Capable Guardians? UN Peacekeeping, Vulnerable Civilians, and Rates of Sexual Exploitation and Abuse

(1)

Master’s Thesis in Peace and Conflict Studies Department of Peace and Conflict Research,

Uppsala University

Capable Guardians?

UN Peacekeeping, Vulnerable Civilians, and Rates of Sexual Exploitation and Abuse

Jack Breslin Spring 2021

Supervisor: Lisa Hultman

Word Count: 15,477

(2)

i | P a g e

Abstract

For decades now the UN has weathered accusations that it has failed in its duty by allowing the scandal of sexual exploitation and abuse [SEA] carried out by peacekeepers to continue under its mandate. While preventative policy has improved and significant progress has been made in the academic literature, there are still gaps in the collective understanding the drivers of SEA. This paper seeks to fill one such gap by exploring the relationship between the presence of vulnerable civilians and the phenomena of SEA. Borrowing from criminological literature, a theory is generated which suggests that greater numbers of vulnerable civilians represent increased access to suitable targets and therefore a greater opportunity to offend. From this, two hypotheses are drawn: that the presence of vulnerable civilians will increase the rate of SEA incidents and that the presence of vulnerable civilians will increase sexual exploitation to a notably greater degree than sexual abuse. Utilising a large-N study of UN peacekeeping missions between 2010 and 2019, this paper seeks to test the hypotheses using both OLS and negative binomial regression models. The outcome of the empirical analysis, while indicating a positive relationship between vulnerable civilians and the occurrence of SEA, are not conclusive enough to confirm the relationship without further testing. While the same is true with regards to sexual exploitation seeing a greater effect than sexual abuse, the results for the second hypothesis reveal a stark dichotomy between the two facets of SEA. This dichotomy is shown to be prevalent across several alternative explanations.

For these reasons, further research is called for not only into the effect of vulnerable civilians but also into the extent to which numerous other factors affect sexual exploitation and sexual abuse differently.

Key Words: Sexual Exploitation and Abuse, UN Peacekeeping, Vulnerable Civilians, Refugees

(3)

ii | P a g e

Acknowledgements

Without a doubt, I would not have been able to complete this thesis without the support of a number of individuals to whom I owe my gratitude. Firstly, I would like to thank my supervisor Lisa Hultman for her excellent guidance which enabled me to turn my initial concept into the piece that it is today. I would also like to thank Angela Muvumba-Sellström for her enthusiastic encouragement of my interest in the study of gender and conflict. Credit is also due to my friends Alanna, Bryan, Jan, Louis, and Maurice for their advice and support throughout the thesis writing process. And last but most certainly not least, I owe everything I have accomplished to the love and support of my family, who have stood beside for all of my endeavours.

(4)

iii | P a g e

List of Abbreviations

CDU Conduct and Discipline Unit CRSV Conflict Related Sexual Violence

DV Dependent Variable GDP Gross Domestic Product

IDP Internally Displaced Persons IV Independent Variable

MINURCA United Nations Mission in the Central African Republic MINURSO United Nations Mission for the Referendum in Western Sahara

MINUSCA United Nations Multidimensional Integrated Stabilization Mission in the Central African Republic

OLS Ordinary Least Squares PKO Peacekeeping Operation SEA Sexual Exploitation and Abuse SGBV Sexual and Gender Based Violence

TCC Troop Contributing Country UN United Nations

UNAMID United Nations – African Union Hybrid Operation in Darfur

UNDMSPC United Nations Department of Management Strategy, Policy and Compliance UNDOF United Nations Disengagement Observer Force

UNDPO United Nations Department of Peace Operations UNHCR United Nations High Commissioner for Refugees UNISFA United Nations Interim Security Force in Abyei

UNMIK United Nations Mission in Kosovo

UNMOGIP United Nations Military Observer Group in India & Pakistan UNTSO United Nations Truce Supervision Organisation

(6)

v | P a g e

List of Figures and Tables

Figure 1 Yearly Total Allegations Reported and Decade Average 4

Figure 2 Distribution of Allegations Against Missions 5

Figure 3 Allegations Reported by Staff Category 6

Figure 4 Allegations by Type 7

Figure 5 Yearly Total Vulnerable Civilians 13

Figure 6 Causal Mechanism 15

Figure 7 Alleged Incidents vs Allegations Reported 18

Table 1 Summary of Variables and Operationalisations for Hypothesis 1 21 Table 2 Additional Variables and Operationalisations for Hypothesis 2 22

Table 3 H1 OLS Regression – Absolute Measure 27

Table 4 H2 OLS Regression – Absolute Measure 29

Table 5 H1 OLS Regression – Geographic Density Measure 46

Table 6 H1 OLS Regression – Population Density Measure 47

Table 7 H1 Negative Binomial – Absolute Measure 48

Table 8 H1 Negative Binomial – Geographic Density Measure 49

Table 9 H1 Negative Binomial – Population Density Measure 50

Table 10 H2 OLS Regression – Geographic Density Measure 51

Table 11 H2 OLS Regression – Population Density Measure 52

Table 12 H2 Negative Binomial – Absolute Measure 53

Table 13 H2 Negative Binomial – Geographic Density Measure 54 Table 14 H2 Negative Binomial – Population Density Measure 55

(7)

1 | P a g e

Introduction

“We also recognize that the United Nations often operates in circumstances of heightened risk […] For example, reports of sexual exploitation and abuse […] occur proportionately more frequently in humanitarian assistance missions that typically involve broad, continuing, and deep engagement with local populations at their most vulnerable.”

UN General Assembly 2017

Sexual violence has been one of the terrible realities of war for as long as wars have been waged, yet it is only in recent decades that it has truly gained any sort of widespread attention in international discourse. This attention has manifested academically in numerous papers and empirical studies on the concept of Conflict Related Sexual Violence [CRSV]. Some works on this topic, many written to the backdrop of CRSV in the Yugoslav and Rwandan conflicts, focus heavily on explaining rape as a ‘weapon of war’ used to shame, dominate, terrorise, cleanse, or otherwise achieve strategic goals (Allen 1996; Card 1996; Diken and Laustsen 2005; Buss 2009;

Danjibo and Akinkuotu 2019). While such a concept can explain CRSV on a systemic level, it is insufficient for explaining variation between cases. Other potential causal factors such as in group socialisation (Cohen 2013), militarised masculinities (Baaz and Stern 2009), and structural inequalities (Davies and True 2015) have all been discussed in the literature as a way to account for such differences. Wood’s (2018) depiction of wartime rape as an act that is influenced by top- down strategic decisions, horizontal socialisation, and individual preferences perhaps best captures the cumulative academic understanding of CRSV. Much of the logic behind the investigations into CRSV can also be found in the smaller field of research into sexual exploitation and abuse in UN peacekeeping missions. While the concept of SEA as a top down strategy is of course absent, theories about masculinities, gender, and socialisation are all notable in the literature. Finally, in both of the above mentioned bodies of literature there is a running underlying assumption that regardless of what other influencing factors are present, individual choice to offend is still a core element determining the occurrence of abuses.

(8)

2 | P a g e

In a way, this paper similarly examines influences on the decision to offend. But rather than examining factors that influence the willingness of an individual to offend, this paper examines the impact opportunity has on the transition from motivation to actual offending. Indeed, the concept that the opportunity to commit crime and occurrence of crime itself are intrinsically linked is well grounded in the literature on criminology. It stands to reason then that sexual offences against vulnerable civilian populations in post-conflict settings will follow a similar logic. As reflected in the opening quote to this paper, the idea that vulnerable civilian populations are particularly at risk of SEA is far from outlandish. Yet of all the numerous studies of SEA that populate the literature, very few, if any, have methodically theorised and tested the exact nature of this relationship. To that end, this thesis contributes to the collective understanding of SEA committed by UN peacekeepers through quantitative investigation of the following question:

Is sexual exploitation and abuse by UN peacekeepers linked to the prevalence of vulnerable civilian populations?

In line with the above research question, this paper proposes two hypotheses. Firstly, that an increased presence of vulnerable civilians will cause a substantial increase in the rate of SEA.

Secondly, that sexual exploitation will see a greater increase as a result of the presence of vulnerable civilians than that of sexual abuse. Primarily, a systematic large-N study of UN PKOs from 2010 to 2019 is utilised to test these hypotheses at a mission-year level of analysis. Using data on refugees and internally displaced persons [IDPs] as an independent variable [IV] and the incident rate of SEA as a dependent variable [DV]; the fundamental relationship is first tested with a bivariate regression. Thereafter, several other variables are introduced for a more rigorous OLS regression. The effect of vulnerable civilians on allegations of exploitation in particular is tested using similar methods. But rather than an incident rate two DV, allegations of exploitation and allegations of abuse, are used. The tests for both hypotheses are then replicated using alternative IVs to account for the size of mission host countries in both land area and population. As a robustness check, the alternative method of negative binomial regression is used on duplicate models.

(9)

3 | P a g e

The initial findings from the OLS regressions on the absolute number of vulnerable civilians show a slight positive relationship with the rate of SEA, though not with a consistent level of statistical significance. Further testing to account for the size of the host country in both population and land mass show an opposite, negative relationship. These findings are consistent across both the OLS and the negative binomial models. The inconsistency of the effect of vulnerable civilians on the rate of SEA means that this paper cannot categorically prove a relationship. Moving on, the results of testing on the relationship between absolute numbers of vulnerable civilians and allegations specifically concerning sexual exploitation indicates a positive and significant relationship.

However, once again this relationship does not persist when tested using IVs accounting for host country land mass or population. The results do however show a consistently greater increase in exploitation than in abuse. It is likely that future studies examining a larger time frame and using more precise, geocoded data, may be able to establish a connection more conclusively by building upon the initial findings of this paper. Additionally, the findings reveal a surprising dichotomy in the ways exploitation and abuse are influenced by various factors.

The remainder of this paper is split into several parts. Part 1 provides a definition of SEA and a series of visualisations of trends in SEA allegations since 2010 in order to provide the reader with an understanding of the concept and the problems which interest researchers. Part 2 introduces previous research on the topic of SEA to assist the reader in understanding the extent of collective knowledge on the topic. Part 3 goes into greater detail on the theory behind this paper. The key concepts of interest are defined, and a causal mechanism is outlined and justified, resulting in the introduction of the hypotheses. In Part 4 the research design is introduced. The scope, operationalisations, and methodology are explained in full alongside a discussion of the data limitations and inherent biases to be aware of. Part 5 encompasses the results of the empirical tests for both the primary variables of interest and the alternative explanations included in the tests while Part 6 presents an analysis of the implications. Finally, the conclusion reiterates the discussions of the paper and suggestions for policy or research in brief.

(10)

4 | P a g e

Part 1: Descriptive Patterns and Trends

According to the UN, SEA is defined in two parts: sexual exploitation, the “abuse of a position of power, or trust, for sexual purposes”, and sexual abuse, the “physical intrusion of a sexual nature, whether by force or under coercive conditions” (UN Secretariat 2003). While there is debatably a degree of overlap between exploitation and coercion, it is generally considered that exploitation covers incidents of a transactional nature. Additionally, any incident concerning a child is considered to be abuse by the UN. In order to gain a full understanding of the upcoming discussions surrounding the causality of SEA, it is useful to first take a detailed look at the trends over the last decade. Important to note is that the numbers represented here are likely a substantial underestimate as many cases go unreported due to fears of stigmatisation, retribution, material loss, or simply a lack of knowledge (Csáky 2008, 12). The following visualisations, creating using data from the UN Department of Management Strategy, Policy and Compliance [UNDMSPC], emphasise the problem of variation in SEA between missions. Such variation is the focus of most academic investigation into the concept. Figure 1 displays the yearly total of allegations of SEA reported to the UN.

(11)

5 | P a g e

As the figure shows, the number of allegations reported each year has a tendency to fluctuate quite extensively. It should also be noted that the total number of reports is not truly reflective of how many incidents occurred in a given year, as many incidents are not reported until one or more years after they occured. On average, there were 69 allegations a year from 2010-2020. For the purposes of this data, an allegation should be understood to represent a single report of SEA in which there may by several victims and perpetrators. The extensive variation in which missions contribute to these total allegation numbers is much more precisely captured in Figure 2.

Of the 177 mission-year observations included in this visualisation, 80 represent an allegation count of zero. With the exception of 2010, for which the median number of allegations was 2.5, at least half of the missions each year received two or less allegations. Again, with the exception of 2010, for which the upper quartile was 10, 75% of mission received six or less allegation each year. What this demonstrates is that outlier missions, those represented by black dots in the figure, account for the vast majority of the total number of allegations seen in Figure 1. For example, the two most significant outliers occurring in 2016 and 2019 can both be attributed to MINUSCA in the Central African Republic.

(12)

6 | P a g e

The reality of only a few missions accounting for most instances of SEA emphasises the importance of empirical investigation into the underlying mechanisms for the occurrence of such misconduct. It is clear that, while SEA is an endemic problem with UN peacekeeping, it does not occur universally. Two further examinations of UN SEA data, regarding the staff category of the offender and the type of offense, are pertinent in this section due to their relevance to both the explanations already present in the literature on the subject and the research design discussed later in this paper. Figure 3 highlights which category of UN staff allegations have been made against for each year: distinguishing between military, police, and civilian staff.

As can be seen from the diagram above, military staff are generally responsible for the largest number of allegations of SEA, with policing staff being responsible for the least. The argument could of course be made that this is a reflection of the significantly larger number of military personnel on missions, which will be discussed later in the research design. The high prevalence of allegations against civilian staff is harder to explain, purely due to the lack of available data on staffing numbers for civilian personnel. Again, this will be discussed in greater detail later in the paper.

(13)

7 | P a g e

Figure 4 provides a visualisation which breaks down overall allegations into those of either abuse or exploitation. As discussed previously, abuse allegations cover non-consensual sex acts and any incident involving children, whereas exploitation allegations generally involve transactional sex and abuses of positions of power.

The diagram makes evident a significant trend shift in the balance of abuse and exploitation allegations over the course of the decade. While exploitation allegations have generally been the most common, abuse allegations experienced a large spike around 2015 before significantly dropping off towards the end of the decade. The spike can be accounted for in part by the increase in cases involving children, always coded as abuse, which occurred around the same time. The reason for this increase in cases involving children is unclear, though similar to the discussion of overall allegations above, a large share of the cases against children are attributable to MINUSCA.

Primarily, the purpose of these visualisations has been to emphasise the importance of studying variation between PKO’s levels of SEA allegations. Additionally, the trends outlined in figures 3 and 4 will inform the discussions in future sections.

(14)

8 | P a g e

Part 2: Previous Research on SEA

Due in part to the study of SEA being a relatively new field of specific academic interest, the literature is not as expansive as some other areas. Despite this, the pieces that are presented in the literature provide a rigorous discussion of the potential mechanisms behind SEA. Additionally, numerous works on peacekeeping and CRSV in general can be used to provide useful insights. To make sense of the academic discourse, this section will approach the literature as belonging to several categories. The first is that of militarised masculinities, which focuses on military socialisation and gender inequality to provide systemic level explanations for violence. The second category deals largely with the concept of peacekeeping economies, explaining how uneven power dynamics contribute to exploitative relationships. The literature in this section is as close as the field has come to analysing vulnerable civilian populations statistically. Thirdly, there is a comparatively smaller body of literature that deals with the lack of accountability and deterrence for offences in UN PKOs as an explanation for the continued failings. And finally, there are discussions in the literature of case specific factors that don’t fit into any of the three categories outlined above, such as host/contributor cultures and conflict intensity. This section concludes with a discussion of the gap in research relating to vulnerable civilians and the ongoing issue of SEA.

Militarised Masculinities

The concept of militarised masculinity is well established in feminist literature on both war and peace time violence committed predominately by men against women and minority groups (see Martin 2005; Whitworth 2007; Bjarnegård and Melander 2011; Park 2016). At its most basic, the concept of militarised masculinity is identifiable as culture which prizes power, honour, and the ability to protect and dominate above all else in both societies and organisations. One of the negative effects of such a culture is the devaluation of both the feminine and the other.

Additionally, the nexus of an emphasis on power and the diminished value of the powerless is likely to lead to situations where violence is perceived as acceptable. In terms of the wider concept of CRSV, militarised masculinity relies on much the same logic as the discussion of rape as a weapon of war introduced earlier. In that it can manifest as both top down official practices designed to ensure control and as a horizontal socialisation process between the soldiers of a company which devalues the vulnerable and praises dominance. While such top-down strategic

(15)

9 | P a g e

explanations are not relevant to the discussion of SEA, considering the UN does not encourage such offenses, the process of the devaluation of the vulnerable and the legitimisation of violence are still highly relevant. Especially considering that military personnel tend to have the most allegations of SEA lodged against them, as shown in Figure 3. To transpose this idea more clearly into the topic of SEA, it could be argued that peacekeeping forces function in much the same way as a national armed force. The inherent militarised masculinities that can manifest so negatively in a normal wartime setting are carried across into peacekeeping, despite the contrary purpose.

However, as is pointed out by Higate (2007) and Wood (Wood 2014), the theory does not sufficiently account for variation in levels of SEA between missions.

While the explanation is in itself insufficient, it has informed several studies of more specific dynamics of PKOs that have a significant effect of the occurrence of SEA. One of the common ideas to emerge is the suggestion that higher proportions of female peacekeepers on PKOs could serve as a means of reducing SEA. Bridges and Horsfall (2009), based on a series of interviews with female Australian Defence Force personnel with PKO experience, suggests that increasing the number of female peacekeeping staff could combat sexual misconduct by male colleagues.

Nevertheless, some authors have expressed scepticism at the suggestion that women should be expected to actively police their male colleagues or somehow take on responsibility for preventing misconduct. (Simić 2010).

Despite the contested nature of suggestions to increase female representation in PKOs, some studies have examined its potential. For example, Karim and Beardsley (2016) theorise that more female PKO staff could counteract the effects of militarised masculinity. Through an empirical analysis they demonstrate that higher proportions of women in military and police contingents consistently have a negative effect on the number of SEA allegations, though the results are not always statistically significant (Karim and Beardsley 2016, 107). UN policy to increase the share of female peacekeeping staff is likely in part motivated by such arguments. Moving on from the gender ratio of contingents, several studies have also tested the effect of gender equality in both host countries and Troop Contributing Counties [TCCs] on SEA. Nordås and Rustad (2013, 527), for example, test the impact of spousal rape laws on the occurrence of SEA. The findings of these studies suggest that spousal rape laws in TCCs have a negative effect but one that is not significant

(16)

10 | P a g e

and that the same laws in the host country in fact have a slight positive effect on SEA. They explain this unexpected result by suggesting that such countries are more attuned to abuse and therefore encourage better reporting. Similar tests using variables such as the prevalence of women in the workforce and secondary enrolment also find slight negative effects on SEA (Karim and Beardsley 2016; Rodriguez and Kinne 2019).

Peacekeeping Economies

While the factors discussed above focus largely on the specific composition of PKOs, be it in terms of gender or country of origin, the presence of operations in and of themselves are present in the literature as a factor influencing SEA. The deployment of a PKO to a conflict country consists of more than simply declaring a mandate and dropping troops on the ground. Headquarters and operating bases need to be established, logistic networks developed, suppliers sourced, and a myriad of other things that enable operation staff to work in-country. The local industries and services that emerge in response to these needs, and the jobs that they create, make up what is referred to as the peacekeeping economy; essentially any economic activity that is created or enhanced purely due to the presence of peacekeepers (Jennings and Bøås 2015). While these economic activities can benefit the local population in terms of jobs and higher wages, there are also clear and undesirable side effects. One of the most significant of these is the growth of the sex industry in areas surrounding peacekeeping bases. Scholars have described how the influx of wealth creates a market for sex work, utilised both by locals and by peacekeepers, in a significant number of PKOs (Jennings 2014; Edu-Afful and Aning 2015). While the growth of this industry can have long term negative effects, such as the entrenching of long term sex tourism industries (Jennings 2010) and the growth of human trafficking networks (Smith and Smith 2011), the greater concern for this paper is the impact such industries have on SEA. As Beber et al. (2017) found through interviews with women in the Liberian capital of Monrovia, there is a high degree of transactional sex paid for by peacekeepers. Their findings indicate that around half of women aged 18-30 in that city had engaged in transactional sex, with more than three quarters of that number having done so with UN personnel (Beber et al. 2017, 3). Such behaviour by peacekeeping staff falls under the category of sexual exploitation introduced above in Part 1 which was also shown in Figure 3 to be easily the largest category of offenses. Considering the above example and the

(17)

11 | P a g e

predominance of exploitation cases it appears clear that there is a substantial link between peacekeeping economies and the occurrence of SEA.

Translating this seemingly obvious relationship into an empirical measure is unfortunately not quite so simple. One method that to some extent captures the concept of the peacekeeping economy is the use of measures for GDP as control variables. With varying levels of significance, lower levels of GDP in a mission host country have consistently been found to have a positive relationship with levels of SEA (Nordås and Rustad 2013; Karim and Beardsley 2016; Moncrief 2017).

Accountability and Deterrence

Another systemic influence on SEA that has been identified in the literature is the general lack of accountability or deterrence for offences. While the UN sets strict values and standards for its personnel, the extent to which this effectively reduces the occurrence of offences is debatable. In technical terms, the UN only has the authority to dismiss and repatriate peacekeeping staff. The onus to punish misconduct instead largely falls of the TCC and the actions taken vary greatly (Miller 2006; Ndulo 2009). For example, of the 196 substantiated cases of SEA involving uniformed personnel between 2010 and 2019 shown on the UNDMSPC (2021) website only 57 resulted in jail time for the involved individuals in their home country. The functional immunity to prosecution by a host country’s legal system and the variable support for zero tolerance from TCCs creates a culture of impunity in PKOs that is certainly linked to the occurrence of SEA (Defeis 2008, 192). To make reference back to the discussion of CRSV from earlier, the lack of effective top-down prohibition of SEA could be seen to function in a very similar way to top-down strategic decisions encouraging CRSV in war.

Empirically, the concept of deterrence and accountability has been tested in a few ways. Firstly, Moncrief’s (2017, 725) examination of the link between previous levels of general misconduct and the occurrence of SEA in particular shows a significant causal pattern. While this strongly suggests that a culture of impunity could be a contributing factor to SEA, it does not quite display the impact of proactive deterrence. For this, one might look at the tests conducted by Neudorfer (2014), who shows that the presence of Conduct and Discipline Units [CDUs] has a significant

(18)

12 | P a g e

and negative relationship with the occurrence of SEA. These teams, established in 2005, provide overall direction and guidance for the conduct of UN staff in the field. Their evident effectiveness as a deterrence for misconduct lends great credence to the arguments that more could be done to actively prevent SEA, as posed by scholars like Defeis (2008). The combination of these two examples testifies that there is in all probability a strong link between failures to adequately deter offenders or hold them accountable, and the continued prevalence of SEA under the jurisdiction of the UN.

Case Specific Factors

Beyond the categories discussed above, there have been a number of other case specific factors empirically proven by researchers to have an impact on the occurrence of SEA. In other words, these factors are more specific to each particular mission rather than having roots in system wide features. The largest cluster of these factors pertain to the culture of the TCCs and host countries.

Regarding TCCs, the freedom of the press and the enforcement of international humanitarian law have both been found to have a significant and negative relationship with the occurrence of misconduct on PKOs (Rodriguez and Kinne 2019, 633). Conversely, the presence of abuses against domestic populations and national corruption in TCCs have been found to have a positive, though not consistently statistically significant, relationship with SEA (Horne, Robinson, and Lloyd 2019, 239). In terms of host country cultures, it has been found that strong rule of law and press freedoms have a negative and significant relationship with peacekeeper misconduct (Rodriguez and Kinne 2019, 633). To some extent, these issues are tied to the same effects of legitimised violence and the devaluation of human life discussed under the militarised masculinities literature. If abuse of civilians at home is commonplace, alongside corruption and contempt for human rights law, then similar behaviour against civilians in a PKO mission-host country should not be surprising. On the other hand, they also relate to the previous discussion on both accountability for misconduct and the underreporting of SEA vis-à-vis the freedom of the press to report misconduct.

Similar to the particularities of TCCs, the nature of conflict in a mission host country has been found to have a notable impact of allegations of SEA. The security situation of a host country has been theorised to influence the extent of peacekeeper-civilian interaction, serving to limit said

(19)

13 | P a g e

interactions when conflicts are more intense. Indeed, according to previous testing, host countries with a higher number of battle related deaths in a given mission-year are shown to have a negative relationship with allegations of SEA (Nordås and Rustad 2013, 527). Conversely, the extent of CRSV in a host country has been shown to have a positive relationship with the occurrence of SEA. Tests for the impact of high rates of CRSV in the conflict preceding a peacekeeping operation have been found to have a positive, though not consistently statistically significant, relationship with SEA (Nordås and Rustad 2013, 527; Karim and Beardsley 2016, 108). Likewise, Neudorfer (2014, 632) shows that when sexual violence is problematic in a host country during the deployment of a peacekeeping operation it is more likely for allegations against peacekeepers to emerge. Both of these examples once again play into the earlier discussions on the legitimising of violence that can influence the behaviour of peacekeepers.

A Gap in the Research

While the above discussion provides only a snapshot of the rigorous past investigations into SEA, it should serve as an adequate starting point for familiarisation with the state of knowledge regarding the phenomenon of SEA. It should also, however, serve to highlight that a substantial research gap exists. Considering that SEA is an endemic and persistent problem within UN peacekeeping, every effort should be made to maximise the collective understanding of the issue in order to enable practical and effective solutions. While the focus of past studies has been both the composition of PKOs and the circumstances of the mission host country, there has been a notable lack of attention paid to one of the major consequences of armed conflict, refugees.

Considering that intense conflict invariably leads to the displacement of civilian populations, it is surprising that so little attention has been paid to what impact larger displaced populations may have on SEA. If displaced civilians are to be considered as vulnerable to abuse, which this paper argues is indeed the case, then the potential causal link seems self-evident. It is this gap in the research that this paper seeks to fill. As the focus of this paper is on the relationship between the presence of vulnerable civilians and the occurrence of SEA, the aforementioned causal factors do not take centre stage in the oncoming theoretical discussion. However, a number of the factors already established in the literature will be introduced in this paper’s more extensive models in order to properly establish the significance of the presence of vulnerable civilians.

(20)

14 | P a g e

Part 3: Theory

Part 3.1: Conceptualising Vulnerable Civilians

As a key component of the relationship under investigation in this paper, the concept of vulnerable civilians needs to be clearly defined. The term itself is used with such frequency by numerous actors referring to an array of situations that it is implicitly understood in the relevant context. As of yet, however, there is no generally accepted understanding of what exactly is meant when civilian populations are referred to as vulnerable in the context of SEA. For the purposes of this paper, vulnerable civilians should be understood to be referring to refugees, internally displaced persons [IDPs], and other populations of concern to the UNHCR. This is of course not intended to refute the fact that non-refugee populations can also be vulnerable to abuses, but rather to help focus the discourse of this paper. Figure 5 provides a visualisation of the global trend in vulnerable civilians under this definition over the last decade, using an aggregation of UNHCR (2021a) data on populations of concern. As is clearly demonstrated, the number of individuals around the world considered to be vulnerable through this definition has risen considerably over the last decade.

(21)

15 | P a g e

Having established what is meant by the term ‘vulnerable civilians’ it is important to now theoretically lay out exactly why refugees and those in refugee like situations should be considered to be particularly vulnerable in the context of SEA. One aspect of the existing literature that can assist in explaining this is the discussion of the vulnerability of children in peacekeeping settings.

Several studies have highlighted that the exchange of food, money, luxuries, or other services for sexual favours is a common dynamic of abuse against children by adults working in humanitarian or peacekeeping settings (Blakemore, Freedman, and Lemay-Hébert 2019; Lee and Bartels 2020).

Indeed, it is not difficult to see this as a tactic of SEA that is also applicable to adult victims in vulnerable situations. The link between refugee populations and poverty is not a new concept, and several studies in the past have empirically established that such populations suffer relatively higher deprivation that other domestic populations (Dershem, Gurgenidze, and Holtzman 2002;

Khawaja 2003). With this in mind it seems even more probable that the same dynamic that takes advantage of the vulnerability of children to exchange goods for sex is also applicable to vulnerable adult refugees. In general terms, by the nature of their displacement people in refugee like situations often end up in situations of power imbalance and dependency. This, in culmination with the loss of their prior support networks, leaves them at risk of abuse. Indeed, it is already well established that refugees are at a heightened risk of sexual violence in their country of origin, during the migration process, and in the host country (Buckley-Zistel and Krause 2017; Araujo et al. 2019). Taken in conjuncture with the discussion of the link between peacekeeping economies and the sex industry from Part 2, it is not hard to draw a theoretical link between vulnerable civilians and SEA.

Part 3.2: Causal Mechanism and Hypotheses

As mentioned earlier in this paper, the idea that the presence of vulnerable civilians might be related to occurrence of SEA would not be met with scepticism by most practitioners. What is less clear is how that relationship might function. By turning to literature from the field of criminology, it is possible to logically outline why such a relationship would exist and how it might function.

From a criminological perspective, the relationship between opportunity and crime has been heavily debated. While earlier works tended to perceive opportunity as purely coincidental, determining only the location of crimes, later studies have more readily embraced the concept of

(22)

16 | P a g e

opportunity as a factor that can actively determine occurrence (Mayhew et al. 1976; Warr 1988;

Clarke 1984; 2012; Beauregard, Rebocho, and Rossmo 2010). While the exact approaches still vary, the concept is now well established in the field. At its most basic level, opportunity can be viewed as one of two necessary components for the occurrence of crime, the other being an individual willing to commit an offense. This of course relies on rationalist assumptions about human behavior, essentially that individuals who offend are consciously making a rational decision to do so. Cohen and Felson (1979, 589) provide a more detailed breakdown, stating that crime requires the convergence of three elements: ‘(1) motivated offenders, (2) suitable targets, and (3) the absence of capable guardians against a violation.’. Furthermore, they argue that ‘the convergence in time and space of suitable targets and the absence of capable guardians may even lead to large increases in crime rates without necessarily requiring any increase in the structural conditions that motivate individuals to engage in crime’.

When the discussions of this criminological literature are applied to the above-mentioned concepts of interest, the building blocks of a causal mechanism begin to emerge. If vulnerable civilians are perceived as ‘suitable targets’, then it is logical that an increase in their population could be perceived as an increase in access and therefore an increase in opportunity. This would serve as the first stage of a causal process. More so if UN peacekeepers are understood to represent both

‘capable guardians’ and potential offenders when their actions betray their responsibilities. From here, the increase in opportunity can be expected to lead to an increase in offences of SEA if perpetrators are assumed to behave according to rationalistic theory. Additionally, Warr’s (2001) understanding of opportunity not as a singular entity but as a collection of numerous factors further justifies investigation of vulnerable civilians as a causal factor operating through an opportunity dynamic in conjuncture with already established explanations. Based on this discussion, the following causal mechanism displayed in Figure 6 is established. Primarily, the mechanism presumes that the decision to commit crimes of SEA in PKOs is partially influenced by the opportunity to do so in the same way as other criminal acts observed by civilian populations domestically.

(23)

17 | P a g e

Figure 6 Causal Mechanism

Referring back to Cohen and Felson’s (1979) argument, it is undeniable that vulnerable civilians would be perceived as ‘suitable targets’. Such individuals are separate from their normal support networks, are often living in financially precarious situations, may not enjoy full legal

protections, and in many cases rely upon PKOs for protection. Essentially, vulnerable civilians represent a prime target for abuse that is easily accessible and potentially without access or awareness of mechanisms to protect themselves from abuses. Therefore, this paper argues that vulnerable civilians should be viewed as part of an opportunity mechanism. The above causal mechanism forms the basis of the first hypothesis to be tested in this paper:

H1: The presence of larger numbers of vulnerable civilians in a mission host country will have a positive and significant relationship with the rate of SEA.

While the above hypothesis lays on the theoretical foundations discussed in this section and should be considered the primary concern of this paper, a second hypothesis is also formulated. As was demonstrated in Figure 4, the number of SEA allegations categorised as exploitation has been steadily rising for several years and now represents a substantial proportion of the overall allegations of SEA. Thus begs the question of why this trend is emerging. Looking back to the discussions on the consequences of peacekeeping economies and the pattern of exchanging goods and services to abuse vulnerable children, bearing in mind that transactional sex is considered by the UN to be a form of sexual exploitation, a potential link begins to emerge.

ACCESS TO SUITABLE TARGETS INCREASES, CREATING MORE

OPPORTUNITIES TO OFFEND

(24)

18 | P a g e

As was demonstrated in Figure 5, the global number of vulnerable civilians has seen a similar upward trend to that experience by allegations of sexual exploitation. Understanding of course that the global trend of vulnerable civilians increases more significant due to the inclusion of non-PKO countries, there is still merit in investigating if the two increases are linked. Thus, the following hypothesis is proposed:

H2: The presence of larger numbers of vulnerable civilians in a mission host country will have a positive and significant relationship with sexual

exploitation to a greater extent than it has with sexual abuse.

The particular wording of this hypothesis to include the comparison of exploitation and abuse is done to provide greater clarity when testing. If a causal link exists between vulnerable civilians and sexual exploitation in particular, then a more substantial increase should be observable in comparison to abuse. Otherwise, it would suggest that the presence of vulnerable civilians only increases SEA in general terms. The following section outlines in technical terms the research design for testing these hypotheses.

(25)

19 | P a g e

Part 4: Research Design

Part 4.1: Temporo-Spatial Scope

In order to conduct tests for the hypotheses outlined in the section above, a dataset has been created specifically for this paper. This dataset covers a ten year range of UN peacekeeping operations from 2010 to 2019 and is coded on a mission-year level of analysis. This time frame was selected in order to make full use of the data on UN SEA allegations made available through the aforementioned UNDMPSC website, which only has data available from 2010 onwards. The decision was made to exclude 2020 from the dataset as several key variables were not yet available for this year at the time of writing and to account for the potential bias of late SEA allegations for 2020 reported in 2021, which will be discussed in greater detail in section 4.2. As part of the primary scope conditions, only missions which are listed as PKOs by the UNDPO (2021a) have been considered eligible for the dataset. Beyond this, several additional exclusions of PKOs have been made. Four missions (UNMOGIP, UNTSO, UNDOF, and MINURCA) have been excluded because they operate across the border regions of more than one sovereign state, while this does not breach the mission-year unit of analysis it does present complications for variables coded using national data. Additionally, UNISFA and MINURSO in Abyei and Western Sahara respectively have been excluded due to a lack of data on vulnerable civilians.

Part 4.2: Operationalisations

Dependent Variable (DV)

As the main concept of interest for this paper, reports of SEA form the basis of the DV being tested for both hypotheses. However, rather than using the total number of mission-year allegations of SEA visualised in Part 1, this paper has made slight alterations in creating the DV. The most significant of these alterations apply to H1 and is based on the important distinction between allegations of SEA and the incidents themselves. As discussed in Part 1, not all incidents are reported in the year that they occurred, which would create inaccuracy if the data were to be used in its original format. In order to correct for this the SEA data has been transformed as per the process outlined by Breslin (2020, 47). Put simply, late allegations received in the subsequent year

(26)

20 | P a g e

are resorted into the year they are alleged to have occurred. Due to the limited access to reports, allegations about incidents more than a year old are excluded as the exact dates are undefined.

Figure 7 visualises the effect of this process and clearly demonstrates the importance of taking into account the time difference between an incident occurring and being formally reported.

In order to also account for the variance in the size of PKOs, this reorganised data is combined with information of average mission-year police and military staff levels to give an Incident Rate variable representing the number of incidents per 10,000 PKO military and police personnel. This is the DV used for testing H1. Civilian staff are excluded from this calculation due to inconsistent access to data. Alternatively, for H2, allegation data is disaggregated to distinguish between allegations purely involving exploitation and allegations purely involving abuse. However, as the data source used to temporally correct the data as per the process above does not distinguish between the type of offense the absolute number of allegations is used instead. In this way, H2 is tested with two IVs: the absolute number of exploitation allegations and the absolute number of abuse allegations

(27)

21 | P a g e

Independent Variable (IV)

In order to conceptualise vulnerable civilians as an IV, this paper uses an aggregation of the five categories used by the UNHCR (2021b) to define their populations of interest. The first category, that of refugees and people in refugee like situations, is defined in the Convention and Protocol Relating to the Status of Refugees (UNHCR 2010) as individuals who ‘owing to well-founded fear of being persecuted … is outside the country of his nationality and is unable or, owing to such fear, is unwilling to avail himself of the protection of that country’. The second category is asylum seekers, understood as individuals who have sought international protection but have not yet had their claims of refugee status determined. The third, and in many cases the largest, category is that of IDPs; meaning persons who have been forced to flee their homes due to violence but have not crossed international borders. Notably, as far as the data collected by the UNHCR is concerned, only IDPs created by conflict are included in this count. Next, under the 1954 Convention Relating to Status of Stateless Persons (UNHCR 2014), individuals not considered a national by any state are defined as stateless persons and are also included as a separate category by the UNHCR.

Finally, the UNHCR provides a measure for other groups or persons of concern who do not fit into the aforementioned categories, but to whom the organisation has nonetheless extended its assistance. This aggregation is presented as the absolute count of vulnerable civilians and acts as the primary IV. Additionally, two secondary independent variables measuring the density of vulnerable civilians in terms of host country geography and population will be used for robustness checks.

Additional Variables

In addition to the primary variables of interest, this paper employs several additional variables in order to better capture the relationship in the wider context of peacekeeping. A log GDP per capita variable for the mission host countries is used as a means of accounting for the economic strength of a host country. This is treated as a confounding variable due to the likelihood of economic stress increasing the number of vulnerable civilians and more extreme effects of peacekeeping economies influencing SEA, as discussed in Part 2.2. Based on the findings from previous studies discussed in the literature review it is expected that this will have a negative relationship with SEA.

In order to account for the security situation of a mission host country, data from the UCDP/PRIO Armed Conflict Dataset (Pettersson and Öberg 2020) is used to create an ordinal variable for

(28)

22 | P a g e

conflict intensity for each mission-year. Borrowing from the methodology of Nordås and Rustad (2013, 527), host countries with no recorded battle conflict are coded as 0 while those with less than 1000 battle related deaths are coded as 1 and those exceeding that threshold as 2. As conflict intensity is likely to influence both the IV, by increasing the number of vulnerable civilians, and the DV, by limiting peacekeeper-civilian interactions in more dangerous countries, it is also treated as a confounding variable. A negative effect on SEA is expected.

Based on both the insight from the criminology literature regarding opportunity from Part 3.1 and Neudorfer’s (2014) work on evaluating the effect of deterrence, this paper also introduces a variable to account for the effect of previous punishment for SEA offenders. Using data from the UDMSPC (2021) on TCC actions in response to UN repatriation after an allegation, a measure for the number of allegations resulting in punishment emerges. Note that this is reflective of the number of cases and not individuals. Cases that result in Jail, Dismissal, Administrative Penalties, Demotion, Financial Sanctions, Forced Retirement, or Suspension are aggregated per mission-year and then lagged by one year. As this would influence the interaction between the IV and DV it is treated as a moderator variable and is expected to have a negative effect.

In part because of the prevalence of allegations involving military personnel, as visualised in Figure 3, an additional control variable is introduced to account for the variance in mission composition. Monthly data on police and military staff levels for each mission is used to create a yearly average for mission size, from which the percentage of staff belong to a military category is then taken to be used as a mission-year variable. It is expected that mission-years with a higher military percentage will see a higher rate of SEA. A similar process is used to create a variable for the percentage of mission police and military staff which are female. Following the results of Karim and Beardsley’s (2016) similar analysis, this is expected to have a negative relationship with SEA. It must be noted that for both the above variables civilian staff are excluded from the measure due to data scarcity. Finally, a count variable for UN PKO fatalities as a result of malicious actions is taken for each mission year. It is expected that missions with higher levels of violence against peacekeepers will have minimised civilian-peacekeeper contact, thereby minimising SEA. For this reason, fatalities from other causes such as illness or accidents are not counted. A summary of these operalisations can be seen overpage in Table 1 for ease of reference.

(29)

23 | P a g e

TABLE 1

Summary of Variables and Operationalisations for Hypothesis 1

Dependent Variable Operationalisation

SEA Incident Rate Reported incidents of SEA per 100,000 mission staff (Military and Police) per mission-year.

Sources: (UNDMSPC 2021; UNDPO 2021b) Independent Variables

Vulnerable Civilians

(Absolute Count) Total Number of Vulnerable Civilians in mission host country per mission year. Measured in thousands.

Source: (UNHCR 2021a) Vulnerable Civilians

(Geographic Density) Number of Vulnerable Civilians per km² of host country’s land mass, per mission-year.

Sources: (UNHCR 2021a; CIA 2021) Vulnerable Civilians

(Population Density)

Number of Vulnerable Civilians per 100 people of host country’s population, per mission-year.

Sources: (UNHCR 2021a; World Bank 2021b) Additional Variables

GDP per Capita (Log) Log GDP per capita of mission host country per mission-year.

Source: (World Bank 2021a)

Security Situation Security situation in host country per mission-year: (0) no conflict, (1) minor conflict, (2) over 1000 battle related deaths.

Source: (Pettersson and Öberg 2020)

TCC Punishment Number of allegations that resulted in punishment by TCC per mission- year. Lagged by one year.

Punishment = (Jail, Dismissal, Administrative, Demotion, Financial Sanctions, Forced Retirement, Suspension)

Source: (UNDMSPC 2021)

Military Percentage Percentage of the average total of mission staff (Military and Police) categorised as Military per mission-year.

Source: (UNDPO 2021b)

Female Percentage Percentage of the average total of mission staff (Military and Police) categorised as female per mission-year.

Mission Fatalities Reported total fatalities of mission staff as a result of malicious acts per- mission year.

Source: (UNOCC 2021)

(30)

24 | P a g e

As discussed above, the DV for the second hypothesis is measured differently to the one displayed in Table 1. Two IVs, one for the number of exploitation allegations and one for the number of abuse allegations, are used. The exact reasoning for this is further explained in the methodology section. While the IVs and additional variables for the second hypothesis are the same as in Table 1, there are two additional control variables added for testing H2. The first is the total SEA allegations per mission-year. This is included due to the substantial variance in total allegations between missions, as was visualized in Figure 2. Missions with a higher number of overall allegations would of course be expected to see a larger number of both exploitation and abuse, making the control variable necessary to get accurate results. Following similar logic, the second control variable is for mission size. This was not included in the variables for H1 because the Incident Rate variable accounts for variance in mission size already. As it is not possible to temporally realign the disaggregated data on exploitation and abuse allegation, creating a rate per staff variable would be inappropriate. Despite this, it is still necessary to control for the variance in mission size. Because of this, it must be noted that the models for H2 will lose a degree of accuracy due to allegations being reported outside of the year in which they occurred. Table2 summarizes the new IVs and additional control variable used in testing H2, all other variables remain the same.

TABLE 2

Additional Variables and Operationalisations for Hypothesis 2

Dependent Variables Operationalisation

Exploitation Allegations Total allegations of Sexual Exploitation specifically per mission-year.

Sources: (UNDMSPC 2021)

Abuse Allegations Total allegations of Sexual Abuse specifically per mission-year.

Sources: (UNDMSPC 2021) Control Variables

Total SEA Allegations Total reported SEA allegations of any type per mission-year.

Sources: (UNDMSPC 2021)

Mission Size Average monthly total of mission staff (Military and Police) per mission- year.

(31)

25 | P a g e

Part 4.3: Methodology

In order to test the hypotheses outlined in Part 3.2 this paper first uses a series of OLS regressions.

The decision to conduct observational quantitative research such as this is guided by beliefs of the positivist paradigm; that is, that reality is singular and tangibly measurable (Reiter 2015). The advantage of such an approach is the ability to use a single, limited study such as this to generalise on the phenomena at large. While this comes at the cost of more fine grained exploration into the causal mechanisms that a comparative qualitative case study would provide, it is more than sufficient for the initial steps in filling the research gap of an understudied concept. For H1, a bivariate regression is initially run with the absolute number of vulnerable civilians used as the IV and the SEA Incident Rate as the DV in order to establish the nature of the relationship outside of any additional factors. Once the relationship has been established, Model 2 introduces both the host country’s log GDP per Capita and Security Situation as confounder variables. Following this, the number of allegations followed by a punishment from the TCC in the previous mission-year is added as a moderator variable. In the final model, three additional variables already established to have a significant effect in the literature are included: the percentage of mission staff categorised as military, the percentage of mission staff categorised as female, and the number of fatalities resulting from malicious acts per mission-year. For H1 to hold true, it is expected that the absolute count of vulnerable civilians in a mission host country will have a positive and significant relationship with the SEA Incident Rate.

As discussed above in part 4.2, two separate IVs are used to the second hypothesis. This is done so that the results of the two sets of models for each IV can be compared in order to test the assertion that vulnerable civilians increase the number of exploitation allegations to a greater extent than they do abuse allegations. While simply running models using the total number of exploitation allegation may show a positive and significant relationship, this would not sufficiently support H2 until compared to the total number of abuse allegations. Similar to the models introduced for H1, both IVs for H2 are initially tested using a bivariate regression to establish the nature of the relationship outside of external influences. Following this, the second model introduces both the total number of SEA allegations and the mission size as control variables. The rationale behind the introduction of these variables, as discussed in part 4.2, is to account for both the variance in the size of PKOs and for the variance in the number of allegations made against

(32)

26 | P a g e

each mission. The third model introduces the same two confounding variables, a log measure of GDP per Capita in the mission host country and the security situation that mission-year, as used to test H1. Following this model 4 incorporates TCC Punishments of SEA allegations the previous year as a moderator variable. Finally, in model 5 the same three additional variables as used to test H1 are introduced. In order for H2 to be credible, it is expected that when the total number of sexual exploitation allegations is used as a DV there will be a positive and significant relationship with the number of vulnerable civilians. Additionally, this should be to a consistently greater extent than the relationship for the total number of sexual abuse allegations. While it is expected that sexual abuse will also have a positive relationship with the IV, this is not a prerequisite for the confirmation of H2.

Robustness Checks

In order to ensure the robustness of the results from the primary tests, this paper employs several alternative methods. For both H1 and H2, the models outlined above are retested with alternative IVs. Firstly, as opposed to the absolute count for vulnerable civilians, the models are run using a variable for the number of vulnerable civilians per square kilometre. In this manner the geographic density of vulnerable civilians in a mission host country is measured, accounting for the variance in land mass. Similarly, an additional alternative IV is tested. In order to account for the variance in population size for host countries a population density variable is created. In this instance that variable represents the number of vulnerable civilians per 100 population in the host country. The scale of these density variables was selected in order to maximise readability in the results, without changing the output.

In addition to running models with the aforementioned alternative IVs, this paper also ensures the robustness of the results by using a negative binomial model to account for the over dispersed nature of the dependent variable. While a zero inflated model could also be used to account for non-normally distributed data with a large number of zero values, as is the case for the DV in this paper, the lack of a solid theoretical explanation for why these zero values occur makes a negative binomial regression preferable. As part of this particular robustness check all previously tested models are rerun using a negative binomial regression, including the two alternative IVs outlined above.

(33)

27 | P a g e Limitations

While the robustness checks will increase the accuracy of results, it is still important to bear in mind the impact of data limitations on investigation into this topic. Primarily, as has been mentioned in prior sections, SEA as a dependent variable suffers from unavoidable reporting bias.

Be it due to fears of stigmatisation, retribution, material loss, or simply a lack of knowledge it must always be presumed that the data available on SEA is an underestimate (Csáky 2008, 12). Beyond this, the manner in which allegations are reported necessitates a mission-year level analysis at the cost of more fine grained results. Fortunately, this melds well with the data on vulnerable civilians from the UNHCR, which is collected on a country-year basis. Additionally, the use of absolute numbers for vulnerable civilians in a mission host country as the IV presents a similar problem.

While variation in both host country and mission size is controlled for in the primary models and again through the robustness checks, it can only hope to capture the hypothesised relationship on a national level. While such a highly aggregated examination is sufficient for the initial investigation into an understudied dynamic of SEA, such as is the case for this paper, it presents the risk of failing to truly capture the intricacies of the theorised relationship. It is also important to reiterate that, while allegations against civilian UN personnel are accounted for in the DV, the lack of data on exact staff numbers means they are excluded from the variables assessing both the proportion of military staff and of female staff as alternative explanations. On the topic of alternative explanations, there is a limit to how many of the established concepts from the literature can be feasibly included in the models run for this paper. As such, there is some extent of omitted variable bias to be aware of when examining the oncoming results. As a final note, while missions operating in subnational regions such as UNAMID and UNMIK are reported separately for the DV it has been necessary to use data on vulnerable civilians from the wider host country as the IV for testing due to the unavailability of subnational level reports.

(34)

28 | P a g e

Part 5: Results

In this section, the results for the key variables of interest to the first and second hypothesis are detailed in full. The initially OLS regression using an absolute measure of vulnerable civilians is discussed first and, thereafter, the results of the robustness checks using alternative IVs with an OLS regression and the negative binomial method for the same set of IVs are analysed. The results for the control variable are discussed later in Part 6.3.

Hypothesis 1

The results of the initial OLS test models for the first hypothesis are displayed in Table 3. When tested using a simple bivariate model with an absolute measure for vulnerable civilians, there is a positive effect of 0.0002 on the rate of SEA. This means that for every additional thousand vulnerable civilians in a mission host country, the rate of SEA is increased by 0.0002. However, this result is not statistically significant. The relationship between the IV and DV remains positive throughout all four models and is observed to be statistically significant only in Model 2 when Log GDP, Host Security Situation, and Host Population are incorporated. In this case, the effect is stronger at an SEA rate increase of 0.002 per thousand vulnerable civilians at a 90% significance level. For Models 3 and 4, the effect of the IV weakens and loses its statistical significance, with the relationship ranging from 0.0005 to 0.001. Despite the lack of consistent significance, the positive relationship between vulnerable civilians and the rate of SEA in each of the four models implies that there is some merit to the hypothesis that the presence of larger numbers of vulnerable civilians in a mission host country will increase the rate of SEA. However, the observed effect is extremely small and the low R²implies that the variables modelled do not sufficiently account for change in the DV. This means that an ironclad statement of support for H1 is not possible.

When a robustness check is carried out for H1 the results become even less promising. If the geographic density of vulnerable civilians in the mission host country is used as a substitute IV, the results are consistently negative across all four models. In the bivariate model the relationship is -0.055 at a 95% significance level. The observed relationship becomes drastically weaker when other variables are introduced in the other models. If instead a measure of the number of vulnerable civilians as a proportion of the host countries total population is used, results are even less

(35)

29 | P a g e

consistent. In the bivariate model, shows a relationship of -0.012 without significance.

Alternatively, in the three other more comprehensive models the relationship is positive and ranges from 0.014 to 0.133. None of the four models returned significance using this IV. Similar to the models in Table 3, all models using the above IVs show an extremely small effect with poor scores for R². The full results for the density models can be seen in Tables 5 and 6 in Appendix A.

As discussed previously, the method of negative binomial regression is also employed as a robustness check. When the absolute measure of vulnerable civilians is used as an IV, the negative binomial results are remarkably similar to those of the OLS models in Table 3. In both the OLS and the negative binomial models the relationship is consistently positive, as well as being

(36)

30 | P a g e

statistically significant at a 90% significance level in Model 2. When using the geographic density IV in the negative binomial tests the results are consistently negative and significant to at least a 90% significance level while the population density measure is consistently negative without significance. The results for these tests can be seen in Appendix A under Tables 7 to 9.

Overall, when the initial results for the first hypothesis are examined there appears to be some implied truth to the claim that vulnerable civilians will result in higher rates of SEA. This is true when measured with either OLS or negative binomial regression models. However, as the effect is so small and not consistently significant this paper cannot categorically claim proof of H1. This is compounded by the geographic and population density robustness checks, which show highly inconsistent results and often contradict the hypothesised relationship.

Hypothesis 2

As previously discussed, the testing for the second hypothesis requires the comparison of regression results when either the number of sexual exploitation cases or the number of sexual abuse cases are used as the DV. This is necessary to definitively say whether the presence of vulnerable civilians has a more substantive impact of sexual exploitation than it does on sexual abuse. Table 4 provides a visual comparison of all five models for both DVs. In the initial bivariate regression both DVs are found to have a significant and positive relationship with the number of vulnerable civilians. In line with the hypothesis, the effect is greater on sexual exploitation allegations than it is on sexual abuse. For the former a 0.002 increase in allegations per thousand vulnerable civilians is observed at a 99% significance level, whilst for the latter it is 0.001 at a 90% significance level. In the latter four models, the relationship between vulnerable civilians and sexual exploitation allegations remains positive and returns a significant result in models 2 and 3.

For sexual abuse, the relationship flips to being negative in all other models, with significance only in the first two. While the R² in the bivariate is exceptionally low, as per expectations from testing H1, it improves dramatically for the latter models to a level far preferable to that observed in H1’s models. This would seem to imply that the second hypothesis has a stronger grounding than that of H1.

(37)

31 | P a g e

Capable Guardians? UN Peacekeeping, Vulnerable Civilians, and Rates of Sexual Exploitation and Abuse