Women in Peace Processes and the effects on Sexual Violence
A quantitative study on how female participation in peace agreements affect the occurrence of post-conflict conflict-related sexual violence
Cornelia Fransson
Uppsala University
Department of Peace and Conflict Research Bachelor’s thesis
Supervisor: Susanne Schaftenaar Fall semester 2020
Word count: 13744
Table of Content
1. Introduction 3
2. Theoretical framework 5
2.1 Terminology and concepts 5
2.2 Previous research 7
2.2.1 Conflict Related Sexual Violence 7
2.2.2 Women in peace processes 10
2.3 The causal mechanism 13
3. Research design 16
3.1 Methodology 16
3.2 Datasets 17
3.3 Operationalisation, validity and reliability 18
3.3.1 Dependent variable 18
3.3.2 Independent variable 20
3.3.3 Control variables 21
3.5 Limitations and potential biases 24
5. Results 26
5.1 Descriptive statistics 26
5.2 Bivariate hypothesis tests: Tabular analysis and chi-square test for independence 27
5.3 Regression analysis 29
5.4 Predicted probability 31
6. Discussion 32
6.1 Alternative explanations 34
7. Conclusion 35
8. Bibliography 37
9. Appendix 43
9.1 R Script 43
_____________________________________________________________________
Abstract
Despite the vast occurrence of conflict-related sexual violence, it is seldom addressed in peace agreements. The overall absence of female delegates to peace processes has been suggested as an explanation, however this suggestion remains empirically untested. This article has quantitatively tested the effect of female signatories to peace agreements on conflict-related sexual violence occurring post-conflict, hypothesising that it will have a mitigating effect on the occurrence. Using data from Krause, Krause and Bränfors as well as PRIO’s Sexual Violence in Armed Conflicts, the hypothesis is mainly tested through logistic regression analysis. The research yields no statistically significant result regarding the relationship. Further research is needed in order to establish covariation between the two variables.
Key words: Peace agreement, conflict-related seuxal violence, post-conflict security, female signatories, female participation, gender equality.
_____________________________________________________________________
1. Introduction
“If sexual violence is not addressed squarely in ceasefire and peace processes, there will be no peace for women” (Egeland 2009). Conflict-related sexual violence is a human rights violation which has gained increased attention since reports of its vast occurrence in the wars in Bosnia-Herzegovina, Rwanda and DR Congo (Carlsson 2016:1). It is a crime that has long-lasting consequences for individuals, societies and peacebuilding. Despite its widespread occurrence, sexual violence is seldom addressed in peace negotiations. One suggested explanation for this is the general lack of female participation in peace processes (Sida 2015). However, this suggestion has never been empirically tested. Therefore, further research is needed in order to understand what effect female participation in peace negotiations actually have on the occurrence of conflict-related sexual violence.
Previous research regarding sexual violence in armed conflicts agree that sexual violence is a deliberate action, and therefore preventable (Dolan, Eriksson Baaz & Stern 2020:1157f.f., Wood 2014:478). Policy-makers often stress the urgency of addressing sexual violence during peace negotiations (UN Peacemaker nd, Jenkin & Goetz 2018:261). However this is seldom achieved in practice. As women and girls often are the main targets of sexual violence, women are assumed to be more likely to address the issue. Sexual violence is presumed to be rooted in gender inequality (UNHCR nd), and gender inequality is suggested to be a major predictor of renewed civil war (Nilsson & Svensson 2020). Meaningful female participation in peace processes has been shown to increase both the durability and quality of peace (Krause, Krause & Bränfors 2018). Therefore, female signatories to peace processes are suggested to decrease the likelihood of sexual violence occurring in the post-confict stage.
Two research gaps have been identified in previous research. The first gap entails separating different stages of sexual violence. Previous research tend to examine sexual violence in armed conflicts as one phenomenon, in all of its conflict stages. This paper instead looks solely to sexual violence prevalent post-conflict. It is suggested to increase our understanding of whether women delegates and certain agreement characteristics have any effect on its occurrence. The second gap regards literature on women in peace processes. Previous
research tends to observe the societal and peacebuilding effects of female participation, however the practical consequences for women is often neglected. Therefore, this thesis aims to examine an issue of which women and girls are the main, but not only, targets. Hence, the following research question will be investigated; How does female participation in peace negotiations affect the occurrence of post-conflict conflict-related sexual violence?
A quantitative analysis will be conducted on 132 observations of post-conflict societies which signed peace agreements between 1991-2011. The independent variable will be measured as female signatories and the independent variable will be measured as post-conflict conflict-related sexual violence occurrence. The relationship between the two variables will be tested through several statistical tests, including descriptive statistics, tabular analysis, chi-square test of independence, predicted probability test, bivariate logistic regression and the main multivariate logistic regression.
The results show that there is no estimated statistically significant correlation between the two variables. The bivariate analyses demonstrate a significant relationship, however this is likely a spurious relationship. Despite the result rejecting the hypothesis, this paper has stressed the value in further investigating the effects of female signatories on wartime sexual violence. Furthermore, different types of qualitative analysis (e.g. process tracing), would be valuable for a possibility to examine an underlying causal mechanism.
The paper will be dispositioned in the following way; Firstly, the Theoretical framework chapter will define central concepts, provide an overview of previous research within the concerned fields and present the suggested causal mechanism. Secondly, theResearch design chapter will describe the methodology chosen for the quantitative analysis. Thereafter, it describes the operationalisation, validity and reliability of each central variable and finishes by discussing limitations and potential biases. Thirdly, the proceeding section Results demonstrate the outcome of all statistical tests conducted. Fourthly, the result chapter is followed by a Discussion, analysing the results and providing alternative explanations.
Lastly, the thesis ends with a Conclusion. The R Script is available in the appendix.
2. Theoretical framework
This section will define the terminology and central concepts used throughout the thesis. It will thereafter present previous research findings within the fields of female participation in peace negotiations as well as conflict-related sexual violence. The chapter ends with a section regarding the testable implication and the suggested causal mechanism connecting the independent and dependent variable.
2.1 Terminology and concepts
The unit of analysis in this study is post-conflict societies. The level of analysis is post-conflict years. The population includes all post-conflict societies which signed peace agreements, whilst the sample includes all post-conflict societies which signed peace agreements between 1989-2011.
The dependent variable is p ost-conflict conflict-related sexual violence, which will be divided into three different categories for differentiation and clarification; sexual violenec, conflict-related sexual violence and post-conflict conflict-related sexual violence. Sexual violence [SV] is defined as “any sexual acts or attempt to obtain a sexual act, or unwanted sexual comments or acts of traffic, that are directed against a person’s sexuality using coercion by anyone, regardless of their relationship to the victim, in any setting (...)” (Jansen 2016:4).Conflict-related sexual violence [CRSV] narrows down the scope by identifying the setting, function and perpetrator of the crime. CRSV are deliberate incidents and/or patterns of sexual violence motivated by political, military or economic objectives to control territory or resources. It occurs in conflict or post-conflict settings or in other situations which concern the political strife. They also have a direct or indirect nexus with the conflict itself, i.e.
temporal, geographical or through a causal link (UN Peacekeeping, nd:a). CRSV is perpetrated by conflict actors, i.e. state forces (military, police and PGMs1) as well as non-state actors (rebels and militias) (Koos 2018:198). The definition does not identify the victim, only the type of perpetrator. However, the victims are often actual or perceived
1 PGMs are pro-government militias, sponsored by the government. They either work alongside government armies or independently to shield civilians from rebel influence (Jentzsch, Kalyvas & Schubiger 2015:755)
members of minority groups (UN Peacekeeping, nd:a). The victim can nevertheless have any or no relation to the perpetrator, though has to have a direct or indirect connection to the conflict.
Post-conflict conflict related seuxal violence [PCRSV] only includes CRSV perpetrated post-conflict. Post-conflict can be defined in different ways, however this paper will define post-conflict as post-signing of a peace agreement. This is not the obvious definition as conflicts do not have to end by peace agreements, therefore others may define post-conflict as when the estimated battle-related death count ascends 25 in a calendar year (UCDP, nd), or with a ceasefire or the initiation of a peace process (Frére & Wilén 2015:2f.f.). As the unit of analysis is post-conflict societies, the actual peace agreements and their outcome are of interest to the study.
The independent variable is female signatories to peace agreements. A peace agreement is a formal agreement between at least two conflict parties, addressing the incompatibility, either by clearly regulating or settling the entire incompatibility or parts of it. A conflict is defined according to the Uppsala Conflict Data Program’s [UCDP] definition of an armed conflict; a contested incompatibility concerning either government or territory, where the use of armed force between the two parties results in at least 25 battle-related deaths in one calendar year.
This study only includes intra-state conflicts, hence the conflict parties must consist of at least one government of a state and one rebel group. Signatories are defined as delegates signing peace agreements (UCDP, nd). Women have formally participated at the peace table as signatories, witnesses, mediators and negotiators (Stone 2014:12). However, this study will focus solely on female signatories, representing governments, rebel groups or civil society society. Signatories are presumed to take active part in official as well as unofficial negotiations and are hence believed to have actual influence over the process and agenda (Krause et al., 2018:987). Therefore, female negotiators, mediators and witnesses are not included in the analysis, as meaningful participation is fully possible but uncertain. It also excludes third-party representatives. One important trade-off with focusing solely on signatories is the scarcity of female signatories. It significantly decreases the number of existing cases for analysis, which will be discussed in section 3.5 Limitations and potential biases. However, since it is suggested to be the strongest indicator of meaningful participation, it could increase the variable’s validity and is therefore considered the best option.
2.2 Previous research
2.2.1 Conflict Related Sexual Violence
“CRSV exploits emotionally charged values about sexuality, virtue, shame and honor. CRSV thereby stigmatizes victims, their families and even communities, and leaves a lasting reminder of humiliation and defeat” (Koos 2018:199). Sexual violence is a form of gender based violence mainly, but not exclusively, targeting women and girls. Besides the physical, psychological and sexual harm it poses to individual victims, it also creates consequences of large scale forced displacement, early marriage and children born of rape risking statelessness and stigmatisation (OSRSG-SVC, nd). Some research suggest that sexual violence likely occurs in all armed conflicts, in varying degrees of severity and prevalence (Wood 2010).
Contemporary conflicts are oftentimes characterized by civilian targeting, increasing the risk of violence against women (St. Germain & Dewey 2012:90). Armed conflict tend to heighten levels of sexual violence drastically, levels which often remain elevated post-conflict. Despite the acknowledged prevalence of sexual violence in armed conflicts, it is seldom addressed in peace agreements. Women’s general absence from the peace table is suggested to be one of the main reasons for its neglect (Sida 2015). Research show that 75 percent of women’s voiced demands in peace processes highlight sexual violence as a special concern, revealing a large gap between lived reality and applied measures (UN Women 2012:12).
It is important to note that men and boys are victims of CRSV as well, likely in much larger hidden statistics than those amongst women and girls. Therefore, men in peace processes should in theory be prone to address the issue as well. However, in the common understanding of sexual violence the perpetrator is male and the victim female, since it is the most commonly known scenario. Therefore, NGOs knowledge and priorities of male victims is often minimal (Cohen & Nordås 2014:421). Due to norms of hegemonic masculinity and socially constructed gender roles of masculinity, situations deviating from the common understanding may increase male victims’ experienced sense of stigma and humiliation.
Hence, this paper will assume that women in peace negotiations are more likely to address the issue of sexual violence.
As mentioned in section 2.1 Terminology and concepts, CRSV distinguishes from other forms of sexual violence in the type of perpetrator and nexus to the conflict. CRSV during and post-conflict is often studied as one phenomenon, where the post-conflict setting alone seldom receives much attention. However, this paper wishes to examine conditions and characteristics in peace agreements which may affect the prevalence of PCRSV. Despite the lack of specific research and analysis on PCRSV, there is some existing data. It has been found that in conflicts between 1989-2009, 13 percent (58 of 446) of all conflict actors were reported as perpetrators of sexual violence in the first five-year post-conflict period. 13 percent (28 of 220) of rebels, 21 percent (25 of 119) of state actors and 5 percent (5 of 107) of PGMs were reported as perpetrators (Cohen & Nordås 2014:425).
As CRSV includes the post-conflict period as well, previous research on the topic is equally valuable for theory building. The complex problem of CRSV has gained increased attention in the last few decades. There has been an observable shift in theoretical focus in scholarly work over the years. For a long time the dominating notion was that CRSV was more or less an inevitable by-product of war. This was later replaced with the framing of CRSV as a deliberate tactic or weapon of war. Benefits with the shift include the vital possibility to punish the crime under international humanitarian law, increased acknowledgement of the severity of the crime and the realization that CRSV as a deliberate action is preventable (Dolan et al., 2020:1157f.f., Wood 2014:478). The prevailing mantra however, is to end the focus on the “rape as a weapon of war” explanation and instead look more thoroughly at organisational structures instead (Davis & True 2015:495). If conflict-related rape occurs solely as a tactic of war, would not all armed actors perpetrate sexual violence? The prevalence and type of sexual violence perpetrated during conflict differs greatly between actors and dyads (Wood 2014:457), hence complementary explanations for CRSV are required.
Several important suggested explanations for CRSV have been developed already. Some of the classic explanations include the previously mentioned CRSV as a weapon of war, an increased opportunity to commit sexual violence during war, rape as “substitution” for sex, ideology and gender norms (Wood 2014:462, Wood 2008). None of the above mentioned should be excluded as the do carry explanatory power, though none succeeds in explaining the variation in violence; why do many armed actors not perpetrate sexual violence (Wood 2014:463f.). Wood argues that variations in leadership institutions, ideology and the culture
in armed organisations could explain the variation in sexual violence (Wood 2009, Wood 2014:466). Some CRSV is explicitly ordered and adopted as a strategy by higher commands within the organisation. However, patterns of CRSV more often occur not as an ordered strategy, but as a practice This is understood as an action which is unordered, though tolerated, by commanders. Motivations behind CRSV vary greatly, reports and studies tell of motives ranging from individual onesto the socialization theory (suggesting that sexual violence can be used to build group cohesion), to strategic motives for political goals (Karuhanga Atuhaire, Gerring, Huber, Kuhns & Ndirangu 2018:5, Cohen & Nordås 2015:878). The continuum theory suggest that CRSV is a continuum of sexual violence that occur in non-conflict settings as well, and should therefore occur in the same way though not in the same prevalence (Wood 2014:463). Regardless of what circumstances trigger sexual violence perpetrated by armed organisations, social interactions between commanders and combatants are relevant for understanding how it could be prevented from within the organisation (Wood 2018:513).
“When an organization does adopt rape as a policy, it may do so for internal rather than strategic reasons, and it may promote its occurence through authorization or institutionalization rather than explicit orders” (Wood 2018:528f.). Effective prohibition of the previously tolerated violence requires strong institutions of leadership and authority. The mere use of unauthorized CRSV may already have undermined the conditions necessary for command and control (Ibid:530). Prohibition and ending tolerance of the use of sexual violence requires strong group cohesion and leadership capability (Cohen & Nordås 2015:878). If the social dynamics in an armed organisation does not allow for authoritative leadership capability, it is difficult to prohibit such crimes. CRSV is incredibly complex, motivated and committed by a mix of individuals and collective (UN OCHA 2008:1).
Therefore, this research suggests that juridical prohibition, international legislation, could have a dissuasive effect on an individual level as well. Conflict-related sexual violence is punishable under international humanitarian law, international human law and international criminal law (UN Peacekeeping, nd:a).
This thesis will base its understanding of CRSV on Elisabeth Wood’s work, as a pattern occurring either as a strategy or a practice. It does not aim to explain the root causes of CRSV or PCRSV, but to examine the context and prerequisites in peace agreements that may affect its occurrence.
2.2.2 Women in peace processes
“Wherever I have served, I always get some rendition of the same line. From UN generals, to Libyan militias, to Afghan religious leaders; ‘As soon as this war is over, we can talk about your women’. And I always have the same response. ‘This war is not going to be over until you include your women’” (Omar 2018). Women’s rights and inclusion is often considered a secondary issue, separate from that of peacebuilding. Fighting in armed conflict is the only social situation in which an entire gender is, and historically has been, nearly completely excluded (Malesevic 2010:275). Consequently, women have largely remained excluded from the peacebuilding process as well, However, women participation is an important cornerstone in reducing the risk of conflict and enhancing stability (Council on Foreign Relations, nd).
Female participation in peace negotiations is a field which has gained increased scholarly attention since the 1990s. An important milestone in the aim for inclusion was the United Nations [UN] Security Council unanimous adoption of resolution 1325 Women, Peace and Security in 2000. A central goal since the resolution has been to increase meaningful female participation and increase gender perspectives in peace processes (United Nations 2000, Nilsson & Svensson 2020:1). Despite the resolution and increased international attention to the issue, the last 20 years has seen a continuous absence of women in formal negotiations.
This points to a gap between policy and reality (UN Women 2012:1). Between 1992 and 2019, women constituted approximately 14 percent of negotiators, 11 percent of mediators and 7 percent of signatories in major peace processes globally (Council on Foreign Relations, nd). This gap is somewhat surprising, given the robust evidence of the effect of women participation on positive peace.
Previous research has found that women’s participation in peace negotiations increases both the durability and quality of peace. Short-term peace is suggested to be 20 percent more likely to last at least two years, whilst it long-term is 35 percent more likely to last at least 15 years (Stone 2014:29). This is mainly explained by two different factors, the first one being women’s tendency to cooperate with civil society organisations [CSO], women civil society organisations [WCSO] in particular. CSO and WCSO inclusion in peace processes makes a peace agreement 64 percent less likely to fail (Nilsson 2012:258). Linkages between female signatories and WCSOs are crucial for inclusion of provisions and informing the negotiation process about civic issues, gendered ones in particular (Krause et al., 2018:1006).
Comprehensive agreements with female signatories show an overwhelmingly better score for implementation of women’s rights, further suggesting that women’s participation increases the durability of peace (Krause et al., 2018:1002). Gender inequality in post-conflict societies is a strong predictor of renewed armed conflict, and CSO and WCSO cooperation can increase chances of strengthening gender equality. WCSOs play a crucial role in societies of varying degrees of gender equality, suggesting that they can create room for dialogue and engagement in the concerned questions even in highly gender unequal contexts (Nilsson &
Svensson 2020:5).
Broadening participation through CSO cooperation is beneficial in several ways. When political and societal actors are included in the peace process, the resulting peace tends to be more sustainable. It can strengthen the effectiveness of the mediation process itself, as well as the quality and sustainability of the agreement. This happens through several different mechanisms; enhanced legitimacy and representation (representation of more groups enhances the legitimacy of the agreement), greater diversity and a broader negotiation agenda (which helps ensure that broader public interests are included while counteracting interests of the political elite) and monitoring the process. Public buy-in during the negotiation stage is suggested to create a pro-agreement atmosphere in the larger society (Paffenholz 2014:8f.f.) as it broadens the set of issues taken into account. One of the main reasons armed groups resort to violence is to contest their exclusion from social, political and/or economic power.
Inclusion of additional actors beyond the main conflict parties (such as CSOs and political parties) in negotiations is crucial in making war-to peace and the political transition more sustainable (Paffenholz & Ross 2015:28f). Increasing CSO cooperation is hence presumed to lessen the risk of violent spoilers during the negotiation and short-term post-conflict stages.
One of the pitfalls of inclusion is that broadening the set of issues at the table also could expand the incompatibility and further complicate the process (Paffenholz 2014:11). There is a dilemma between gaining support for the peace process to increase its legitimacy and the complication of negotiations and ensuring its efficiency. The worst case scenario with CSO inclusion and the increased complexity is that divisions may hinder reaching an agreement at all. Increased legitimacy could however also work in favour of implementation prospects (Nilsson 2012:248).
The second explanatory factor for the positive relationship between female participation and the durability of peace is women’s tendency to focus on provisions with regards to societal
development. Studied agreements without female signatories included more provisions regarding military reform and withdrawal of troops, whilst the compared agreements with female signatories included a maximum of provisions with regard to political, social and economic reform (Krause et al., 2018:1001, Karim Issifu 2020:58), which increases the chances of positive peace. The highly gendered consequences of war can explain the equally gendered understandings of peace and security. As the majority of combatants are male, men tend to die at higher rates from the direct consequences of conflict, in battle-related deaths.
Women on the other hand run a higher risk of dying from the indirect consequences post-conflict, such as the breakdown in social order, economic devastation, human rights violations, from infectious diseases or from lethal conflict-related sexual violence (O’Reilly, Ó Súilleabháin & Paffenholz 2015:5). Women often face a continuum of violence, related to the conflict and continuing post-conflict. This paper suggests that it therefore is crucial to examine the post-conflict consequences of war from a gendered perspective.
Research has also shown that peace agreements with gender sensitive language and inclusion of gender provisions cannot be equated with meaningful female participation. The peace agreements with the most holistic reference to women and gender tend to suffer the most from implementation failures, since there is little agreement between the conflict parties (Krause et al., 2018:1006), and little reference to the actual context. Therefore, female participation cannot be replaced with an expressed feminist agenda. Other studies have found that women participation is likely to generate at least one gender provision in an agreement, however it does not seem to have an effect on wider implementation of numerous gender provisions. The suggested explanation for this is the participating individual's will, capacity and circumstances to actively put gender provisions on the agenda (Tengbjer Jobarteh 2018:48). The recognition of women as a heterogenous group, individuals with different interests and circumstances, emphasizes the need for an ethical comment regarding the topic.
Seemingly establishing traditional gender roles as the root cause and explanation for women’s efficiency and “usefulness” in peace processes requires essential disclaimers for further analysis. This paper does not, similarly to other feminist research in the field, suggest that women are inherently more peaceful. Neither does it claim that women are a homogenous group sharing common interests, needs or experiences of conflict. Women in peace processes are often denied the right to meaningful participation and ought to promote the interest of the party, however it is likewise possible that forwarding gender issues is not in
their personal interest. Most importantly, it does not claim that women’s historical subordination is beneficial for the peace process. Women’s traditional roles as nurturers, in both family and societal settings, in combination from their general exclusion from the battle-field has created a perception of women being more peaceful. Men on the other hand are considered less peaceful and less caring for social needs. This paper assumes that these socially constructed gender roles are harmful in both directions. Female inclusion should never be considered simply a means towards positive peace, but an absolute necessity and a given for adequate representation of the people. As past research has provided evidence for women’s increased likelihood of CSO and WCSO inclusion as well as increased focus on societal and developmental provisions, these results are the outline of the study.
Additionally, this paper considers all conflict parties and their representatives rational actors.
More specifically, it presumes all actors to function according to the subjectivist decision theory; making rational choices to the best of their knowledge (Frahm 2019:3). A rational choice is defined as “an optimal action or strategy, i.e., maximize their expected utilities”
(Ibid, 92).
2.3 The causal mechanism
This thesis aims to research PCRSV, perpetrated after the signing of peace agreements. It aims to examine traits of peace agreements that could have deterring effects and hinder the perpetration of sexual violence. The explanatory variable is suggested to be female signatories. The research aims to answer the research question How does female participation in peace agreements affect the prevalence of post-conflict conflict-related sexual violence?
The independent variable is female signatories, and the dependent variable is PCRSV. This section will explain every step in the suggested relationship’s causal mechanism.
Previous research has provided both quantitative and qualitative evidence for women’s meaningful participation in peace negotiations leading to more durable and qualitative peace (Krause et al., 2018, O’Reilly et al., 2015, Paffenholz 2014). Gender inequality is one of the main predictors of renewed civil war (Nilsson & Svensson 2020), and female participation is expected to increase the likelihood of addressing gendered consequences of conflict through CSO and WCSO cooperation (Krause et al., 2018, Nilsson & Svensson 2020). Gender equality is also suggested the prevalence of sexual violence (World Health Organisation nd, ActionAid UK, Gender and Development Network, Womankind, International Planned
Parenthood Federation & Orchid Project nd). Therefore, female participation in peace negotiations is suggested to have an effect on levels of PCRSV. The reasoning of this claim is twofold, as visualised in Figure 1 below.
Figure 1. The causal mechanism.
The first line of reasoning suggests that female signatoreis increase the likelihood of addressing sexual violence during peace negotiations, if prevalent during onflict. As previously mentioned, 75 percent of women’s raised concerns in peace processes regard sexual violence (Sida 2015). The leads to the second step in the causal mechanism. The increased likelihood of voicing concerns of CRSV is suggested to raise awareness and engage discussions on the topic. This could increase the chances of CRSV being covered and monitored as a potential violation of the peace agreement and that the perpetration of sexual violence is excluded from any amnesty provisions (UN Peacemaker nd). Discussing sexual violence may also shed light on the different types of crimes CRSV actually can constitute under the ICC’s Rome Statute. Since all individual actors and armed organisations are presumed to act according to the rational choice theory, the increased awareness is suggested to make PCRSV too costly due to fear of repercussions. This fear can work on both an individual and on a group level. On a group level, higher commanders can formally prohibit the use of PCRSV. However, this requires strong authority and leadership (Cohen & Nordås
2015:878). If this does not exist within the organisation, higher commands do not have the authority to break chains of predisposed patterns of violence. However, this paper argues that the mechanism can work on an individual combatant level as well.
The second line of reasoning, which works complementary and simultaneously with the first line, concerns women’s tendency to cooperate with CSOs and WCSOs (Nilsson & Svensson 2020, Paffenholz 2014). WCSO cooperation further increases the likelihood of addressing CRSV during peace negotiations. Since women are not a homogenous group and not all women share the same experiences nor personal interest in addressing gender issues during negotiations, it is not guaranteed that female signatories are to address sexual violence.
WCSO inclusion in peace processes is therefore presumed to increase the likelihood of putting CRSV on the agenda, if prevalent during the conflict. This leads back to the assumption that increased pressure on negotiating parties to address sexual violence will elevate the chances of highlighting repercussions of the crimes. Female delegates are also more likely to cooperate with CSOs. CSO cooperation increases the legitimacy of the process, exerts public pressure on conflict parties whilst simultaneously broadening the agenda. It also produces a greater pro-agreement atmosphere (Paffenholz 2014, O’Reilly et al 2015). In accordance with the rational choice theory, and the assumption that all actors want to maximize their own utility, deviating from the peace agreement is presumed to be even more costly if there is wide public support for the agreement. Due to the two complementary suggested lines of reasoning, the prevalence of PCRSV is suggested to be lower in post-conflict societies where women operated as signatories to the peace agreements than those without female signatories. The paper will hence examine the following hypothesis;
H1: Female participation in peace negotiations decreases the risk of post-conflict conflict-related sexual violence occurring.
In summary, this paper suggests that the participation of female signatories to peace agreements leads to lower levels of PCRSV. The explanatory mechanism is twofold and is expected to work complementary. Women are more likely than men to voice concerns regarding CRSV. They are also more likely to include WCSOs in the process, which further increases the likelihood of addressing sexual violence during negotiations. The highlighting of sexual violence is presuemd to reenforce repercussions against the perpetration of CRSV.
Women are also more likely to cooperate with CSOs, which can enhance a public pro-agreement atmosphere. With all of the above taken into consideration, the rational
conflict actors (on an individual and organisational level) are facing tougher consequences (or have an increased understanding of them), in terms of both juridical and societal repercussions. As rational actors, they are presumed to want to avoid that. Therefore, female signatories are presumed to decrease the risk of PCRSV. The suggested causal mechanism is based on evidence presented in section 2.2. Previous research.
3.
Research design
This chapter will present and motivate the research design chosen to examine the research question. Firstly, it will describe the methodology used for analysis. Secondly, it will describe the datasets where data is collected from. Thirdly, it will discuss and motivate the operationalisations, validity and reliability of all important variables, i.e., the independent, dependent and control variables. Lastly, it will discuss limitations and potential biases of the paper.
3.1 Methodology
This study aims to answer the research question How does female participation in peace negotiations affect the occurrence of post-conflict conflict-related sexual violence? and will do so through a quantitative analysis. “Broader quantitative studies must first reveal the trends before an analysis of the intricacies of individual cases can be conducted” (Stone 2014:11). Since there is no previous research focusing solely on the peace agreement characteristics affecting PCRSV, a quantitative study is preferable for observing trends and tendencies. Quantitative research has a better ability for hypothesis testing and has a higher potential to isolate the effects of the IV of the DV due to the larger sample size than that of qualitative research. Large-N studies can establish whether there is a covariate relationship and are more likely to yield generalizable results, however they cannot ensure causality (Brancati 2018:101f.f., 302). A mixed methods approach would have been preferable in order to build a more accurate theory and produce greater confidence in the analysis (Ibid 112), however due to time constraints of the assignment it will not be feasible to conduct. The population of interest is all-post conflict societies which signed peace agreements, and the sample is all post-conflict societies peace agreements between 1989-2011. The sample hence includes a total of 82 peace agreements (Krause et al., 2015:994). The study takes the four causal hurdles into thorough consideration and will discuss them in relation to the proposed relationship. The causal hurdles are covariation, isolation, causal mechanism and time order.
Firstly, establishing a causal mechanism between the two variables is not feasible as it is a large-N study. The study may produce a credible causal mechanism, however it cannot establish causality (Kellstedt & Whitten 2018:64). Secondly, time order should not be an issue to establish as the theory and logic of coding has cleared that hurdle (Ibid 62). The DV, PCRSV occurrence, is bound to occur in a post-conflict setting. As post-conflict is defined as a post-agreement signing stage, and the DV will be lagged and take values only for post-agreement years, PCRSV cannot occur prior to the agreement. Therefore, the hypothesis performs well on the time criterion. Thirdly, covariation will be tested through bivariate hypothesis tests. A tabular analysis will be conducted. A tabular analysis is an appropriate choice of hypothesis testing, as it is suitable for two categorical variables. It will examine the distribution of the DV values across values of the IV (Ibid 162f.f.). In order to obtain a p-value, a chi-square test of independence is also suitable for two categorical variables (Statistics Solutions, nd). Covariation will also be examined through a bivariate logistic regression. Fourthly, isolation is the criteria which establishes that there are no confounding variables affecting the relationship (Ibid 97). It will be examined through a multivariate logistic regression. Logistic regressions are suitable for two binary variables, as they are one type of generalized linear models. Results are expressed in log odds of the odds ratio (Sperandei 2013:12). The logistic regressions will provide the research with an indication of the statistical significance of the relationship. It will ensure some degree of isolation, at least in the context with the selected control variables, however it cannot account for other potential confounders. Hence, it will only ensure isolation within the chosen scope. Lastly, a predicted probability test will examine the effect of the IV on the DV in a pre-decided situation, where all continuous covariates take their mode value and the ordinal ones take their mean value (UCLA, nd). This will produce results of the procentual likelihood of PCRSV occurring in the given scenario. The results are easily interpreted, though are only applicable to the specific situation.
3.2 Datasets
Data has been merged from two separate datasets; the Krause, Krause and Bränfors’
Replication dataset [Krause] (Krause et al., 2018) and PRIO’s Sexual Violence in Armed Conflict 2.0 dataset [SVAC] (Nordås & Cohen, 2019). The Krause one is the main dataset, including information on peace agreements between 1989-2011. Due to drastic changes in the
international environment and a paradigm shift in peacemaking after the Cold War, the authors decided to only include data from 1989 and onwards. The dataset is based on information from the Peace Accord Matrix and UCDP. It was created for a mixed-methods study, testing the relationship between women’s participation in peace agreements and the durability and quality of peace. It includes a total of 82 peace agreements in 42 intra-state conflicts (Krause et al., 2018:987f.f). The independent variable, female signatories, as well as all control variables (power-sharing provisions, comprehensive peace agreement, democracy, UN peacekeeping missions and leftist ideology) are available through the Krause dataset. The unit of analysis in the dataset is peace agreements, hence there is a dyadic lay-up of the dataset.
The SVAC 2.0 dataset includes information on reported sexual violence in armed conflicts between 1989-2015 and contains information about the prevalence and different forms of sexual violence perpetrated. Data has been coded from Amnesty International, Human RIghts Watch and the US State Department annual reports. These are the three most commonly referenced sources in human rights literature (Cohen & Nordås 2019:10). The dataset includes all conflicts that were active, or within five years of cessation, within the timeframe.
The unit of analysis in the SVAC dataset is conflict-actor-year (Ibid 4f.).
3.3 Operationalisation, validity and reliability
3.3.1 Dependent variable
The dependent variable post-conflict conflict-related sexual violence is operationalised as the occurrence of sexual violence perpetrated by conflict actors (state armies, reel groups or PGMs) against individuals outside the organisation, after the signing of a peace agreement to an intra-state conflict. The operationalisation identifies state armies, rebel groups and PGMs as perpetrators of CRSV. The theoretical definition does not define the victims of sexual violence, however the operational definition only includes violations against individuals outside the own organisation. This is in accordance with the SVAC dataset (Cohen & Nordås 2019:4). The demarcation has not been explained in any publication related to the dataset. The operationalisation does not explicitly state the function or motive of violence, nevertheless only including victims from outside the perpetrators in-group could indicate that there is a political motive behind the perpetration, regardless of its goal. This is in accordance with the theoretical definition of CRSV and is hence not considered an issue
for validity. The post-peace agreement element and the fact that the dataset only includes the first five-post conflict years ensures a temporal nexus to the conflict.
PCRSV is a binary variable, either taking the value(0) = Absent or (1)= Present. Therefore, it will measure whether any female participation has any effect on PCRSV occurrence, instead of its prevalence. In the SVAC dataset sexual violence is an ordinal variable, ranging from 0-3. This thesis decided to use a binary variable instead, to avoid any problems concerning the authors or the original sources potential biases whilst describing and categorising, as that could have jeopardized the variable’s reliability. SVAC categorised CRSV as either (0)=
None, (1)= Isolated, (2)= Numerous or (3)= Massive (Cohen & Nordås 2019:9f.f.). Hence, category 1-3 have been coded as (1) in a dummy variable for this paper. Missing information was coded -99 and was later removed from the dataset in the analysis. Moreover, SVAC coded CRSV into three separate columns divided by source. In this paper, the three columns are juxtaposed into one, taking the value (1) if any of the three sources reported any sexual violence. As the Krause dataset only includes peace agreements, all cases from the SVAC dataset that merge into the Krause one will be conflicts ending in peace agreements.
Validity is the most important feature of a measure, indicating that the variable accurately measures what it is intended to measure (Kellstedt & Whitten 2018:114). The operationalisation does not differ majorly from the theoretical definitions, which indicates strong face validity. Excluding sexual violence perpetrated by armed organisation representatives against individuals within the organisation arguably strengthens the variables content validity. Content validity measures whether the essential elements are included, and whether the irrelevant ones are excluded (Ibid 115). By excluded sexual violence committed against individuals within the organisation, it ensures a function of the violence. The function is an important part of the theoretical definition, hence content validity is strengthened.
One important issue with the dataset is that the 2.0 version lacks some information. In comparison to the prior version of the dataset, it contains additional years but lacks some variable information due to data availability constraints. Therefore, information regarding sexual violence perpetrated by PGMs is not included for years 2010 and 2011 (Cohen &
Nordås 2019:4), which could reduce the face validity somewhat, and therefore will be further discussed in section 3.5 Limitations and potential biases. As it concerns very few cases, it arguably does not compromise validity to a large extent.
The concept’s reliability refers to the likelihood of reaching the same results if tested in the same situation again, i.e. the consistency of the variable outcome (Kellstedt & Whitten 2018:112, Heale & Twycross 2015:66). One of the issues with the dependent variables reliability is the three source system for data collection by the SVAC dataset. By creating a binary dummy variable instead of the ordinal one available in the SVAC dataset, this thesis avoids the risk of measurement bias from the SVAC categorising. However, the risk of measurement bias from the original sources remains. The three source system will be further discussed in section 3.5 Limitations and potential biases.
3.3.2 Independent variable
As mentioned in section 2.1 Terminology and concepts, the independent variable is narrowed down to female signatories to peace agreements as signatory status is the strongest indicator of meaningful participation (Krause et al., 2018:987). The operational definition of the independent variable is hence a minimum of one woman signatory (representing a government, rebel group or civil society) to a full or partial peace agreement of an intra-state armed conflict. This definition captures the essential features; the role of female participants which is of interest as well as the requirements for peace agreements to be included in the dataset. Krause et al collected data on female signatories by cross-checking a list of signatories from the UCDP Dyadic Dataset with information from the UN Peacemaker database in order to verify their sex (Krause et al., 2018:994). In accordance with the Krause dataset, only peace agreements to intra-state conflicts are included, as a majority of modern conflicts are intra-state. Furthermore, the definition only includes full and partial agreements.
It therefore excludes process agreements, since they are connected to the negotiations but not the agreement itself. The authors also decided to exclude all durable partial agreements which were followed by a final one, to avoid overestimating the impact of female participation.
Similarly, if women signed partial agreements though not the full one, the partial agreements are excluded and the full agreement is coded as signed by a female delegate (Krause et al., 2018:993).
Female signatories is a binary variable and will be coded as either (0)= Absent or (1)=
Present. A minimum of one woman has to participate for the variable to take the value 1.
Arguably, it would have been interesting to see the IV as an ordinal variable, in order to examine the potential effects of broader versus small-scale participation. However, the Krause dataset only includes a binary variable and due to time constraints the thesis cannot
introduce additional datasets. Therefore, this study will examine whether there is an observable effect of any female signatory participation on any PCRSV occurrence.
The independent variable’s validity is estimated to be strong. The theoretical and operational definitions cohere to a very large extent, therefore it has strong face validity. Likewise, content validity is presumed to be strong as only the essential elements are included in the operationalisation. One potential challenging aspect to content validity is the fact that the independent variable is dichotomous. However, as the thesis only aims to examine the effect of any female participation, it is presumed not to make a major difference. Having a binary variable can actually avoid jeopardizing the variable’s reliability, in terms of measurement bias as discussed regarding the dependent variable. It hence faces no issue with objectivity in scoring values.
3.3.3 Control variables
This research will use five different control variables in order to examine the treatment effect of the IV on the DV. The control variable will be used in the multivariate logistic regression and examine each factor’s effect on the outcome variable. The five control variables are power-sharing provisions, comprehensive peace agreement, democracy, UN peacekeeping missions and leftist ideology. All four are available through the Krause replication dataset (Krause et al 2018).
Power-sharing provisions is defined as the number of military, political and territorial power sharing provisions (political arrangements for sharing power) included in a peace accord , and refers to the peace agreement itself. Previous research has found a significant relationship between power-sharing and the durability of peace (Hartzell & Hoodie 2003, Jarstad &
Nilsson 2008, Bell 2018). Power-sharing institutions have higher judicial independence rates post-conflict vis-á-vis countries that have no or few power-sharing provisions. This is explained by them having an effect on mitigating feelings of insecurity and enhancing checks and balances that constrain political and conflict actors (Hartzell & Hoodie 2019:652).
Therefore, the power-sharing variable is related to the DV, as checks and balances are presumed to enhance actors accountability post-conflict. It is an ordinal variable, coded as an additive measure ranging from 0-3 based on the number of provisions in a peace agreement (Krause et al., 2018:995). As the Krasuse dataset has a dyadic framework, there are several observations for each conflict-year. Therefore, in the initial stage of coding when subsetting
data, the mean of all values for each year is selected, since this is an ordinal variable. The variable’s face validity is strong, as it is operationalised in accordance with the Krause dataset, hence it measures exactly what it is intended to. As there is not much room for interpretation, the variable is estimated to have strong reliability and produce replicable results.
Comprehensive peace agreements [CPA] are peace agreements that include all conflict dyads, in contrast to dyadic agreements in which at least one of the actors are excluded (UCDP, nd). Peace agreement is defined in accordance with the theoretical definition in section 2.1 Terminology and concepts, as a formal agreement between the conflict parties, addressing the incompatibility, either by clearly regulating or settling the entire incompatibility or parts of it (Ibid). CPAs tend to be more durable, since conflict actors face a prolonged negotiation phase and tend to resolve a wider set of issues in numerous partial agreements which are later ratified in a final accord (Joshi & Quinn, 2015, Krause et al., 2015:995). Resolving a wider range of issues could increase the likelihood of addressing CRSV, if prevalent before or during conflict. CPA is a dichotomous variable, coded either (0)= Not comprehensive agreement, or (1)= Comprehensive peace agreement. When merging data, the maximum value of CPA for each conflict-year will be selected, as it is a binary variable. CPA is suggested to have strong validity and reliability, as it is very straightforward.
If all conflict dyads are included in the agreement, the agreement is comprehensive. It does not risk measurement bias, and the variable measures what it is intended to.
Democracy is a control variable on which Krause et al collected data from the Political Instability Task Force dataset (Marshall, Gurr and Jaggers 2016, Krause et al., 2018:996).
The dataset defines democracy as “three essential, independent elements. One is the presence of institutions and procedures through which citizens can express effective preferences about alternative policies and leaders. Second is the existence of institutionalized constraints on the exercise of power by the executive. Third is the guarantee of civil liberties to all citizens in their daily lives and in acts of political participations. Other aspects of plural democracy, such as the rule of law, systems of checks and balances, freedom of the press, and so on are the means to, or specific manifestations of, these general principles” (Marshall, Gurr and Jaggers 2016:15). Research has shown that democracies rarely engage in widespread perpetration of sexual violence (Wood 2006:332). Democracy is a dichotomous variable, coded either (0)=
Non-democratic, (1)= Democratic or (2)= Fully democratic. The variable is lagged by one
year in the Krause dataset to avoid endogeneity problems. The values in the Political Instability Task Force dataset range from 1-10, however the Krause dataset has coded all values below five as non-democratic, 5-9 as democratic and 10 as fully democratic (Krause et al., 2018:996). For merging, the mean value of democracy is submerged for each conflict-year, since it just like the power-sharing variable is an ordinal variable. Democracy is a very thick concept, entailing many different dimensions, hence it is difficult to operationalise. The variable’s face validity is considered strong as it is defined according to the original dataset. Reliability is compromised by the ordinal nature of the variable. As democracy has been categorised and re-categorized by the Krause dataset, it risks measurement bias.
UN peacekeeping missions are peacekeeping missions employed by the United Nations in areas where civilian security is compromised, in order to facilitate peacebuilding and provide physical security. Its operations are effective means of peacebuilding mechanisms, monitoring and working against impunity, oftentimes in cooperation with provincial, national and international partners. Peacekeeping missions also have the mandate and capacity to provide multi-disciplinary measures to prevent and respond to CRSV (UN Peacekeeping nd:a, UN Peacekeeping nd:b). Therefore, UN peacekeeping missions are presumed to have an effect on PCRSV. It is a binary variable, coded either (0)= Absent or (1)= Present. The variable is coded time-variant as it changes over time. Some peacekeeping operations have a specific focus on combating CRSV during and after conflict, though that cannot be specified from the variable. The presence of any peacekeeping operations can however have a mitigating and dissuasive effect on PCRSV, hence the validity of the concept is not compromised. When merging data, the maximum value of UN peacekeeping missions for each conflict-year will be submerged, as it is a binary variable. The variable simply tells whether peacekeeping operations were present or not, therefore both validity and reliability are suggested to be high.
Leftist ideology is a control variable defined as rebel groups with Marxist ideology (i.e., socialist, communist, Maoist, or Marxist-Leninist) (Wood & Thomas 2017:39). Previous research has found evidence for leftist groups being less likely to perpetrate sexual violence than other groups (Sarwari 2020). Ideological strive for equality, including that between sexes, and the increased likelihood of including female combatants in leftist armies are used as explanatory factors (Wood & Thomas 2017:31). Leftist ideology is a binary variable
coded either (0)= No leftist ideology or (1)= Leftist ideology. When merging data, the maximum value of leftist ideology for each conflict-year will be submerged, as it is a binary variable. The variable is suggested to be both valid and reliable, as it measures only what it is intended and does not risk measurement bias. Therefore, it would likely yield the same results if repeated.
3.5 Limitations and potential biases
Due to data and time constraints, this study has been forced to make some trade-offs. Some of the potential limitations and decisions concerning them are complex and require further discussion and motivation, which is what this section is for. These limitations are important to keep in mind for the forthcoming result section. The first and perhaps most crucial problematic concerns the dependent variable. Sexual violence is a very difficult phenomenon to study, as it likely occurs in high hidden statistics due to the sensitive nature of the crime.
Hence, available data may not be reliable. Although numbers may not be reported in a reality-reflecting manner, this paper and its selected datasets does not consider strictly numbers of victims. The SVAC dataset explicitly states that it cannot be used to estimate numbers of victims, and has twofold criterias for the categorisation of violence. It is coded from qualitative information, which categorises either based on numbers of reported victims or by criterias regarding reported violence (Cohen & Nordås 2019:9f.f.). Relying on reports is very language and jargon dependent, and it could theoretically differ majorly between different sources. It is however considered the only viable option, due to the complexity of the issue.
The SVAC dataset has coded information from the three most commonly referenced human rights sources; Amnesty International [AI], Human Rights Watch [HRW] and the US State Department (Cohen & Nordås 2019:10). All three are major human rights organisations dependent on their perceived objectivity, however it is possible that they still work under some bias, jeopardizing the reliability of the DV. Underreporting is the most commonly cited potential bias in data collection of CRSV. For example, male victims were only reported for about 1 percent (72 of 7286) of all cases in the SVAC dataset (Cohen & Nordås 2014:421), which certainly does not reflect reality. AI and HRW are NGOs which might have incentives to overestimate numbers and use strong language in their reports for the sake of aid, publicity or gained support. The US State Department on the other hand represents a country, the
country with the strongest military force in the world (Global Fire Power, 2020), and could theoretically have incentives to underestimate numbers and use milder language regarding their own or ally forces, and likewise overestimate figures for opponents. Publications related to the SVAC dataset discuss the concern of measurement bias and have taken several measures in order to limit them. These include multiple data sources for triangulation, training of the coder teams, testing of alternative data collection processes, and documenting coding decisions in standardized manuscripts (Cohen & Nordås 2014:422, Cohen & Nordås 2019:12f.). By creating a binary variable, the research only examines whether any CRSV has been reported, not in what frequency, which increases reliability.
As mentioned in operationalisation section 3.1.1 Dependent variable , the 2.0 version of the dataset does not include information regarding violence perpetrated by PGMs for the years 2010 and 2011. Due to time and data availability, the authors did not update the figures for PGMs. As the dataset does not define which actors perpetrate violence in the 2.0 version, it is impossible to know what figures have gone missing. This thesis argues that this is not an issue, since the research examines CRSV perpetrated by any conflict actor. As only 5 percent (5 of 107) of PGMs were reported as perpetrators of sexual violence for the years 1989-2009 (Cohen & Nordås 2014:425), year 2010-2011 likely does not contain enormous figures that would have made a significant difference for the quantitaive analysis. Since it only concerns two years and data is limited as it is, limiting the entire scope to >2009 would not be feasible.
The only potential effect is an underestimation of sexual violence perpetrated in those particular years, and whilst it is not desirable, it will not risk overestimating the effects of female participation.
The decision to only include female signatories in the independent variable is based on it being the strongest indicator of meaningful participation (Krause et al., 2018:987). Other peace table delegates, such as mediators, negotiators and witnesses, can naturally participate equally directly and meaningfully, however it is more uncertain. Only including female signatories however significantly decreases the number of available cases with female participation. However, female signatories were used as the independent variable in the Krause et al article stemming from the replication dataset, therefore the data is not too limited. It should be kept in mind for the analysis that larger samples tend to lower the p-value, and increase our confidence in generalizability (Kellstedt & Whitten 2018:164).
However, this paper argues that it is more important to have valid measurements.
5. Results
This section will present the results from the quantitative analysis conducted in R Studio.
Firstly, it will present the initial descriptive statistics. Secondly, it will show the results from a tabular analysis and a chi-square test of independence. Thirdly, it will present the logistic bivariate and multivariate regressions. Lastly, it will include the results from a predicted probability test.
5.1 Descriptive statistics
When merging the Krause and SVAC datasets together into a final dataset, it includes 165 observations of 33 peace agreements between 1991 and 2011. The final dataset includes the independent variable with values of female signatories, the five control variables and the dependent variable with values of PCRSV occurrence. Arguments signed in 1989 and 1990 were hence excluded. It includes a total of 11 agreements with female signatories in 5 different peace processes. The included cases are Guatemala (1996), DR Congo (2003), UK/Northern Ireland (1998), El Salvador (1992) and Libya (2003). The Krause dataset included six different processes (Krause et al., 2018:994), however Papua New Guinea (2001) was excluded when merging.
Summary statistics provide an overview of the sample data. Table 1 below provides summary statistics for all variables. Column N shows how many observations are available per variable. The last observation for each conflict-id has been removed from the PCRSV variable, in order for the lag variable not to accidentally take the first value of the next conflict. Therefore, the variable (DV) PCRSV (lagged) has 33 fewer observations than the others. Data skewness measures the symmetry of the data. Real life data is often not symmetrical, and this is the case for this research as well. The independent variable has right skewed (positive) data, as the variable mean is higher than the median. The data skewness is valued at ~2.14, indicating highly skewed data. Only 22 of 165 observations take the value 1 for female signatories, therefore it is not surprising. Skewed data is not necessarily a problem, and as the general distribution of observation is similar to that in the Krause dataset, it should not be a problem for regression. The dependent variable has a much more normal distribution
of values, A skewness test shows ~0.03 for the DV. It is not identical with the measures of central tendency, thus it is not a completely normal distribution but approximately symmetric.
The distribution of the observations will be visualised in the next section. The DV has a mean of 0.492, indicating that PCRSV occurred in almost half of the sample conflicts. The IV has a mean of 0.133, showing that women were not signatories to most of the agreements in the sample.
Table 1. Summary statistics.
5.2 Bivariate hypothesis tests: Tabular analysis and chi-square test for independence
A bivariate hypothesis test is used to examine covariation between the IV and DV. It examines whether the variables are related and whether the relationship is likely to be generalizable to the larger population of cases (Kellstedt & Whitten 2018:161f.). From an initial cross-tabulation (a two-way frequency table) in Table 2 below, post-conflict societies signing agreements with female signatories appear to have lower occurrence of PCRSV than those without female signatories. PCRSV occurred in 6,5 percent of the cases with female signatories, and in 18,46 percent of the cases without female signatories. It could hence be concluded that PCRSV is 11,96 percent more likely to occur in peace agreements where female signatories are absent, which establishes covariation. It also appears that PCRSV is absent in armed conflicts to a larger extent than it is present.
Table 2. Tabular analysis (two-way frequency table) of PCRSV and female signatories.
Figure 2 below visualises the results from the tabular analysis, which is useful for better understanding the distribution of observations. Jittered scatter plots are not accurate in the placement of each observation, but rather gives an indication of their location. As both variables are binary, the plot somewhat randomizes the observation location in order to avoid overplotting caused by discreteness (ggplot2, nd). Otherwise, there would be only four visible points. Therefore, the x and y axes do not match the scope. Nevertheless, they are a good way of understanding the distribution of values in a dataset.
Figure 2. Jittered scatterplot of observations
As an extra control and to obtain a p-value, the thesis will also include a chi-squared test for independence, which is suitable for two categorical variables. The p-value (p stands for
Female signatories (0) Female signatories (1)
PCRSV (0) 53 61
PCRSV (1) 14 4
probability) is a value ranging from 0-1 which shows the probability of finding the relationship between the IV and DV if there were no actual relationship between them in the larger population. Thus, a low p-value indicates a high confidence interval, though not causality (Kellstedt & Whitten 2018:164, McLeod 2019). The chi-square test returns a p-value of ~0.03. As 0 is the lowest number, this result indicates a high confidence interval.
The value is statistically significant on a 95 percent level, since it is below 0.05. Both tests have hence supported the hypothesis, and the covariation criteria is cleared. However, bivariate tests do not control for confounding variables. Therefore, the main logistic regression will be conducted both bi- and multivariate in order to include variables as well.
5.3 Regression analysis
Results from the logistic regressions are available in Table 3 below, including a bivariate logistic regression (Model 1) and a multivariate logistic regression (Model 2). As the y-axis in a logistic regression is confined to the probability values 0 and 1, it is (contrary to in linear regression), all coefficients are expressed in log odds. The outcome should hence be treated as the outcome of a probabilistic model, which estimates the coefficients for an equation which predicts the probability of either 0 or 1. By taking the log odds of odds ratio, a “logit”
function will create a more linear relationship. Coefficients are therefore expressed in a format showing the model as if they were linear (Starmer 2018). The total number of observations has been reduced to 132, as the DV is lagged and has removed the last of each conflict-year.
