<<

Master’s Thesis in Peace and Conflict Studies VT2017 Department of Peace and Conflict Studies Uppsala University

Fighting Norms: , Masculinity, and the Dynamics of in Conflict

By: Mikael Kataja Supervisor: Lisa Hultman

ACKNOWLEDGEMENTS

All my love to my fantastic friends Alden, Christie, Luci and Rickard (calm down, its in alphabetical order) who provided countless laughs and snacks throughout all this. And to my family for providing support, and sometimes food, when I needed it. But perhaps most of all, thanks to my supervisor Lisa for helping me, guiding me and pushing me to do this. I really couldn’t have done it without you.

2 ABSTRACT 4 1. INTRODUCTION 5 2. PREVIOUS RESEARCH 7 2.1 Gender equality 7 2.2 Conflict intensity 8 2.3 One-sided violence 9 3. THEORY 11 3.1 Definitions 11 3.2 Masculinities 12 3.3 Causal mechanism 14 3.4 Hypotheses 17 4. RESEARCH DESIGN 19 4.1 Unit of analysis 19 4.2 Independent variable 19 4.3 Dependent variables 21 4.4 Confounding- and control variables 21 4.5 Data 22 4.6 Method 23 5. RESULTS 26 6. ANALYSIS 35 6.2 Illustrative cases 38 6.2.1 Janata Vimukthi Peramuna 38 6.2.2 Groupe Islamique Armé 40 6.1 Limitations 42 6.3 Future research 45 7. CONCLUSIONS 45 8. REFERENCES 48

3 ABSTRACT

Gender equality is an emerging field in conflict studies, but so far research has mainly focused on the onset of conflict. However, strong correlations between low levels of gender equality and the risk for conflict has been found, which gives encouragement for further research. Other fields have explored the field of gender equality more extensively, and has for example found relationships between attitudes towards gender equality and the prevalence of violence in personal relationships. Building on those findings, this thesis attempts to advance the field of gender equality in conflict studies by examining how attitudes towards gender equality affects the dynamics of violence during conflicts, with the hypothesised relationship that more positive attitudes towards gender equality results in lower levels of violence. This is done through a statistical study of 172 selected cases from different regions and with different ideologies. The suggested mechanism for this relationship is that of traditional gender norms, and in particular traditional ideals of masculinity which are closely associated with violence. While not lending strong support for the all the hypothesised relationship, the results gives some reason for optimism for future research.

4 1. INTRODUCTION

Gender is becoming a more and more prominent topic of research in the field of peace and conflict studies, and even more so in regards to policy. The United Nations Security Council Resolution 1325 has gained a lot of attention in the past few years and is now a core of the Swedish foreign policy. Gender inequality has been hailed as one of the strongest indicators of probability of conflict within a state, and the role of women in peace processes is high on the agenda in the foreign policy of states and within the United Nations. However, gender is often interpreted as synonymous with women, with focus subsequently almost uniquely being on the impact of conflict on women. While this is an often overlooked perspective in a field dominated almost exclusively by men, and incredibly important to highlight, it misses the true dynamics of the societal norms, expectations, and behavioural pattern that affects men and women differently. By instead focusing on gender as a driver of conflict and violence, this study hopes to contribute to nuancing the image of what gender means for conflict research and policy.

While some biological differences exist between the sexes and might explain a part of the variance in behaviour, most researchers now agree that it is the culturally constructed gender that has the biggest impact on attitudes and behaviour (Tickner 1992; Goldstein 2001). Culturally or socially constructed gender means that we are shaped by the expectations that society and people have on us, and that the differences between men and women therefore comes from societal pressure. A is not violent, tough or ambitious because of his biological sex, and women are not caring by birth, but these ideals are instilled in men and women because of societal expectations on gender. This also means that constructed gender norms can also be reconstructed and changed, which has important policy impact.

Previous research on gender equality and conflict has mainly focused on the onset of conflict, showing that there is a strong correlation between national gender equality and the risk of conflict (Melander 2005; Caprioli 2005). It has also been shown that individuals’ attitudes towards gender equality also affect the general tolerance against other people and the willingness to go to war (Melander & Bjarnegård 2017). The next step from there would be to dive inside the conflict and study how gender equality and attitudes towards gender equality affects the dynamics of the conflict, in particular the dynamics of the violence in the conflict and the use of one-sided violence. The field of one-sided violence has covered variables like strategic choices (Kalyvas 2006), level of democracy (Hultman 2012) and the recruitment process (Weintstein 2007), but so far little

5 attention has been paid to gender and gender equality. This study aims to fill the gaps in both fields by asking the question: how does attitudes towards gender equality affect the dynamics of violence in civil war? A few different theories will be the jumping off points to answer this question. A study from Norway about gender equality and violence in relationships has shown a correlation with more gender equal decision-making in a relationship and a reduction in violence, which would indicate that attitudes towards gender equality has an effect on an individual’s inclination to use violence (Holter 2013). Similarly, a study on political protest organisations in the Middle East found that groups with positive attitudes towards gender equality was more likely to use non-violent methods in their protests, which would indicate that attitudes towards gender equality also has an effect on group level strategic choices (Asal et al, 2013).

The proposed mechanism of how this happens is a change in traditional gender norms, and in particular a change in what is considered to be masculine and not. Traditional gender norms are closely associated with a strict gender hierarchy, where those men who live up to society’s ideals of masculinity are at the top. Women, men who do not live up to those norms, and others who do not fit the ruling gender norms are, and should be according to the gender hierarchy, dominated by those on the top (Goldstein 2001; Bjarnegård and Melander 2011). Traditional masculinity is also closely associated with power and violence, meaning that violence often is an acceptable method to dominate others and impose ones will on others. As a result, domestic violence and homophobia is often widespread in countries with strong traditional gender norms (Theidon 2009).

When applied to the context of rebel groups in conflict, these theories would suggest that in a society with positive attitudes towards, both individual rebels and rebel groups would be less inclined to use violence. However, as all these groups are armed rebel groups involved in active conflicts, they are still expected to use violence at some level. As such the biggest reduction is expected to be found in the use of one-sided violence as this in many ways goes against the norms of gender equality and non-violence. Three hypotheses are formed to help us answer the research question. First, attitudes towards gender equality are expected to reduce the overall violence in the conflict, and thereby also reducing the number of battle related deaths in the conflict. Second, one-sided violence is expected to be reduced as attitudes towards gender equality becomes more positive, based on previously mentioned studies that has shown that positive attitudes towards gender equality reduce propensity for violence both at the individual and at a group level. In sum attitudes towards gender equality is theorised to have an effect on both battle related deaths and one-sided violence, but the the effect may be equal on both dependent variables. One-sided violence is more of an ideological and

6 strategic choice, which is why the effects of attitudes towards gender equality is expected to have a greater on one-sided violence. To illustrate this effect, and also as a way to avoid possible confounders, the ratio between one-sided violence and total dead will be calculated. Our third hypothesis is therefore that a country with more positive attitudes towards gender equality will have a lower ratio of one-sided violence compared to countries with less positive attitudes.

The data used for this study was collected from four different datasets; the Women in Armed Rebellion Dataset, Varieties of Democracy, and two datasets from Uppsala Conflict Database Program on battle related deaths and one-sided violence. The results were generated through regressions on the 172 available observations. In the end, clear support was only found for one of the hypotheses. Battle related deaths was consistently reduced as a result of attitudes towards gender equality, but in relation to one-sided violence the results were not clear enough to draw any substantial conclusions. However, the tendency of the results gives hope for future research into the subject.

2. PREVIOUS RESEARCH

2.1 Gender equality

In his article from 2005, Melander lays out comprehensive evidence that gender equality at a state level reduces the risk of intrastate conflict onset, while looking at both and at a more societal level through the ratio of men and women in higher education. He finds stronger support for constructivist argument that it is not necessarily the biological sex of for example leaders that determines the risk of conflict onset, but rather that gender norms are the determining factor (Melander 2005, 706-707). One of the arguments he presents against that female leaders would be more pacifist than male is that societies dominated by masculine ideals are more likely to elect women that agree with those ideals. The biological sex would matter less in that scenario, and the gendered norms in society more. (Melander 2005, 711).

In a later article, he returns to the often discussed gender gap that indicates that there is a difference between men and women in how conflict is viewed. He convincingly shows that the effect of biological sex is negligible, and that attitudes towards gender equality in both men and women plays a much stronger effect (Melander & Bjarnegård 2017). He specifically looks at how hostile attitudes

7 towards minorities and other nations are affected by one’s attitude towards gender equality, which logically would also affect one’s attitude towards going to war. This also further strengthen the constructivist argument that the biological sex carries little importance in this context and that the socially constructed and the gender norms in society are the important factors. Similarly, a study in Arab countries about different solutions to the Israeli-Palestinian conflict found a strong correlation between those supporting gender equality and those favouring a diplomatic solution to the conflict (Tessler & Warriner 1997, 275).

A great deal has been written about gender equality and violence in personal relationships. A study made in Norway found that couples that were gender equal in their decision-making experienced much less violence in their relationship, as well as much lower risk of violence against children (Holter 2013, 58). The study also found that the level of violence was closely connected to the distribution of power in the relationship. In families where the mother held more power she would also be more violent. However, when the man held most of the power he was also the more violent, and those families were more violent in general than those were the mother was in charge (Holter 2013, 62). This indicates that power is closely tied to gender norms, especially to masculine gender norms, which in turn is also associated with violence.

Some research has also been done on the group level. Asal et al. explored the link between attitudes towards gender equality and different forms of mobilisation, and found that groups that actively supported gender equality were more likely to use non-violent methods of protests compared to groups who did not support gender equality, regardless of other ideological beliefs (Asal et al 2013, 313). The strong correlation between gender equality ideology and non-violent form of protests led the authors to the conclusion that ideologies of gender equality are at least as important as other more established indicators such as ethnicity, race and class when discussing organisational behaviour.

2.2 Conflict intensity

There are a couple of variables that have been theorised to have an effect on the level of violence in the conflict and the conflict intensity. Eck studied the mobilisation along ethnic lines and found that ethnic conflicts indeed are more likely to escalate from an armed conflict to a war compared to other types of conflicts (Eck 2009). There has been a lot of research in to the field of natural resources and its effect on conflict. While there is strong evidence to suggest that an abundance of

8 lootable resources increases the risk for conflict, no clear pattern has been found when it comes to conflict intensity. There have, however, been some findings that suggests that the existence of lootable resources such as oil, gas or diamonds in the conflict zone risk to increase the severity of the conflict (Lujala 2009). Easy access to lootable resources could also change the dynamics of a rebel group, giving them the opportunity to provide more material incentives for potential members but also risk attracting more opportunistic rebels, which have been shown to increase certain types of violence in conflicts (Weinstein 2007). Another variable to consider is the presence of external support to the rebel group. Rebellions has a high start-up cost, and resources is required to compensate leadership enough to be able to initiate a serious recruitment process. A base of support needs to be built as quick as possible, and the group must be armed to a sufficient degree to survive the first uncertain phase of the rebellion. The primary enemy of a rebel group is usually the government, which is almost always stronger, better organised and better equipped, at least when the group is newly organised. By infusing cash, equipment and military expertise, an external actor can greatly help to reduce the start-up cost of a rebel group, even out the strengths between the government and the rebel group, and improve the chances of the group for surviving the initial phases of the conflict (Testerman 2015, 534). The external support has primarily been shown to increase the duration of conflict as the rebel group becomes more evenly matched with the government, and it becomes harder for the government to quickly defeat the rebel group. But it could also have an effect on the intensity and the level of violence in the conflict. The support from an external actor gives the rebel group opportunities to grow bigger and become better equipped than they could have been without support, which all gives them better possibilities to inflict damage on their enemy as they gain a higher capacity for violence. This could then lead to an increase in conflict intensity.

2.3 One-sided violence

There are many different reasons for why groups employ one-sided violence against civilians, which can be simplified and summed up in two categories: group-level strategic factors and individual’s motivations. Group-level strategic factors can be selective punishment as a way to control civilians, for example by a group that is too weak to provide positive incentives to support the group (Wood 2010, 604). It can also be used to reduce the capacity of the other group, for example by undermining logistics and access to resources or attacking civilians to show that the government is weak and unable to protect its citizens, thereby weakening their constituency both politically and physically (Sullivan 2012, 380; Fjelde and Hultman 2014) In addition, it can be used as a way to

9 increase the cost of winning for the other party, especially by rebels that have low chances of winning (Boyle 2009; Hultman 2007; Valentino 2004). It is militarily cheap to attack civilians, but can increase the cost for standing their ground for the party responsible for protecting the civilians, thereby increasing their bargaining power. Similarly, attacking civilians can also be a way for rebels to signal their resolve to the other party. Killing civilians carries a high political cost in many societies, and by doing so they signal that they will not shy away from these types of tactics to achieve their goals (Hultman 2007, 209-10).

But not all violence is a strategic choice by the group. It can also be a more organisational issue, an agent-principal problem where the group cannot exert sufficient control over the action of their members. Rebel organisations that rely on material incentives to their members are less dependent on the civilian population, and therefore don’t have consider the effects on the civilian population in their action. They also tend to attract more opportunistic individuals, compared to groups that rely on social endowments who are more dependent on the support from civilians and often attract more highly committed individuals (Weinstein 2007, 196). Groups that rely on social endowments also tend to build stronger institutions, with ways to collect information and structures to punish behaviour that divert from orders and the strategy of the group. Without access to the same information and less mechanisms of control, groups that rely on material incentives are more likely to display a more inconsistent behaviour. Without the same threat of punishment, opportunistic individuals have more opportunity to attack civilians and loot or rape for their own winning, without sanctions from the group (Weinstein 2007, 205).

As seen here, the area of gender equality is not new to the field of conflict studies. A lot has been written, and it is a growing area of research, but so far it has been limited to mainly study the onset of conflict, or in the case of Asal et al. studying tactics but not extending it to armed conflict. Other fields, such as , psychology and gender studies have explored the connection between gender equality and violence, but then focusing on personal relationships and specific contexts. So while gender has proved to be an important variable in relation to conflict, and other fields of research have suggested that low levels of gender equality or negative attitudes towards gender equality likely increases violence in personal relationship, very little has been written about the effect of gender equality on the levels of violence or the dynamics of violence within conflicts. Similarly, a lot has been written about violence in conflicts, both conflict intensity and one-sided violence, and many different possible explanations has been tested, but again very little has been written about the effects of gender equality and attitudes towards it. The following section will

10 attempt to bridge these two areas of study and lay out a theoretical model of the relationship between attitudes towards gender equality and the violence within a conflict.

3. THEORY

3.1 Definitions

Gender equality can mean a great many different things, from political representation, to equal pay, to sharing household work fairly. In this study the focus is specifically on attitudes towards gender equality and less on the actual level of equality. While there of course is an interplay between norms of equality and actual equality, it is possible for gender equality to exist on paper through for example political representation while society at large still dominated by inequality and (Asal et al 2013, 308). Measuring attitudes towards will therefore bring us closer to actual equality that affects individual’s actions and behaviour, which is the focus of this study. Attitudes towards gender equality gives indication of the norms in society, and what is expected of each gender. If a large part of society supports gender equality, it is reasonable to assume the same people will be more open to alternative gender norms than the traditional of the warrior man and the soft woman. The attitudes towards gender equality will be captured in our independent variable, the ratio between men and women in secondary education, which is measured on a scale from 0 to 2. This will be expanded further in the section for research design.

The term traditional masculinities is frequently used in this text, and requires some attention. The term traditional can sound problematic when studying a broad range of cases from different regions of the world and different cultures. It is true that norms, including norms of masculinity, differ from region to region, country to country or even village to village, but it has been shown that some attributes are common enough in most cultures to generalise. These are broad concepts such as fatherhood, physical strength, acceptance of violence, being a provider, and toughness, that as a concept can be found in most cultures, and even if the application of those concept might differ the general idea is similar (Goldstein 2001, 264).

This study revolves around violence, which is a concept with many possible different meanings. While many forms of violence, such as sexual violence, would be of interest when looking at the effects of attitudes towards gender equality, this study will for methodical reasons that will be

11 expanded on in a later stage only focus on lethal violence.

3.2 Masculinities

When we claim that traditional gender norms lead to more violence, we must also examine what about the traditional gender norms that is leading to violence. While the change in both norms of masculinity and is important for progress, when looking for a reduction in violence we are better off focusing only on masculinity and theories about the social construction of masculinities. One of the most influential theories is the one of , an type of masculinity that dominates society and influences the behaviour of men in said society. It is an important distinction that it is an ideal type and not necessarily found in actual persons. Rather, it is something men has to relate to and either strive to become or reject. Hegemonic masculinity is only one among a plurality of masculinities in society, what defines it is that it is dominant over the others (Connell & Messerschmidt 2005, 832). One common example in western society is the archetypical man who is physically strong, sexually active, rational and unemotional, not afraid to use physical violence, a protector of the family, and a provider (Hooper 1999, 477). While some of these attributes can be seen as positive, such as being a provider, it is important to note that even these types of different expectations on the sexes contributes to the unequal gender norms and ultimately to oppression (Connell & Messerschmidt 2005, 840-841). The expectation that a man should be the provider also implies that his partner should not earn an income and is therefore subordinate to the man and provider. The traditional type of masculinity is in general higher in the hierarchical order, above for example homosexual men, feminine men, women, etc. Local and regional differences exist and the hegemonic masculinity may look different in different places, but the hierarchical order remains.

Often, however, the dominant masculinities follow some similar patterns. The man is often expected to be the provider for his family, as is sexual prowess. Physical strength is often highly regarded and the ability to resort to violence if needed (the warrior) is considered masculine (Esuruku 2011, 26). The warrior ideal exists almost universally across cultures across the globe, closely tied to masculinity and with certain elements commonly present. These include , endurance, strength, and honour. A strong sexuality can sometimes also be included in these elements. These traits are often instilled during rites of initiations for boys to become men, either through tribal rites, military training, or informal socialisation between individuals. The preparation for war is central in this training, and instilling warrior-ideals is often synonymous with becoming

12 a man. In that sense, a man is something that you become by living up to certain ideals, and not something you are born as (Goldstein 2001, 264). Through training boys are learned to distance themselves from everything that is regarded as feminine, thereby cementing the hierarchy between masculinity and femininity. This include the suppression of emotions and learning to deal with both physical and emotional pain. Crying or showing emotion is often deemed feminine and therefore not allowed, limiting their emotional response to violence or simply suppression which could lead to a diminishing mental health (Goldstein 2001, 268–69). The emotional and caring work is left for women, which further entrenches the traditional divisions between the genders (Melander 2005, 698; Goldstein 2001, 269). An integral part is the domination of hierarchically lower status men and women. As a result, homophobia is often rampant in countries with strong hegemonic masculinities, and women’s roles and freedom in society is limited and constrained. These concepts of toughness has proven problematic in post-conflict settings where unemployment is high and former fighters feel emasculated by defeat, leading to high level of and domestic violence (Esuruku 2011, 26).

A similar concept is that of militarised masculinity, a specific type of masculinity that encourage men and boys to act tough to be “real men”. Violence and dominance are inherent in these norms, and are acceptable ways to dominate other groups and opponents, as well as an acceptable way to achieve manhood ((Lwambo 2013, 50). Those who fail to live up these expectations of manhood will be shamed and regarded as less of a man, and thereby lowering their status in society. Less masculine men and women are hierarchically lower than those men who live up to the norms of militarised men, and should therefore be dominated by the militarised men. The structure of domination and hierarchy can also extend outside the gender norms and include all who belong to the out-group such as other ethnic groups or political opponents. This is in many ways very similar to hegemonic masculinity, but is more specific in the role of violence in masculinity and how it is used to oppress. Societies where militarised masculinities are prevalent and accepted will also be more unequal and more violent as a result of this (Bjarnegård & Melander 2013, 7-8; Theidon 2009).

Across most cultures the concepts of masculinity and femininity are very rigid, with little to no overlap. There is of course great diversity within genders, both biological and culturally created gender, but the strict dichotomy of what is masculine and feminine forces that diversity in to constrained ideals with little room to “be different” (Goldstein 2001, 252). Men obviously uphold these ideals by accepting and empowering the gender roles and gender hierarchies, but women also

13 plays a large role in this. There are several recorded instances of women shaming men for not being masculine when being reluctant to fight in a war. One example was a widespread campaign in Britain during World War I where women handed out white feathers to young men who were not serving in combat. A similar campaign took place in Britain during World War II, which prompted the government to issue badges to the able men who were exempt for medical reasons (Goldstein 2001, 272). Countless of similar examples can be found in different cultures all across the globe and across different time periods. This phenomenon is also applicable to other contexts than fighting in wars, just like men can shame each other for not being masculine enough, women can also view some men as “not real men” if they don’t live up to the expected ideals of masculinity.

These types of masculinities have also proven to play a role in recruitment to rebel groups. Men who feel unable to live up societal expectations of masculinity have in some cases been more susceptible to recruitment to rebel groups as well as to commit more violence at home. Joining a rebel movement can be a way to provide a sense of identity or to gain wealth and status that is otherwise hard to get (Wright 2014, 2). This can for example happen as traditional sources of income disappear or unemployment increases. When men no longer are able to live up to expectations of being a provider, an alternative can be to resort to violence in a rebel group as a way to assert their masculinity. An example of this is rural Uganda where the traditional ways to attaining manhood, such as marriage, fatherhood, and being a provider for the family, became more difficult to live up to as a result from displacement and poverty (Wright 2014, 11). As a result, joining the military or a rebel movement has became a more appealing option as a way to both access resources and to live up to the traditional masculine ideals (Wright 2014, 2). While selective incentives such as a salary and status of course is a vital reason behind joining armed movements in these cases, it is likely that masculinity norms are interacting with this.

In the following section ideas will be tied together and formulated as a mechanism of how masculinities can affect the dynamics of violence in a conflict.

3.3 Causal mechanism

Positive attitudes towards gender equality has a documented effect on individual’s acceptance of violence and tolerance towards others, on the use of violent or non-violent tactics by groups, and on the probability of conflict onset at a national level (Holter 2013; Asal et al 2013; Melander 2005). One reason for this is the changing of gender norms. Traditional and militarised masculinities are

14 not compatible with gender equality as they are built on dominance and hierarchy. When attitudes towards gender equality starts to improve, this will also mean that those traditional gender norms are becoming less prominent. As traditional gender norms, and militarised masculinity in particular, is closely associated to norms of violence, the expected pattern is that a society with less traditional gender norms will be less violent overall.

Gender norms have the potential to affect conflict dynamics in two ways. First they will affect individuals view on violence as an acceptable method to achieve goals, the threshold to to use violence will increase as individuals subscribe less to traditional gender norms. This will affect how individuals act within the rebel groups, what tactics and strategies are acceptable and what acts they are willing to perform. If traditional gender norms are less prominent in group leadership as well, this is likely to affect the group dynamic even more. It will also affect the different recruitment tactics available for the group. Playing on and leveraging traditional gender norms as a way to recruit mostly young men has been a successful tactic in several instances. As traditional measurements of manhood become harder to achieve due to poverty or conflict, violence can be an acceptable method to gain income, status, and manhood. When traditional masculinities become less prominent, the traditional avenues to manhood, such as having a wife and starting a family, might become less important as other avenues emerge. Using violence to achieve one’s goal also becomes less acceptable, making armed rebel groups less interesting for those who are trying to assert their masculinity. The types of tactics used for recruitments will affect both the types of individuals that joins, as well as the overall group dynamic of the rebel organisation. Those who are attracted to a rebel group because of the potential opportunity to gain wealth, status and manhood are likely to be more opportunistic than those recruited on more ideological grounds.

The causal mechanism is illustrated in the Figure 1.

15

Individuals

subscribe less to violent masculinities Positive attitudes Less Lower levels of towards traditional violence in gender masculinities conflict equality in society Rebel groups less influenced by violent masculinities

Figure 1. Illustration of causal mechanism

As all the cases in this study are armed rebel organisations involved in some sort of conflict, deaths will happen regardless of ideological tendencies of the group or the society the conflict takes place in. Other, more peaceful, avenues might have been explored previously in the conflict, but all of these countries and rebel groups has decided that armed conflict is now the best way forward. Therefore, regardless of how positive the attitudes towards gender equality are in a country, there will be certain levels of violence present. However, our theory would suggest that attitudes towards gender equality will still reduce the level of violence used in the conflict compared to conflict with less positive attitudes towards gender equality and more traditional gender norms. While there are many other factors that can affect the level of violence and intensity of a conflict, it can still be expected to see a reduction in the amount of battle related deaths in the conflict as a result of an improvement in attitudes towards gender equality and more progressive gender norms.

But considering that all of the cases are involved in an active conflict, we know that the groups we study already are involved in violence which could limit the effect seen on battle related deaths. They have already accepted violence as a method to reach their goals, which is likely to affect the impact of attitudes towards gender equality. The real effect is expected to be found when looking at one-sided violence. While also not uncommonly found on conflicts, one-sided violence is not always an obvious part of a conflict like battle related deaths is. One-sided violence is more of a strategic choice or an effect of the dynamic of the rebel group, which means that the level of one-

16 sided violence used will reveal something about the rebel group. As more positive attitudes towards gender equality is expected to also reduce the acceptance for violence, it can also be expected that rebel groups in countries with more positive attitudes will be more hesitant towards using one- sided violence. While using lethal violence towards your enemy on a battlefield might be accepted, killing civilians and bystanders as a way to increase the bargaining position of the group, send a signal, or for personal gain for opportunistic rebels is less likely to be acceptable.

To capture the different effect on battle related deaths and on one-sided violence, and to avoid using absolute numbers which can be affected by many other variables, the ratio of one-sided violence and total amount of deaths will also be measured. This will show something about how prominent the use of one-sided violence and the nature of the actors in the conflict. As battle related deaths could possibly be high in a country with positive attitudes towards gender equality because of other factors, it is still expected that the ratio of one-sided violence and the total amount of dead will be lower in that country as this study will propose that the biggest effect will be found in the level of one-sided violence used.

3.4 Hypotheses

There are many reasons to believe that more positive attitudes towards gender equality and less traditional gender norms will affects the violence in a conflict. As has been seen in several examples from previous research, gender equality and attitudes towards gender equality has a reducing effect on the inclination to use violence both on an individual level in personal relationships and on group level strategic choices when it comes to forms of protest mobilisations. One possible mechanism for this is the breaking up of traditional gender norms, in particular traditional expectations on masculinity which often is built on violence and domination. Norms of gender equality removes the inherent hierarchies that exists in traditional gender norms, and are built on fairness and acceptance instead of a strict hierarchy where women and men who did not live up to the ideals of masculinity were to be dominated by those who did. As these norms start to change, I expect violence to become less acceptable as a method to enforce one’s will. In the context of conflict, this means that we would expect to see lower levels of violence overall in the conflict. This leads to our first hypothesis:

H1: More positive attitudes towards gender equality will lead to lower levels of battle related deaths in a conflict.

17 Battle related deaths, however, is only one part of the violence committed in a conflict. As previous research has shown, one-sided violence and attacks against civilians is often a part of the strategies used by rebel groups in conflict, for example to improve their bargaining position relative to the government they fight. It can also be a principal-agent problem without strategic values, where opportunistic rebels engage in one-sided violence for their own personal gain, moving the choice from the group level to the individual level. I expect more positive attitudes towards gender equality and lower levels of traditional masculinity to have an effect on both these levels. Firstly, if the norms of gender equality are shared by most members of the group, violence against civilians is expected to be less accepted as a group level strategy to increase the leverage of the group. Secondly, groups that no longer utilise masculinity as a tool for recruitment are expected to attract less opportunistic rebels, thereby reducing the one-sided violence used by individual rebels for personal gains. From this, the second hypothesis is derived:

H2: More positive attitudes towards gender equality will lead lower levels of one-sided violence.

Attitudes towards gender equality is expected to have an effect on both battle related deaths and on one-sided violence. I do not, however, expect the effect to be the same. It is important to remember the unit of analysis at this point, that all the groups that we are studying are armed rebel organisations that are involved in an active conflict. This means that all these groups have chosen to use violence in their struggle to achieve their goals, and that violence in some forms is an acceptable tool in their arsenal. The difference between battle related deaths and one-sided violence is that one is unavoidable part of the concept of armed struggle and civil war, while one particular strategy employed by certain groups that believe it to be a tactically good choice. It is known that battle related deaths will happen as that is a part of an armed conflict, and a prerequisite for these groups to be a part of this study. One-sided violence, on the other hand, is a strategy that while in some cases having some strategic benefits also goes against much of the normative foundation of gender equality. It can therefore be expected that the effect of attitudes towards gender equality to have a greater impact on one-sided violence than on battle related deaths. It is also important to take into account that factors such as the length of the conflict or whether or not it is an ethnic conflict can affect both battle related deaths and one-sided violence, making absolute numbers less interesting to measure. Instead the ratio between battle related deaths and one-sided violence will be measured to understand how big part of a one-sided violence plays in the conflict. This leads to our third and final hypothesis:

18 H3: More positive attitudes towards gender equality will lower the ratio between one-sided violence and the total number of dead in a conflict.

The next section will be dedicated to detail and explain the methodological choices that has been made in this study, and to show how the mentioned hypotheses will be researched.

4. RESEARCH DESIGN

4.1 Unit of analysis

The main unit of analysis is armed rebel groups that are involved in an active conflict. This means that all of the 173 observed rebel groups was involved in an active conflict that reached at least 25 battle related deaths in a year. Out of those 173 groups, 84 of the groups were actively using one- sided violence to a level that reached the threshold of 25 dead per conflict year. This means that there will be a lot of zeroes in dependent variable for one-sided violence, as well as in the dependent variable for the ratio of one-sided violence of the total amount of dead in the conflict. The unit of analysis is also not measured per conflict year but rather as an average over the span of all the active years of the conflict, and all variables are adapted to this fact, for example by calculating the mean value over all active years. While time varying data with individual observations per year would give more observation to work with, it does not provide much different information or information of value. Measuring the yearly change in the independent variable over the span of the conflict will most likely not tell us much, but rather be an effect of the conflict. It would also make cross-case comparison more difficult as we now are measuring change within cases. The interesting variation is between cases and not within them, which is also how we will make our comparison. This makes the mean value over the span of the conflict an appropriate variable to use.

4.2 Independent variable

The main independent variable used in the study is the ratio of men and women in secondary education (labelled Secondary Educ. Ratio in the study). The ratio of men and women in secondary education will act as an indicator of the attitudes towards gender equality in the country. While not a perfect indicator, this will give an indication of how gender norms are viewed in the country, and therefore tell us something about attitudes towards gender equality. A country with a high

19 proportion of women in secondary education will also subscribe less to traditional gender norms where the man is expected to be the sole provider and the woman is expected to stay home. A country with more women in secondary education will be more open to alternative roles for women where they are given more opportunities to control their own lives, a cornerstone in gender equality.

When using education as an indicator, there is a risk that what actually is measured is the governments attitudes towards gender equality and not the peoples’ attitudes towards it. Allowing women to attend school to a similar degree as men could tell us that there is a government policy in place to allow it and not necessarily that it actually is widely accepted by the people. However, it is also reasonable to assume that a high ratio of women in higher education also means that there is some level of acceptance of it. Without acceptance from the people, and in particular from fathers in patriarchal societies, women most likely would not have been allowed to attend higher education regardless of government policy unless it was extremely strictly enforced.

There are other indicators that are commonly to measure attitudes towards gender equality, such as domestic violence and violence in general against women in society. However, these risk being to similar to what we measure in the outcome. If violence is used both as the independent and dependent variable we essentially are using violence to explain violence, which will not yield very interesting results. One benefit of using education as the indicator instead of for example domestic violence is that it tells something about expected gender roles as well as attitudes towards gender equality, while domestic violence mostly gives an indication about attitudes towards women and men’s hierarchical position in society. A high ratio of women in secondary education tells us that women are given, at least theoretically, similar opportunities as men and are most likely allowed to take on different roles than the traditional mother and nurturer.

As a robustness check, an alternative measure for gender equality will also be used to ensure that our independent variable is reliable. The alternative measure is an index of women’s civil rights, measuring among other things freedom of domestic movement, property rights, and access to the judiciary for women (Coppedge et al. 2016).

20 4.3 Dependent variables

The main dependent variables used in the study are the yearly average of battle related deaths in a dyadic conflict (labelled BRD, yearly in the study), the yearly average of deaths through one-sided violence in a conflict (labelled OVS, yearly in the study), and the ratio of one-sided violence of the total amount of dead in the conflict (labelled OVS, ratio in the study). The first two variables, the yearly average for battle related deaths and one-sided violence share the same threshold of 25 deaths per year and actor for a conflict to count as active, anything below that is disregarded (Themnér 2016). The ratio of one-sided violence and total amount of dead was calculated by the author. There are of course many ways to measure conflict violence other than deaths, for example sexual violence, non-lethal physical violence, deaths from starvation caused by the conflict, or unorganised riots. The main appeal for using lethal violence for measuring our dependent variables is the practicality of it. It is by far the easiest to code with little room for misinterpretation, which makes it a good measurement for cross-case comparison.

4.4 Confounding- and control variables

Two confounding variables are included as a way to isolate the effect of gender equality. The first is the level of democracy in the country, measured as to what extent a country has achieved the ideals of a liberal democracy, putting focus on constitutionally protected civil liberties, rule of law and an independent judiciary, among other things (Coppedge et al. 2016). The second is the development level in the country, measured as the natural log of the national per capita GDP (Wood and Thomas 2017). Both of the variables could have an effect on both our independent and our dependent variables.

The control variables are all of the nature that they can be assumed to have an effect on the dependent variable. The variables used are whether forced recruitment has been a part of the recruitment strategies for the rebel group, whether or not women have had active combat roles in the rebel group, if the rebel group is weak relative to their opponent (a binary outcome coded as ‘very weak’) than the state in the Non-state Actor Dataset (NSA)(Cunningham, Gleditsch, and Salehyan 2013), and the duration of the conflict (Wood and Thomas 2017).

21 4.5 Data

Three different datasets were merged to gather enough data to answer the research question. The main independent variable, as well our control variables, was collected from the Women in Armed Rebellion Dataset (WARD). WARD was developed by Reed M. Wood and Jakana L. Thomas to study the variation in participation of women in armed rebel movements. It attempts to understand under what conditions women join rebel movements, and what role the ideology of the group plays in whether or not women are incorporated in the armed forces of the rebel movement. The dataset contains a sample of 211 rebel organisation that were active between 1979 and 2009 and where evidence could be found that women had participated actively. Groups where not selected from a specific geographical area but are spread across all continents. (Wood and Thomas 2017).

Two of the dependent variables were collected from the Uppsala Conflict Database Program, specifically the UCDP Dyadic Dataset that measures battle related deaths in dyadic conflicts (Harbom, Melander, and Wallensteen 2008; Melander, Pettersson, and Themnér 2016) and the UCDP One-sided Violence Dataset v1.4 that measures one-sided violence (Eck and Hultman 2007.; Melander, Pettersson, and Themnér 2016). Both datasets measure the number of dead per conflict year. The dyadic dataset measures specifically within one dyad, so the total amount of dead per year specifically between two parties in a conflict. A conflict can of course be larger than two parties, so there are instances of several dyads within the same conflict. The cut off for both datasets are 25 deaths per year, anything below that does not count as a conflict and is therefore not measured (Themnér 2016). The third dependent variable, the ratio of one-sided violence and total amount of dead, was calculated by the author. The third and final dataset used is Varieties of Democracy (V-dem), a large dataset focused on different measures and indicators of democracy (Coppedge et al. 2016). Both of our control variables was collected from this dataset.

Each dataset contained at least one variable of interest, and to make this study possible the four datasets were merged. As they were developed to study different variables and had slightly different parameters, some data was lost in the process of merging. WARD, for example, focuses on the time-period 1979 to 2009 while UCDP covers 1989 to 2015. This means that we lose all observations from 1979 to 1989 and from 2009 to 2015. WARD is also a sample of conflict dyads, and as such the majority of the observations from UCDP will not be matched and therefore dropped. After the merges, a total of 173 observations are left for one-sided violence, battle related deaths, and for the ratio of one-sided violence of total deaths. Loss of data always problematic, and in this case a total of 38 observations was lost. The majority of observations was dropped because

22 they did not fit the selected time frame. However, tThe dropped observation does not seem to be related which means that the final dataset should not have systemic bias.

One problem with combining different datasets developed for different purposes is that the characteristics does not always match. For example, UCDP measures over time while WARD measures an average over the span of the whole conflict. To deal with that, the UCDP-data was calculated in to the mean value per year of the whole conflict. To test the third hypothesis, a ratio between one-sided violence and total dead during the conflict was calculated. V-dem also has yearly measurements, but only the year of the start of the conflict was used as it was deemed the most significant for the study, the level of democracy or the level of civil liberties would like be affected negatively by the conflict and therefore risk skewing our results.

4.6 Method

As both the dependent variables are count variables, it makes sense for to use a count model to create our results. When looking at the descriptive statistic, we can see that the standard deviation is quite a lot bigger than the mean for both dependent variable, which is an indication of overdispersion. To deal with this a binomial model is used, in this case a negative binomial regression. This also improves our model by weighing the increases in the dependent variables. The step from 25 dead civilians to 100 is expected to be a lot bigger than from 3000 to 3075 in real life, but that does not show up in the data. Using a model that weighs results slightly will help us get more accurate results.

One of the bigger drawbacks of any quantitative method is proving the mechanism. With a large- N study like this we can at best show a correlation between attitudes towards gender equality and lower level of one-sided violence, without being able to prove that the proposed mechanism actually is in play. To deal with this, control variables and confounding variables are included as a way to isolate the effect of the independent variable on the dependent variable. If any potential effects remain after the inclusion of the other variables it has been proved that the proposed mechanism could be true, but it will not be possible to prove without more a more detailed small- N study.

Another problem is the direction of causality. While the expected causal chain can be described and a correlation possibly could be shown, it unfortunately is not possible to decide the causal direction of the variables using this method and this data. It cannot with absolute confidence be

23 said that that change in norms is what is affecting the reduction in violence in civilians, nor can it with absolute confidence be said that it isn’t the conflict itself that is causing a change in norms on society. By using education as indicator for norms, however, the uncertainty of this will hopefully be reduced. A change in norms rarely brings about instant results, and it would most likely take a few years before actual change would be seen in society as a result of that change in norms. This is especially true when it comes to education. In the beginning of the norm change a few progressive individuals and families would perhaps break the tradition and send their daughters to school, and over time as the norms become more established more and more families would dare to follow suit. A country with a high level of women in secondary education therefore most likely saw a change in norms several years before the time of measurement, which would help us establish the temporal order.

When looking through the dataset one outlier was found that potentially could skew the results. It is the case of AFDL (Alliance des forces démocratiques pour la libération du Congo) in the Democratic Republic of Congo (Zaire) between 1996 and 1998, where an estimated 35 000 civilians were killed over a two-year period. While this case may have important implications for our study, there are also some good arguments to exclude the case. One of the main points is that about half of the approximately 40 000 rebels actually were Rwandan soldiers that had ventured across the border. A large part of the victims were also Rwandan refugees that had fled across the border, which prompts the question of whether this was a Rwandan or Congolese conflict, the stated goals of the group was to topple the Congolese government but its actions obviously was influenced by Rwandan politics and Rwandan actors. To be completely sure that our results are not skewed, and based on the complicating background factors, that observation will be excluded from our study. Regressions were made including this observation, but it did not seem to affect the results in a significant way.

Table 1 provides an overview of the main variables used in this study. In total we have 172 observations after dropping the case of AFDL in DRC, meaning 172 different rebel groups who were active in a conflict where either the one-sided violence or battle related deaths reached the threshold level of at least 25 deaths in one year. Battle related deaths are labelled BRD, yearly and one-sided violence is labelled OVS, yearly and ratio respectively. When it come to our alternative measure for gender equality, an index for women’s civil liberties, the available number of observations are 146. A similar number is found for our other variable from the V-dem dataset, the level of democracy.

24

Table 1. Descriptive statistics

As mentioned previously, we can see that the standard deviation is quite a lot larger than the mean for both the battle related deaths and deaths from one-sided violence. This is an indication of overdispersion in the data, and prompts us to use negative binomial regression as our method for our results.

In the following section, the results from the study will be presented.

25 5. RESULTS

To first probe whether there is any relationship between our independent and dependent variables, we plot a graph to visualise our data. This will only be done our two count variable, one-sided violence and battle related deaths. Our main independent variable, the ratio of men and women in secondary education, is split on the median (0,698) in to two different categories – low and high, where ‘Low ratio’ is all the countries with a low ratio (below the median) of women in secondary education and ‘High ratio’ is all the countries with a relatively high ratio (above the median). The mean values of our dependent variables are calculated for both categories. As we can see, there is a clear difference between the two categories. The countries with a high ratio of women in secondary education seem to have a lot fewer deaths as a cause of battle related deaths, which gives us reason to believe that we could find support for our first hypothesis. The same goes for one- sided violence where there is a clear difference between the countries with a high ratio of women in secondary education compared to those who does not, suggesting that our second hypothesis might also be true.

Figure 2. Graph of the distribution of our dependent variables over a dichotomous independent variable.

26 However, the terms low ratio and high ratio are arbitrary in this case. Without knowing the actual distribution for the ratio we know very little about what actually is a high number, and even then we will only know whether it is relatively high compared to other cases within our sample. This is a relatively small sample of countries in conflict, and compared to countries outside of our sample we would perhaps not be able to say that these ratios are high.

Figure 3. Histogram over the distribution of the independent variable.

As we can see in Figure 3, there is a clustering of observations towards the low end of the scale. The variable spans between 0, which would represent a complete domination of men in secondary education, to 2, which would represent a complete domination of women in secondary education. Less than a fifth of the cases reaches at least an even distribution, and less than a tenth has more women than men in secondary education. Very few cases have more women than men in secondary education, and no country reaches a ratio above 1.2. This, however, is not completely unexpected. We know that countries with lower levels of gender equality are more likely to see the onset of conflict, and as all our cases are involved in conflict it is logical that they overall would have relatively low levels of gender equality. This is important to keep in mind as continue our study, while some cases may be relatively high seem to be significantly higher than others the general level in our sample is quite low, which can also be seen in our median value of 0,686. Now that we know

27 a little more about our data, the next step is to properly test all three hypotheses through regressions.

First off we will test our first hypothesis that the amount of battle related deaths in a conflict will be negatively affected and reduced by attitudes towards gender equality in the country. To test this, we will employ four different models. Model 1 is simply our independent and dependent variable, to first give us any indication if there is a relationship between the two. The second model adds two possible confounding variables, the level of democracy in the country as well as the level of development. The third model is the same as model two, but with our alternative measure for gender equality as the independent variable instead of the secondary education ratio. For the fourth model we also add a couple of control variables to the initial measure for gender equality, variables that could affect the amount of battle related deaths; if forced recruitment was used by rebels, if the rebel group was weak compared to their opponent, the duration of the conflict, and finally a dichotomous variable of whether or not the rebel group had a high amount of female fighters in the group.

28

Table 2. Regression of battle related deaths, yearly average

As we can see in the first model, attitudes towards gender equality do seem to have a negative effect on the amount of battle related deaths, which is also significant at the 90% confidence level. In the second model our confounding variables both have a significant effect on the battle related deaths. The negative effect that attitudes towards gender equality had disappeared with the inclusion of confounding variable, and is no longer significant. This would indicate that most of that negative effect actually came from the development level and the level of democracy, both of which are highly significant. It is also worth to note that the number of observation is lower than in model 1. Model 3, in which we introduced an alternative measure for gender equality, gives us a different result than our original measure does. Gender equality is significant at the 99% level, and has a larger negative effect on battle related deaths than our original measure for gender equality had. In

29 model 4 we move back to our original measure. We can see that forced recruitment seems to increase the number of battle related deaths, in accordance of theories of social cohesion and violence (Cohen 2013b), but little else has changed and our independent variable is not significant.

1 Table 3. Size of the effect for battle related deaths

As a robustness test the size of the effects was calculated, which can be shown above. The calculations are based on the fourth model of our regression, and the variables are the same as in that model, although in this we show the variable names instead of variable labels. The top row describes what the data is. The first symbol b is the coefficient, which is the same as in our regression. z and P>| z| is our z and p-values respectively. The symbol % is the change in percent that one step on our independent variable will have the dependent variable. %StdX is the change in percent that one standard deviation on our independent variable will have on our dependent variable, and SdofX is the standard deviation.

When looking at our independent variable, it seems to have a quite small effect on the level of battle related deaths. When moving from a school system completely dominated by men to a

1 Strength of the effect was calculated and generated in Stata with the command listcoef, created by Scott J. Long and Jeremy Freese. For more information, see Long and Freese, 2006.

Explaination for variable names: mean_femse~o is the ratio of women in secondary education, v2x_libdem is our democracy index, lngdppc is the development level. forced_rec~t, veryweak, duration, female_com~h are our control variables of forced recruitment, relative strength of the group, the duration of the conflict, and whether or not there is a high level of women in combat roles.

30 completely equal distribution between men and women, the effect would be a 15 % decrease in battle related deaths, which is not huge effect when looking at what the change in our independent variable actually is. A change of one standard distribution, which is 0,234, would yield a decrease of 3,7 %. However, just as in our regression model, the variable fail to reach significance and any substantial interpretation should be approached with some scepticism. Our confounding variables have stronger effects, especially the index for democracy. A full step on that index would have an effect of 92,3 % reduction in the amount of battle related deaths. One step on that index, however, represent a change from an extreme autocracy with absolutely zero freedoms, liberties and rights to a perfect democracy without flaws, which means that we should not read in too much into that number. The more interesting change comes from the change of one standard deviation, which is 0,188, that would have an effect of a 38,3 % decrease in battle related deaths, which is still a strong effect. It is also interesting to note the strong and significant effect of forced recruitment, 111,1 % increase from one step on the variable and 45,5 % increase from one standard deviation.

To test the second hypothesis, the same independent- , confounding- , and control variables will be used, but the dependent variable will be changed to the yearly average of deaths from one-sided violence to see if attitudes towards gender equality has any effect on it.

31

Table 4. Regression for one-sided violence, yearly average

The results of our second regression give us a little less clear relationship between the independent and dependent variables. Our primary indicator of attitudes towards gender equality is strongly negative in our first model but has a positive direction in model 2 and model 4. It does not reach significance in either of the models. Our confounding variables in model 2 has, as expected, a negative direction, but unlike in the battle related deaths-regression they do not reach significance It is also noteworthy that the variable for democracy changes direction and shows a possible relationship when including the control variables. When including the alternative measure for gender equality with the confounding variable it shows a small positive relation, similar to that of our original measure, but like the other variables it does not reach significance. The fourth model

32 with our control variables tell us little new as neither of the control variables are significant or of a relevant size, but again it is interesting to note the change in the variable for democracy.

Table 5. Size of the effect for one-sided violence

The same robustness test as was made for our first regression is used again here. For an explanation of the variable names and the table, see Table 3. In this model, the attitudes towards gender quality seem to have a very strong positive effect on one-sided violence, with one step on our index increasing the amount of deaths from one-sided violence by 203,2 % and one standard deviation increasing it by 29,7 %. This is in stark contrast of the hypothesised relationship between the variable, but once again it fails to reach significance and should therefore be read with caution. The effect of our confounding variables is slightly smaller on one-sided violence compared to the effect they had on battle related deaths and they both fail to reach significance.

Our third and final hypothesis relates to the ratio of one-sided violence of the total number of dead in the conflict (one-sided violence and battle related deaths). The same models and variables are applied to this dependent variable, just like the previous regressions. However, as this dependent variable is not a count variable and has quite different characteristics as the other two, this one being a ratio of something, we will use a regular OLS regression instead of negative binomial regression for the following regression.

33

Table 6. Regression for one-sided violence, ratio of total dead in conflict

Our third regression fits in the previous pattern of insignificant results. The ratio of women in secondary education has a negative effect of the ratio of one-sided violence in a conflict in the first model, but turns direction and shows a positive relationship when including the confounding variables as well as control variables. The index for democracy shows a consistently positive direction across all models, while the development level shows a consistently negative direction. Additionally, the alternative measure for gender equality is positive but not significant, just like the original measure which indicates that both variables measures a similar dynamic.

The next section will be dedicated to analysing and discussion the results, as well as discussion potential problems with the study and possible alternative explanations.

34 6. ANALYSIS

When it comes to the first hypothesis, that of battle related deaths, some support was found. Both measurements for gender equality had a significant effect on the amount of battle related deaths, although our main indicator lost its significance when including the confounding variables of development level and level of democracy. The alternative measure remained very robust, even with the inclusion of confounding variables. We can therefore say that we have found enough proof to accept our first hypothesis, gender equality does seem to have an effect on the amount of battle related deaths generated in a conflict. For our second and third hypotheses, relating to one- sided violence, the results were more ambiguous. The effect from our main measurement for gender equality was consistently negative is the same relationship that was hypothesised, but the effect disappeared with the inclusion of confounding and control variables. This was also true for our alternative measure for gender equality. Our second hypothesis can therefore be rejected; not enough evidence was found to support the claim that more positive attitudes towards gender equality would reduce the level of one-sided violence in a conflict. The initial effect of our independent variable, however, gives us some optimism that a relationship between the two exists. As the effect disappeared with the inclusion of confounding variables, the relationship between attitudes towards gender equality, level of democracy and development level should be explored before we completely disregard the hypothesis.

So how do these results relate to our proposed mechanism? While this study cannot claim to have found full support for our mechanism, especially when relating it to one-sided violence, it would be too early to disregard even those hypotheses without having examining the actual process closer. From previous research we know that attitudes towards gender equality do have an effect on individuals’ actions, as Holter (2013) showed in the study about violence in relationships. Couples that were more gender equal in their decision-making saw less violence in their relationship while also reducing the risk of violence against any potential children in the relationship. This would suggest that an individual’s inclination to use violence as method to enforce their will is reduced as attitudes towards gender equality becomes more positive. The study, however, focused on a very specific context of two individuals with a close personal relationship. The actions that a person takes in that setting is very different from the context of civil war, and it seems like the clear effect that attitudes towards gender equality had on violence in personal relationships does not translate as clear when it comes to the actions of a rebel in civil war. The same goes for the strategic choices and tactics employed by the groups. As Asal et al. showed in their research in to what tactic that

35 protest movements use, groups supporting gender equality are more likely to employ non-violent tactics than groups that did not, regardless of other ideological beliefs. This would indicate that gender equality indeed does have an effect on the tactics used by groups in protest against government. Once again, however, this pattern does not seem to translate as well when looking at the use of one-sided violence as a tactic in civil war. Because of the methodological choices made for this study it is not possible to draw many conclusions about the proposed mechanism. A relationship fitting our theory was found in hypothesis 1, a relationship that possible could exist in relation to hypothesis 2 as well, but whether this happened through our proposed mechanism or in some other way is not possible to tell with the data available.

One of the biggest differences between these scenarios and concepts borrowed from previous research and the situations and concepts on which they were applied on in this study on is the presence of civil war. A civil war is of course a major event that will have an effect on the mechanisms of society. A phenomenon that has a certain effect on a peaceful society might very well have a different effect or no effect at all in a society wrapped up in a civil war, which could be an explanation for why our theoretical claims seemed did not find proper support when applied on our cases. Hultman (2012) found a similar pattern when researching the effect of democracy on one-sided violence in conflict. Democracy has a well-documented effect on the onset of conflict and there has been research suggesting that terrorism is more prevalent in democracies. She suggests that democracies will see more one-sided violence as they are more vulnerable for coercion through violence, especially against civilians. The social contract that is a democracy is built on the fact that the government can and will protect its citizens, and when a rebel group attacks the civilians of a democracy they expose an inability from the government to uphold their end of the contract. An attack on civilians is an attack on the credibility of the government, which sometimes can be the purpose of such attacks. Moreover, democratic governments have a more limited arsenal of responses at their disposal for dealing with rebels. An authoritarian government can strike with little regard to how the public will react, but a democratic government will have to adapt their response after the norms in society. This is something that can be taken advantage of by rebel groups (Hultman 2012). This pattern can be seen in this study as well. Democracy had a consistently negative effect on battle related deaths, but when looking at both one-sided violence and the ratio of one-sided violence and total deaths it had a small positive effect, which would be in accordance with Hultmans findings.

A similar logic could be applied to the relationship between attitudes towards gender equality and

36 one-sided violence. There is a documented relationship between attitudes towards gender equality and the onset of conflict, but when it comes to one-sided violence it does not seem to have a clear effect, as seen in this paper. This could be because the same incentives for the rebel group to attack civilians are there, and the same constraints on the government exists. In a country where positive attitudes towards gender equality is prevalent and violence is not accepted, there is a strong expectation for the government to protect its citizens as they have put their faith in the governments monopoly on violence. When failing to protect its citizens, similarly to effect on seen on democracies, the government has let down the trust put in them which could be more damaging than battle related losses to their military. Additionally, the norms of gender equality and non- violence puts constrains on the governments ability to retaliate which again could be an incentive for rebel groups to employ one-sided violence.

The effect of attitudes towards gender equality, similarly as the effect of democracy, might therefore be negated by the presence of a civil war. The arguments laid out in the causal story of this study puts forward the idea that more positive attitudes towards gender equality will make it less likely for violence to be committed, essentially saying that the more one believes in gender equality and progressive gender norms the less likely a person is to initiate violence. While this still might be true, and indeed seem to be in relation to battle related deaths, those arguments might actually relate more to the onset of conflict than it does to the violence within conflict. Once a conflict has become active and the rebel group has become involved in the fighting, that threshold of not committing violence might already have been passed. As the old cliché says, “All’s fair in love and war”, and perhaps we are seeing the effect of that. Once the barrier for going to war has passed, the norms of what is acceptable and justifiable behaviour are changed or even disregarded, and a more war-like mentality is adopted. This study has argued that the norms of gender equality is fundamental enough withstand the shock of conflict, and even influence the behaviour within the conflict. For that to happen, however, the norms of gender equality must be deeply ingrained in society and the people, and strong enough to withstand the onset of conflict. How strong the norms have to be is hard to say, and might depend very much on the situation, but it is reasonable to believe that even some of the countries in our study with relatively high ratio of women in secondary education or a high level of women’s civil rights are quite new to the norms of gender equality. Depending on other factors, such as the severity of the conflict and the reasons behind it, the norms might therefore not be ingrained in society enough affect the dynamics of the conflict and might instead be discarded for more violent norms related to war and conflict. Additionally, if holding Melanders conclusions that high levels of gender equality reduces the risk of conflict onset,

37 we should consider the level of gender equality in the cases of this study to be low as conflict started in all of them, or that there is something special about these cases that is causing the onset of conflict. Indeed, as was shown in Figure 3, few countries had an equal distribution of men and women and there was a clear bias of more men. Future research on the topic might benefit from a larger sample including more non-cases and more cases with an objectively high level of gender equality.

Another interesting finding is that a high level of female rebels in combat roles did not have a negative effect on the level of violence, which has been hypothesised some (Card 1996, 12–13; Wood 2009, 141). There is always a risk that this is a measurement issue and that the problem lies more in our variable than in the actual phenomenon, but there is also a chance that it is a real relationship, or lack thereof. This would be another indicator that biological sex is less important and that women aren’t inheritably less violent than men. More research must be made before any conclusions can be drawn from these results, but if it could be shown that more female fighters does not reduce violence then it can also be shown that it is not biological sex but rather outside circumstances like group dynamics, ideologies, strategies and convictions that affect the behaviour of individuals in conflict, in line with arguments made by Cohen (Cohen 2013a).

6.2 Illustrative cases

Analysis of our data as a whole will only be able to tell us so much, to truly understand the mechanisms at work it is necessary to dive into the cases and find out what different variables can potentially affect the outcomes. A proper case study could be designed to get to the core of the mechanisms at play, but for this study a few illustrative cases will be selected to show the complexity of a conflict situation and what other variables could have had an effect on the results of this study.

6.2.1 Janata Vimukthi Peramuna

Our first case is that of Janata Vimukthi Peramuna (JVP), or the People’s Liberation Front, in Sri Lanka from 1989. The JVP was first formed in 1965 as a Marxist revolutionary movement that evolved in to a full-fledged political party in the 1970s. In 1971 JVP was involved in an armed uprising against the government with a base of mostly young people, the unemployed, and the rural south. The uprising gained control of a few regions but was defeated by the government within a few weeks. Up to 30 000 rebels were killed during the uprising according to JVP themselves, while others place the number between 1200 and 10 000 killed. A new rebellion was launched in 1989 in

38 connection to the elections the same year, but at this point JVP had shifted from a Marxist revolutionary message to a more Singhalese-nationalistic rhetoric and the group strongly opposed itself a recent peace accord that according to them gave too much concessions to the Tamils in the north, as well as opposing the presence of Indian peacekeeping troops in the county, both of which resonated strongly with nationalistic sentiments in the country. The second insurrection was more successful than the first and quickly gained support from the population, as well as saw several military victories against the government. One-sided violence was central tactic used by JVD to deter people from voting as well as reducing the base of support from the government, as it mainly was civilians that was deemed to be supporters of the government in any way that was targeted. The violence was less ethnically motivated by mainly focused on political opponents. A wide array of different means was used to kill civilians, including publicly burning opponents, including politicians, journalists and businesses. JVD also purposefully targeted the families of members of the national army as a way to reduce morale and deterring soldiers from fighting against JVD. However, the opposite was achieved as the violence against the families of soldiers sparked a counter-insurgency that eventually completely defeated the JVD (UCDP 2017b; Utrikespolitiska Institutet 2017)

The case of JVD directly contradicts our theory. Sri Lanka during this conflict has the second highest ratio of women in secondary education (1,086), while also having a very high ratio of one- sided violence (0,690). Our other variable for gender equality, the index for women’s civil liberties, was also high at the start of the conflict at 0,697 and was constantly at around 0,7 the decade before the conflict which shows that the positive attitudes towards gender equality existed before the onset of the conflict (Coppedge et al. 2016) This also means that our first measure, the ratio of women in secondary education, is a reliable measurement in this case and that the number is not inflated as a result of the conflict. This illustrates that there are other factors present in this case that might have a greater influence on one-sided violence than the attitudes towards gender equality. When it comes to conflict it Sri Lanka it is hard and not wise to not consider its history of ethnic tensions when researching possible impacting variables on dynamics of other conflicts. One of biggest factor for the resurrection of the conflict with JVP was the ongoing peace negotiations with the Tamil minority in the north, which JVP and much of the Sinhalese population considered to give too much authority to the Tamils. So while the roots of JVP was Marxist revolutionary, and its goal was still state capture and the installation of a socialist government, it had in large also transformed in to a nationalist movement. Research has found that leftist groups are slightly more likely to employ violent means of protest, arguing that because the ideology calls for an overthrow of

39 existing political structures they are more likely to use terrorism as a tactic (Asal et al 2013, 308). While not mentioned specifically in that research, an increase in terrorism as a tactic would also increase the amount of dead from one-sided violence in the conflict, which could be one contributing reason for the high number we find in this conflict. While the one-sided violence was not specifically targeting based on ethnicity, the JVP movement was strongly nationalistic in their rhetoric and goals. There has been some research suggesting that as an ideology often strives to affirm and strengthen traditional gender structures (Wood & Thomas 2016, 36), which could suggest that while Sri Lanka as a country displayed positive attitudes towards gender equality, JVD as a movement might not be as supportive of gender equality and might even try to reverse the steps that had been taken towards gender equality. Without a proper case study, however, this line of reasoning will not be much more than speculation, but it does illustrate that there is a local context to consider as well as many other important variables in addition to attitudes towards gender equality that should be considered to truly understand the outcome from our study.

6.2.2 Groupe Islamique Armé

Our second case comes from Algeria. After gaining its independence from France in 1962, power was giving to the political party FLN by the former colonial power. FLN was the only party allowed and Algeria became a one-party state. In the 1980s discontent grew in the country and more and more voices was raised in favour of turning Algeria into an Islamic state. The discontent grew into demonstrations and riots, which ultimately forced the government make democratic reforms and allow more political parties to exist and compete for state power. One of the parties that was created was FIS (Front Islamique du Salut, Islamic Salvation Front). The party quickly gained popularity and was the clear winner of the first round in the national elections of 1991. However, before the second round could be held a military coup was launched and successfully claimed the state power. The election was cancelled, the parliament was dissolved, and FIS was banned and disbanded. Protests against the military government emerged and quickly escalated, leading to the formation of several armed rebel groups that challenged the government. One of the groups formed was GIA (Groupe Islamique Armé, or Armed Islamic Group). The GIA begun their armed struggle in 1993 with the aim to overthrow the government, buy progressively became more focused on installing a strong Islamic regime at all levels of society as well as a world-wide theocracy. The group received support from Iran and Sudan at different time periods, and the Algerian government also accused Morocco, Libya, Syria and Sudan for harbouring members of the group. Many of the fighters of GIA had previously fought against the Soviet occupation in , and Al-Qaida had been involved in arming and training the rebels. The government

40 was also supported by outside states such as Tunisia and Egypt, and later the United States of America after September 11, 2001. As such, this conflict was quite internationalised. The dynamics of the conflict changed over time from being mainly regular combat between armed opponents to including more attacks against civilians, including large-scale massacres. This came after the GIA issued a fatwa against the Algerian people, accusing the population as a whole of abandoning Islam. The activity of the group died down over time as supporters and leaders of the group was arrested, and the Algerian government claimed to have arrested the last leader of GIA in 2006 (UCDP 2017a; Utrikespolitiska Institutet 2017).

The dynamic of this conflict is more in line with our theorised relationship between attitudes towards gender equality and the violence in a conflict. Algeria during the time of conflict had a fairly high ratio of women in secondary education at 0.956 and the ratio of one-sided violence and total amount of dead was 0,165, which is in agreement with our theory one-sided violence would have a less prominent role in more gender equal countries. However, during the course of the conflict over 1500 civilians was killed through one-sided violence which still is a substantial amount. A clear change of tactics can be seen after a few years when the group shifted from almost exclusively attacking military targets to killing civilians at a large scale. This could possibly be attributed to a strategic choice to change tactics when the initial strategy of regular warfare did not achieve the desired results against a superior enemy, but it could also be attributed to radicalisation of the group that happened over time. The group began as a more moderate Islamic movement which focused mainly on overthrowing an unlawful government and the creation of an Islamic state through state capture, but expanded their goals as the conflict continued and became more extreme in their ideology. This could be a result of the outside support from other states, as well as the influx of foreign fighters who could have been radicalised abroad before returning to Algeria. The clear turning point in the radicalisation and the shift of tactic came when the group changed their target from what they perceived to be an unlawful government to a warfare against the Algerian people for what they called their abandonment of Islam. Researchers has suggested that religious movements are more likely to use violent tactics in their struggle, disregarding the consequences of their actions and embracing a ‘cosmic war’-mentality where they fight for the good and the just, and the and ungodly enemies must be defeated at all costs (Juergensmeyer 2003). This seem to be similar to the perspective that was adopted by the GIA over the course of the conflict, and could be one of the explanations of the increase in one-sided violence. The external support could have contributed to both the radicalisation of the group as well as an overall increase of violence, which could in turn have affect the level of one-sided violence used (Testerman 2015).

41 Regardless of this, there is a correlation between the level of gender equality in the country and the level of one-sided violence used, which gives some support to our theory.

Two cases have been presented, one giving support to our hypotheses while the other did not. The case of JVP in Sri Lanka, while showing positive attitudes towards gender equality, also saw high levels of one-sided violence. This shows that the relationship between attitudes towards gender equality and one-sided violence is more complicated than a straight relationship, and that there are many other variables that could affect the outcome. Further specifying under which condition attitudes towards gender equality has an effect and when it does would be a good topic for future research. The second case, GIA in Algeria, showed that attitudes towards gender equality can be correlated to reduced levels of one-sided violence. The high ratio of women in secondary education corresponded with a relatively low ratio of one-sided violence, but the mechanisms of this relationship is not clear with the data available for this study. This, in addition to the increase of one-sided violence as the group became more ideologically entrenched gives reason to further explore and specify the mechanisms of gender equality and the dynamics of violence.

6.1 Limitations

The alternative measure for gender equality, the index of women’s civil liberties sometimes generated slightly different results from our main measurement at times. This could suggest that there is a possibility that we did not correctly measure what we were looking for in attitudes towards gender equality, creating a reliability problem. Capturing attitudes through a very concrete indicator that are a couple of theoretical steps away from the actual attitudes is always hard, and we risk measuring something else than the actual attitudes or other things in addition to the attitudes. While not feasible in this case, a more specific study that asked directly about actual attitudes would have been preferable. However, the difference between the two measures was not of such a size to cause major concern. The biggest difference between the variables was found the regression for battle related deaths, on model 2 and model 3 where the confounding variables caused our original measure to turn positive and insignificant, while our alternative measure was significantly negative. Apart from this, they seem to capture approximately the same relationship as the direction of the coefficients mostly was in the same direction, and only minor differences was found in the size of the coefficients, which is to be expected when using two different variables.

A potentially bigger problem is the temporal order. As mentioned in previous sections there is a

42 risk that causal direction is the opposite and that it is the conflict itself that is affecting the norms in society. It is well documented that women often gain status and are empowered during conflicts as they have to fill roles in society that otherwise are reserved for men, something that could affect for example the ratio of women in secondary education while the conflict is ongoing (Demeritt et al 2014, Hughes & Tripp 2015). While it is possible that it follows our proposed logic and that norms in society change first which then affects the rate of women in secondary education, it is also true that this chain of events is more something we would expect in a conflict-free society. For our assumption to be true, these changes would have had to happen before the onset of the conflict, to ensure that the pattern of more women in secondary education is not a result of the conflict. Without the control variable of ratio before the conflict, which the data at hand did not allow for, there will an uncertainty of whether or not this simply is because of a lack of men in higher education rather than a purposeful increase of women present, as a result of a large amount of young men being actively involved in the conflict.

To control for this possible issue, our alternative measure for gender equality was included. The index for women’s civil liberties is measured in both actual legislation and how that legislation translates in to actual freedoms for women in their everyday lives. This reduces the problem of wondering whether this is a phenomenon created by the conflict. An actual legislation reflects an institutionalisation of a norm that in some form is present in society. For it to become institutionalised it needs to pass through several layers of bureaucracy, be agreed upon by decision makers and be formulated by legal expert, a process that often takes considerable time regardless of what type of government is in place. This means that it is more likely to be a deliberate decision by a group of decision-makers, and not a consequence of external factors like the presence of a conflict. It also means that it is likely that the process started or was implemented before the conflict started as a government involved in an armed conflict often won’t prioritise legislating women’s rights. The results from the different measures for gender equality generated similar results which would tell us that the temporal order is not a major problem in this study.

Another issue with the design of the study could potentially be the levels of analysis. The effects of attitudes towards gender equality we are looking for are found on the group- and individual levels, but the level the level that this is measuring attitudes towards gender equality on is at a national level. There is therefore not a direct link between our independent variable and the effect we are looking for in the dependent variable. There are a couple of theoretical steps between those two that could diffuse or distort the effect that attitudes towards gender equality might have.

43 Optimally we would measure the attitudes towards gender equality at the group or individual levels within the rebel groups, thereby closing the gap between cause and effect and reducing the risk of distortion from other variables. By measuring at the national level we also make the assumption that all, or most, citizens in society share the same attitudes towards gender equality. The reality, however, is that there is a wide spectrum of different opinions in society and that we at best get an average of all those different opinions. The problem with that is that those who are active in rebel movements may differ in opinions, norms, and behaviour from the average citizens. There is a quite high threshold that an individual has to cross before joining a rebellion, and there is most likely a strong will to change the status quo. The average citizen will not uproot his or her whole life to join an armed struggle against the government, which could tell us that the average rebel might differ from the average citizen. This does not necessarily mean that they would differ when it comes to attitudes towards gender equality, but it does mean that we might not be able to equate the set of norms and attitudes shared by society as a whole with the set of norms and attitudes that exists in rebel organisations.

An argument could be made for using for example sexual violence instead of lethal violence, based on the fact that sexual violence is more theoretically connected to gender equality and gender norms than lethal violence is. There are many other reasons for using lethal violence that is not connected to gender equality, while gender equality plays a very large part of the mechanisms behind sexual sexual violence. The downside, however, lies in the practicality. The concept of sexual violence is much vaguer and can include a lot more different things than lethal violence, making coding the variable harder. With more room for interpretation for what exactly the variable means the risk increases that the study will not be as precise in what exactly is compared, especially when making a large-N study. It runs the risk that what is reported as sexual violence in one country will not be reported as the same in another country, making comparison between cases more difficult. But perhaps a bigger problem is the stigma that usually follows sexual violence. It is a common theme that a large number of cases of sexual violence is not reported or underreported, to a degree that can differ from country to country and culture to culture, making cross-case comparison difficult. This is also the reason that lethal violence is used instead. While it might be a slightly less theoretically relevant measure, there is little stigma involved and almost no room for different interpretations, meaning that data is readily available for a large number of cases and thereby making it a good measure for a large-N comparison. With the availability of good data, however, this could be a very interesting area for future research to explore.

44 6.3 Future research

This study has given indication that there is a relationship between attitudes towards gender equality and the dynamics of violence in conflict. To further explore this relationship, future research would do well to test this at a group or individual level. As seen by the results of this study, using indicators of gender equality at a national level carries the risk of generating to general results to study a very specific outcome. Changing the level of analysis to the group level would better capture the actual attitudes towards gender equality within the group and not relying on the assumption that national tendencies are transferred to the rebel group as well. It would also provide a picture of how the dynamics of gender norms and masculinity works within the rebel group, and help us to better understand what effect it has on the groups behaviour. This could be done both as a large-N statistical study as well as a comparative case study. By aggregating data taken from many different rebel organisations we can better show the size of the relationship, and by further specifying confounding and control variables the effect can be pinpointed better. But to truly understand the mechanism and in-depth case study should be made. The case study allows for a view of the mechanism from the inside and provide a better understanding what does have an effect, as well as when and under what conditions. Without the insights from a case study, we are left to make assumptions and create hypotheticals. There is in large still a relatively unexplored area of conflict research, and it is my strong belief that further research into this area could drastically change and improve our understanding of the dynamics of violence in conflicts. Finally, it would be of interest to apply the theories and mechanisms used there to sexual violence as well, as that concept is theoretically closer to attitudes towards gender equality.

7. CONCLUSIONS

This article has focused on the relationship between attitudes towards gender equality and the dynamics of violence in a conflict, two fields has been studied extensively as separate phenomena but never in a systematic way in relation to each other.

Three hypotheses were developed to understand the relationship between attitudes towards gender equality and the dynamics of violence in conflict. First, attitudes towards gender equality was theorised to have a reducing effect on violence in the conflict overall, which was measured in the amount of battle related deaths in the conflict. This hypothesis found some support, gender

45 equality seemed to reduce the amount of battle related deaths which would suggest that more gender equal countries would experience less violent conflict. This was especially true for our alternative measure for gender equality that remained significant after the inclusion of confounding variables. Because of the necessity of battle related deaths in an armed conflict, and because of the stigma of killing civilians, attitudes towards gender equality was theorised to have a greater effect on one-sided violence than on battle related deaths. The second hypothesis was therefore that one- sided violence would be reduced as attitudes towards gender equality became more positive. While some indications of this relationship was found, it was not significant enough to draw conclusions and the effect found disappeared with the inclusion of confounding variables. Not enough support was found to confirm this hypothesis. The expectation was that gender equality would have an effect on both battle related deaths and one-sided violence, but that the effect would be stronger on one-sided violence. To capture this effect, the ratio of one-sided violence and total amount of deaths was calculated with the expectations that the ratio would decrease as attitudes towards gender equality became more positive. This was our third hypothesis. Similar to the second hypothesis, however, not enough evidence was found to support this. In spite of the lack of evidence for the hypotheses about one-sided violence, some optimism about the relationship still remains.

Two possible issues related to the design of this study was presented, one issue of reliability and one issue of temporal order. Our main independent variable aimed to capture attitudes towards gender equality through the ratio of men and women in secondary education, which could be problematic if you are not sure that our indicator is capturing the desired phenomena. To control for this, an alternative measure for gender equality was included. These measures showed a similar relationship in most cases, which would tell us that our original measure for attitudes towards gender equality is reliable. The second potential problem related to the temporal order. As conflict itself often has an effect on women’s status in society, and our data did not allow us to control for the ratio before the onset of conflict, there was a risk that a high ratio of women in secondary education was a result of the conflict itself. This problem was also controlled for by the inclusion of our alternative measure. As that measure included a much more institutionalised form of gender equality, which is less vulnerable to the effect of conflict, as similar result in both measures meant that the temporal order likely was in the expected direction.

The aim of this study was to fill the gaps in both the area of gender equality in conflict studies as well as in the field of one-sided violence. The study was successful in showing that there is a

46 relationship between the suggested variables and thereby has taken a step to fill the gap, but a lot more research remains before we can consider the gap to be filled. Two possible direction exists for future research. One is to expand the large-N study to include more observations as well as specify the variables further to more clearly illustrate the relationship. The second is to make an in- depth case study to find out the dynamics of how the variables are related and affect each other, as well as being able to better illustrate the mechanism.

In conclusion, this thesis has shown that a relationship between the attitudes towards gender equality and the dynamics of violence in a conflict do exist. However, the relationship remains largely unexplored and more research is needed to fully understand it. My hope is that this thesis has been able to convey the potential of this area of research, as it is my strong belief that more research into this could greatly enhance our understanding of the dynamics of violence in conflict.

47

8. REFERENCES

Asal, Victor, Richard Legault, Ora Szekely, and Jonathan Wilkenfeld. “Gender Ideologies and Forms of Contentious Mobilization in the Middle East.” Journal of Peace Research 50, no. 3 (2013): 305–18. Bjarnegård, Elin, and Erik Melander. “Disentangling Gender, Peace and Democratization: The Negative Effects of Militarized Masculinity.” Journal of Gender Studies 20, no. 2 (June 2011): 139– 54. doi:10.1080/09589236.2011.565194. ———. “Pacific Men: How the Feminist Gap Explains Hostility.” The Pacific Review, January 18, 2017, 1–16. doi:10.1080/09512748.2016.1264456. Boyle, Michael J. “Bargaining, Fear, and Denial: Explaining Violence Against Civilians in Iraq 2004– 2007.” Terrorism and Political Violence 21, no. 2 (March 31, 2009): 261–87. doi:10.1080/09546550902765565. Caprioli, Mary. “Primed for Violence: The Role of Gender Inequality in Predicting Internal Conflict.” International Studies Quarterly 49, no. 2 (2005): 161–178. Card, Claudia. “Rape as a Weapon of War.” Hypatia 11, no. 4 (1996): 5–18. Center for History and New Media. “Zotero Quick Start Guide,” n.d. http://zotero.org/support/quick_start_guide. Cohen, Dara Kay. “Explaining Rape during Civil War: Cross-National Evidence (1980–2009).” American Political Science Review 107, no. 3 (August 2013): 461–77. doi:10.1017/S0003055413000221. ———. “Female Combatants and the Perpetration of Violence: Wartime Rape in the Sierra Leone Civil War.” World Politics 65, no. 3 (July 2013): 383–415. doi:10.1017/S0043887113000105. Connell, R. W., and James W. Messerschmidt. “Hegemonic Masculinity: Rethinking the Concept.” Gender & Society 19, no. 6 (December 2005): 829–59. doi:10.1177/0891243205278639. Coppedge, Michael, John Gerring, Staffan I. Lindberg, Svend-Erik Skaaning, and Teorell, Jan. V-Dem Codebook v6. Varieties of Democracy (V-dem) Project, 2016. Cunningham, David E., Kristian Skrede Gleditsch, and Idean Salehyan. “Non-State Actors in Civil Wars: A New Dataset.” Conflict Management and Peace Science 30, no. 5 (November 2013): 516–31. doi:10.1177/0738894213499673. Demeritt, Jacqueline H.R., Angela D. Nichols, and Eliza G. Kelly. “Female Participation and Civil War Relapse.” Civil Wars 16, no. 3 (July 3, 2014): 346–68. doi:10.1080/13698249.2014.966427.

48 DeRouen, Jr, Karl. “The Duration and Recurrence of Civil War.” In Routledge Handbook of Civil Wars, 300–310. New York: Routledge, 2014. Eck, Kristine. “From Armed Conflict to War: Ethnic Mobilization and Conflict Intensification.” International Studies Quarterly 53, no. 2 (2009): 369–388. Eck, Kristine, and Lisa Hultman. “One-Sided Violence Aganist Civilians in War: Insights from New Fatality Data.” Journal of Peace Research 44, no. 2 (2007): 233–46. Enloe, Cynthia H. Maneuvers: The International Politics of Militarizing Women’s Lives. Berkeley: University of California Press, 2000. Esuruku, Robert Senath. “Beyond Masculinity: Gender, Conflict and Post-Conflict Reconstruction in Northern Uganda.” Journal of Science & Sustainable Development 4, no. 25 (2011): 25–40. Fjelde, Hanne, and Lisa Hultman. “Weakening the Enemy: A Disaggregated Study of Violence against Civilians in Africa.” Journal of Conflict Resolution 58, no. 7 (October 2014): 1230–57. doi:10.1177/0022002713492648. Goldstein, Joshua S. War and Gender: How Gender Shapes the War System and Vice Versa. Cambridge: Cambridge University Press, 2001. Harbom, Lotta, Erik Melander, and Peter Wallensteen. “Dyadic Dimensions of Armed Conflict, 1946- 2007.” Journal of Peace Research 45, no. 5 (2008): 697–710. Hicks, Madelyn Hsiao-Rei, Uih Ran Lee, Ralph Sundberg, and Michael Spagat. “Global Comparison of Warring Groups in 2002–2007: Fatalities from Targeting Civilians vs. Fighting Battles.” Edited by Tom Denson. PLoS ONE 6, no. 9 (September 6, 2011): e23976. doi:10.1371/journal.pone.0023976. Holter, Øystein Gullvåg. “Masculinities, Gender Equality and Violence.” Masculinities and Social Change 2, no. 1 (2013): 51–81. Hooper, Charlotte. “Masculinities, IR and the ‘gender Variable’: A Cost [Hyphen] Benefit Analysis for (Sympathetic) Gender Sceptics.” Review of International Studies 25, no. 3 (1999): 475–491. Hughes, M. M., and A. M. Tripp. “Civil War and Trajectories of Change in Women’s Political Representation in Africa, 1985-2010.” Social Forces 93, no. 4 (June 1, 2015): 1513–40. doi:10.1093/sf/sov003. Hultman, Lisa. “Attacks on Civilians in Civil War: Targeting the Achilles Heel of Democratic Governments.” International Interactions 38, no. 2 (April 2012): 164–81. doi:10.1080/03050629.2012.657602. ———. “Battle Losses and Rebel Violence: Raising the Costs for Fighting.” Terrorism and Political Violence 19, no. 2 (April 6, 2007): 205–22. doi:10.1080/09546550701246866.

49 ———. “Violence against Civilians.” In Routledge Handbook of Civil Wars, 289–99. New York: Routledge, 2014. Juergensmeyer, Mark. Terror in the Mind of God: The Global Rise of Religious Violence. Berkeley: University of California Press, 2003. Kalyvas, Stathis N. The Logic of Violence in Civil War. Cambridge: Cambridge University Press, n.d. Long, J. Scott, and Jeremy Freese. Regression Models for Categorical Dependent Variables Using Stata. 2nd ed. College Station: Stata Press, 2006. Lujala, Paivi. “Deadly Combat over Natural Resources: Gems, Petroleum, Drugs, and the Severity of Armed Civil Conflict.” Journal of Conflict Resolution 53, no. 1 (2009): 50–71. Lwambo, Desiree. “‘Before the War, I Was a Man’: Men and Masculinities in the Eastern Democratic Republic of Congo.” Gender & Development 21, no. 1 (March 2013): 47–66. doi:10.1080/13552074.2013.769771. Melander, Erik. “Gender Equality and Intrastate Armed Conflict.” International Studies Quarterly 49, no. 4 (2005): 695–714. Melander, Erik, Therése Pettersson, and Lotta Themnér. “Organized Violence, 1989–2015.” Journal of Peace Research 53, no. 5 (September 2016): 727–42. doi:10.1177/0022343316663032. Ottmann, Martin. “Rebel Constituencies and Rebel Violence against Civilians in Civil Conflicts.” Conflict Management and Peace Science 34, no. 1 (January 2017): 27–51. doi:10.1177/0738894215570428. Reiter, D. “The Positivist Study of Gender and International Relations.” Journal of Conflict Resolution 59, no. 7 (October 1, 2015): 1301–26. doi:10.1177/0022002714560351. Sullivan, Christopher Michael. “Blood in the Village: A Local-Level Investigation of State Massacres.” Conflict Management and Peace Science 29, no. 4 (2012): 373–396. Tessler, Mark, and Ina Warriner. “Gender, Feminism, and Attitudes toward International Conflict: Exploring Relationships with Survey Data from the Middle East.” World Politics 49, no. 2 (1997): 250–81. Testerman, Matthew. “Removing the Crutch: External Support and the Dynamics of Armed Conflict.” Studies in Conflict & Terrorism 38, no. 7 (July 3, 2015): 529–42. doi:10.1080/1057610X.2015.1016312. Theidon, Kimberly. “Reconstructing Masculinities: The Disarmament, Demobilization, and Reintegration of Former Combatants in Colombia.” Human Rights Quarterly 31, no. 1 (2009): 1– 34. Themnér, Lotta. UCDP Dyadic Dataset, Version 1-2016. Uppsala: Uppsala Conflict Database Program (UCDP) Department of Peace and Conflict Research, Uppsala University, 2016.

50 Tickner, J. Ann. Gender in International Relations: Feminist Perspectives on Achieving Global Security. New York: Columbia University Press, 1992. UCDP. “GIA.” Database. UCDP Uppsala Conflict Database Program. Accessed May 14, 2017. http://www.ucdp.uu.se/#actor/538. ———. “JVP.” Database. UCDP Uppsala Conflict Database Program. Accessed May 15, 2017. http://www.ucdp.uu.se/#/actor/281. Utrikespolitiska Institutet. “Landguiden: Algeriet, Modern Historia.” Algeriet, Modern Historia. Accessed May 14, 2017. https://www.ui.se/landguiden/lander-och-omraden/afrika/algeriet/modern- historia/. ———. “Landguiden: Sri Lanka, Modern Historia.” Sri Lanka, Modern Historia. Accessed May 15, 2017. https://www.ui.se/landguiden/lander-och-omraden/asien/sri-lanka/modern-historia/. Valentino, Benjamin A. Final Solutions: Mass Killing and Genocide in the Twentieth Century. Ithaca: Cornell University Press, 2004. ———. “Why We Kill: The Political Science of Political Violence against Civilians.” Annual Review of Political Science 17, no. 1 (May 11, 2014): 89–103. doi:10.1146/annurev-polisci-082112-141937. Warren, Karen J., and Duane L. Cady. “Feminism and Peace: Seeing Connections.” Hypatia 9, no. 2 (1994): 4–20. Weinstein, Jeremy M. Inside Rebellion: The Politics of Insurgent Violence. Cambridge: Cambridge University Press, 2007. Wood, E. J. “Armed Groups and Sexual Violence: When Is Wartime Rape Rare?” Politics & Society 37, no. 1 (March 1, 2009): 131–61. doi:10.1177/0032329208329755. Wood, R. M. “Rebel Capability and Strategic Violence against Civilians.” Journal of Peace Research 47, no. 5 (September 1, 2010): 601–14. doi:10.1177/0022343310376473. Wood, Reed M, and Jakana L Thomas. “Women on the Frontline: Rebel Group Ideology and Women’s Participation in Violent Rebellion.” Journal of Peace Research 54, no. 1 (January 2017): 31– 46. doi:10.1177/0022343316675025. Wright, Hannah. “Masculinities, Conflict and Peacebuilding: Perspectives on Men through a Gender Lens.” Saferworld, 2014.

51