Female Fighters

Empirical Evaluation of the Effects of Female Combatants

The results of the analyses presented in chapter 4 support the strategic logic of female recruitment I advanced in chapter 1, which linked war costs and the acute resource constraints they impose to rebel leaders’ willingness to utilize female combatants. However, these analyses provide little insight into the impact that the incorporation of female combatants has on the groups that ultimately recruit them. Chapter 2 directly engaged the question of the strategic implications of female combatants, arguing that the decision to mobilize women for war, particularly in large numbers, can effectively increase the size of the rebel group’s combat force. I further highlighted the potential role that the presence of female combatants can play in shaping observer attitudes toward the movements that employ them, and in assisting these groups in securing support from diaspora communities and transnational activist networks. In this chapter, I therefore evaluate the empirical validity of these arguments. In order to assess the hypotheses presented in chapter 2, I combine group-level analyses using information on female combatants taken from WARD, which I discussed in detail in chapter 4, with a novel survey experiment. In the following sections I describe the research designs used to validate these hypotheses and present the results. Overall, the findings support the claims advanced in chapter 2.

Research Design

Each hypotheses stated in chapter 2 requires a distinct analysis. In order to examine Hypotheses 5 and 7, which focused on the effects of female combatants on rebel troop mobilization and external support, respectively, I rely on conventional regression analyses of cross-sectional data that take the rebel group as the unit of analysis. In both cases, I use data from WARD to measure variation in the presence and prevalence of female combatants across the rebel groups in the sample. While WARD measures served as the dependent variables in chapter 4, in this chapter they become the primary independent variables of interest in models predicting the number of troops under a rebel group’s command and the likelihood that a group successfully obtains support from transnational actors. I examine the validity of the relationship between external audience attitudes toward a rebel group and the presence of female fighters, which was formalized in Hypothesis 6, using data from a novel survey. Because each analysis employs different variables and utilizes different statistical methods, I discuss each in more detail below.

Recruitment, Replacement, and Rebel Military Capabilities

Hypothesis 5 asserts that rebel groups that incorporate larger numbers of female fighters into their ranks will be able to field larger and more capable combat forces compared to those that eschew women’s participation. The logic for this argument is straightforward: increasing the pool of potential recruits should expand the number of troops under a group’s command. To test this hypothesis, I rely on ordinary least squares (OLS) regression to evaluate the relationship between an estimate of rebel group troop size and a number of covariates. In this analysis, rebel troop size reflects the natural log of the estimated average number of troops available to a given rebel group as reported in the Non-state Actor (NSA) Dataset (Cunningham, Gleditsch, and Salehyan 2009, 2013). Because the hypothesis explicitly posits that increasing the proportion of female fighters in a rebel group should increase its overall size and material capability, I use the variable female combatant prevalence as the primary variable of interest in each model. As discussed in chapter 4, this categorical indicator represents a coarse estimate of the proportion of the rebel combat force that was composed of women.

In chapter 4 I also discussed the various alternatives measures of female combatant prevalence used in WARD, including both a “high” estimate and a measure excluding cases in which female fighters were strictly limited to suicide bombings (e.g., Hamas, the PIJ, and Boko Haram). In addition to the “best” estimate of this indicator, I also assessed models using these alternative estimates. Assessing the influence of the measure that excludes cases where female combatants were restricted to only the role of suicide bombers is important because the inclusion of these cases might bias the results against the hypothesized relationship. Specifically, suicide bombers are comparatively rare, representing only a handful of any given group’s overall troop numbers. The inclusion of women in only these roles would therefore result in a negligible expansion of the overall force, though it might spur recruitment overall by shaming reluctant men to join the movement.

I also account for a number of confounding variables in the models. First, I account for whether or not the group was involved in a secessionist conflict with the state. Such conflicts typically pit an aggrieved minority ethno-nationalist group against the state. Previous studies suggest that the combination of shared cultural identity and collective grievances facilitates political mobilization and that political conflicts centered on identity issues escalate more rapidly (e.g., Eck 2009; Gurr 1993). As such, rebel groups involved in secessionist conflicts have a built-in mobilization advantage, which may allow them to field more troops than rebellions that mobilize along other dimensions. On the other hand, because the autonomy-seeking group is often a numerical minority in the state, they might ultimately field smaller forces even if they are more easily able to mobilize troops. I am thus agnostic about the effect of this variable. It is a binary indicator coded 1 if the conflict focused on claims of territorial autonomy, and coded 0 if otherwise. Data for this indicator come from the UCDP Dyadic Dataset (Harbom, Melander, and Wallensteen 2007).

I also include the variables population size and GDPpc to account for state-level structural factors that might influence rebel mobilization capacity. I include the former because rebel groups involved in conflicts in countries with large populations should theoretically have access to larger a pool of potential recruits. As such, these countries might be more likely to host larger rebel movements. The latter variable is included because wealthier states possess greater resources with which to suppress rebellion and/or to address the grievances that drive it. Both mechanisms should theoretically suppress the supply of available recruits. Wealthier countries should therefore on average host numerically smaller and comparatively less capable rebel groups. The respective variables reflect the natural log of the average size of the population and average estimated per capita GDP of a given country during the years in which the group was active. Data for both variables are taken from Gleditsch (2002).

I control for the state’s political institutions because repressive, autocratic states are more likely to create grievances among the population and to deny them legitimate means through which to address those grievances compared to democratic states. The presence of democratic institutions should help minimize support for the rebellion because individuals can express their grievances at the ballot box rather than in the risky act of rebellion. Much as rebellion is less common in democracies, rebellions that occur in democratic states should be comparatively weaker than those that arise in autocratic states. The binary variable democracy indicates whether or not the country in question had a (nominally) democratic political structure for the majority of the years during which the group was actively involved in a violent conflict with the state. I use the democracy variable from the “Democracy and Dictatorship” dataset (Cheibub, Gandhi, and Vreeland 2009). According to this measure, democratic states hold routine and competitive direct or indirect multiparty elections for the executive and legislature. I create this variable by taking the modal value of the government’s democracy score for the years in which the group was involved in armed rebellion against the state.1 I also include a control for the size of the state military force. State troop size reflects the natural log of the average estimated number of personnel in the state’s military force during the years in which it was involved in armed conflict with the rebel force. This variable is taken from the National Material Capabilities Dataset, v. 5.0 (Singer, Bremer, and Stuckey 1972).

Forced recruitment is a binary indicator accounting for whether or not abduction or other forcible recruitment strategies were ever employed during the conflict in which the group fought. I include this measure because the use of these tactics is explicitly intended to increase the number of troops available to rebels and is frequently employed when the resource demands experienced by the group outpace the inflow of voluntary recruits. As such, it represents an alternative (or perhaps complementary) strategy to female recruitment. Moreover, as noted in chapter 4, groups that engage in forced recruitment are more likely to include female combatants. Including this variable helps to account for the possibility that forced recruitment rather than the decision to recruit female combatants drives any observed relationship between the prevalence of female combatants and the military strength of the rebels. Overall, and to the extent it is successful, rebels that engage in forcible recruitment should be able to field larger numbers of troops than those that do not. As discussed in chapter 4, this measure is adapted from Cohen (2013a).

The binary variable rebel external support indicates whether or not the rebel group under observation received military or nonmilitary support from a foreign state at any point during the conflict. These types of support include the provision of troops, military hardware or weapons, financial support, logistical support, the use of territory, and other types of assistance. I include this variable because external support can provide rebels with the equipment, knowledge, and space to recruit, train, and arm troops. These assets should facilitate the expansion of the rebel force. I likewise include the measure transnational constituency, which indicates that a nonstate transnational entity provided some form of tacit or explicit support for the rebel group.2 These constituencies are often diaspora communities that fund-raise on behalf of their ethnic kin and provide them additional tangible and intangible benefits. However, they also include global religious communities and international solidarity networks (Salehyan, Gleditsch, and Cunningham 2011).3 Groups that received support from such actors are coded as 1, and groups that received no such support are coded as 0. Both indicators of external support are taken from the NSA Dataset (Cunningham, Gleditsch, and Salehyan 2009, 2013)

Finally, I include a control for the length of time the group was actively involved in violent conflict with the state. The variable duration is the log-transformed value of the number of years between the group’s initiation of violent conflict with a government and its cessation of those hostilities as reported in the UCDP Dyadic Dataset. I include this variable because rebel groups tend to begin life from a position of relative weakness and expand their capabilities over time (provided they survive) (e.g., Bapat 2005; Cunningham, Gleditsch, and Salehyan 2009). Thus, duration should be positively correlated with a group’s overall size and mobilization capacity.

I present the results from a series of OLS models in table 5.1. Overall, these results support the contention that rebel groups that recruit female combatants ultimately field larger combat forces than those that exclude them. Across the models, the coefficients on the various versions of female combatant prevalence are positive and statistically significant, suggesting that groups including larger proportions of female combatants on average have larger numbers of troops than those that have fewer or no female combatants. This result is intuitive since removing artificial constraints on the supply of troops should facilitate mobilization. Assuming that significant numbers of male recruits do not eschew recruitment explicitly because the group contains female combatants, a group composed of 10 percent female combatants would arguable have foregone (more or less) that proportion of their force had they decided against recruiting women into the group’s combat force.

It is possible that some number of male recruits might abandon the group or refrain from joining due to the presence of female fighters in the ranks. Yet, the anecdotal evidence discussed in chapters 2 and 3 suggests that, if anything, the presence of female combatants serves as an important recruitment device for armed groups. There is little evidence for a contrary, dampening effect of female combatants on recruitment. In addition to their direct influence on rebel troop size, female combatants are also likely to indirectly benefit rebel recruitment efforts. Principally, the visible presence of some number of women in combat roles—even in relatively small numbers—is likely to compel reluctant men to join the rebellion. Their presence may also exert a demonstration effect that generates additional female recruitment. Regardless of the specific mechanism, these results suggest that the decision to recruit female combatants on average increases the number of troops available to the rebel movement. This provides support for Hypothesis 5.

In order to better understand the substantive impact of female combatant prevalence on rebel troop size, I present the predictions from Model 4 in figure 5.1.4 The figure illustrates the predicted value of rebel troop size (y-axis) over the categories of female combatant prevalence (x-axis). The results suggest that a one-category increase in the measure of female combatant prevalence is expected to produce a change of 0.244 in the logged troop estimate. An average group with no female combatants would be expected to have an estimated logged troop value of approximately 8.04 (~3,100 troops); however, increasing female combatant prevalence to the low, moderate, and high categories increases the expected troop size to roughly 8.29 (~4,000 troops), 8.53 (~5,100), and 8.78 (~6,600), respectively. These results imply that the effect of incorporating female combatants on rebel troop size is nontrivial. Consistent with the motives that prompted their recruitment, female combatants appear to produce more robust rebel groups, at least in terms of troop size.

Table 5.1 Effect of Female Combatants on Rebel Troop Size

	Model 1	Model 2	Model 3	Model 4
Female combatant prevalence (best)	0.370 (0.098)**	0.224 (0.091)**
Female combatant prevalence (high)			0.252 (0.086)**
Female combatant prevalence (excluding suicide bombers)				0.244 (0.091)**
Secessionist	0.495 (0.264)*	0.011 (0.294)	−0.045 (0.289)	0.012 (0.290)
Population size†	−0.301 (0.113)**	−0.198 (0.108)*	−0.186 (0.107)*	−0.192 (0.105)*
GDPpc†	−0.558 (0.141)**	−0.529 (0.155)**	−0.548 (0.151)**	−0.518 (0.151)**
Democracy	−0.629 (0.241)**	−0.550 (0.286)*	−0.532 (0.282)*	−0.561 (0.285)*
State troop size†	0.246 (0.108)*	0.128 (0.087)	0.127 (0.086)	0.127 (0.086)
Forced recruitment		0.421 (0.219)*	0.381 (0.213)*	0.426 (0.219)*
Rebel external support		0.619 (0.199)**	0.627 (0.195)**	0.608 (0.199)**
Transnational constituency		0.445 (0.222)*	0.462 (0.219)*	0.467 (0.221)*
Duration†		0.236 (0.093)**	0.212 (0.096)*	0.225 (0.092)**

F	6.09	12.26	12.49	12.12
R2	0.22	0.37	0.38	0.37
BIC	747.97	655.07	650.90	653.76
N (groups)	210	192	192	192

**p < 0.01

*p < 0.05 (one-tailed test)

†natural log

Note: Coefficients from ordinary least squares with standard errors (clustered on conflict country) in parentheses.

Figure 5.1 Predicted effect of female combatant prevalence on rebel troop size

While figure 5.1 shows the predicted changes in troop values over the categories of female combatant prevalence, it reveals little about the effect of recruiting female combatants on overall rebel troop strength relative to other recruitment strategies. As I argued in both chapters 1 and 2, recruiting female combatants represents one of several potential mobilization strategies by which rebel leaders seek to meet their human resource needs, expand the size of the rebellion, and increase the odds that the group achieves its political objectives. Forcible recruitment represents another strategy rebel leaders routinely employ to acquire human resource inputs during periods of excess demand.5 As such, it is informative to compare the effect of forced recruitment and female combatant recruitment on rebel troop size.

In order to evaluate the comparative effects of these strategies, I rerun Model 4 but replace the prevalence variable with a binary variable indicating whether or not the group utilized female combatants. By using binary indicators as proxies for each recruit strategy, I can more easily compare their effects on troop size. The results imply that the effects are very similar: the coefficient for the binary female combatants variable is 0.490 (p = 0.002) while the coefficient for forced recruitment is 0.419 (p = 0.028). Furthermore, the predictions from the model suggest that groups that engage in forced recruitment have approximately 5 percent more troops than those that do not utilize this strategy.6 By comparison, groups that recruit female combatants have approximately 6 percent more troops than those that fully exclude women from combat. Thus, the substantive effect of these two recruitment decisions is quite similar. Recall, however, that the results presented in the last chapter suggest that forcible recruitment is correlated with the prevalence of female combatants. This highlights the potential similarity in motivations behind the strategies. Rebel movements recruit female combatants and engage in forcible recruitment in order to secure human resources, and these results suggest that both represent nominally effective strategies in this respect. Nonetheless, the evidence suggests that they represent two distinct strategies, and the presence of female combatants is not contingent on the use of forcible recruitment.7

Turning to the other control variables, secessionist conflict is significant in a single model and switches signs across the specifications. This suggests that groups engaged in separatist conflicts are numerically no larger and no more militarily capable than their center-seeking counterparts. GDPpc is also negative across the specifications and statistically significant throughout. This implies that rebellions in wealthier states are on average smaller and less militarily capable than those in poor states. This result is not surprising given that wealthier states are more likely to possess the resources to effectively suppress rebellions and can devote substantial resources to addressing the grievances that drive political violence and incentivize rebel recruitment in the first place (should they choose). Democracy is also negative and significant across the models, suggesting that rebels challenging nominally democratic regimes are generally numerically smaller than those fighting nondemocratic governments. As with the GDP indicator, the result is intuitive.

By contrast, population size is surprisingly negative across the specifications. It achieves significance in each of the models. While the result is unexpected, it is possible it obtains because larger populations allow governments, which are often better able to mobilize resources (often by compelling service), to expand their military capacity rapidly when needed. If this were the case, we would expect rebel forces to be comparatively smaller in states with larger militaries. In other words, states could convert population size into coercive power, thereby suppressing rebellion. The results, however, do not appear to support this claim. State troop size is positive across the models but only achieves statistical significance in one of them, suggesting that the size of the state’s military has limited influence on the size of the rebel group. This ambiguous relationship likely reflects the interdependent nature of the variables: states increase the size of their military in response to rising rebel threats.

The results suggest that external support provided either state or nonstate actors is positively correlated with the size of the rebel group. Moreover, rebel external support and transnational constituency are both significant across the models in which they are included. These results are largely anticipated since groups that receive material support are more likely to possess resources that enable them to effectively recruit from the population. Previous studies have also proposed that a group’s ability to secure support from external actors—particularly international IOs and NGOs—is often perceived as a signal of legitimacy by domestic constituencies (e.g., Bob 2005). To the extent this is true, groups with such support may enjoy superior mobilization capabilities, resulting in larger troop sizes. Lastly, duration is positive and significant across the models, suggesting that the size of rebel movements tends to increase over time—or at the least that longer-lived rebellions field larger numbers of troops. This is unsurprising given that potential supporters are more likely to mobilize in support of a group once it has demonstrated its ability to survive and impose costs on the state.

These results provide support for Hypothesis 5 and are consistent with the primary arguments made in chapter 2: rebel groups that recruit female combatants are better able to address their human resource needs and ultimately are able to amass larger combat forces than those that exclude women. An admitted limitation of this analysis is that the data are not time-series and therefore cannot accurately assess the causal relationship proposed in Hypothesis 5. Specifically, it is possible that the groups in the sample that ultimately elected to recruit female fighters already fielded larger than average combat forces. If this was the case, the relationship observed here might be spurious or stem from endogeneity among the variables. In light of this limitation, these results should be treated as preliminary and should be reevaluated in the advent of data that can better account for the causal ordering posited in the argument. Nonetheless, they are consistent with both the argument presented in chapter 2 and the anecdotal evidence presented in chapter 3.

In the following sections I address additional arguments made in chapter 2. Specifically, I evaluate the effect of female fighters on audience attitudes toward the groups that employ them. I also consider whether the presence of female fighters increases the likelihood that a group actually receives support from external actors.

Female Fighters and Audience Attitudes

Hypothesis 6 in chapter 2 asserted that the presence of female fighters increases domestic and international audiences’ overall sympathy and support for the rebels and their goals. In order to assess the validity of this claim, I rely on evidence from a survey experiment intended to examine the effect that awareness of the presence of female combatants in a hypothetical armed group exerts on a respondent’s perceptions of the group. The instrument used in the experiment was designed in collaboration with Devorah Manekin, to whom I am indebted for allowing me to share a portion of our work here. The survey experiment was deployed in multiple waves and to multiple samples of respondents.8 A pilot version was deployed in spring 2016 to a sample of 450 undergraduate students enrolled in online classes at Arizona State University (ASU). After making very minor modifications in the wording of the treatment used in the pilot study, the survey was again deployed to a national sample of more than 1,800 online respondents in summer 2017.9

While neither sample is completely representative of the demographics of the U.S. population, both nonetheless capture substantial variation in the demographic indicators normally viewed as relevant to experimental studies of this type (e.g., gender, age, race, and so on).10 Most importantly, in each wave of the survey the treatment was randomized, ensuring that the individual characteristics of the respondents had no influence on the likelihood that a given respondent received the treatment. Moreover, in both cases there is nominal balance across the relevant demographic categories, indicating that a roughly equal number of male/female, white/nonwhite, and liberal/conservative respondents received the treatment versus the control. Because the experiment utilizes a convenience sample of U.S. respondents, the results are not necessarily representative of populations in other locations. They nonetheless suggest that respondents viewed the hypothetical armed resistance movements somewhat more favorably when they were made aware that the group included a substantial number of female combatants. This result provides some preliminary support for arguments regarding female combatants and (external) audience support.

The experiment consisted of a simulated news article describing a fictional conflict involving an armed self-determination movement named the Kaftarian National Front (KNF). After completing a battery of demographic and individual difference measures and reading introductory text describing the survey and providing instructions, all participants received the same narrative describing a rebel group attempting to establish an independent homeland. Specifically, subjects received the following description of the conflict:

For several months, the Kaftarian National Front (KNF), a separatist movement located in South Asia, has been engaged in a violent struggle against the government that has claimed hundreds of lives on both sides. The conflict has rapidly escalated in recent months, and experts believe it has the potential to spill over into neighboring countries, potentially destabilizing the region and creating severe humanitarian costs.

Human rights organizations have expressed serious concerns over the deteriorating human rights situation in the country. They accuse the government of committing many serious human rights abuses and using harsh counterinsurgency tactics that violate international humanitarian law, including the execution of civilians suspected of supporting the KNF and the indiscriminate bombing of villages in KNF-held areas.

Participants were then randomly assigned to an article that described either a mixed-gender rebel group in which one-third of the fighters were reportedly female or to a nearly identical article including only references to male combatants. Nearly identical photographs of a group of fighters accompanied the articles. Specifically, the photograph accompanying the article describing the mixed-gender group included a female combatant in the frame, while the female fighter was cropped from the frame of the photograph included in the all-male article. Thus, the photographs provided to the treatment and the control groups differed only in terms of the inclusion of a female combatant.

Compared to the argument put forth in chapter 2, the treatment used in this experiment is extremely subtle. Indeed, with the exception of hair length and facial features, the male and female combatants in the photos are extremely similar. Both assume the same poses, have similarly stoic expressions on their faces, carry the same weapons, and wear the same uniforms. Yet, propaganda materials and media coverage are often intended to promote audience interest, and the former are explicitly intended to shape the attitudes of the observer. As discussed in chapter 2, such materials therefore highlight the exceptional nature of women’s participation in combat and attempt to frame their participation in ways that maximize its impact on audience perceptions, including explicit efforts to accentuate and discuss the feminine characteristics of female fighters. The experiment used here provides the audience only with knowledge that female fighters participated in the rebel force in substantial numbers and provides an image including a single female combatant. It thus allows respondents to interpret their participation independently. The virtue of this more subtle treatment is that it permits a clear test of the underlying impact that the simple presence of female fighters has on audience attitudes rather than the effect of any additional persuasion or contextualization.

Hypothesis 6 posits a direct and unconditional relationship between female combatants and audience attitudes; however, it is possible that different characterizations of the movement (and the combatants included in it) also shape audience attitudes. Particularly, as discussed in chapter 2, the presence of female combatants might ameliorate some of the adverse effects of negative characterization of the group. This would be indicative of success in rebel efforts to create a more sympathetic counternarrative in the face of government efforts to characterize the group as nothing but thugs and terrorists. In order to examine this conditional effect, the survey experiment included an additional manipulation. In addition to random assignment to gender treatment condition, respondents were also randomly assigned to one of two conditions describing rebel interactions with the civilian population.11 In the first, the rebel group was described as having positive relations with civilians, including building schools and operating medical clinics. In the second, the group was characterized as abusive toward civilians, engaging in looting, forced recruitment, and other abuses. The addition of this condition provides insight into whether negative descriptions of the group’s behavior condition any effects that the presence of female combatants might exert on audience perceptions of it.

The dependent variable used in the analyses is a categorical measure of respondents’ self-reported support for the group and its goals. Support is a five-point categorical measure based on responses to the following question: “Based on the information you read, to what extent do you support or oppose the KNF and its goals?” The scale provided to the respondents ranged from “strongly support” to “strongly oppose.” The variable is reverse coded in the analysis for ease of interpretation: higher values of the dependent variable reflect greater support.

The variable reflecting the treatment condition for female combatants is labeled mixed gender, while the condition describing whether or not the group abused civilians is labeled abusive. Because I am primarily interested in the effect of the gender treatment, I focus primarily on it and view the other treatment as a moderating variable that potentially conditions the relationship between the gender treatment and audience attitudes. I first present a simple difference of means tests for mixed gender and then present results from a series OLS models.12 I include both treatment conditions in each of the regression models. In addition, in one specification for each sample I include an interaction term created by multiplying the two treatment variables. In these models, I also include controls for relevant demographic and attitudinal characteristics that might influence respondent support for the group. The first of these is respondent sex, which is coded 1 if female, and 0 otherwise. I include the variable race, which is coded as 1 if the respondent identified as white and 0 if they identified as any other racial group. Age is a categorical indicator coded by decade. Education is a six-category indicator ranging from “did not complete high school” to “post-graduate degree.” Because the student sample is composed only of respondents enrolled in university courses, I exclude this measure from the analyses of the pilot study. Liberal is a binary indicator denoting that the respondent identified as liberal.13 Income is a four-category indicator reflecting self-reported household income ranging from less than $25,000 to greater than $75,000.14

The results from the experiment, which I present in table 5.2,15 provide preliminary support for one of the central arguments explicated in chapter 2: female fighters positively influence audience attitudes toward the groups that employ them, making them more supportive of the group and its goals. Models 7 through 9 report results for the pilot study conducted on the student sample, while Models 10 through 12 report the results for the larger national sample. The results of both simple t-tests (Models 7 and 10) and the OLS models including a number of control variables (Models 8 and 11) reveal a positive and statistically significant treatment effect for mixed gender in both the student sample and the national sample. In both samples, the respondents who received the treatment condition highlighting the presence of female combatants in the hypothetical armed group were more favorably inclined to the group and its goals compared to those who received the description of the group that did not reference female combatants. Overall, the results therefore support the argument that the presence of female combatants can positively influence audience attitudes toward the group and its goals.

Table 5.2 Effect of Female Combatants on Audience Attitudes

Student sample

National sample

Model 7

Model 8

Model 9

Model 10

Model 11

Model 12

Mixed gender

0.327

(0.102)**

0.317

(0.100)**

0.103

(0.140)

0.086

(0.052)*

0.086

(0.049)*

0.092

(0.070)

Abusive

−0.485

(0.100)**

−0.696

(0.140)**

−0.609

(0.049)**

−0.604

(0.070)**

Mixed gender*

abusive

0.426

(0.198)*

−0.012

(0.099)

Sex (female)

0.088

(0.104)

0.080

(0.103)

−0.036

(0.051)

−0.036

(0.051)

Race (white)

−0.077

(0.104)

−0.071

(0.102)

0.079

(0.057)

0.079

(0.057)

Age

–0.011

(0.073)

–0.010

(0.073)

−0.056

(0.015)**

−0.055

(0.015)**

Education

0.022

(0.020)

–0.022

(0.020)

Liberal

0.197

(0.100)*

0.198

(0.100)*

0.104

(0.053)*

0.104

(0.053)*

Income

−0.005

(0.042)

−0.003

(0.041)

−0.019

(0.024)

−0.019

(0.029)

5.86

5.76

21.48

18.09

0.09

0.10

0.09

BIC

1,280.5

1,281.85

5,098.69

5,106.58

429

1,741

*p < 0.05 (one-tailed test)

Note: Difference in sample means (t-test) (Models 7 and 10) and coefficients from ordinary least squares regression models (Models 8–9 and 11–12) with standard errors in parentheses.

The other treatment condition, abusive, is also statistically significant across the models. This result (unsurprisingly) indicates that respondents were less likely to express support for the hypothetical armed group when it was described as acting violently toward the local population. Interestingly, the effect of the interaction of the treatment conditions varies across the two samples (Models 9 and 12). In the student sample, the interaction is significant and positive, while it is negative and insignificant in the national sample. These results therefore provide mixed evidence for the potential influence of female combatants to ameliorate the adverse influence of negative characterizations of the group on observer support for it.

In order to more clearly illustrate the effect of the treatments and to more closely examine the conditional effect of abuse on the relationship between the presence of female combatants and respondent attitudes, I present the marginal effects from the models in figures 5.2 and 5.3. Figure 5.2 shows a direct relationship between the presence of female combatants and respondent support for the group and its goals based on the results from Models 8 and 11. The figure illustrates the effect of mixed gender (y-axis) on the level of the respondent support (x-axis). The left panel of the figure shows the effect within the student sample while the right panel shows the effect within the national sample. As noted above, in both samples the treatment effect is significant, with those respondents exposed to the conditions highlighting the presence of female combatants expressing greater support for the group and its goals. However, the scale of the effect differs across the samples. In the student sample, exposure to the treatment increased support by roughly a third of a point (on the five-point scale) while in the national sample it increased respondent support by only approximately one tenth of a point. While the effect size is marginal, the effect is nonetheless discernible. Moreover, the effect size is both unsurprising and largely consistent with the argument.

Figure 5.2 Predicted effect of mixed-gender treatment on respondent support

In a real-world setting, respondents are exposed to the presence of women in war zones only in passing or only for brief moments, such as when they skim a news article online or watch a segment on cable news. A single, limited instance of exposure to an image of a female combatant would not be expected to produce a substantively large effect on respondent sentiment. However, repeated exposure to such images might have such an effect. Indeed, if the initial exposure to such images or information pique observer interest or provoke sympathy, it may encourage that observer to seek out additional information on the case or to devote greater attention to subsequent media reports about it. In this sense, initial exposure may lead to additional exposure, which may in turn lead to a deepening of the initial sympathetic sentiment. The design of this study prevents testing this speculative corollary hypothesis, but attention to this question might be warranted in future studies. For now, it is sufficient to say that the results provide some support for Hypothesis 6.

Figure 5.3 Conditional effect of mixed-gender treatment on respondent support

I report the predicted effects of the interaction of the treatment conditions in figure 5.3. They provide mixed support for the proposition that female combatants exert an observable ameliorative influence on the adverse influence of negative descriptions of the group on respondent attitudes. As in the previous figure, the left panel illustrates the effect in the student sample while the right illustrates the effect in the national sample. Both panels demonstrate that respondents exposed to the abusive condition expressed lower support for the group, regardless of whether they were exposed to the mixed gender treatment condition. Moreover, the effect of abuse is significant regardless of exposure to the gender treatment. However, at least in the case of the student sample, the effect of mixed gender varies across the other treatment condition. Respondents exposed to the female fighter narrative and image (triangle) were generally more supportive of the group than were those exposed to the male-only narrative and image (circle), and the difference between the two is substantially larger in the presence of the abuse condition. Specifically, while the description of abuse drove down support, for both all-male and mixed-gender rebel groups, the decline is much steeper in the case of the hypothetical all-male group. In other words, there is some evidence that female fighters offset the negative characterization of the group. This effect is not apparent in the national sample. Consequently, these results provide only limited support for this assertion. Nonetheless, as a whole the results provide support for the arguments presented in chapter 2.

Securing External Support

Hypothesis 7 asserts that rebel groups that include female fighters are more likely to receive support from external actors than those that exclude women. The argument that produced this hypothesis focused primarily on the ability of female fighters to shape external audience perceptions of the group and its goals. This claim received some support from the previous set of analyses. Rebels often strategically highlight the presence of female fighters in their ranks in their propaganda materials and public relations efforts. More specifically, by focusing on (and sometimes exaggerating) women’s participation in the group, rebels hope to present a more positive image to both domestic and international audiences and to counter the government’s ubiquitous efforts to characterize its goals as illegitimate and to portray the movement as a band of thugs and terrorists. Additionally, owing to their (perceived) novelty, the presence of female fighters is more likely to attract attention from these audiences, thereby increasing the visibility of the group and its goals. To the extent that the presence of female fighters assists rebel groups in garnering sympathy, interest, and support for its goals from outside observers, it might also increase the odds that the group receives tangible benefits or tacit support from transnational actors such as diaspora communities, NGOs, or other solidarity organizations.

To my knowledge there has been no effort to quantitatively assess the factors that predict transnational nonstate actor support for rebel groups.16 Moreover, few cross-national measures of external nonstate actor support for rebel movements exist. However, the NSA Dataset includes the variable transnational constituency, which I discussed briefly above. Given the dearth of systematic data on this topic, this variable arguably represents the best available proxy for the phenomenon of interest. In the following analyses, I use two versions of the basic variable. The first accounts for both tacit and explicit support. Tacit support generally includes statements of sympathy or solidarity by a community or activist group and may include lobbying, protests, and other demonstrations of support on behalf of the group. Explicit support indicates that transnational actors provided material support to the rebels, such as economic funding, supplies (e.g., food, medical supplies, clothing, and weapons), or fighters. The variable transnational constituency represents both tacit and explicit support and is coded 1 if the group under observation received any support from a diaspora community, transnational advocacy network, or other similar organization during its campaign against the state, and 0 otherwise. The related variable transnational constituency (explicit) is coded 1 for those cases that received explicit support, and 0 otherwise.

I also estimate one model that explicitly examines the effect of female combatants on the provision of material support to the rebellion by diaspora communities. While few good indicators of such support exist (presumably because such support is often covert), the UCDP External Support Dataset, v. 1.0 (Högbladh, Pettersson, and Themnér 2011) provides some information about such support. Using information from the dataset, I construct the variable diaspora support, which is coded as 1 for any group that received tangible support from a diaspora community composed of the same ethnic or national group as the rebels, and 0 otherwise. Because diaspora groups are almost exclusively composed of defined ethnic, religious, or national communities, I limit the sample of cases in this model to rebel groups advancing nationalist claims.17 This eliminates the large number of cases for which I would not anticipate the presence of a diaspora community of any kind.

The primary independent variable in these analyses is the binary female combatants measure. I present results using two versions of the measure, one that includes cases of female suicide bombers and one that excludes them. I include both because much of the previous research on this topic has focused heavily on the media coverage of female suicide bombers and audience responses to them. It is therefore useful to ensure that the relatively rare occurrence of women in such roles is not the driving force in any observed relationship between female combatants and external support. I rely on the binary indicator, which simply accounts for the presence or absence of female combatants in a group, rather than the prevalence measure largely because rebel propaganda efforts do not require large numbers of female combatants. Thus, a relatively small number of female fighters may suffice to attract support from an external audience. For instance, images of female fighters are common in the propaganda materials of the PFLP, FRELIMO, and the MPLA, even though these groups employed only small numbers of women in combat operations. Given that most observers are likely to be unaware of the true extent of women’s participation in a given group, reports of their participation may be sufficient to shape the perceptions of external actors. Nonetheless, clear evidence of the large-scale participation of women in the organization, as was the case with the EPLF, FMLN, and LTTE, may serve as an indication of the depth of support for the group within the base community and as a signal of its resolve to external audiences. I therefore estimate models using the prevalence indicator as a robustness check.18

I include a number of control variables in the analyses. First, I control for the relative strength of the rebel group. Previous studies have suggested that external actors are unlikely to extend support to groups that are particularly weak because they have a very low likelihood of victory (Gent 2008; Salehyan, Gleditsch, and Cunningham 2011). While this argument has previously been employed to explain support from external states, it likely applies to transnational advocacy networks as well. While these groups often support “underdogs,” they are nonetheless likely to consider the relative strength of an actor when deciding whether or not to invest resources to support it. Extremely weak actors with a high likelihood of defeat simply represent a poor investment of the external group’s finite resources. In addition, larger, more capable groups are more likely to have the capacity to engage external actors, solicit their support, and coordinate activities with them. For these reasons, I include the binary measure much weaker, which is coded 1 if a group was coded as such in the NSA Dataset, and 0 if otherwise.

I also account for whether the group exercised effective control over some portion of territory within the conflict state. Such control may encourage external support because it signals one type of rebel strength and facilitates rebel ability to mobilize resources away from government interference (Salehyan, Gleditsch, and Cunningham 2011). Territorial control is a binary indicator reflecting whether the group exercised a moderate or high degree of control over some territory outside of the government’s control within the conflict state. I further include the variable central control to account for whether or not the group had a clear central command structure that could exercise a high degree of control over the group’s forces. I include this measure because groups with more clearly defined command structures are likely to possess superior organizational capabilities and may be able to effectively utilize them to build relationships with transnational actors. Both measures come from the NSA Dataset.

I also control for the ideological orientation of the rebel group under observation. I include variables reflecting the ideologies used in previous analyses because transnational constituencies are often organized along ideological lines. For example, diaspora communities and kinship groups located in neighboring states are among the most common type of transnational constituency. Given that diaspora communities predominantly support the autonomy aspirations (or other goals) of their kinship group, I control for whether the group espoused a nationalist ideology. It is worth noting that transnational organizations focused on ending colonialism have also supported armed resistance movements that espoused a nationalist ideology. For instance, the ANC, SWAPO, FRELIMO, ZANU, and numerous other nationalist movements benefited from support provided by anticolonial solidarity movements located in the United States, Europe, and elsewhere. Indeed, nearly 60 percent of rebel groups in the sample that received support from transnational constituencies espoused some form of nationalist ideology. Global religious movements also represent a significant number of contemporary transnational constituencies (roughly 40 percent). For instance, transnational Islamist movements have often provided verbal support as well as material assistance to Islamist insurgencies around the globe. Somewhat surprisingly perhaps, the data suggest that leftist movements rarely receive support from a transnational constituency, accounting for only about 8 percent of the groups receiving such support. I therefore also include the control variables leftist and religious ideologies in the analyses.19

I also control for multiple state-level characteristics that potentially facilitate or impede rebels’ ability to secure foreign support. I include the variable democracy because transnational advocacy groups, especially those promoting democracy and human rights, may be reluctant to extend support to armed movements rebelling against democratically elected governments, even if they agree with some of their goals or share a common ideology. I thus anticipate that rebel movements located in democratic countries will be less successful in securing transnational support. As noted above, this binary indicator comes from the “Democracy and Dictatorship” dataset.

Finally, I include controls for development level and population size. The variable GDPpc reflects the natural log of the average per capita GDP of the conflict country. I include this measure because rebel movements located in wealthier countries may have greater access to technologies that allow them to more easily engage external communities. For instance, communication, travel, and the transfer of funds are often easier between individuals or groups located in more-developed countries compared to those in less-developed countries. I include population size because large populations may reduce the external visibility of individual groups. In a large country with many competing groups and causes vying for international recognition, it may be difficult for outside observers to effectively connect with and champion the cause of any one group. I therefore anticipate a negative relationship between population size and external support.

Due to the binary nature of the dependent variables, I employ probit models with standard errors clustered by country to account for error correlation across observations within the same country. Results from these regression analyses are reported in table 5.3. Models 13 through 15 report the results of models in which the dependent variable includes both explicit and tacit support from a transnational constituency, while Models 16 through 18 report results of models that include only explicit support in the dependent variable. Model 19 presents the results of the analysis using only the subset of nationalist rebellions and the measure of diaspora support as the dependent variable. Across all of the model specifications, the measures of female combatant presence or prevalence are positively signed and statistically significant. Notably, the coefficient for female combatants obtained from Model 15, which uses the version of the measure that excludes cases in which women were employed only as suicide bombers, is actually larger than the coefficient in Model 14. Consequently, it appears that the presence of women in more traditional combat roles exerts a substantively larger effect. Overall, however, these results suggest that groups that contain female combatants are more likely to garner the support of transnational actors. As such, these results support Hypothesis 6. As I discussed in chapter 2, rebel propaganda is often targeted at exactly such audiences, including solidary networks abroad and co-ethnic diaspora communities. The intention of these materials is to gain sympathy from external audiences, including sympathetic organizations and their members overseas, in the hopes of acquiring material or financial resources or lobbying and activism on their behalf. While fairly cursory, these results are largely consistent with those arguments.

Table 5.3 Effect of Female Combatants on Transnational Constituency Support

Transnational constituency

Transnational constituency (explicit)

Diaspora support

Model 13

Model 14

Model 15

Model 16

Model 17

Model 18

Model 19

Female combatants (best)

0.621

(0.189)**

0.533

(0.199)**

0.538

(0.184)**

0.408

(0.192)*

1.483

(0.755)*

Female combatants (excluding suicide bombers)

0.555

(0.208)**

0.399

(0.210)*

Nationalist

0.727

(0.229)**

0.733

(0.220)**

0.734

(0.218)**

0.584

(0.229)**

0.511

(0.246)*

0.509

(0.243)*

Leftist

−0.395

(0.254)

−0.458

(0.280)

−0.496

(0.291)*

−0.211

(0.314)

−0.342

(0.358)

−0.362

(0.367)

−0.222

(0.398)

Religious

0.825

(0.257)**

0.846

(0.260)**

0.936

(0.255)**

0.627

(0.247)**

0.511

(0.265)*

0.585

(0.252)**

−0.257

(0.639)

Much weaker

−0.410

(0.261)

−0.442

(0.272)

−0.453

(0.280)

−0.151

(0.291)

−0.314

(0.285)

−0.321

(0.292)

−1.078

(0.467)*

Territorial control

−0.069

(0.108)

−0.053

(0.106)

−0.058

(0.105)

0.082

(0.107)

0.105

(0.109)

0.102

(0.108)

0.115

(0.162)

Central control

−0.012

(0.111)

−0.009

(0.108)

−0.004

(0.109)

0.077

(0.132)

0.087

(0.136)

0.093

(0.137)

−0.289

(0.205)

Democracy

0.122

(0.215)

0.124

(0.220)

−0.160

(0.239)

−0.161

(0.242)

0.901

(0.632)

Population size†

−0.128

(0.067)*

−0.128

(0.070)*

0.006

(0.069)

0.005

(0.070)

0.224

(0.129)*

GDPpc†

0.236

(0.131)*

0.262

(0.132)*

0.356

(0.130)**

0.379

(0.131)**

0.111

(0.282)

Wald X2

47.05

64.18

66.09

33.65

58.64

53.18

14.80

N (groups)

325.39

329.24

329.84

261.51

267.67

268.42

89.01

BIC

250

248

250

248

111

**p < 0.01

*p < 0.05 (one-tailed test)

†natural log

Note: Coefficients from probit models with standard errors (clustered on conflict country) in parentheses.

Figure 5.4 presents the predicted probabilities of observing support from a transnational constituency. The left panel shows the predictions based on the results in Model 15, while the right panel shows the prediction from Model 18. According to the predictions, a rebel group with no female combatants in their ranks has roughly a 25 percent chance of receiving some form of support from a transnational nonstate actor. However, this probability increases to just over 45 percent for groups that include female combatants—thus, the presence of female combatants increases the chance that a group will secure external support by approximately 20 percent. The substantive effect of female combatants on obtaining explicit support, which includes material support such as funding and supplies, is smaller, but nonetheless apparent. Groups with no evidence of female combatants have about a 12 percent of securing explicit external support, while those with female combatants have a roughly 22 percent chance of acquiring this support. Thus, the substantive effect of female combatants on explicit support is about half of what it is for any form of support. While these results are preliminary, and should be replicated using more detailed data on the sources of support, they nonetheless suggest that female combatants exert a substantial impact on the group’s success in mobilizing transnational support.

Figure 5.4 Predicted effect of female combatants on transnational constituency support

While the results are consistent with the hypotheses and robust across a variety of model specifications, it would be difficult to argue that the presence of female combatants is the primary reason that diaspora communities or transnational solidarity networks begin supporting a rebel movement. The conditions that fundamentally determine the onset or existence of such support are more likely related to the strength of existing networks between the rebel group and those communities, the ideological interests of the actors, and their strategic objectives. However, these relationships have generally been assessed only in a descriptive and theoretical sense or through studies of specific cases (e.g., Bob 2005; Nepstad 2004). Additional research is therefore necessary to identify the factors that systematically influence the likelihood of transnational actor support for rebel movements. The arguments outlined in chapter 2 and the results presented herein identify one factor that potentially increases the likelihood that these groups extend and maintain their support. By casting the movement in a more sympathetic light and making their goals appear more legitimate, female combatants may, at the margins, influence the extent to which these networks voice support for armed resistance movements.

Before turning to the discussion of the control variables, it is useful to address one additional issue related to the relationship between female combatants and external support. The argument about rebel ideology presented in chapter 1 suggests that groups espousing fundamentalist religious ideologies and those that rely on constituencies that embrace traditional gender norms are less likely to recruit female combatants, even when this strategy might allow them to ameliorate resource demands. A similar relationship is likely to obtain with respect to the role that the visible presence of female combatants plays in helping rebel groups secure external support. Specifically, while images of women in arms and the perceptions of sacrifice associated with them might encourage support and sympathy from many audiences, these images would not necessarily be expected to prompt positive responses from audiences that view the presence of female fighters as antithetical to their worldview. Just as traditionalist domestic constituencies might rebuff groups that recruit women in large numbers, transnational actors that hold deeply conservative or fundamentalist beliefs would be unlikely to support foreign rebel groups that highlight women’s participation in the ranks.

To assess this corollary hypothesis, I rerun Model 15 above but include an interaction term composed of the female combatant measure and the religious ideology indicator. The expectation is that the effect of female combatants on transnational support is limited to groups that do not espouse religious ideologies. There should be no effect—or possibly a negative effect—for fundamentalist religious rebels. The results of these additional analyses are consistent with this expectation. While the interaction term is negative and insignificant, plotting the marginal effects of the interactions from these models reveals that the effect holds only in the case of nonreligious rebel organizations. As demonstrated in figure 5.5, for rebel groups that do not embrace fundamentalist religious ideologies, the presence of female combatants produces a statistically significant 20 percent increase in the probability that they will secure external support. By contrast, religious fundamentalist rebels receive virtually no additional benefits in securing support—the extremely wide confidence intervals around the coefficient estimate indicate that the effect is not significant. This result comports with the relationships identified in chapter 1 and provides additional evidence of the differential effects of female fighters in religious and nonreligious rebellions.

Figure 5.5 Marginal effect of female combatants on transnational constituency support by rebel ideology

Several of the control variables exert a consistently significant effect on the odds of external support; however, others appear to have a limited influence. First, the coefficient on much weaker is negative and statistically insignificant across most of the models. Thus, while the relative strength of a rebel group has previously been shown to predict support from foreign states, it does not appear to influence the odds that it receives support from transnational actors. Territorial control also fails to achieve significance in any of the specifications. This result is somewhat surprising, given that the ability to control territory should facilitate a group’s ability to build and maintain relationships with external actors. Central control also appears to have no statistically significant influence on the existence of transnational support for a rebellion. This might imply that external support is more a function of external actors’ basic impressions of the group and the conflict than the result of rebel leaders’ organizational capabilities and ability to cultivate networks outside of the conflict state. Democracy likewise fails to achieve statistical significance in the models and switches signs across the specifications. Thus, while rebel organizations may enjoy some benefits from the greater civil and political freedoms associated with more democratic states, these freedoms do not appear to enhance groups’ ability to solicit external support. Neither, however, does it appear that the presence of a democratic government significantly deters external nonstate actors from offering support for rebels.

While several other group-level predictors are insignificant, the ideological variables included in the model appear to exert significant and substantively meaningful effects on the odds that a group secures support. The results suggest that movements that adopt religious or nationalist ideologies are particularly likely to again support from a transnational nonstate actor compared to other movements.20 Global religious movements, particularly global Islamist movements in recent years, and diaspora communities are arguably more likely to support the struggles of their co-nationals or religious brethren abroad compared to other groups. Leftist groups, however, are apparently less likely to receive such support, though the variable only attains significance in a single specification. It is possible that global Marxism exerts a weaker influence on transnational activist support because leftist groups were already heavily predisposed to receive support from communist and socialist states, at least during the Cold War. A cursory review of major solidarity movements in the United States and Europe demonstrates that there were some efforts to support the struggles of leftist movements such as the FMLN. However, these efforts were dwarfed by the level of support provided to predominantly nationalist anticolonial movements such as ZANU, SWAPO, and the ANC.

Lastly, the results also suggest that rebels in wealthier states may be more likely to receive support from transnational actors than those engaged in conflicts in less wealthy states, while states with larger populations may be less likely to receive such support. GDPpc is positive and significant in four out of the five models in which it is included. This is perhaps because wealth facilitates communication and mobilization between activists and the conflict state and those located abroad. Finally, the effect of population size varies substantially across the models. While it is negative and achieves statistical significance in the models predicting any transnational support, it becomes positive and insignificant in the models predicting only explicit support. It then becomes positive and significant in the model using diaspora support as the dependent variable. It is difficult to explain what drives such changes, but in at least some contexts the variable appears to be a relevant predictor of constituency support.

Discussion and Conclusion

In chapter 2, I argued that recruiting female fighters and highlighting their presence and participation in the group benefits rebels in a variety of ways. First, it expands the supply of available recruits, allowing rebels to field larger, more capable military forces. Second, it helps rebels build sympathy and legitimacy in the eyes of external audiences, leading to increased support for the group and its goals. These factors should in turn increase the likelihood that the group receives support from external nonstate actors (e.g., diaspora communities or transnational organizations). While these benefits may be more likely to accrue to rebel groups that include female fighters, I do not argue that securing foreign support is a primary motive for the initial decision by rebels to recruit women and deploy them in combat. Rather, as I argued in chapter 1, a mixture of ideology and resource demands are probably the most influential factors in that decision. Nonetheless, once the group has incorporated women into its fighting force, its leaders often recognize the potential role the image of female combatants can play in garnering sympathy and support for the movement. Many rebel groups therefore strategically highlight female fighters in their propaganda and recruitment materials as a means of attracting external (as well as domestic) support.

The results presented above provide support for these arguments. First, I find that rebel groups that include larger numbers of female combatants are also more likely to field a larger number of troops compared to groups that exclude female combatants or recruit them in only very small numbers. Second, I find evidence that the presence of female combatants positively influences audience attitudes toward the groups that employ them. The results from a survey experiment designed to probe this relationship suggest that simply being made aware of the presence of female combatants in the groups improved respondent support for the group and its goals. Third, I argued that, given the influence of female combatants on observer attitudes, rebel movements that include female fighters should be more likely to attract support from transnational solidarity organizations and diaspora communities than those that exclude them. The results of multiple quantitative analyses support the arguments regarding the relationship between female combatants and external support for the rebel group. Specifically, they provide empirical evidence that transnational nonstate actors such as diaspora communities and international solidarity networks are more likely to extend support to rebel groups in which female fighters actively participate.

As a final note, it is important to reiterate that expanding the supply of available troops and acquiring foreign support are not end goals for these groups. Rather, successfully achieving these objectives is valuable to rebels because they shape the trajectory of the conflict and influence the group’s likelihood of survival and the odds that it will eventually achieve broader political and military goals. Thus, to the extent that female fighters assist the group in achieving these lower-order objectives, they should indirectly influence the group’s ability to survive and achieve its broader objectives. In other words, the argument and results imply that not only do female fighters provide specific strategic benefits such as a larger supply of troops and increase likelihood that the group receives external support, they may also indirectly influence the outcome of the conflict.