The Distribution and Consequences of Sexual Misconduct by Peacekeepers in Haiti: A Cross-Sectional Epidemiological Analysis

Background From 2004 to 2017, the UN sanctioned The United Nations Stabilization Mission in Haiti (MINUSTAH) to address Haiti’s political instability. Reports of sexual misconduct perpetrated by peacekeeping personnel, including children fathered by peacekeepers, were made during the course of MINUSTAH. To date no epidemiological studies have analyzed the distribution and consequences of sexual misconduct perpetrated by MINUSTAH peacekeepers. This research will: (1) Explore the association between location (rural, semi-urban, urban) and the sharing of narratives about sexual misconduct perpetrated by MINUSTAH peacekeepers; (2) Investigate the relationship between sharing narratives about sexual misconduct and the desire to engage with the UN/MINUSTAH. Methods In 2017, a cross-sectional survey was administered by Haitian research assistants using SenseMaker®, a rapid mixed-methods data collection tool that allows participants to share a narrative on a topic of interest. In total, 2541 self-interpreted narratives pertaining to the experiences of women/girls in relation to the UN were collected across Haiti. Secondary data analysis using log-binomial regression modeled the relationships of interest. Statistical analysis was performed using SAS 9.4. The non-overlapping 95% confidence intervals demonstrate that the RR point estimates comparing sexual misconduct to positive/neutral experiences were statistically different between the personal and community strata. Regarding personal narratives, narrators of stories about experiences of sexual misconduct had 4.52 (95% CI: 3.34, 6.12) times the probability of rejecting the UN/MINUSTAH compared to narrators who shared positive/neutral experiences. Among narratives about community members, narrators of experiences about sexual misconduct had 2.51 (95% CI: 2.00, 3.14) times the probability of rejecting the UN/MINUSTAH compared to narrators who shared positive/neutral experiences. These results suggest that first-hand experiences of sexual misconduct perpetrated by peacekeepers had the most extreme effect on the probability of rejecting the UN/MINUSTAH. Due to a lower sample size in the family/friends stratum (N=215), the imprecise CI (95% CI: 1.35, 7.11) surrounding the point estimates for the family/friends stratum prevented meaningful inferences regarding the heterogeneity of effect.


Introduction
. To address SEA by its personnel, in 2003 the UN introduced the Zero Tolerance Policy, which recognized and defined two forms of sexual violence within peacekeeping contexts: sexual exploitation and sexual abuse (SEA) [1]. The UN considers SEA a category one offense-the most serious type of misconduct (K. M. Jennings, 2008). Since 2007, 117 allegations of SEA against MINUSTAH peacekeepers have been formally reported to the UN (12). In addition between 2010 and 2020, 34 SEA-related paternity claims have been formally laid against MINUSTAH peacekeeping personnel, with 8 of those having been positively confirmed (13).

Peacekeeper Immunity
According to the Status-of-Forces Agreement (SOFA), military members of national contingents are given immunity from host state jurisdictions for criminal acts undertaken during their official capacity (14)(15)(16)(17). Instead, as outlined by the Memorandum of Understanding (MoU) between troop and police contributing countries (TPCC), disciplinary authority over national military contingents is the responsibility of TPCC (15,18,19). However, it is common for member states to fall short of implementing the MoU, thereby leading to criminal immunity without prosecution in the TPCC (20).

Prevalence of transactional sex
A 2011 cross-sectional study in Libera estimated that 72% (95% CI: 68% to 76%) of women aged 15-25 had engaged in transactional sex (21). Risk factors positively associated with transactional sex included: low educational attainment, low income, increased duration of the relationship, early sexual debut, multiple sexual partners, and previous exposure to sexual violence. A second cross-sectional study in Liberia, this time in 2012, analyzed transactional sex specifically between Liberian women aged 18 to 30 and foreign military personnel deployed to the United Nations Mission in Liberia (UNMIL) (22). In this sample, 56% of Liberian women had engaged in transactional sex and 44% of these did so with peacekeepers. For every additional 1,000 peacekeepers deployed, the probability of a Liberian woman engaging in her first transactional sexual encounter increased by 3%.

Risk factors for reported sexual abuse and exploitation
Using data made publicly available by the UN, Nordas and Rustard (23) examined the variation in reported SEA allegations across different PO between 1999 and 2010. Risk factors positively associated with reported SEA included mission-level factors (type of mandate -i.e. humanitarian assistance and focus on gender equality, high levels of violence during conflict, and large mission size) as well as host-country factors (high level of sexual violence preceding conflict, low GDP per capita, and absence of spousal rape law). Similarity, Moncrief (24) analyzed PO data from 2007 to 2014 from a variety of publicly available sources and concluded that increasing mission size was significantly associated with greater SEA allegations. Furthermore, increasing host-country GDP per capita was significantly related to reduced SEA allegations (Moncrief, 2017). In addition, the disciplinary erosion of the contingent, measured by contingents with reports of misconduct unrelated to SEA, was found to be positively associated with SEA allegations.

Protective factors against sexual abuse and exploitation
Using UN PO data (2006)(2007)(2008)(2009)(2010)(2011), Karim and Beardsley (25) examined the effectiveness of deploying female peacekeepers with respect to reducing SEA allegations. Results indicated negative correlations between the deployment of female peacekeepers and: (i) the number of deaths during PO and (ii) GDP per capita of the host county, suggesting that female peacekeepers were more likely to be deployed to safer areas with that are more economically secure. Karim and Beardsley (26) also investigated the effectiveness of female peacekeepers and gender equality on the number of SEA allegations between 2009 and 2013. The authors demonstrated that an increased proportion of female peacekeepers was negatively correlated with SEA allegations. Gender equality measures of TCPP (measured by primary school and labour force gender ratios) were also negatively correlated with SEA allegations. Lastly, Karim (27) assessed Liberian perceptions of female peacekeepers using micro-data, concluding that civilian contact with female peacekeepers was positively associated with favourable perceptions of them. However, there were no significant associations with perceptions of improved security with respect to SEA, and female peacekeepers faced barriers accessing local civilians, especially women and girls.

Geographic distribution of peacekeeping economies and reports of SEA
Peacekeeping economies have been well studied (28)(29)(30)(31)(32) and refer to "Economic activity that either would not occur or would occur at a much lower scale and rate of pay, without the international presence, of which a UN peacekeeping mission is a central component". (K. M. Jennings, 2014, p. 315) In Haiti, socio-economic vulnerability and ubiquitous gender inequality result in a range of gendered survival strategies that disproportionately affect women and girls, such as trading sex for money, food, or security (i.e. transactional sex) (33). In this context, the influx of foreign, well-paid, and predominantly male peacekeepers magnifies the gendered economies of sex work, transactional sex, and domestic labour. Through their employment in peacekeeping economies, women and girls face an increased risk of SEA; they work in unregulated, informal, precarious, and/or illicit sectors that have no legal protection; face unequal power relations vis-à-vis the peacekeeper employer; and are financially dependent on transactions with peacekeepers to sustain their livelihoods (34). In this analysis we consider the differential effect of urban, semi-urban, and rural peacekeeping bases in relation to peacekeeping economies and risk of SEA. Given that women and girls in rural Haiti face greater socio-economic disparities compared to their urban counterparts, we hypothesize their participation in peacekeeping economies will be greater, and the occurrence of SEA will be higher, in rural locations, compared to urban and semi-urban locations.

Exposure to sexual misconduct and future engagement with the UN
Gathering contextual information from civilians is often essential for the success of PO, and yet it can be challenging for peacekeepers to gather this information when they are not cohesively integrated into the host community. The cooperation and engagement of local civilians has been shown to depend on the degree to which local communities positively perceive peacekeepers (35). Local civilians may not have access to reliable information about peacekeepers (36) and instead, may rely on personal, vicarious, and community-level experiences regarding peacekeepers' day-to-day activities. In fact, civilian knowledge of SEA has been shown to negatively affect perceptions of peacekeepers and the UN, thereby reducing cooperation and engagement with PO (35,37,38). Thus, we hypothesize that Haitians exposed to sexual misconduct by MINUSTAH peacekeepers will be less likely to engage with the UN, compared to Haitians who are not exposed to sexual misconduct by peacekeepers.

Contribution of Research
This research provides an in-depth and contextualized understanding of SEA in Haiti. Based on community perceptions (as opposed to formal UN allegations), the analysis allows a grounded understanding of personal, vicarious, and community-level exposure to SEA in urban, semi-urban and rural regions, while also highlighting how these experiences impact Haitians' willingness to engage with the UN. The results have implications for how/where the UN focuses its SEA prevention interventions and resources and provide further evidence regarding how perceptions of SEA impact host-community cooperation and engagement with the UN and future PO.
[1]The "zero-tolerance policy" defines sexual exploitation as: "the actual or attempted abuse of a position of vulnerability, differential power, or trust for sexual purposes" and sexual abuse as: "actual or threatened physical intrusion of a sexual nature, whether by force or under unequal or coercive conditions" (46). Transactional sex, which is also prohibited, is defined as "the exchange of money, employment, goods, or services for sexual activity between beneficiaries of assistance and peacekeeping personnel" (46).

SenseMaker Survey
SenseMaker is a mixed-methods data collection tool and research methodology (39). After collecting micro-narratives related to a phenomenon of interest, SenseMaker requires participants to share their related feelings, attitudes, and perspectives by responding to a series of analytical questions. (40).
We used three prompting questions to elicit narratives about how women and girls interact with MINUSTAH peacekeepers; no questions explicitly mentioned sexual misconduct perpetrated by peacekeepers.
After recording a narrative, participants self-interpreted the shared experiences by plotting their perspectives spatially using dichotomous sliders (dyads) or between three possible options (triads) ( Figure 1 and Figure 2 in Appendix B). SenseMaker then assigns quantitative values to the plotted self-interpretations. Participants also answered six multiple choice questions pertaining to demographic characteristics (age, gender, education level, etc.) and narrative characteristics (emotional tone, protagonist, frequency of occurrence, etc.). This study design has also been described in other publications (41,42) The SenseMaker survey was originally drafted in English, then translated to Kreyol and independently back-translated to English. Translation discrepancies were resolved by agreement between the two translators. We pilot tested the survey in Haiti among 54 participants who provided feedback, which was used to refine the survey prior to implementation. Haitian community partners from the Institute of Social Work and Social Science (ETS), the Commission of Women Victims for Victims (KOFAVIV), and Bureau des Avocats Internationaux (BAI) contributed to the survey formation process. To support qualitative analysis, the narratives were audio recorded and transcribed and translated from Kreyol to English. The survey is presented in Appendix A.

Sampling
Twelve Haitian research assistants recruited participants in cities, towns, and villages within a 30km radius of the 11 selected MINUSTAH bases (Cité Soleil, Charlie Log Base, Tabarre, Léogâne, Cap Haiïtien, Saint Marc, Gonaïves, More Casse, Fort Liberté, Hinche, and Port Salut). When the 30km radiuses for two bases overlapped, the research assistants alternated between the two bases.
Accordingly, survey collection was grouped for: Charlie Log Base and Tabbare, Saint. Marc and Gonaïves, Morne Casse and Fort Liberté. To purposefully select MINUSTAH bases we used a combination of community partner knowledge, published peacekeeping data, geographical location (north, south, central plateau, metropolitan areas), base staffing characteristics, and dates of operation. The number of participant narratives sampled per site are depicted in Table 2.
At each location, the research assistants recruited a convenience sample of individual participants, aged 11 and older, from community settings including, markets, post offices, commercial settings, and bus stops. The survey was administered in Kreyol using the SenseMaker application on iPad Mini 4s between June and August of 2017. The research assistants were members of two community partner organizations: ETS and KOFAVIV. The last two authors facilitated a four-day research assistant training on the study design, research protocol, SenseMaker, research ethics, iPad use, data upload, management of adverse events, service referral, and role-play activities for survey delivery.

Geographical distribution of sexual misconduct narratives
Each narrative was coded into a binary outcome according to whether the narrative mentioned or described sexual misconduct. Narratives that mentioned any sexual interaction with peacekeepers including SEA, long-term sexual relationships perceived as consensual, children fathered by peacekeepers, sexual harassment, and sexual assault-were identified as being about sexual misconduct. First-hand experiences of sexual interactions with peacekeepers as well as narratives about the experiences of family/friends, and experiences of other community members were included in the sexual misconduct category. Lastly, sexual interactions between MINUSTAH peacekeepers and Haitian men and boys were also included in the outcome.

Consequences of experiencing sexual misconduct
The outcome was derived from the dyad question asking about willingness to engage with the UN: The primary exposure variable for the second regression model was the subject-matter of the narratives. In addition to the narrative subject-matter, the emotional tone of the narrative (identified by the participant) was used to create the exposure levels. Narratives were grouped into three exposure groups, based on the predominant subject-matter and emotional tone: (i) Positive and neutral narratives of the UN/ MINUSTAH that did not address wrongdoings perpetrated by peacekeepers (referent group; examples include: employment offered by peacekeepers, safety improvement, public service, humanitarian assistance, friendships); (ii) Narratives addressing sexual misconduct and MINUSTAH peacekeepers; or (iii) Narratives about other UN wrongdoings and negative perceptions of the UN unrelated to sexual misconduct (examples include: cholera outbreak, use of tear gas, violence against civilians, water supply contamination, littering, kidnapping, goat stealing, motor-vehicle collisions, improper waste disposal).

Data Analysis
The unit of observation for the two models was the narrative. Multivariate log-binomial regression modeled the relationships of interest. In the first model, a multi-variate log-binomial regression investigated the relationship between geographical locations in Haiti (rural, urban, semi-urban), and the presence of community-level narratives about sexual misconduct perpetrated by MINUSTAH peacekeepers. A multivariate log-binomial regression model also analyzed the relationship between the narrative subject-matter (sexual misconduct, other UN wrongdoings, positive/ neutral experiences with the UN) and desire to reject the UN/MINUSTAH. To create the most parsimonious model and identify any potential confounders, a backwards elimination method was used with a liberal cut-off criterion of p<0.15 during the model building. The data were analyzed using SAS 9.4.

Confounders
Five categorical variables pertaining to participant demographics and two categorical variables pertaining to the narrative characteristics were considered either as potential covariates, confounders, or effect modifiers. Chi-Squared tests assessed the relationship of each variable with the exposure and outcome variables using a 0.05 level of significance. Parameter estimates corresponding to the exposure variable were compared between crude and adjusted models.
Confounding was considered to be present when the percent difference between the parameter estimates was ≥ 10%. For the model examining the geographical distribution of SEA, identified potential confounders included gender, who the narrative is about, emotional tone, income, and age group. Statistically significant relationships existed between the each of the confounders and (i) geographical location (exposure) as well as (ii) whether a narrative about sexual misconduct was shared (outcome). For the model examining the desire to engage with the UN/MINUSTAH, the potential confounders identified a-priori were: gender, who the narrative was about, emotional tone, age group, and location. Statistically significant relationships existed between each of the confounders and (i) Narrative subject-matter (exposure) as well as (ii) Rejection of the UN/MINUSTAH (outcome).

Effect modification
In both models, we measured effect modification for two variables: gender and who the narrative was about. Emotional tone of the narrative was also assessed for effect modification for the first objective but not the second objective since emotional tone was used to create the referent category of the exposure variable (positive/neutral stories unrelated to wrongdoings). Effect modification was tested by including interaction terms in both models with a standard cut-off of p<0.05.

Covariates
For the model examining geographical distribution of SEA, we included relationship status, education, and number of narratives told as covariates associated with sexual misconduct (outcome), without being associated with geographical location (exposure). Location (the eight individual cities, towns, and villages categorized into urban, rural, and semi-urban), education, income, and the number of narratives told were included as covariates associated with the subject-matter of the narrative (exposure) in the model examining the desire to engage with the UN/MINUSTAH. Furthermore, relationship status predicted the desire to engage with the MINUSTAH/UN (outcome) in the second model.

A-priori power
Detectable difference calculations at 80% power and an alpha level of 0.05 were performed prior to data inspection and analysis. At 80% power and an alpha level of 0.05, the smallest detectable difference ranged from 1.685 (most conservative) to 1.170 (most liberal).

Ethics
Research ethics approval for the cross-sectional survey and the present secondary analysis was obtained by Queen's University Health Sciences and Affiliated Teaching Hospitals Research Ethics Board (#6020398). Trained Haitian research assistants introduced the study to participants in Kreole using a standard script and informed consent was obtained prior to the start of the survey. No compensation, financial or otherwise, was provided. To mitigate the risk of psychological distress, the survey intentionally did not ask participants to share experiences of sexual violence. As a community partner, KOFAVIV was able to offer psychological support and counselling when necessary, and referral information for legal advice was provided. No identifying information was collected, and participants were requested to not use the names of real people when reciting their experiences. To ensure data security, all completed surveys were uploaded from the tablets to a secure server and then were subsequently permanently deleted from the iPads.

Results
In total, 2541 self-interpreted narratives were collected from 2191 unique participants. After completing the cross-sectional questionnaire, participants were given the option to share and selfinterpret an additional narrative (up to a maximum of four). Of the 2541 self-interpreted micronarratives n=276 (10.9%) were second stories, n=37 (1.46%) were third stories, and n=10 (0.39%) were fourth stories shared by participants.

Characteristics of the narrators (participants)
The majority of participants were male (69.6%), single/never married (60.5%), below the age of 35 years (66.1%) and attained an income that was considered "average" (63.3%), meaning they had access to two to three household items. Additional participant demographic characteristics are presented in Table 1 Table 2, 74% of narratives about sexual misconduct were shared by Haitian men, this mirrors the general sex distribution noted among the participants (Table 1). Compared to narratives that did not address sexual misconduct by MINUSTAH peacekeepers, narratives addressing sexual misconduct were more likely to be collected in Port Salut and Cité Soleil, more likely to be "about someone else in my community", and less likely to be reported as first-hand experiences experienced directly by the participant (Table 2). Moreover, narratives addressing sexual misconduct were more likely to have a "negative" and "very negative" emotional tone, as identified by the participant. Participants were also more likely to respond that "others must hear this story and take action" if they shared a narrative addressing sexual misconduct. Lastly, Uruguayan peacekeepers were more likely to be identified in narratives about sexual misconduct. [Insert Table 2] Geographical distribution of narratives about sexual misconduct As illustrated in Table 3    Consequence of experiencing sexual misconduct on the willingness to engage with the UN In general, after adjustment, we found a positive association between experiences of sexual misconduct and the desire to reject the UN/MINUSTAH. A similar positive association was also noted between experiences of sexual misconduct unrelated to sexual misconduct and the desire to reject the UN/MINUSTAH. Interestingly, the association between subject-matter of the narrative and desire to reject the UN was modified by who the narrative was about (p-value= 0.0113 for interaction term).
This statistically-significant interaction reveals that while general positive associations are noted, the magnitude of the association differs by who experienced the misconduct.
The heterogeneity of effect was identified between the personal stratum and community stratum. Regarding personal narratives, narrators of stories about experiences of sexual misconduct  310 narratives were removed from this analysis due to incomplete covariate information (N total =2231) Interaction term (narrative*who is the story about) p-value=0.0113 Log-binomial model is adjusted for: narrator income, narrator gender, narrator relationship status and the narrative's emotional tone Effect of adjustment variables on probability of rejecting the UN/MINUSTAH is depicted in Table  7.

Discussion
The purpose of this research was to generate empirical understandings of peacekeeper-perpetrated sexual misconduct through secondary analysis of 2541 self-interpreted narratives obtained from Haitian community members who lived in proximity to MINUSTAH peacekeeping bases.

Geographic distribution of sexual misconduct perpetrated by peacekeepers
Multi-variate log-binomial regression was used to model the probability of participants sharing a narrative that addressed sexual misconduct perpetrated by MINUSTAH peacekeepers as a function of location. After adjusting for narrator age, emotional tone of the narrative, and who the narrative was about, the probability of sharing a narrative about sexual misconduct was greatest in rural locations, followed by urban locations. Narratives shared in semi-urban locations had the least probability of mentioning sexual misconduct.
Peacekeeping economies may explain why narratives shared in rural locations had a greater probability of mentioning sexual misconduct compared to urban locations. Given women and girls in rural Haiti face greater socio-economic disparities compared to their urban counterparts (43), their participation in peacekeeping economies was hypothesized to be greater, potentially leading to a variety of sexual interactions with peacekeepers: transactional sex, commercial sex work, long-term relationships, and SEA. Thus, we expected the occurrence of sexual misconduct to be greater in rural locations, compared to urban locations.
However, the mechanism of peacekeeping economies does not consider the conduct and discipline of peacekeepers within each base. The rural-effect may also be explained by the differential implementation of sexual misconduct deterrence measures such as: the zero-tolerance policy, SEA reporting mechanisms, surveillance, and peacekeeper discipline. The certainty of punishment has been identified as an important factor in deterring potential SEA perpetrators during PO (44).
The model included narratives of children fathered by peacekeepers within the sexual misconduct outcome. The mechanism of peacekeeping economies does not fully address the existence of children fathered by MINUSTAH peacekeepers. The rural-effect may be more fully explained by considering differential access to resources and information between urban and rural locations, such as contraception needs and access to abortion. Narratives that mention children fathered by peacekeeper might be shared to a greater extent in rural locations because women and girls in rural Not only could deterrence measures such as CDUs be implemented, but also interventions that address the social and economic factors that increase the vulnerability of experiencing SEA.
Approaches to preventing and responding to SEA exist on a spectrum, from targeting peacekeepers to targeting local women and girls (45). By recognizing the geographical distribution of SEA, actionable intervention areas can be identified. In the absence of an evidence-based approach with respect to the allocation of deterrence and prevention policies and programs, sexual misconduct perpetrated by MINUSTAH peacekeepers has the potential to further tarnish the UN's legitimacy in host countries. Exposure to everyday experiences with peacekeepers will shape civilian perceptions of peacekeepers, which in turn affects the desire to engage with peacekeepers (35). Local civilians may not have access to reliable information about peacekeepers (36); they rely on personal, vicarious, and community-level experiences regarding the day-to-day activities of peacekeepers. Thus, experiences with peacekeepers influence perceptions, thereby impacting decisions related to engaging with the UN/MINUSTAH.

Multi
Gordon and Young found that exposure to abuse perpetrated by peacekeepers negatively affected perceptions related to the effectiveness, abusiveness, and benevolence of peacekeepers (35).
Negative perceptions of peacekeepers were found to reduce cooperation with peacekeepers, where local community members have deeply entrenched perceptions, sentiments, and attitudes that diminish cooperation and trust. This is likely to affect the legitimacy of future UN operations in Haiti, thereby destabilizing long-term peace-building and democratization agendas.

Limitations
The results must be interpreted with several limitations in mind. First, convenience sampling has the potential to introduce selection bias and the gender imbalance noted in the sample (70% males, 30% females) is the result of participants' self-selection out of the study.

Ethics approval and consent to participate
Research ethics approval for the cross-sectional survey and the present secondary analysis was obtained by Queen's University Health Sciences and Affiliated Teaching Hospitals Research Ethics Board (#6020398). Trained Haitian research assistants introduced the study to participants in Kreole using a standard script and informed consent was obtained prior to the start of the survey.

Consent for Publication
All authors provide their consent for publication. No identifying participant information was obtained or published in this research.