes it difficult to verify the raw data independently. Moreover, representations of talk between the researcher and respondents and aspects of speech, such as stutters, pauses, verbal and non-verbal cues, and involuntary vocalisations, are lost. This diminishes independent assessment of the social interaction during the focus groups and weakens the validity and replicability of the study.
Fourth, the gender breakdown of respondents in the focus groups also has implications. Women ex-combatants were over-represented. While the inclusion of more women made more visible the experiences of females in the armed groups, their greater presence may have signalled to respondents that I was interested in female participation, gender equality or women's rights. This cue could have influenced the mixed-gender focus groups. The men may have been reluctant to speak up and the women could have been more forthcoming. Male-only focus groups were sources of information about particular incidents of sexual violence, whereas mixed gender sessions were not.
Fifth, a discussion of limitations should also address the secondary sources used to estimate the patterns of abuse of the armed groups (see Salehyan, 2015). The time period covered with regard to sexual violence events by the actors, 1998-2011, does not correspond to the civil war in Burundi, 1994Burundi, -2008, which is the period of study for this article. This mismatch is due to the parameters of the original study (Muvumba Sellström, 2015b) and should be noted. On the quality of the event reporting itself, several problems with the objectivity, comprehensiveness and consistency of the sources affect the reliability of this data (Cohen & Hoover Green, 2012;Öberg & Sollenberg, 2011). Variation in the actual reports (in depth, language, style and frequency), as well as among the sources themselves (in their institutional expertise, resources, objectives and interests), suggests that these data have inherent irregularities. Reporters, local news stringers, and watchdog representatives may not have had similar resources, skills, authority and capacity to cover events consistently. Perpetrators may not always be easily identifiable by their armed group affiliation. In general, sexual violence is under-reported by victims, also during peacetime (Alcorn, 2014). Comparison with other data sources such as SVAC (Cohen & Nordås, 2014) may show that FNL committed more sexual violence or abuses of a particular type not covered or captured in the event reporting and my estimation.
Sixth, the two groups were not perfectly comparable. Behaviours by each could have been influenced by events within the conflict, the positions of their adversaries, or other explanations. Alternatively, the lack of knowledge about rules against sexual violence by CNDD-FDD members could be explained by its larger size, the many historical cleavages, and it may have been prone to internal, institutional inconsistencies. Social distance between the top and bottom of the chain of command within CNDD-FDD would have been greater than within FNL, since the former was a more complex organisation. This would have implications both for stigmatisation and prohibitive normative work, due to differences in social interactions.
Finally, the scope of this article is limited to particular types of non-state armed groups and to sexual violence that is not ordered or authorised. It is not meant for generalisation to all actors using organised armed violence or all forms of sexual violence. Further research on stigmatisation processes by state military or other non-state armed actors is necessary. Comparisons with other armed groups that have prevented sexual violence will enhance the evidence base. Future similar research on stigmatisation processes, if bearing in mind the limitations of this study, could generate more reliable and valid data. This is necessary, if the relationships between armed group institutional processes and patterns in wartime sexual violence are to be understood.