Gender Equality - the Key to Conflict Resolution?
A Quantitative Study of How Gender Equality in Rebel Groups Affect the Likelihood of Peace
Alma Estrada Metell
Uppsala University
Department of Peace and Conflict Research Bachelor´s thesis
Fall semester, 2019
Table of Content
1. Introduction ... 4
2. Theoretical Framework ... 6
3.1. Terminology and Concepts ... 6
3.2. Previous Research and Gap ... 8
3.3. The Causal Mechanisms ... 12
4. Research design ... 15
4.1. Choice of Method ... 15
4.2. Dataset ... 15
4.3. Operationalization: Independent Variable ... 17
4.3.1. Validity and Reliability: Independent Variable ... 19
4.4. Operationalization: Dependent Variable ... 20
4.4.1. Validity and Reliability: Dependent Variable ... 20
4.5. Operationalization: Control Variables ... 21
4.5.1. Validity and Reliability: Control Variables ... 24
4.6. Model Specification ... 24
5. Results and Analysis ... 26
5.1. Descriptive Analysis ... 26
5.2. Logistic Regression Analysis ... 28
6. Discussion ... 34
6.1. Limitations and Alternative Explanations ... 36
7. Conclusion ... 38
8. Bibliography ... 39
9. Appendix ... 45
9.1. Robustness Tests ... 45
9.2. Bivariate Regression Tests: Control Variables ... 46
Abstract
Despite an increasing interest in questions regarding gender and peace, little is still known about the effects of gender equality on the outcome of armed conflicts. In this study, I explore
how gender equality norms within rebel groups affect the likelihood that armed conflicts are terminated through negotiated settlements. My main claim is that rebel groups with strong
gender equality norms are more likely to reject the use of violence and pursue peaceful conflict resolution. This claim is tested in a quantitative study incorporating data from the WARD and UCDP Conflict Termination datasets, demonstrating gender distribution within rebel groups and the manner in which conflicts end. Gender equality norms are measured as both female combatants’ prevalence and female leadership prevalence. Based on the findings
from a logistic regression analysis, it is demonstrated that higher female combatants’
prevalence in rebel groups increases the likelihood of armed conflicts ending through negotiated settlements.
Key words:
Gender equality, female combatants, female leaders, armed conflict, rebel groups, negotiated
settlement, peace agreements
1. Introduction
It has been said that “you can no more win a war than you can win an earthquake” (Legacy, 2000). Indeed, even if a war is won, it has consequences of such a magnitude that it is
reasonable to ask if it is ever worth the struggle. Considering the high cost of war, it is logical to assume that rational actors should prefer negotiations as a mean to solve conflict, rather than engaging in the hazard of war (Fearon, 1995: 380). However, despite the obvious benefits of peace, few actually pursue and succeed with conflict resolution (Kreutz, 2010;
Downes, 2004). I argue that a part of this variation can be explained due to differences in gender equality norms. Stronger gender equality norms should transmit tolerant attitudes and reluctance for the unnecessary use of violence (Bjarnegård, Melander, 2017; Caprioli, 2005;
Wood, Ramirez, 2018). Rebel groups with strong gender equality norms should therefore be more willing to negotiate, and further do this with greater efficiency. To test this claim, I will investigate the following research question; How does gender equality in rebel groups affect the likelihood that an armed conflict is terminated through a negotiated settlement?
Out of all armed conflicts between 1990-2005, only one out of five ended through negotiated settlements (Kreutz, 2010: 246). This creates an intriguing puzzle for researchers as well as for policy makers. For researcher within the field of peace and conflict, it is crucial to understand why and how armed conflicts end. This study contributes to the research field by advancing the knowledge regarding the likelihood of negotiated settlements. Theoretically, a novel explanation is provided regarding the relationship between gender equality norms in rebel groups and their decisions to pursue peace. Empirically, the relationship between gender equality norms and negotiated settlements is tested in a comparative large-n study using quantitative methods. Provided that this is a quantitative study, the external validity is high, and results are thus generalizable (Gerring, 2006: 38). This makes the results relevant for policy makers as well, as information is provided regarding why some rebel groups are more likely than others to respond to peace initiatives.
Previous research has indeed demonstrated that gender equality, in addition of being a
fundamental human right, have indirect benefits for society at large. It is believed, and
empirically supported, that gender equality correlates with lower rates of armed conflict
onsets, state repression, and other violent acts (Melander, 2005a, 2005b; Hudson et al, 2009;
Caprioli, 2003, 2005; Koch, Fulton, 2011; Schaftenaar, 2017). However, little research has been conducted on how gender equality affects the outcome of conflicts – and in particular, how it affect the likelihood of a negotiated settlement. Furthermore, the literature on gender and peace almost exclusively focuses on gender equality at the country level or within a government of a state. Gender equality within rebel groups thus remains an under-researched subject, especially in regard to its consequences (Henshaw, 2017: 122; Henshaw et al, 2019;
1106).
Gender equality is the obtainment of equal rights, responsibilities and opportunities for men and women. When gender equality norms are weak, men and women are ascribed
stereotypical gender roles disallowing them certain behaviors (Caprioli, 2003; Bjarnegård et al, 2019; Wood, Ramirez, 2018). In the context of armed conflict, women are limited to supportive and dependent roles while men are encouraged to use violence (Goldstein, 2001:
251; Bouta et al, 2004: 13f). Weak gender equality norms thus provide a justification for violence and increases the likelihood of its usage in armed conflict (Melander, 2015;
Bjarnegård, Melander, 2017; Bjarnegård et al, 2019). When gender equality norms are strong however, stereotypical gender roles are contested, and with that so is the use of violence (Wood, Ramirez, 2018; Caprioli, 2003, 2005). Strong gender equality norms within rebel groups should therefore transmit positive attitudes towards peace. This should in turn increase the likelihood that the group decides to pursue peace, which makes the attainment of a
negotiated settlement more likely.
The analysis in this study builds on a large sample including 182 conflict dyads over the time period 1979-2013 (Thomas, Wood, 2017, 2019; Kreutz, 2010, 2016). Gender equality norms will be measured as female combatants and female leadership prevalence in rebel groups and negotiated settlements as the successful attainment of a peace agreement. In addition, seven control variables are included with connections to both the independent and dependent variable. In order to test the claim in this paper, the quantitative method of logistic regression will be utilized. A descriptive analysis will also be conducted as an initial step in the analysis.
Based on the empirical analysis, support is found that gender equality norms, when measured as female combatants’ prevalence, is connected to a higher likelihood of negotiated
settlements. No relationship was however found between female leadership prevalence and
negotiated settlements. Several robustness tests were also conducted in order to increase the
confidence in the results and the main findings remained robust. More research is however needed to either confirm or disconfirm the claim that gender equality increases the likelihood of negotiated settlements. Future research should especially test the causal mechanism, which is not investigated in this study. This could be done by investigating the peace process more in-depth within rebel groups that possess strong gender equality norms.
This paper begins with the chapter theoretical framework. The main concepts are here
defined, and previous findings are presented. Thereafter, the causal mechanism is explained in detail, clarifying how gender equality norms in rebel groups increases the likelihood of
negotiated settlements. The proceeding chapter, research design, presents the choice of method, the dataset, the operationalizations of the variables, and the model specification.
Following, the chapter results and analysis will be presented. This chapter begins with a descriptive analysis of all variables in the study, followed by an analysis of the results from the logistic regression. The next chapter, discussion, summarizes the study as a whole. The results and its limitations are also discussed further. The paper ends with a conclusion.
2. Theoretical Framework
In this chapter, I will start by defining the terminology and central concepts used in the study.
I will thereafter present previous research and the research gap, which will be followed by a specification of this study’s contribution. The chapter ends with a section on the causal mechanisms.
3.1. Terminology and Concepts
The unit of analysis in this study is conflict dyads. According to the Uppsala Conflict Data Program (UCDP), a dyad can be defined as “two armed and opposing actors” (UCDP,
2019b). Since this study specifically focuses on rebel groups, a scope condition is that a dyad
includes 1) a government and 2) a rebel group. A rebel group is a formally organized group
that uses violence as part of an armed conflict. In order to be formally organized, the group
has to have an official name and use armed force (Ibid). Finally, an armed conflict is defined
as “a contested incompatibility that concerns government and/or territory where the use of
armed force between two parties, of which at least one is the government of a state, results in at least 25 battle-related deaths in one calendar year” (Ibid).
The independent variable in this study is gender equality norms. A norm refers to what is perceived as normal in a certain context. They function as general rules that are socially enforced (Horne, 2018). Gender equality can be defined as the moment when men and women enjoy “equal rights, responsibilities, and opportunities” (Bouta et al, 2004: 5). What this concerns more specifically is highly contextual, and the focus here will be on the armed conflict setting. The meaning of gender is socially constructed roles that men and women are expected to ascribe to (Ibid: 3). Gender roles vary both between and among men and women.
They are not rigid and can thus be contested both by individuals, groups, and societies (Ibid).
The ability to question gender roles is however dependent on the context as well. It will in this study be assumed that gender roles are contested to a higher degree when gender equality norms are strong (Caprioli, 2003: 199). Gender is furthermore not the only defining
characteristic of men and women. Rather, gender interacts with a number of attributes which all intersects in order to shape an individual’s behavior (McClintock, 1995). Gender is however believed to be a more influential attribute in general, as it has been argued that gender-based beliefs have societal consequences beyond the individual (Hudson et al, 2009:
12). It is also important to note that men and women are not the only existing genders, albeit the most common ones (Lussenhop, 2018: 197). Due to the scope of this study, only gender as a category is included in the theoretical explanation and the only genders considered are men and women.
The second key concept in this study is a negotiated settlement to an armed conflict. A negotiated settlement refers to an agreement between the main parties in the armed conflict.
The agreement should both address the underlying conflict incompatibility and terminate
armed violence (Kreutz, 2010: 245). The key is here that both parties are willing to pursue
peaceful rather than violent tactics to solve their conflict. Indeed, the negotiations process
towards a settlement is also part of the parties’ conflict, albeit without the use of armed force
(Pillar, 1983). A negotiated settlement is only one out of several ways in which a conflict can
be terminated. In this study, an armed conflict will be deemed as terminated when it no longer
results in 25 battle-related-deaths in a year (Kreutz, 2010). That is, when a year of active
conflict, in which more than 25 people are killed in battle, is followed by a year in which
fewer than 25 people are killed in battle. This paper does not consider whether the conflict
remains terminated or not. The sustainability of peace is thus outside the scope of this paper.
The variation of interests is between armed conflicts that ends through negotiated settlements, and armed conflicts that do not.
3.2. Previous Research and Gap
It has been assessed that a negotiated settlement is both an outcome and a process (Kalin, 2018: 37f). It is an outcome as it portrays an ending to a conflict, different from other
outcomes such as victory or loss. It is a process as it consists of different phases that all have to be realized in order for the outcome to occur (Ibid). A negotiated settlement consists of three different phases; the decision to negotiate, the reaching of an agreement, and the implementation of the agreement (Findley, 2013; Walter, 2002). The attainment of a
negotiated settlement has traditionally been explained by two grand theories within security studies; ripeness theory and bargaining theory (Kalin, 2018: 39f). These theories will
following be explained in greater detail. Thereafter, I will assess how the research on gender equality could contribute to the understanding of what increases the likelihood of a negotiated settlement.
Ripeness theory focuses on the first phase in the peace process; the initiation of negotiations.
The founder of ripeness theory, William Zartman, claims that “parties resolve their conflict only when they are ready to do so” (2001: 8). That is, in a moment when the conflict is ripe for resolution. Ripeness is dependent on two criteria. First, both conflict parties have to perceive that they are in a “mutually hurting stalemate”. A hurting stalemate is a deadlock in which the parties realize that the continuation of violence no longer is profitable (Ibid; Pruitt, 2005). Second, both parties have to perceive that there is a “way out” of the conflict. This is typically described as an attractive alternative to violence that motivates the parties to put down their arms (Ibid; Pruitt, Kim, 2004: 189f). Ripeness is thus affected by a subjective cost and benefit analysis by the main actors in the conflict (Zartman, 2001; Kaplow, 2016).
When ripeness theory has been investigated in empirical research, focus has been on what
makes a ripe moment more or less likely to occur. Key findings state that warring parties that
are equally matched in strengths are more likely to get stuck in a hurting stalemate. This, in
turn, increase the chance of a negotiated settlement (Clayton, 2013; Hultquist, 2013;
Cunningham et al, 2009). Another important finding concerns the importance of third parties, as they can make the main parties perceive that they are in a stalemate (Zartman, 2001: 15).
Third parties can either take on a more passive role and offer their help as facilitators during negotiations. As facilitators, they help the main parties communicate and can help re-phrase conflict issues in order to move negotiations forward (McDermott, 2009; Pruitt, Kim, 2004:
232f; Beardsley et al, 2006). Third parties can also take on a more active role and attempt to manipulate the parties into perceiving a stalemate (Beardsley et al, 2006; Pruitt, Kim, 2004:
238f). This can be done through the use of sanctions and rewards, and thus demands that the third party holds leverage over at least one of the conflict parties (Ibid). While third parties generally make a negotiated settlement more likely, the settlement is believed to be more sustainable when the parties agree upon it by themselves (Pruitt, Kim, 2004: 238, Nathan, 1999).
Bargaining theory has in common with ripeness theory that it centers its argument on a cost and benefit analysis. However, rather than explaining when conflict resolution occurs, bargaining theory explains how conflict resolution occurs (Pillar, 1983: 3). According to the theory, armed conflict is in itself a part of the bargaining process (Ibid; Fearon, 1995). Violent tactics are thus conducted with the same purpose as peace talks; to strike a bargain that
maximizes utility for the perpetrator (Pillar, 1983). The key to conflict resolution is therefore to make the conflict parties perceive peace as more beneficial than violence. According to Fearon (1995), this can only be achieved when the warring parties trust each other enough to commit to a negotiated settlement. Due to this argument, bargaining theorists also believe in the importance of third parties. The focus is however on the implementation phase, rather than the negotiation phase. Walter (2002) claim that third parties play a crucial role as peace guarantees, as they can make promises to maintain peace if the conflict parties would apart from the peace agreement. This can in turn help the parties overcome problems of credible commitments.
Taken together, both ripeness and bargaining theorists outline agent-based explanations as to
what increases the likelihood of a negotiated settlement (Pruitt, Kim, 2004: 183f; Zartman,
2001; Pillar, 1983). As discussed, the focus has been on what affects the decision-making
process of the main actors in the conflict, also called their resolve (Ohlson, 2008: 140). An
actor’s resolve to either continue or end armed conflict is partly affected by the cost and
benefit analysis already discussed. In addition, the actors’ belief systems are of importance, as
it helps to sort out the alternatives considered before making a decision (Ibid; Mintz, 2004).
The main focus within the literature has been on third party involvement, on the relative power distribution between the parties, and on the type of conflict in question (Walter, 2002;
Pearson, 2011; Cunningham et al, 2009; Kalin, 2018: 37f). All of these areas are linked to the likelihood of a negotiated settlement, as they affect the main parties’ willingness to pursue peace. Little focus has however, until very recently, been given to how gender equality norms affects the likelihood of a negotiated settlements. Gender equality is believed to make a society more peaceful in general (Caprioli, 2005; Melander, 2005a). Findings show that gender equality within the government result in less military spending and less violent repression (Caprioli, 2003; Melander, 2005b; Koch, Fulton, 2011). Similarly, higher gender equality at the country level decrease the likelihood of armed conflict onset (Melander, 2005a;
Caprioli, 2005; Schaftenaar, 2017). It is therefore evident that gender equality should be included in the study of conflict resolution.
While the covariation between gender and peace has been confirmed in several studies, little is known about the mechanism behind this. Some believe that the relationship exists due to a gender gap (Koch, Fulton, 2011; Brooks, Valentino, 2011). As such, men and women are believed to possess different attitudes, with women in general being more peaceful than men (Conover, 1994; Smith, 1984). This can in turn be explained by either an essentialist or a constructivist argument. Essentialists believed that differences between women and men can be explained by biology (Togeby, 1994; Fite, Genest, Wilcox, 1990), while constructivists believe that the same differences are a result of a socialization process (Goldstein, 2001: 251;
Caprioli, 2003). The gender gap as a whole has however been questioned in recent studies, as it has been argued that the gender and peace relationship is better explained by a feminist gap (Bjarnegård, Melander, 2017). According to this argument, feminist attitudes rather than gender explains why gender equality transmits peace (Bjarnegård, Melander, 2017;
Schaftenaar, 2017). Both women and men that hold feminist attitudes should thus be more peaceful (Ibid).
Provided that gender equality reduces violent conflict norms, it should affect how a conflict is
terminated. Recent findings have indeed demonstrated that an increase in female legislators
increases the chance that an armed conflict is terminated through a negotiated settlement
(Best et al, 2019). An issue with these findings is however that they only focus on one side of
the conflict - that being the government. This is a general trend within the negotiations
research, which is partly supported by findings that the government is more likely to initiate peace talks than are rebel groups (Pearson, 2011). However, a conflict consists of more than one party, and a negotiated settlement cannot be achieved unilaterally (Zartman, 2001). More research should therefore be conducted on how the government’s opponent, which typically is a rebel group, affect the outcome of the conflict.
Research on rebel groups have tended to be gender blind. As such, a group’s behavior has been analyzed without taking gender into account, or only taking male gender into account (Henshaw et al, 2019: 1091; Henshaw, 2017: 1f). Compared to common beliefs, there are many conflicts where women have made important contributions as both combatants and leaders (Ibid; Trisko, 2015). However, large scale comparative data have not been available demonstrating this until very recently (Thomas, Wood, 2017; Henshaw et al, 2019; Henshaw, 2017). This creates yet another gap in the research, as few quantitative studies on gender equality in rebel groups have been conducted. This is especially true in regard to what
consequences this may have for the conflict dynamics and its outcome (Henshaw, 2017: 122;
Henshaw et al, 2019; 1106). Only one study has to my knowledge tested this by investigating whether female inclusion in rebel groups increased the likelihood of rebel victory
(Braithwaite, Ruiz, 2018). No statistically significant relationship was however found (Ibid).
Based on the specified gaps in the preceding sections, there is a need to investigate the relationship between gender equality in rebel groups and the likelihood of negotiated
settlements. Several scholars have indeed suggested that such a relationship should exist and requested quantitative investigations of the subject (Henshaw et al, 2019; Henshaw, 2017).
Two contributions will therefore be made; one theoretical contribution and one empirical contribution. First, I make a theoretical contribution to the research field by offering a novel explanation as to why gender equality in rebel groups should be connected to a higher likelihood of negotiated settlements. In contrast to previous studies, the focus is here on gender equality norms, rather than biological sex (Best et al, 2019; Braithwaite, Ruiz, 2018).
Second, I make an empirical contribution by investigating the implications of gender equality
in rebel groups using quantitative methods.
3.3. The Causal Mechanisms
The theoretical argument in this paper primarily builds on Goldstein’s (2001) theory of gender roles during war and Mintz’s (2004) poliheuristic theory on actor’s decision-making process. The causal mechanism consists of two steps; 1) strong gender equality norms results in a positive attitude towards peace, and 2) positive attitudes towards peace increases the likelihood that a rebel group will make decisions to pursue peace. These steps will here be explained in greater detail. The variation of interest is between rebel groups that hold strong gender equality norms, versus groups that hold weak gender equality norms.
In the context of war, gender equality norms are strong when men and women are allowed to participate in the conflict under the same conditions. A significant majority of all combatants and leaders in rebel groups are men (Bjarnegård et al, 2015). This corresponds to Goldstein’s (2001) notion of stereotypical gender roles during conflict
1. When stereotypical gender roles are prevalent, masculinity becomes synonymous with qualities that defines a good warrior.
These qualities are physical courage, endurance, honor, strength and skill (Ibid: 266). Taken together, they all prepare the man for the use of violence. Men are therefore expected to use violence and punished when not doing so (Ibid: 251f). This punishment typically takes the form of shaming, an act that both men and women participate in (Ibid: 272). Men who otherwise would question the use of violence are therefore encouraged to fight. As a consequence, violence becomes a justified act (Ibid: 251f).
The stereotypical masculinity described above is only one out of two mutually reinforcing gender identities. Men are thus only expected to comply with the notion of masculinity during war as long as this is contrasted by a stereotypical notion of femininity (Goldstein, 2001:
301). When women participate in armed conflict as part of a rebel group, they typically take on supportive or dependent roles, rather than combatant or leader roles (Bouta et al, 2004:
13f; Henshaw, 2017: 117). As supporters, women help the rebel group with non-fighting tasks. As dependents, women are included in the rebel group due to family ties with the male combatants (Bouta et al, 2004: 14). The stereotypical gender roles separate men and women and is thus the foundation for gender inequality (Caprioli, 2003). The contestation of these
1 The gender norms discussed in this paper are broad generalizations, as no universal definition of masculinity nor femininity exists (Nagel, 1998: 245f). The norms discussed here are however believed to be hegemonic, which supports this generalization (Caprioli, 2003: 199).
roles allows women and men to receive the same rights, responsibilities and opportunities (Ibid). It will therefore be assumed that when stereotypical gender roles are contested, a rebel group holds stronger gender equality norms than when these norms are complied with.
In accordance with the preceding argument, gender equality norms grow stronger when gender stereotypes are less prevalent. This has important implications for the conflict since the preservation of gender roles provides a justification for violence (Caprioli, 2003;
Melander, 2015; Goldstein, 2001; Wood, Ramirez, 2018). The first step in the causal mechanism is therefore; strong gender equality norms results in a positive attitude towards peace. This occurs as the abolishment of stereotypical gender roles should result in men no longer feeling pressured to use violence in order to avoid punishment in the form of shaming (Goldstein, 2001: 251f; Bjarnegård et al, 2019). The use of violence should therefore no longer be rewarded nor seen as valuable in itself. The change in gender equality norms should affect the attitudes of men and women equally, as the key is not the biological sex but the attitudes towards gender equality in general (Bjarnegård, Melander, 2017).
When gender equality norms grow stronger in rebel groups, the pressure towards men to use violence should decrease. The rebel group should thus be more reluctant towards the use of violence, as an important factor contributing to its justification vanishes. The questioning of the use of violence should produce cognitive changes in the actor’s belief systems which affects how they make their decisions. A belief system affect how information is selected and interpreted. In turn, this affect how decisions are made (Ohlson, 2008: 141; Herrmann et al, 1997: 406). In accordance with poliheuristic decision-making theory, it is here assumed that actor’s make their decisions based on a two-step process (Mintz, 2004). First, they sort out all available alternatives and, based on their belief system, removes the ones which are not deemed acceptable to them on a critical basis. Second, based on the remaining alternatives they make the decision which maximizes their expected utility. Actors’ decisions are thus still assumed to be rational. However, the subset of alternatives on which decisions are based differs depending on the actor’s belief systems.
Based on the previous argumentation, gender equality norms should affect how rebel groups
make their decisions. Strong equality norms should increase the preferences for peace, and
weak norms should increase the preferences for violence. The second step in the causal
mechanism is therefore; positive attitudes towards peace increases the likelihood that a rebel
group will make decisions to pursue peace. This has two effects on the process towards a negotiated settlement. First, it increases the likelihood of a ripe moment as it is more likely that the rebel group will perceive that they are in a hurting stalemate. As previously stated, a stalemate in an armed conflict occurs when the actors perceive that violence no longer is profitable and therefore seek peaceful means through which their conflict can be solved (Pruitt, 2005; Zartman, 2001). Since violence is costlier for groups with strong gender
equality norms, it is more likely that they will perceive this stalemate. Second, the decision to pursue peace by rebel groups with strong gender equality norms should also have a positive effect on peace negotiations. When the rebel group has stronger preferences for peace, they should be more flexible in their demands (Hopmann, 1995). Flexibility refers to the
willingness to make concessions and is necessary if an agreement is to be reached (Ibid).
More flexibility should thus increase the chance of negotiation success, and as such of a negotiated settlement. The causal mechanism has been visualized in figure 1.
Figure 1. The causal mechanism: the relationship between gender equality norms and negotiated settlements.
To summarize, when stereotypical gender roles are deconstructed, violence becomes delegitimized. Therefore, when a rebel group has strong gender equality norms, they gain a positive attitude towards peace. This makes it more likely that the rebel group will decide to pursue peace, which in turn increases the likelihood of a ripe moment and the likelihood of negotiation success. Taken together, this supports the claim that strong gender equality norms in a rebel group increases the likelihood of a negotiated settlement. This claim will be tested through two hypotheses. The hypotheses differ as they measure gender equality norms
differently. Since the formulation of the hypotheses are dependent on the operationalization of the independent variable, they will be specified first in the proceeding chapter.
Positive attitude towards peace
decision to pursue peace
Negotiated settlement Strong gender
equality norms in rebel groups
4. Research design
This chapter presents the research design which will test the theoretical claim in this paper.
First, the choice of method will be described and motivated. This will be followed by a presentation of the dataset used to conduct the analysis. All variables used in the dataset will thereafter be operationalized. In relation to the operationalizations, validity and reliability will also be discussed. The chapter will end with a more in-depth description of the method and the models used in the quantitative analysis.
4.1. Choice of Method
The method in this paper has been chosen with the aim to examine how gender equality norms affects the likelihood of a negotiated settlement. In order to anwser this question, this study would in accordance with the criteria for causality have to 1) demonstrate that there is covariation between the main variables, 2) isolate the relationship from alternative causes, 3) demonstrate that the events occur in the correct time order, and 4) test the causal mechanism in order to describe how the variables are connected to each other (Kellstedt, Whitten, 2013:
55). Time order is not considered an issue in this particular study as the dependent variable is the outcome of conflicts. By definition, this cannot affect gender equality norms in rebel groups during the conflict. The causal mechanism is further excluded since it is not common to test how two variables are connected before knowing whether they are connected at all (Beach, Pedersen, 2013). A first step to answer the research question is thus to investigate whether a relationship exists, and whether this relationship can be isolated. With this in mind, a comparative large-n study using quantitative methods will be conducted. This is furthermore the most appropriate method since results from quantitative studies have higher external validity (Gerring, 2006: 38). Observed patterns within the sample used in this study can therefore be generalized to the larger population.
4.2. Dataset
I have merged data from two datasets; the Women in Armed Rebellion (WARD) dataset
(Wood, Thomas, 2017, 2019), and the UCDP Conflict Termination (UCDP-CT) dataset
(Kreutz, 2010, 2016). Both WARD and the UCDP-CT are the most extensive datasets
available in their respective fields (Henshaw et al, 2019; Wood, Thomas, 2017, 2019; UCDP, 2019a). Furthermore, an updated version of WARD has been used which includes new data on female leadership prevalence. This version was conducted by Henshaw et al in 2019.
While the population in this study is all armed conflicts, the dataset used only includes conflicts that has already been terminated. Since ongoing conflicts still has the potential to end through a negotiated settlement, it is problematic to assume that these stand in opposite to successful negotiations. Thus, while the focus on only terminated conflicts limits the sample, it makes the comparison of cases more valid.
As previously stated, the unit of analysis in the dataset is conflict dyads between a
government and a rebel group. The UCDP-CT dataset includes all terminated conflict dyads from the UCDP dyadic dataset, covering the time period of 1946-2013 (Kreutz, 2016;
Pettersson, 2019). The WARD dataset also bases their sample on the UCDP dyadic dataset (Thomas, Wood, 2017, 2019). Data in WARD is however only available for the time period 1979-2009 (Ibid). Due to this, all cases before 1979 has been excluded. Conflicts that were terminated after 2009 are still included as it is assumed that the data in WARD is relevant for these cases as well
2. The WARD dataset further excludes all dyads where the government opponent is a military faction as well as cases that lack sufficient information
3(Thomas, Wood, 2019: 1). The sample is thus limited due to missing values for some cases. Based on these constraints, the sample in this study covers 182 conflict dyads between the time period 1979-2013.
The compatibility between the WARD and the UCDP-CT dataset is good, since they utilize the same identification variable for conflict dyads and rely on similar definitions for the inclusion of cases (Kreutz, 2010; Thomas, Wood, 2017). Data has furthermore been collected systematically and transparently
4, which makes the sources more credible
5. There are
however two issues in regard to the WARD dataset. The first issue concerns the sample
2 The data in WARD is time invariant. It thus describes the general prevalence of female combatants and/or leaders rather than the prevalence during a specific year.
3 Typically, cases with low information are small groups and/or groups that only existed for a short amount of time (Thomas, Wood, 2019: 1).
4 In the WARD dataset, observations are only included if they can be confirmed in at least three independent sources (Thomas, Wood, 2017). The UCDP uses similar measures to ensure source reliability, and further attempts to trace reports back to the primary source (UCDP, 2019c).
5 Considering that all data utilized in this analysis comes from secondary sources, it is important to adopt a critical perspective in relation to the sources. This is especially true since the phenomenon of interest is armed conflicts, which inevitably involves some levels of uncertainty (UCDP, 2019c; Öberg, Sollenberg, 2011).
which, ideally, should be randomly selected from the underlying population if full population data is not available (Kellstedt, Whitten, 2013: 140). The sample in this study is however dependent on data availability. Since some cases have been systematically excluded, this affects to what extent the results can be generalized. Another issue is that the data for the independent variable is time invariant (Thomas, Wood, 2017, 2019; Henshaw et al, 2019).
This study thus assumes that the level of female combatant and/or leader prevalence remains the same across the entire time period. While these limitations should be kept in mind, the WARD dataset remains the most extensive data project on gender distribution within rebel groups that currently exists (Ibid). It is as such the most appropriate data source for this study.
4.3. Operationalization: Independent Variable
The independent variable in this study is gender equality norms in rebel groups. Two indicators have been specified in order to measure this; female combatants’ prevalence and female leadership prevalence. These indicators demonstrate how well men and women in rebel groups comply with stereotypical gender roles. Women that comply with stereotypical femininity do not typically engage in battle nor take on leader roles (Goldstein, 2001: 301).
Higher prevalence of female leadership and/or female combatants therefore indicates that gender stereotypes are weaker. As previously stated, these gender stereotypes uphold the separation of women and men. When the stereotypes are contested, gender equality norms should become stronger (Caprioli, 2003). A focus on gender roles in rebel groups is therefore adequate as a measurement of gender equality norms.
Women are considered as combatants in rebel groups when they actively use armed violence in a conflict. The operational definition was taken from the WARD dataset where female combatants are defined as “female members who underwent military training, received combat arms, and directly participated in organized violence on behalf of the organization in any capacity during the conflict” (Thomas, Wood, 2019: 2). In order for an estimate to be made, three independent sources have to confirm that female combatants either were, or were not, prevalent. All cases where women were only used as suicide bombers are furthermore excluded (Ibid)
6. While there are still no exact numbers of female combatants in rebel groups, WARD include a rough estimate of the proportion of female combatants. The variable
6 This is further motivated in the section “Operationalization: Control Variables”.
includes four categories; no prevalence (0%), low prevalence (<5%), moderate prevalence (5%-20%) and high prevalence (>20%) (Ibid). In the analysis, the variable will be recoded to a dichotomous variable. No prevalence and low prevalence will be coded as a 0, while moderate to high prevalence will be coded as a 1. The distribution of the values in the dataset used for this study is visualized in figure 2. The reason why low prevalence is coded as a 0 is to make sure that the groups that include female combatants do this systematically. Based on this operationalization, the following hypothesis will be tested;
(H1) When a rebel group systematically includes women as combatants it is more likely that an armed conflict is terminated through a negotiated settlement.
Figure 2. Bar-chart over female combatants’ prevalence within rebel groups active in armed conflict.
Women are considered as leaders in rebel groups when they actively influence the group’s decision-making process. The operational definition comes from the extended WARD dataset.
Female leadership was here defined as “any position in which women exercise direct control over other participants, strategy, policies, and/or ideology within the organization” (Henshaw et al, 2019: 1095). Similar to the female combatant measurement, three independent sources have to confirm the existence of female leadership before an estimate is made (Ibid). In the WARD dataset, female leadership was coded as a dichotomous variable indicating whether female leaders existed or not. According to the dataset, approximately 30% of all rebel groups includes female leaders. For the purpose of this study, the variable is coded in the same way as in the original dataset. A shortcoming of this measurement is that no difference is made
62%
18%
12%8%
0 20 40 60 80 100 120 140 160
No Yes
Systematic Inclusion of Female Combatants
No evidence Low Moderate High
between groups that incorporates female leaders to a low respectively high degree
7. Based on this operationalization, the following hypothesis will be tested;
(H2) When a rebel group includes women as leaders it is more likely that an armed conflict is terminated through a negotiated settlement.
4.3.1. Validity and Reliability: Independent Variable
When a large-n study is conducted it is especially important to ensure that reliability is high.
Otherwise, the cases cannot be compared to each other (Kellstedt, Whitten, 2013: 101;
Sundberg, Harbom, 2011: 92f). Reliability ensures that the same thing is measured for all cases. By measuring gender equality norms through female combatants and female
leadership, this study focuses on something concrete. In contrast to other measurements, like ideology, female combatants/leaders are unambiguous concepts that should be interpreted equally across different cases. Reliability should therefore be high. One possible issue is however that some cases where women were acting like combatants/leaders will be excluded.
In practice, it is difficult to differentiate between combatant/leader roles and supporter roles, since different tasks are blurred together (Thomas, Wood, 2019; Bouta et al, 2004: 101f).
Thomas and Wood (2017, 2019), equal to Henshaw et al (2019), adopt fairly conservative definitions of what constitutes a female combatant and/or leader. The threshold for the inclusion of cases is therefore high. This means that the score for all cases should be
somewhat underestimated, albeit in the same way. As such, reliability is still deemed as high
8(Sundberg, Harbom, 2011: 99f).
There is typically a tension between measurement validity and reliability. As such, the same measurement characteristics that increases reliability may decrease validity (Kellstedt, Whitten, 2013: 101; Sundberg, Harbom, 2011: 92f). Gender equality is a multifaceted concept, and it is therefore best captured through the use of multiple indicators and rich descriptions (Coppedge, 1999). One of the main shortcomings of quantitative studies is therefore that it cannot adequately measure thick concepts, since rich descriptions cannot be provided for all cases without seriously jeopardizing reliability (Ibid). Considering the
7 This issue is further developed in the “Limitations and Alternative Explanations” section.
8 The coders behind the WARD dataset has conducted replication tests in order to ensure high reliability (Henshaw et al, 2019: 1095).
complexity of the concept “gender equality norms”, validity is in this study somewhat compromised. That said, female combatants and female leadership directly address compliance with stereotypical gender roles. As discussed, when women are allowed to be combatants and/or leaders gender stereotypes are contested, which strengthens gender equality norms. Therefore, the indicators correspond well with the definition of the main concept
9(Bouta et al, 2004: 5). Validity is further strengthened as two indicators are used for measuring one concept. Given that this is a large-n study, validity is thus deemed as high.
4.4. Operationalization: Dependent Variable
The dependent variable in this study is negotiated settlements. In order to measure this, the variable will be operationalized as the successful attainment of a peace agreement. The operational definition was taken from the UCDP-CT dataset (Kreutz, 2010, 2016). A peace agreement was there defined as an agreement signed by all main conflict parties that either resolves or regulates the conflict incompatibility. The agreement must also end military activity and be publicly accepted by all warring parties (Ibid: 245). In the dataset used for this study, peace agreements will be coded as a 1, accounting for the outcome peace agreement.
The remaining categories for conflict outcome will be coded as a 0, corresponding to other outcome
10. In the dataset, approximately one out of five conflicts ended through a peace agreement. An issue with the data in this analysis is that some dyads reappear in the dataset.
For example, some conflicts are first terminated due to low activity and later due to a peace accord (Ibid). This becomes an issue since the data for the independent variable is time invariant (Thomas, Wood, 2017). There is thus only one observed value for each dyad across all years it was active. Each dyad will therefore only be included once. For the dyads that have more than one outcome, only the final outcome will be included. The choice to focus on the final outcome is based on the assumption that this outcome should be more decisive than the first one.
4.4.1. Validity and Reliability: Dependent Variable
9 The definition of all main concepts can be found in the section “Terminology and Concepts”
10 The other outcomes included in the dataset are victory for side A or B, low activity, ceasefire agreement and actor cease to exist.
By definition, a negotiated settlement result in something that is publicly available; an official document or an official statement representing an agreement between the warring parties (Kreutz, 2010: 245). This distinctive feature indicates that the coding process should be fairly simple, and reliability is therefore deemed as high (Sundberg, Harbom, 2011: 97f; Kellstedt, Whitten, 2013: 99f). The operational definition further corresponds well with the main concept, and validity is therefore also high. An issue is however that the choice to only focus on peace agreements may result in some cases of negotiated settlements being excluded. This mainly refers to ceasefire agreements, which is defined as when the main conflict parties agree to end violence, but do not solve the conflict incompatibility (Kreutz, 2010: 245).
Ceasefires are always a part of more comprehensive peace agreements (Ibid). They are however not necessarily in themselves evidence of peaceful attitudes among the main actors.
In can be theorized that an actor would agree on a ceasefire in order to gain time to re-group.
If this were the case, the ceasefire agreement would not reflect a willingness to pursue peace, which is the key of this paper’s theoretical argument. It is therefore appropriate to only focus on peace agreements in the main analysis, despite the fact that some cases will be excluded. A robustness test where the dependent variable is recoded to include both peace and ceasefire agreements will however be conducted in order to account for this validity issue.
4.5. Operationalization: Control Variables
Since this is an observational study, alternative factors affecting the independent and
dependent variable have to be taken into consideration. I have selected a total of seven control variables. Five of these measure rebel group characteristics, and two measure characteristics at the country level. The purpose of the control variables is foremost to avoid omitted-variable bias (Kellstedt, Whitten, 2013: 87, 210). A variable is thus included when there are theoretical reasons to suspect that it has an effect on both the dependent and independent variable (Ibid).
The first three control variables all concern the ideology of the rebel group. Research has established that groups are more or less likely to include women in combatant and/or leader roles depending on their ideological stance (Thomas, Wood, 2017; Henshaw et al, 2019).
Groups with leftist ideology are in general more likely to include female combatants and/or
leaders, while religious groups are less likely to do so (Ibid). Ideology is furthermore
connected to legitimacy which in turn can have an effect on the likelihood of negotiated
settlements (Heger, Jung, 2017: 1217). It has been established that religious groups are more likely to use extreme tactics during armed conflict, which make them less legitimate in the eyes of their opponent (Berman, 2009). This should decrease the likelihood of a negotiated settlement (Heger, Jung, 2017: 1217). In order to control for the most common ideologies, I include three dichotomous variables in the analysis; Leftist ideology
11, Islamist ideology
12, and Religious ideology
13. All variables were taken from the WARD dataset.
The fourth control variable is rebel capacity. Several studies have demonstrated that rebel groups with better military capabilities are more likely to receive concessions from the state, and as such solve conflict through negotiated settlements (Clayton, 2013; Cunningham et al, 2009). While there are no consistent findings suggesting that rebel capacity affect the
prevalence of female combatants and/or leaders, it can be theorized that weak groups are less capable to recruit members overall, which would have an effect on female recruitment as well. I therefore include the dichotomous variable rebel weakness
14which reflects the relative military capacity of the rebel group in relation to the government. The variable was taken from the WARD dataset.
The fifth control variable is gender equality at the country level. Previous research has established that gender equality is connected to overall higher levels of peace (Best et al, 2019; Melander, 2005a; Hudson et al, 2009). It is thus possible that gender equality at the country level could facilitate conflict termination through negotiated settlements (Best et al, 2019). Gender equality is in addition connected to female inclusion in rebel groups. Higher gender equality in society increases both the chance that rebel groups will be willing to incorporate women, as well as the chance that women will be willing to take on non-
conventional roles (Thomas, Wood, 2018). In order to control for gender equality, I include the continuous variable secondary education ratio, measuring the difference between female- to-male student enrollment in secondary school. This measures gender equality since equal
11 Leftist ideology refers to all rebel groups that incorporates a Marxist ideology (socialist, communist, Maoist, Marxist-Leninist). When this ideology was prevalent, the variable was coded as a 1 (Thomas, Wood, 2017).
12 Islamist ideology refers to all religious rebel groups that advocates for the implementation of Sharia laws and/or the establishment of a state based on Islamic principles. When this ideology was prevalent, the variable was coded as a 1 (Thomas, Wood, 2017).
13 Religious ideology refers to rebel groups whose primary goals in armed conflict is to advance the interest of a specific religion. The variable Religious ideology refer to groups that incorporates a religious ideology but not an Islamist ideology. When this ideology was prevalent, the variable was coded as a 1 (Thomas, Wood, 2017).
14 Rebel weakness refers to how strong the rebel group is measured as the number of troops under the groups command, relative to the strengths of the government. The variable was coded as 1 when the rebel group is much weaker than the government (Thomas, Wood, 2017).
access to education reflects a belief that men and women should be provided the same
opportunities in society (Ibid; Schaftenaar, 2017). When women are provided education to the same extent as men, they are in addition allowed to take on new gender roles, other than that of the caretaker (Thomas, Wood, 2018). Since the data of the independent variable is time invariant, the measure represents an average score of the secondary education ratio during the time the armed conflict was active. The variable was taken from the WARD dataset.
The sixth control variable is also measured at the country level, reflecting the regime type of the government that the rebel group is challenging. Democratic regimes should be more likely to solve conflict through negotiated settlements, considering that the democratic system incorporates several non-violent conflict resolution mechanisms (Wallensteen, 2015: 139f). It can also be theorized that regime type is connected to female incorporation in rebel groups.
Since democracy provides civil liberties and equal opportunities for political participation, women are typically provided better opportunities in democratic systems compared to autocratic systems (Bjarnegård, Melander, 2011: 141). Similar to the preceding
argumentation, this increase the likelihood that women will take on less conventional roles in general, which in turn should increase the prevalence of female combatants and/or leaders in rebel groups (Thomas, Wood, 2018). In order to control for this, the continuous variable regime type will be included. The variable is a scale ranging from 10 to -10, where a higher number corresponds to a more democratic regime
15. In order to avoid endogeneity, the variable was lagged one year behind the time of the conflict outcome. The variable was taken from the Polity IV project (Marshall et al, 2018).
In addition to the variables discussed above, I also include one control variable that accounts for the strategies used by rebel groups. Some argue that the recruitment of women is merely a tactical choice, as female recruits brings several strategic benefits. To the very least it
increases the total number of recruits, and in the cases where terrorist tactics are used it increases the efficiency of the group (Henshaw et al, 2019: 1092; O’Rourke, 2009). This occurs as women are believed to be more effective in carrying out different terrorism acts, mainly due to the peaceful female stereotype (O’Rourke, 2009). If the main purpose of female
15 The variable is measured as the difference between the democratic and autocratic characteristics of a regime.
Democratic characteristics are the existence of institutions through which citizens can make demands, existence of constraints on the exercise of power, and the existence of civil liberties. Autocratic characteristics are lack of political competition and political freedoms (Marshall, Gurr, Jaggers, 2018).
combatants and/or leaders is efficiency it would create a validity problem, as the indicators then would reflect something else then gender equality. In order to account for cases where women were incorporated due to tactical reasons, I include the variable suicide tactics. The variable indicates whether a group used suicide terrorism as part of their conflict strategy (Thomas, Wood, 2017). The variable was coded as dichotomous and taken from the WARD dataset.
4.5.1. Validity and Reliability: Control Variables
The control variables in this study has been collected from the WARD dataset and from the Polity IV dataset. The variable from the Polity IV dataset, regime type, is a particularly difficult variable to code as the concept is very thick. However, the Polity scale is one of the most recognized and commonly used measurement of the variable and thus an appropriate source for this study (Melander, 2005b; Hultquist, 2013; Heger, Jung, 2017; Marshall et al, 2018). Another issue is that the variables from the WARD dataset originally are from other datasets. The sources used are the Terrorist Organization Profiles (TOPs) dataset, the Nonstate Armed Groups (NAGs) dataset, the World Bank, and the Chicago Project on Security and Terrorism (Thomas, Wood, 2017). Validity and reliability are thus somewhat compromised given that the WARD control variables are not primary sources. However, the variables had the same purpose in WARD as they have in this study. That is, to describe the characteristics of rebel groups active in conflict, or in the case of the variable secondary education ratio, to describe the characteristics of the country in which the conflict occurs.
Given this, validity and reliability for the variables should be more or less as high as in the original study.
4.6. Model Specification
In order to test the hypotheses, both a descriptive analysis and regression analysis will be conducted. The descriptive analysis will demonstrate how values are distributed, and the regression will test how the variables are related to each other. This will be done through the use of logistic regression. This is the most appropriate method considering that the dependent variable (Y) is dichotomous. Since the values of Y cannot fall below 0 or above 1, the use of linear regression would violate important assumptions (Kellstedt, Whitten, 2013: 189f;
Pampel, 2000: 9). When logistic regression is used, the changes that the independent variable
(X) produces in Y are expressed in logged odds. This means that the relationship between the variables is expressed in a nonlinear manner, as X produces a larger effect on Y near the middle values, compared to a smaller effect close to Y’s floor and ceiling (Pampel, 2000: 5f).
The issue with this is however that the coefficients in the regression cannot be interpreted directly. While logged odds express the direction of the relationship and the statistical
significance level, they do not demonstrate the substantive significance (Ibid). Thus, provided that statistically significant relationships are found, the logged odds will be transformed into predicted probabilities. The aim with this is to interpret the magnitude of the relationship in question
16.
Three different confidence levels will be used to interpret statistical significance; 90% 95%, and 99%. While the 95% confidence level is the general standard within social science, the 90% level have been included considering the small sample size. When the sample size is small, statistically significant results are in general less likely (Kellstedt, Whitten, 2013: 141).
In order to avoid that significant results are overlooked, both the 90% and the 95% level will be considered. Of course, this generates uncertainty in the other direction, as some
relationships may be interpreted as statistically significant when they really are not. In recognition of this, several robustness tests will be conducted in order to strengthen the findings further. First, the operationalization of the dependent variable will be changed as to include ceasefire agreements in addition to peace agreements. As previously stated, ceasefire agreements can be defined as settlements where the warring parties agrees to end violence, while not solving the underlying incompatibility (Kreutz, 2010). This also reflects a willingness to pursue peace by the main parties, albeit not to the same extent as peace agreements. In the second robustness test, the coding for female combatants will be changed back to the original scale ranging from 0-3
17. The variable will be treated as continuous in the robustness test. This is a common and usually efficient way to include ordinal variables in regressions, despite the fact that the assumption of linearity is violated as there is not equal distance between the scale steps (Williams, 2019). In the final robustness test a larger sample will be used. In order to increase the sample size, the variable for female leadership was excluded, as there were many missing values for this variable in the original dataset
18.
16 The logic behind predicted probabilities is further discussed in the “Logistic Regression Analysis” section.
17 0 = no evidence of female combatants; 1 = <5% of all combatants are female, 2 = 5-20% of all combatants are female, 3 = >20% of all combatants are female (Thomas, Wood, 2017, 2019).
18 When this variable was excluded, the sample increased to 211 cases.
A final consideration in this study is the goodness of fit for the different models. The
goodness of fit will be demonstrated in a pseudo R squared estimate, where a higher number indicates a better model of fit. The number cannot be interpreted directly and should only be used to compare models against each-other (Veall, Zimmermann, 1992). No claims will therefore be made regarding how much variation in the dependent variable that the models account for. General considerations in regard to the goodness of fit has however been considered. The first consideration concerns the risks with excluding relevant variables, which results in an under-specified model. When this occurs, the effects from the omitted variables are likely to be captured by the remaining variables. The results are then likely to become biased (Kellstedt, Whitten, 2013: 89, 210). The second consideration concerns the risks with including too many variables, which results in an over-specified model. When unnecessary variables are included, the model is likely to lose precision. This affect the standard errors and may result in misinterpretation of the statistical significance (Frost, 2019;
Allen, 1997). In order to strike a balance between these two considerations, several models will be included in the analysis. Further concerns in regard to how these models were specified will be discussed in the “Logistic Regression Analysis” section.
5. Results and Analysis
This chapter presents the results from the empirical analysis. The chapter starts with a descriptive analysis of all variables. This is followed by the logistic regression analysis, including a total of seven models testing the hypotheses. In addition to the main analysis, the results from the robustness tests will be presented and the logged odds from the main model will be converted into predicted probabilities.
5.1. Descriptive Analysis
The dataset used for this analysis includes a total of 182 dyads, active between 1979-2013.
All dyads have been terminated and all consist of a rebel group and a government of a state.
The dataset includes two independent variables measuring gender norms through the presence
of either female leaders or combatants, seven control variables, and one dependent variable
measuring the likelihood of a peace agreement. The summary statistics for the variables have
been listed in table 1. As demonstrated, the distribution of the values is negatively skewed for all variables. It should be kept in mind that this could affect the results. When comparing the values in the dataset used in this study with the original datasets, there is a small discrepancy between the distributions of values
19. This is however to be expected considering that the sample sizes differ between the datasets. Considering that the general distribution of values still is similar, this should not be an issue.
Table 1. Summary statistics for all variables in the dataset.
Statistic N Mean St. Dev. Min Max
Peace agreement (DV) 182 0.225 0.419 0 1 Female combatants (IV) 182 0.203 0.404 0 1 Female leadership (IV) 182 0.319 0.467 0 1 Leftist ideology 182 0.192 0.395 0 1 Islamist ideology 182 0.137 0.345 0 1 Suicide tactics 182 0.104 0.307 0 1 Rebel weakness 182 0.357 0.480 0 1 Secondary education ratio 182 0.689 0.259 0.117 1.608 Religious ideology 182 0.049 0.217 0 1
Regime type 182 -0.187 5.798 -9 10
In addition to the analysis of each individual variable, table 2 and 3 exhibits statistics for how the indicators for gender equality norms relate to the dependent variable. As demonstrated in table 2, 35% of all groups that include female combatants to a moderate/high degree solve conflict through peace agreements. This can be compared to the portion of all dyads that ends in peace agreements, which is only 22.5%. This indicates that there indeed is a positive relationship between female combatants and peace agreements. There is however no similar indication for the second indicator, as only 19% of all groups that include female leaders solve conflicts through peace agreements (see table 3). Since this is less than the portion of all dyads that ends through peace agreements, it is likely that no positive relationship between
19 15% of all armed conflicts ended through peace agreements according to the UCDP-CT dataset (Kreutz, 2010, 2016). 15% of all rebel groups incorporated female combatants systematically according to the WARD dataset (Thomas, Wood, 2017, 2019). 28% of all rebel groups incorporated female leaders according to the updated version of the WARD dataset (Henshaw et al, 2019)
