How does the location of rebel-held territory shape insurgent relations with civilians? We argue that when rebel groups control territory domestically, they are strongly incentivized to cultivate mutually beneficial relations with civilians living in their territory and limit their violence against them, while insurgencies with foreign territorial control are incentivized to deploy violence against civilians to gain compliance and extract resources. We test this hypothesis in three ways: a quantitative analysis of all insurgencies from 1989 to 2003 followed by a qualitative case illustration and the synthetic controls method that leverages the mostly exogenous acquisition of foreign territory by the Kurdistan Workers’ Party (PKK) after the 1991 establishment of the northern Iraq no-fly zone. Our results strongly support our hypothesis. These findings shed light on potential broader patterns of civilian victimization by insurgents, and the conditions under which insurgents may strive to limit civilian casualties and provide governance.
Since the 1970s and for decades beyond, the secessionist Oromo Liberation Front (OLF) cultivated both a robust military force and strong ties to the civilian population. Throughout OLF-controlled territory in the Oromia region of Ethiopia, it developed education and healthcare institutions, built roads, promoted agricultural development and created semidemocratic civil-military institutions organized around the Oromo concept of gada (Jalata 1993, 164–75; US Bureau of Citizenship and Immigration Services 2001). Aside from one political office in Mogadishu, the OLF lacked foreign bases (Cunningham, Gleditsch, and Salehyan 2012, 205), and enjoyed neither foreign sponsorship nor access to natural resources. Although the OLF regularly caused up to 1,000 battle deaths per year, the group rarely used direct violence against civilians.
In contrast, Boko Haram (primarily operating in Nigeria) massacred as many as 2,000 people in a single January 2015 attack.1 Previously, Boko Haram had killed over 5,000 civilians, including almost 2,000 people in 2014.2 Like the OLF, Boko Haram lacks foreign support or access to natural resources. But unlike the OLF, at the time of the attacks Boko Haram controlled little Nigerian territory and instead held bases in Cameroon, despite Cameroonian efforts to expel them.
Existing research on insurgent civilian victimization can easily explain the OLF’s behavior but not Boko Haram’s. Scholars have focused on in-group socialization processes (Cohen 2013a, 2013b; Humphreys and Weinstein 2006) or structural factors incentivizing the adoption of certain strategies (Azam 2006; De La Calle and Sánchez-Cuenca 2012; Eynde 2011; Humphreys and Weinstein 2006; Kalyvas 2006, 2008; Salehyan, Siroky, and Wood 2014; Weinstein 2006; Wood 2010). Nonetheless, despite the proliferation of research on this topic, the influence of the location of territorial control on insurgent violence remains largely unexamined.
Thus, we present a midlevel theory for how foreign territorial control causes increased rebel civilian victimization. We argue that the acquisition of foreign territory alters insurgents’ organizational composition and strategic calculus in two critical ways. Territorial control is a military asset that enhances the latent lethality of rebel groups, but foreign territorial control also incentivizes rebels to use violence, as opposed to governance or social service provision, as a means of extracting resources and generating civilian cooperation. Therefore, the location, as well as the mere existence of territorial control, is crucial to understanding civil conflict dynamics.
We test this hypothesis by evaluating the effect of foreign territorial control on the number of civilians killed by rebels in civil conflicts from 1989 to 2003, controlling for both insurgency-level and state-level characteristics. To improve causal identification, we present an instrumental variables analysis and a case study of the Kurdistan Workers’ Party (PKK) that combines qualitative evidence with a synthetic controls test. This leverages the PKK’s exogenous acquisition of foreign territory due to the implementation of the no-fly zone over northern Iraq. The exogeneity of assignment to treatment makes the PKK an excellent case to identify both the effect of foreign territorial control, and to highlight the causal processes undergirding subsequent civilian victimization.
Quantitative and qualitative evidence strongly supports our hypothesis. We find that foreign territorial control causes insurgents to kill twice as many civilians as the mean insurgent group.3 Not only does our study present a novel explanation for territorial control and civilian victimization, we make an empirical and methodological contribution by directly confronting endogeneity concerns and implementing both design- and model-based solutions that allow us to better identify the causal effect of foreign territory.
How Territorial Location Affects Civilian Victimization
Existing scholarship has identified two broad factors contributing to insurgent civilian victimization. The first set of explanations identifies in-group socialization processes (Cohen 2013a, 2013b; Humphreys and Weinstein 2006; Manekin 2013), while the second emphasizes structural factors that shape and constrain the strategies rebel groups deploy within a civil conflict (De La Calle and Sánchez-Cuenca 2012; Kalyvas 2006; Salehyan et al. 2014; Schneider, Banholzer, and van der Haer 2011; Weinstein 2006). Though the first set of explanations is important, we focus on structural level factors that incentivize or constrain rebel group victimization of civilians.
We argue that when insurgents acquire foreign territory–outside the state they seek to govern—it incentivizes them to wield violence against civilians to extract resources (e.g., materiel, compliance, and information). Conversely, when insurgents control territory domestically—within the state they seek to govern—they are incentivized to limit violence against civilians and establish governance to legitimate and regularize resource extraction. Although the mechanism behind this hypothesized effect is not unknown (see, e.g., Salehyan et al. 2014; Weinstein 2006), our theoretical framework contributes new analytical leverage allowing researchers to explain patterns of governance and civilian targeting in circumstances poorly understood by existing theories, such as rebels’ choice of governance versus civilian predation in refugee camps, or the Islamic State’s (IS) seemingly dual strategy of looting in some locations and governance in other, highly similar locales. In the following, we develop a comprehensive theory of how foreign territorial control produces higher levels of insurgent victimization of civilians. We argue that it is not simply control of territory that affects violence against civilians but, importantly, the location of this territory.
Two strategies of resource extraction
To acquire the resources insurgencies need to survive and succeed, rebels may use two strategies of resource extraction: governance to secure resources through civilian cooperation, or coercive extraction of resources, intelligence, and compliance. When insurgents provide governance and services, the group exchanges peace, security, and order (in addition to other services) for “taxes” from the population under the group’s control. Levi (1989, 52–55) labels this relationship “quasi-voluntary compliance”: to legitimize and regularize the transfer of resources from the same population over the long term, insurgent leaders provide services like justice or schools. In return, civilians voluntarily provide rebels with resources, broadly defined. Quasi-voluntary compliance gives rebels a reliable long-term stream of resources as an alternative to coercive extraction (Eynde 2011; Levi 1989, 54). Moreover, quasi-voluntary compliance legitimizes rebels’ political authority and may help them achieve their long-term goals (Fazal 2013; Mampilly 2011, 8; Stewart 2015). Thus, insurgencies such as the OLF, the Farabundo Martí National Liberation Front (FMLN), and the People’s Liberation Army (PLA), as well as many other leftist groups, have sought to cultivate quasi-voluntary compliance by limiting violence and providing services.
Despite the benefits for insurgents of quasi-voluntary compliance through governance, it entails sizeable up-front costs. First, it may be difficult to gain control of any territory without considerable costs in life and materiel (Wood 2010). Once insurgents establish territorial control, they need further resources and personnel to develop institutions to administer governance (McColl 1969, 614). Due to these high fixed costs, it may take considerable time before insurgents see a return on their initial investment. Thus, to generate quasi-voluntary compliance, insurgents must become “stationary bandits” who hold and invest in territory for the long term (Olson 1993).4
Violence against civilians represents an alternative strategy for rebels to extract resources and civilian cooperation (Stanton 2013) and is a strategy employed by both state (Valentino, Huth, and Balch-Lindsay 2004) and insurgent combatants. First, violence is much cheaper in the short term than quasi-voluntary compliance (Kalyvas 2006, 165): comparatively few resources are required to raid a village or extort people for information. Second, violence against civilians yields immediate benefits: instead of building schools and hospitals to win the hearts of the population, insurgents may take what they need at that moment and compel civilian cooperation on pain of death.
Although violence against civilians yields quick benefits at low cost, the “roving bandit” strategy also entails high long-term costs. If insurgents victimize civilians, civilian support for the group declines (Lyall, Blair, and Imai 2013).5 Additionally, killing civilians from rebels’ potential pool of recruits and resources significantly reduces future rebel strength and latent capacity over time (Eynde 2011). What this suggests is that although violence may quickly yield the resources rebels need, if this strategy is sustained over the long term, its effectiveness over time may decline.
Thus, in conflict contexts, insurgents are often incentivized to provide social services and limit civilian victimization to generate quasi-voluntary compliance. This allows rebels to access a steady stream of resources and recruits, engenders popular support and cooperation, prevents government collaboration, and may help insurgents achieve long-term strategic objectives. Yet quasi-voluntary compliance imposes high up-front costs, and rebels may only reap its benefits over the long term. Conversely, violence against civilians yields immediate access to resources and recruits at low cost, but this access will be inconsistent; violence may limit insurgents’ ability to achieve their goals over the long term. How insurgents extract resources and generate cooperation is thus a strategic choice between violence and quasi-voluntary compliance. Though rebels may switch between these two strategies during the course of conflict, in the following section we explain how the acquisition of two different types of territory—territory with the potential political community and territory without it—structures the incentives of rebel groups to favor one strategy over the other.
Resources in civil wars
As both political and military actors, insurgencies utilize a portfolio of different resources to achieve their goals. These resources range from the material—recruits (Berman and Laitin 2008; Weinstein 2006), territory, money, weapons, food, medicine—to the intangible—information (Berman, Shapiro, and Felter 2011), cooperation, and compliance (Kalyvas 2006).
Territorial control is one of the most important military resources insurgents can acquire, conferring a number of military benefits: Rebels have a space to train and hide. They can propagandize to attract recruits and support. They can stockpile their financial resources and weapons or develop alternative methods of financing. They can plan and prepare for attacks or attempt to cultivate (more) foreign backers. Territorial control is a tremendous military asset for insurgents, and control of this resource should enhance any group’s strength and latent capacity for violence (whether against civilians or the state).
Further, when members of a rebel group’s political community reside in rebel-controlled territory, this can yield the further benefits of consistent collaboration, cover, information, recruits, revenue, food, medicine, and weaponry.6 The political community is the group of people the insurgency claims to represent and aims to one day govern. For secessionist rebels, some part of the political community resides within an existing state. For center-seeking rebels, all people within the existing state are members of their political community (though they may favor particular groups within the larger community). The rebel group can use this community as a fixed and finite source of resources for the duration of the conflict. And while the resources that the political community possesses are important, extraction is not costless. If insurgents use force to extract resources, they shrink their pool of supporters, in turn reducing the number of resources the rebel group will be able to extract both in the present and over the long term. This discourages the use of violence and, in turn, encourages the establishment of quasi-voluntary compliance.
Conversely, when no members of an insurgency’s political community live in the territory it controls, the rebel group has unhindered access to the military benefits associated with territorial control. They are able to train, recruit, recoup, stockpile financial and military resources, plan, and equip themselves without any reciprocal commitments to their political community. Yet unless the insurgency is able to secure a committed and powerful foreign backer, without its political community, rebels will be able to acquire few other resources. Because no political community exists within their territory, rebels have little incentive to pay the costs of governance and are instead incentivized to act as roving bandits relying on violence to extract resources.
This suggests two types of territorial control: territory containing the political community and territory without this population. The two conditions imposed by the type of territory rebel groups control incentivizes insurgencies to adopt different strategies to extract the resources they need. Although ethnic kinship or ideology may help insurgents access these resources, ideology alone is often a weak motivator; instead, rebels rely on either governance and services or coercion to extract resources (Wood 2010). Thus, rebels rely on physical inducements—services and violence (or the lack thereof)—to secure the goods they need.
When insurgents control territory with the political community, they seek to hold the territory, extract resources from civilians and to establish control over them, while denying control to the government. To reliably access the resources from civilians without undermining long-term supplies of resources and recruits, rebels need (at minimum) the tacit consent of the political community living within that territory. This support may take time to generate, lengthening insurgents’ time horizons. Moreover, rebels hope to one day govern the potential political community, further incentivizing tactics of legitimation that require longer time horizons.
With longer time horizons, insurgents are incentivized to become stationary bandits (Olson 1993), minimizing civilian predations (Stanton 2013) and relying on a governing strategy that generates quasi-voluntary compliance to extract the resources (Levi 1989) associated with territorial control. By adopting this strategy, insurgents generate legitimacy with the population they seek to govern, allowing combatants regular and voluntary access to resources. Moreover, this form of territorial control is essentially necessary for rebel victory. Therefore, territorial control with the political community should be associated with reduced civilian victimization.
Territorial control with the political community typically occurs domestically, within the state that insurgents seek to govern. Thus, when rebels control domestic territory, they typically control territory with the political community. Though not universally true (e.g., refugee camps),7 we expect that insurgencies controlling domestic territory will be less likely to victimize civilians.
The case of the People’s Front for the Liberation of the Occupied Arabian Gulf (PFLOAG) in Oman from 1968 to the mid-1970s is instructive. The leftist PFLOAG launched its campaign in 1968, quickly seizing territory among peasants in rural Oman. To consolidate its control, the PFLOAG provided security, education, and medical services to local civilians. The population in turn sustained the PFLOAG with resources, recruits, and information (Halliday 2002, 37–90, 374–75, 385). Likewise, in the early years of the Syrian Civil War, the “moderate” Syrian opposition gained territorial control and began governing and providing social services in its domestic territory.8
Conversely, when rebel groups control territory without the political community, they access the military benefits associated with territorial control without the associated costs of developing a governance system to extract additional resources from civilians. In some cases, this territorial control without the political community may also facilitate foreign assistance such as food, medicine, weapons, and intelligence (Byman 2005, 2008). Yet international backing is not universal; even for insurgencies with a foreign backer, sponsors may find it difficult to consistently deliver resources to a war zone, or the sponsor’s domestic politics may suddenly halt state sponsorship.9
Therefore, when rebels control territory without the political community, insurgents enjoy the military benefits associated with territorial control but will face challenges in consistently cultivating additional resources typically provided by civilians, or fostering widespread collaboration with and control over the political community. While territorial control without the political community allows rebel groups unrestrained access to the benefits associated with territorial control, it alone cannot provide all the resources an insurgent group needs to thrive and succeed.
Instead, when insurgents need additional resources that the territory itself cannot provide, they may look to civilians elsewhere, using territory without members of the political community as an “exit option.” Because these civilians are not a fixed source of resources for the rebel group, the insurgency can use violence to quickly extract resources and compliance from one locale, before moving on to another. This mobility incentivizes insurgents to use coercion as a short-term means of securing resources and compliance (Levi 1989). Adopting violence to quickly capture resources causes greater civilian victimization. Because violence can foster control or elicit information, victimization may extend even to members of the political community. However, these insurgents have recourse to an exit option if they engender too much resentment in another area. With this form of territorial control, insurgents lack incentives to foster quasi-voluntary compliance to legitimize the extraction of resources from civilians over the long term (Beardsley, Gleditsch, and Lo 2015) and can raid town after town for goods before eventually returning to foreign sanctuaries.
Taken together, this suggests that insurgencies controlling territory without the political community are “roving bandits” (Beardsley, Gleditsch, and Lo 2015; Olson 1993). Because these roving bandits as insurgents are removed from their political community, they have no incentive to invest in long-term governing institutions and limit civilian victimization. Instead, because predation and coercion are comparatively cheap and efficient strategies that allow insurgencies immediate access to resources, rebels are incentivized to rely on violence over quasi-voluntary compliance as a strategy for extracting resources and cooperation. Once insurgents with this type of territory have extracted resources (e.g., recruits, weapons, money, information, food, or medicine), they are free to return to their base, unencumbered by any responsibilities to the civilian population. Finally, because insurgents lack information and control in the areas where they seek to govern, they may rely on indiscriminate violence against civilians, rather than networks of local collaborators (Berman, Shapiro, and Felter 2011; Kalyvas 2006), to secure information and compliance even within their own political communities.
The Islamic State’s (IS) behavior reflects these incentives. Currently, IS uses the depopulated desert spanning Iraq and Syria as a sanctuary. Some have argued that IS “prefer[s] to give up territory” and operate as “disparate mobile desert units that converge on a target when needed.” As a result, IS fights “as if they were pirates, with the desert being their sea,” treating “the cities and towns they have captured as ports of call, for booty and resupply,” and placing “more emphasis on cowing the populations they control through intimidation rather than adopting a conciliatory manner” (Kazimi 2015). IS relies on the desert as a form of territorial control without the political community, enjoying the military benefits associated with the mobility and relative safety the desert provides without any attendant civilian responsibilities. To extract necessary resources, IS victimizes civilians in towns and villages, even members of its political community.
Territorial control without members of the political community most likely occurs when rebel groups hold foreign territory.10 Foreign territorial control separates rebels from the population they seek to govern—the political community. Thus, rebels with foreign territorial control are incentivized to rely on coercion to extract resources from civilians, even sympathizers (e.g., coethnics, coreligionists, or peasants), before returning to foreign sanctuaries, free from civilian dependents. Ultimately, rebels with foreign territorial control both receive the military benefits associated with territory, increasing their latent lethality, and are also incentivized to use violence against civilians.
To summarize, insurgents may rely on two strategies—providing governance and services to generate quasi-voluntary compliance or coercive violence—to extract resources (e.g., territory, cooperation, information, food, weapons, medicine, recruits, money). Each strategy has certain costs and benefits, and the structural circumstances in which rebels operate alter the payoffs associated with each strategy, in turn incentivizing rebels to deploy one strategy over the other. When insurgents control the territory with their political community, they are incentivized to establish long-term extractive institutions. Moreover, by controlling territory with members of the insurgent’s political community, rebels interact with the very people they wish to govern over the long term. Thus, they tend to limit violence against civilians. Conversely, when insurgents control the territory without their political community, they lack these incentives, rendering coercion an attractive short-term resource extraction strategy for rebels with enhanced military capabilities but little prospect of secure relationships of quasi-voluntary compliance with civilians who are either outside their political community or beyond their territorial control.
As an illustrative example, the Liberation Tamil Tigers of Eelam (LTTE) began a campaign to create a separate Tamil state on the same island as Sri Lanka in 1983. In the early years of conflict, the LTTE failed to establish domestic territorial control and instead controlled foreign bases in India—that is, territory removed from its political community of Sri Lankan Tamils (Mampilly 2011, 94, 105). The LTTE of this time was known for its brutality and targeting of civilians, including coethnics, leading Mampilly (2011, 109) to conclude that “the LTTE sought to establish its dominion by repeatedly punishing those perceived to be sympathetic to [other] factions. This was especially true during the first four years of the war (1983–87), when the group controlled little territory” in Sri Lanka and was instead primarily based in India. Once the LTTE began targeting its Indian benefactors who had established a peacekeeping force on Sri Lanka in 1987, India withdrew its support of the LTTE, and the rebel group lost its foreign bases by 1990 (2011, 160). The LTTE faced a crisis: disband or alter its insurgent strategy. The LTTE turned to a strategy of quasi-voluntary compliance, occupying territory in Sri Lanka vacated by Indian peacekeepers, and reemerging with its first nonsecurity related governing institutions in 1994 (111), while limiting its civilian predations relative to its previous period of foreign territorial control (109). While merely illustrative, the case of the LTTE demonstrates how the nature and location of certain territorial holdings incentivizes rebels to rely on one method of resource extraction over the other, incentivizing violence against civilians when rebels control foreign territory and quasi-voluntary compliance when insurgents control domestic territory.
Thus, we expect to observe significantly higher civilian casualties caused by rebels controlling foreign territory, especially relative to insurgencies controlling domestic territory. More formally stated, we hypothesize that:
Rebel groups with foreign territory will cause more civilian casualties than groups with domestic territory and groups with no territory.
In this way, territory without the political community, or foreign territorial control, mirrors the effects of lootable resources (Weinstein 2006) and financial aid (Salehyan et al. 2014) on insurgent groups. When rebels control lootable resources or have a foreign backer, it reduces their reliance on civilians for support and resources over the long term. As a result, insurgents are much more likely to use predation and violence to take what they need from civilians. We argue that the same mechanism proposed by Weinstein (2006) and Salehyan et al. (2014) applies to foreign territory; however, we develop a new theoretical framework that explains how the location of territory and the people residing within it, activates this mechanism.
Research Design and Data
Our research design incorporates two components. First, we evaluate the global relationship between foreign territorial control and civilian casualties quantitatively, supplemented by an instrumental variables analysis. Second, to better identify the causal effect of foreign territory and to highlight the plausibility of the causal process outlined in our theory, we leverage the exogenous assignment of foreign territory to the PKK by the 1991 establishment of the northern Iraq no-fly zone. We provide qualitative evidence of the processes by which foreign territory affected the PKK’s capacity and incentives for violence against civilians. Finally, using data from other contemporaneous insurgencies, we construct a synthetic counterfactual history of PKK violence in the early to mid-1990s and compare the performance of this “synthetic control” with that of the observed history of the PKK.
Large-N statistical analysis
Data and model specification
We use the non–state actor data set (Cunningham, Gleditsch, and Salehyan 2009) as a base to create a 2,297 observation insurgency-year data set. To operationalize territorial control without the political community, we use the measure Foreign Territory from the non–state actor data set,11 coded as “1” if the insurgency has access to any foreign territory and a “0” otherwise. The distribution of rebel territorial control types is shown in Table 1. Our outcome variable is the natural log of civilian fatalities caused by the insurgent group from the Uppsala Conflict Data Program (UCDP) one-sided fatalities data set (Eck and Hultman 2007). Because the one-sided fatalities data offer high, low, and “best” estimates of fatalities, we use the “best” estimates in our analysis.12 We use the log of this measure to avoid errors in estimation from highly skewed observations.13Caption
|Foreign Territory||No Foreign Territory||Total|
|No domestic territory||105||73||178|
Note. Cross-tabulation of the number of groups controlling foreign territory only, domestic territory only, both domestic and foreign territory, and neither. Note that the proportion of rebels with only Foreign Territorial control is very high (about 40%, the plurality of cases), suggesting that our theory is generalizable.
As controls, we include several insurgency-level variables from the non–state actor data set beginning with our operationalization of territorial control with the political community: Domestic Territory, coded as “1” if the insurgency controlled territory domestically and “0” otherwise. To account for the overall intensity of the conflict, we include the natural log of battle deaths from the PRIO annual battle deaths data (Lacina and Gleditsch 2005).14 We include Central Command Strength as a measure of rebel command and control (coded as “low,” “moderate,” and “high”) because insurgents with a strong central command might be more disciplined, implying decreased civilian victimization (Weinstein 2006). Because more militarily capable rebels may cause more total deaths, we include the ordinal variable Rebel Strength, ranging from “much weaker than the government” (coded as “1”) to “much stronger than the government” (coded as “5”). We include the variable Rebel Size, a logged measure of the number of insurgents, because casualties may simply be a function of group size. Finally, we include the measure Nonmilitary Support because financial resources may cause recruitment of more opportunistic rebels, implying more violence against civilians (Weinstein 2006). All observations corresponding to “endorsement” or “nonmilitary” support from a foreign state are coded as “1” signifying nonmilitary support. Observations corresponding to “no support,” “endorsement; alleged military,” “military,” or “troops” are coded “0.”
We also include three state-level controls. First, we include Population Density from the World Development Indicators (2012) because higher density may mean that rebel groups have to operate in heavily populated areas, which increasing civilian casualties. Following Fearon and Laitin (2003), we include Income (logged GDP per capita) to measure state capacity. We also include Income Growth (annual change in per capita GDP) to account for the effect of economic growth. Both GDP measures are from the Penn World Tables 7.1 (Heston, Summers, and Aten 2012). All country-level predictors are lagged by one year. All regression results are estimated with OLS. We include both country and year fixed effects, unless otherwise specified.
The results of model 1 in Table 2 are consistent with our hypothesis. Model 1 presents the results of the fully specified model, with country and year fixed effects. The coefficient of Foreign Territory is positive and significant at the 95% level. Models 2–4 in Table 2 show that the relationship between foreign territory and greater civilian victimization is robust to the incremental inclusion of control variables and alternative fixed effects. These results are consistent with our hypothesis that foreign territorial control increases the number of civilian fatalities.Caption
|Central command strength||.04||−.01||.11|
|Country fixed effects||Yes||Yes||No||No|
|Year fixed effects||Yes||Yes||Yes||No|
Note. Standard errors in parentheses.
*. p < .10.
**. p < .05.
***. p < .01.
Notably, Domestic Territory has a consistently negative relationship with civilian casualties. Although not statistically significant in all models, this finding is consistent with the expectation that domestic territorial control incentivizes insurgents to at least attempt to limit civilian casualties and generate quasi-voluntary compliance through governance.
Figure 1A shows the marginal effects of different forms of territorial control on civilian casualties in Model 1 when all variables are set to their mean or median values. Because our outcome variable is log-transformed, substantive interpretation of our regression results is less straightforward than those from a conventional OLS regression. Thus, we present exponentiated marginal effects. Interpretation of figure 1A is analogous to odds ratios in logistic regression. Marginal effects are presented as ratios of the geometric mean of predicted civilian casualties. Thus, at mean and median values of other predictors, our model predicts that insurgent groups holding only foreign territory cause more than twice as many civilian casualties than the mean insurgent group. Conversely, ceteris paribus, groups holding only domestic territory are predicted to cause 30% fewer civilian casualties than the mean insurgency. Further, Wald tests indicate that these predicted values are statistically distinct from one another. Therefore, there is a statistically significant and positive effect on civilian casualties when controlling foreign territory, and this effect is statistically significantly distinct from all other configurations of territorial control.
Our results are robust to many different specifications presented in the appendix. These include using bootstrapped standard errors (table B.1), using alternative fixed effects specifications (table B.2), omitting outliers and influential observations (table B.3), jackknifing (table B.4), re-estimating results using alternative (table B.5) and multiple ordinal codings of our independent variable (table B.6),” including additional control variables (table B.7), including external support and long-term objective (tables B.8–B.10), including alternative external support operationalizations (table B.11), including a lagged-DV (table B.12) and reestimating the results with the (un-logged) count of civilian casualties with a Poisson estimator (table B.13).15 The effect of foreign territorial control is consistently positive, large, and significant across all these specifications, increasing our confidence in these findings.
Instrumental variables analysis
Next, we address concerns that our regression results may suffer from omitted variable bias or endogeneity. That is, rebels that strategically employ violence against civilians may be more likely to seek, or be granted, foreign territory. Similarly, an unobserved variable might explain variation in both foreign territorial control and levels of civilian violence. To address these concerns, we present an instrumental variable analysis.
An effective instrument both strongly predicts the endogenous explanatory variable and satisfies the exclusion restriction. The instrument must strongly correlate with the treatment (foreign territory), but can only impact the outcome variable (civilian casualties) through foreign territory. We instrument for foreign territorial control with first, the logged total length of land borders (in kilometers) of the rebel group’s target state, drawn from the “boundary data set” (Furlong and Gleditsch 2003), and second, both the log total border length and the total number of that state’s neighbors, drawn from the Correlates of War Direct Contiguity Data (Stinnett et al. 2002). Both instruments attempt to capture exogenous sources of variation in the availability of foreign territory to rebels: the longer a state’s borders or the greater its number of neighbors, the more accessible border regions in neighboring states will be to rebels, independent of the dynamics of their conflict with the government. Further, total border length or the number of bordering states is not likely to affect rebel targeting of civilians other than through their effects on the likelihood of rebel group’s controlling foreign territory.16 This is not to say that border length may not also impact the likelihood for a state to experience conflict onset or other wartime dynamics, but our primary concern is whether border length impacts an insurgency’s likelihood to target civilians in any way other than through insurgent’s foreign territorial control.
We present the results of our instrumental variables analysis in Table 3. In models 1 and 3, we instrument for rebel foreign territorial control with the target state’s total border length. In models 2 and 4, we instrument for foreign territorial control with both total border length and the total number of neighbors. The inclusion of an additional instrument enables both an additional check on the robustness of our results as well as a test of the (joint) exogeneity of our instruments.17 The results are substantially similar across all model specifications in Table 3, as well as the results presented in Table 2.Caption
|Central command strength||.03||−.49*||.03||−.49*|
Note. Bootstrapped standard errors in parentheses.
*. p < .10.
**. p < .05.
***. p < .01.
The results in Table 3 increase our confidence in the validity of the relationship between foreign territorial control and rebel violence against civilians, further bolstered by the regression diagnostics (Table 4). The Kleibergen-Paap LM and F-statistics indicate that there should be little concern about weak instruments. Moreover, the Anderson-Rubin F and χ2 tests indicate that except in the marginal case of model 4 (two instruments without country-level predictors), our results are robust to weak instruments. Results from the Hansen J-test of overidentifying restrictions suggest that our instruments are jointly exogenous. Finally, Table 5 presents the results of the first-stage regression analysis and the large and statistically significant effect of the border length on foreign territorial control suggests that border length is not a weak instrument.Caption
|K-P underidentification test LM-stat||13.43||12.45||14.78||13.51|
|K-P weak identification test F-stat†||99.96||106.91||50.02||54.65|
|Anderson-Rubin Wald F-test||4.39||7.82||2.49||4.17|
|Anderson-Rubin χ2 test||4.65||8.19||5.75||9.54|
|LM Instrument redundancy test||18.00||.16|
|Hansen J-test of overidentifying restrictions||.53||.59|
Note. p-values are in parentheses.
† . Critical value ≥19.93.
View Table ImageCaption
|Domestic territorial control||.23***||.19**||.23***||.19**|
|Strength of central command||.03||.03||.03||.03|
|Rebel force size||−.00||−.00||−.00||−.00|
|Total border length||−5.94***||−5.80***||−5.96***||−5.72***|
|Number of neighbors||−.00||.01|
Note. Robust standard errors in parentheses.
*. p < .10.
**. p < .05.
***. p < .01.
Moreover, the estimated effect of Foreign Territory in our instrumental variables analysis is substantially larger than the corresponding effect in our base model shown in Table 2. This suggests that any omitted variable or endogeneity concerns about our base model are likely to bias against our finding. As shown in figure 1B, the results in Table 2 and figure 1A may substantially understate the positive effect of foreign territorial control on rebel civilian victimization.
The instrumental variables analysis suggests that even when accounting for potential omitted variable bias or endogeneity, foreign territorial control is associated with a significant and substantial rise in one-sided violence against civilians, consistent with our theoretical expectations. In contrast but consistent with the results presented in Table 2, domestic territorial control remains consistently negatively (though insignificantly) associated with rebel civilian victimization. This lends further credence to our theoretical expectation that foreign territorial control transforms insurgent organizations and incentivizes insurgents to use violence against civilian populations.
Case study of the Kurdistan Workers’ Party (PKK)
To better identify the processes through which foreign territorial control leads to greater civilian victimization, we present a case study combining qualitative and quantitative methods. We present a historical overview of the Kurdistan Workers’ Party (PKK) and then use the synthetic controls method to evaluate the effect of the exogenous acquisition of foreign territorial control (our operationalization of territory without the political community) on the PKK’s level of civilian victimization in Turkey. This combination of design- and model-based inference techniques allows us to isolate and quantify the causal effect of foreign territory on civilian casualties, to specify counterfactual outcomes in a transparent manner, and to account for temporal heterogeneity in a single case. Because there may be several causal pathways through which foreign territorial control makes insurgents increasingly violent against civilians, we use the PKK case to highlight several of these mechanisms, particularly how the PKK’s acquisition of foreign territory incentivized the group’s use of violence to extract resources from civilians, even potential supporters. Our multimethod design thus allows us to identify the causal effect of foreign territorial control on PKK civilian victimization and to contextualize the historical processes driving this effect.
The PKK case offers excellent causal leverage for several reasons. First, when a rebel group gains foreign territory, this process is typically endogenous to the civil conflict itself. Territorial control without the potential political community is not typically assigned exogenously, complicating causal inference about its effects. Thus, we leverage the exogenous imposition of the no-fly zone in northern Iraq and the subsequent control of this space by the PKK (Aydin and Emrence 2015, 58). We argue that the no-fly zone’s creation was orthogonal to the PKK insurgency in Turkey, allowing for clear causal inference. The United States and its allies implemented the no-fly zone after the Gulf War due to a humanitarian crisis caused by Saddam Hussein’s repression of an Iraqi Kurdish rebellion unrelated to the ongoing conflict in Turkey. Although Turkey agreed to the implementation of the no-fly zone, Turkish protestations about its impact on the PKK were muted and not influential.18 Therefore, Turkey’s conflict with the PKK did not influence the creation, timing, or extent of the no-fly zone. Because the creation of the no-fly zone was unrelated to the PKK-Turkey conflict, the PKK’s acquisition of foreign territorial control is reasonably exogenous, facilitating inference about the causal effects of PKK foreign territory.
Second, though Kurds live in northern Iraq, the PKK did not consider Iraqi Kurds as part of its political community and focused exclusively on Turkish Kurds without (at the time) aiming to create a transnational Kurdish state (Aydin and Emrence 2015, 37). We can thus be assured that the territory the PKK acquired in 1991 is indeed “foreign” territory without the potential political community. Consistent with our theory, the PKK victimized both Turkish and Iraqi Kurdish civilians after acquiring this territory. Third, while the PKK used northern Iraq as a sanctuary and base to launch military operations in 1985, pressure from the Turkish government led to a crackdown on the PKK’s presence in the region, and from 1986 to 1991 the group was absent from northern Iraq (Van Bruinessen 1988). Moreover, the PKK never fully consolidated territory with its potential political community (meaning domestic territory, or territory with Turkish Kurds), so we can isolate foreign territorial control’s effects on PKK’s civilian victimization.
Ultimately, because of the exogeneity of the PKK’s acquisition of foreign territory, its absence in northern Iraq immediately before the imposition of the no-fly zone, and its lack of domestic territorial control, we are confident that observed increases in PKK civilian victimization can be causally attributed to PKK acquisition of foreign territory.
The PKK was founded in the early 1970s by a group of Marxist nationalists at a university in Ankara (Van Bruinessen 1988, 40–50). The group formed in response to several contemporaneous sociopolitical factors in Turkey: pervasive Kemalist ideology forced Kurdish identity out of the public sphere, Kurds faced substantial social and economic discrimination, and the Turkish government’s reaction to Kurdish activism was seen as harsh and heavy-handed (Davis et al. 2012, 100–101).
The PKK waited until 1984 to launch a military campaign against the Turkish state, declaring an independent, socialist Kurdistan (Davis et al. 2012, 100; Van Bruinessen 1988). In the intervening period, leader Abdullah Öcalan cultivated ties with foreign governments, particularly Syria, that would help to sustain the PKK in its early years, securing funding and a safe haven to train PKK recruits (Cornell 2001). Öcalan also established relations with Palestinian rebel organizations, and PKK guerrillas and Palestinian fighters trained together (Van Bruinessen 1988).
Throughout the 1980s, consistent with Maoist strategy, the PKK explicitly modeled itself after successful communist insurgencies (Mao 1961), seeking to first weaken state control, then establish and consolidate “liberated zones” housing large rebel formations in which the PKK could replace the state as the legitimate authority (Gunter 1997, 38–40). PKK leaders saw “the initial phase of [the] war [as] a propaganda battle, in which [they] tried to gain the trust and respect of the people and prove they could stand up to the state” (Marcus 2007, 119). Speaking at the PKK’s 1986 conference, Öcalan emphasized the group’s commitment to establishing zones of domestic territorial control to cultivate positive ties with Kurdish civilians and expressed frustration at the group’s inability to do so by that time (Gunes 2013, 108; Marcus 2007, 108). Thus from 1984 to 1990, stories spread about the PKK’s honesty and commitment to the creation of an independent Kurdistan and its “willingness to take into account the demands and criticisms of the people it wanted to represent … an important factor in the group’s growing popularity” (Marcus 2007, 119).
If the PKK’s strategy of violence throughout the conflict can best be understood as an “interaction between combatants and resources,” the PKK’s behavior in this period reflected increasing access to resources, primarily in the form of military and economic goods, new recruits, a compliant population, and legitimacy (Aydin and Emrence 2015, 3, 24–26). In the late 1980s and early 1990s, the PKK’s primary targets were officers of the Turkish state (DeRouen and Heo 2007), but civilians were sometimes targeted in the early years of the insurgency (Van Bruinessen 1988). Moreover, the PKK alienated potential supporters by instituting a program of conscription (Marcus 2007, 117). However, by the end of the group’s 4th congress in 1990, the PKK declared its intention to focus on military targets and limit attacks on civilians (Gunter 1997, 49; Marcus 2007, 118–19), while ending its practice of forced conscription in order to bolster its legitimacy with Kurds in Turkey and abroad. On the eve of the Gulf War, the PKK’s popularity among Kurds was increasing rapidly: the organization attempted to cultivate positive relations with Turkish Kurds through limiting civilian casualties, establishing liberated zones with governing institutions, and avoiding the conscription of Kurdish coethnics. Before 1991, the PKK conscientiously took steps to establish positive relations with coethnics and limit violence against them, in anticipation of securing territory with the potential political community (domestic territory).
The aftermath of the 1991 Gulf War brought unexpected opportunities, changing the PKK leadership’s strategic calculus. Thereafter, the PKK abandoned its traditional Maoist strategy in favor of targeting Turkish-Kurdish civilians (many of whom were educators or were related to state bureaucrats). Saddam Hussein’s brutal response to the Iraqi Kurdish uprising that followed Iraq’s defeat in the Gulf War created a massive humanitarian crisis in northern Iraq and southeastern Turkey. Subsequently, the United States established the northern Iraq no-fly zone in March of 1991 to protect Iraqi Kurdish civilians from Iraqi forces. As a result, “northern Iraq became a power vacuum, which coincided nicely with the aims of the PKK. Öcalan’s organization soon based its operations there, and by 1994 it had managed to deny the Turkish state effective control of large tracts of its southeastern territory” (Cornell 2001). From this base of operations in mountainous northern Iraq, the PKK continued to launch military operations against the Turkish state (Cornell 2001; Turkish Democracy Foundation 1996, 25, 27).19
At its foreign base, the PKK cultivated a variety of financial resources, drawing on contributions from the Kurdish diaspora in western Europe and establishing lucrative activities in drug and human trafficking (Roth and Sever 2007, 901–20). Simultaneously, thousands of young Turkish Kurds flocked to the PKK, attracted to the group’s military successes (Marcus 2007, 160). As a result, foreign territorial control was a tremendous military asset to the PKK allowing it to “roam freely on the border, recruit new members, and acquire weapons” (Aydin and Emrence 2015, 58).
Despite the favorable environment and increased financial support, the PKK terminated attempts to establish institutions for the independent Kurdish state it had attempted to create in the late 1980s and in 1990. Rather than use its increased financial and military resources to fully consolidate control of “liberated zones,” the PKK instead directed its resources to building up its military forces (Marcus 2007, 180–81). Even though the PKK had the opportunity to develop a parallel Kurdish Assembly in northern Iraq to rival the Turkish Parliament and had convened potential delegates, by 1992, the Assembly was dissolved (Marcus 2007, 209). At the height of its strength in 1992, the PKK had given up on the idea of holding and defending liberated territory (Marcus 2007, 206) and did not create a viable, autonomous, and rival administration to govern and thereby wrest control of the region from the Turkish state. In other words, the PKK’s foreign territorial control served as a military asset for the group but did not incentivize the PKK to generate legitimacy through limiting civilian predations and establishing governing institutions.
Instead, the PKK focused on using violence to extract resources, compliance, and collaboration from Turkish Kurds and reintensified its attacks on civilians after having abandoned its civilian targeting strategy before 1991. Thus, the majority of significant PKK attacks on civilians occurred after 1991 (Rodoplu, Arnold, and Ersoy 2003, 154–55). The group used these predatory tactics because it “sought to compel the population to cooperate with its aims” (2003, 157) and specifically attacked fellow Turkish Kurds (Eccarius-Kelly 2011, 104–13) to delegitimize the government, rather than establishing an alternative source of authority. The PKK targeted coethnics “to intimidate them in order to derive active and passive support. Therein, the PKK … used extortion and armed propaganda toward its own population in the region to maintain financial/logistical and electoral support during the process. The PKK … also applied selective violence against government officials assigned into the region (e.g., teachers, mayors, imams), claiming its authority” (Unal 2012, 434). For example, the PKK killed 34 teachers and proclaimed a ban on all schools in the region in 1993. Some schools were burned, and hundreds more were closed as a result of PKK threats (Marcus 2007, 219). By 1994, PKK statements codified these anticivilian tactics into a coherent strategy of “all-out revolutionary war in response to the enemy’s war of destruction … all economic, political, military, social, and cultural organizations, institutions, formations—and those who serve in them—have become targets. The entire country has become a battlefield” (Gunter 1997, 49). Civilians everywhere were now fair game.
The effect of the PKK’s acquisition of sanctuary in northern Iraq on the group’s treatment of civilians was immediate and dramatic (fig. 2). To ensure these apparent effects are not an artifact of our choice of data, we present data from two separate measures of insurgent civilian victimization. Figure 2A shows civilian fatalities caused by the PKK between 1989 and 1995, from the UCDP One-Sided Violence Data Set (Eck and Hultman 2007). Figure 2B shows data on the number of separate PKK attacks on civilian targets from 1986 to 1995, drawn from the START Global Terrorism Database (GTD) (LaFree and Dugan 2007). In both, the effect of foreign sanctuary is clear: a large and immediate increase in PKK civilian victimization. The GTD data allow us to disaggregate attacks by whether the victims were armed (fig. 3A) and the target type for each attack (fig. 3B). After 1991, the number of attacks on nongovernment civilian targets increases dramatically. These data support reports that after the acquisition of foreign territory, the PKK sought to target civilians to extract resources (especially compliance) from Turkish Kurds.
In 1995 the scope and intensity of PKK attacks provoked the Turkish government into repeated invasions of northern Iraq. The PKK suffered heavy casualties, and in the year that followed, Öcalan eventually called for a cease-fire (Minorities at Risk Project 2004). In 1999, Öcalan was captured—a serious blow to the PKK. Violence subsided until 2004, when the PKK reorganized under a new name (KONGRA-GEL, later KCK), developed a more robust, legal political party, and leveraged the chaotic political situation in Iraq to once again use the Turkish-Iraqi border as a base for military operations (Davis et al. 2012, 111–12; Eccarius-Kelly 2011, 113). Violence resumed until 2009, before Öcalan once again called for a cease-fire. Since 2009, Öcalan has called for, then retracted numerous cease-fires, as the Turkish government has implemented political reforms to benefit the Kurds in Turkey, including greater political freedoms (Davis et al. 2012, 117–18).
As compelling as the case description is, the evidence is insufficient to quantify the causal effect of foreign territorial control (territory without the political community). Moreover, given the inherent data limitations of a single treated case, a difference of means or analysis of variance test is likely to lack sufficient power to evaluate the effect of foreign territorial control with much confidence. Finally, there may be reason to suspect that the pretreatment PKK may not be the most appropriate counterfactual for the posttreatment PKK. For example, it is conceivable that the near-contemporaneous collapse of the Soviet-led Eastern Bloc, leading both to the reemergence of ethnic and sectarian conflicts in the former communist world, as well as a flood of cheap weapons from Cold War arsenals, could have caused a general rise in rebel lethality toward civilians, potentially confounding the results presented above. Thus, we turn to the method of synthetic controls, developed by Alberto Abadie and colleagues (Abadie, Diamond, and Hainmueller 2010, 2011, 2014; Abadie and Gardeazabal 2003), designed to address precisely these concerns.
The synthetic controls method attempts to address two problems associated with observational data with a limited sample: accurate counterfactuals and nonrandom assignment to treatment. Researchers cannot observe the same unit with and without treatment, nor can they compare the treated case with a randomly assigned control case. The synthetic controls method creates a synthetic counterfactual drawing on data from contemporaneous, nontreated cases (insurgencies).20 Thus, while we cannot directly observe a counterfactual PKK that did not receive foreign territory, this method creates a virtual control unit, weighted by both prespecified variables, as well as analogs to country and time fixed effects, to be as similar as possible to the pretreatment (pre-1991 acquisition of foreign territory) behavior of the unit of interest (the PKK).
Second, following Abadie et al. (2010), we employ placebo tests to determine whether the putative treatment effect can be observed in other, untreated cases. Analogous to a placebo treatment in a medical trial, we apply the “treatment” to each contemporaneous insurgency, constructing a synthetic control and comparing its performance to its real-world counterpart. In other words, we synthetically treat each insurgency with foreign territorial control and measure its putative effect. This controls for any unobserved factors that might contribute to a general rise in civilian victimization (such as the end of the Cold War) by all insurgencies operating at that time. Thus, if the observed “treatment effect” on the case of interest is sufficiently distinctive, our confidence in its validity is consequently increased.
The synthetic controls test optimizes the fit of the control to the observed unit across the pretreatment period (1989–90) and then compares the outcomes of the observed and synthetic units in the posttreatment period (1992–95). We use a predictive model as directly comparable as possible to the one employed in our regression analyses. Thus, we employ the “best” estimates of total civilian fatalities from the UCDP one-sided fatalities data set (Eck and Hultman 2007) as our outcome variable.21 We also use the same measures22 of domestic territorial control, rebel group size, an ordinal measure of rebel group strength, and a dichotomous measure of external military support.23 Additionally, we include population density from the World Bank World Development Indicators, and GDP per capita from the Penn World Tables (Heston, Summers, and Aten 2012). Variable and unit weights used in the construction of the synthetic control are presented in the appendix (tables C.1 and C.2, respectively).
Additionally, the synthetic control procedure inherently includes both time effects and an analogue to unit fixed effects (accounting for fixed or slow-moving unit-level heterogeneity). To utilize this procedure, however, we can only include insurgencies that existed for the entire duration of the analysis. Due to data limitations and to maximize the number of contemporaneous cases, we limit our analysis to 1989–95. The relevant comparison pool is any ongoing insurgency that was observed during the entire optimization period.
We interpret synthetic controls results visually: the larger the separation between the virtual control case (represented by the dashed line) and the observed behavior (represented by the solid line), the greater the effect of the treatment (acquisition of foreign territorial control, represented by the vertical dotted line). Figure 4 displays results from our synthetic controls estimation of PKK civilian victimization in Turkey from 1989 to 1995. We evaluate the closeness of the fit between observed and synthetic units by calculating the mean squared predictive error (MSPE) in the pretreatment period. Smaller errors indicate better fit. The extremely close fit of the pretreatment period (reflected in the very small pretreatment MSPE) indicates both that our model predicts civilian casualties well and that the synthetic PKK represents a reasonable counterfactual in the posttreatment period, increasing our confidence in the validity of the apparent treatment effect.24
The placebo plots presented in figure 5 gives us further confidence in our results. The placebo plot reflects the predicted civilian deaths in all donor universe insurgencies as though they had also acquired foreign territory in 1991. The large effect of treatment on PKK military activity relative to most control cases is immediately apparent. Indeed, closer examination of the simulated treatment on the donor cases does not suggest that unobserved factors cause all insurgencies operating in 1991 to behave in the same way as the PKK. This increases our confidence that the observed treatment effect on the PKK is not the result of some unobserved factor leading to increased one-sided violence by rebels everywhere. Thus, our synthetic controls analysis supports our hypothesis that the PKK’s exogenous acquisition of foreign territory caused increased civilian victimization.
Discussion of PKK civilian victimization
Evidence from the PKK case study is consistent with the predictions of our theory. Prior to 1991, the PKK sought to establish mutually beneficial relations with civilians to access critical resources by limiting civilian casualties to generate quasi-voluntary compliance. Although the group sometimes targeted nonmilitary personnel, by the late 1980s Öcalan had publicly committed to avoiding civilian casualties. Moreover, consistent with incentives for securing quasi-voluntary compliance, the PKK had committed to establishing liberated zones and alternative political institutions: an “insurgent-state” in southwest Turkey.
However, following the PKK’s exogenous acquisition of foreign territory (territory without the potential political community) in 1991, its strategy shifted in parallel with its changing incentives. Rather than cultivating ties with civilians and establishing itself as the legitimate governing authority in Kurdistan, the PKK ignored or abandoned opportunities to create parallel institutions or provide social services to win over civilian support. Because the PKK abandoned the idea of holding territory in Turkey, it had no incentive to use quasi-voluntary compliance as a strategy of resource or recruit mobilization. Radu (2002, 147) argues that “absolute power matters far more to Öcalan than the aspirations and welfare of the people he claims to lead.” True or not, PKK acquisition of foreign territory removed even the most instrumental of incentives for its leaders to consider civilian welfare. Instead, the PKK used its foreign territory to train recruits, expand propaganda campaigns, and develop sources of additional financing—from state sponsors, coethnics around Europe, or smuggling (Davis et al. 2012, 100–18; Roth and Sever 2007) and used violence against Turkish Kurds or other civilians to extract additional resources, recruits, and compliance within Turkey. This qualitative evidence is strongly supported by our synthetic controls results.
What is notable about both the quantitative and qualitative findings is that after 1991, there is a sharp increase in the number of casualties caused by the PKK, consistent with the hypothesis. Between 1994 and 1995, however, the number of observed casualties declines precipitously. This suggests that although the organization was able to enhance its military and financial capacity, these benefits were short-lived. We attribute the brevity of the PKK’s military effectiveness to the fact that foreign territorial control did little to enhance the political apparatus within Turkey and had no effect on the PKK’s ability to create a viable political alternative to the Turkish government in southeast Turkey’s predominantly Kurdish region, further underscoring the idea that using violence and coercion to extract resources (especially compliance and collaboration) yields quick benefits immediately but is seriously flawed if sustained for the long term.
In this article, we hypothesize that insurgents with foreign territorial control cause increased rebel civilian victimization. To test our hypothesis, we used a large-N statistical analysis and leveraged the exogenous acquisition of foreign territorial control by the PKK due to the 1991 establishment of the northern Iraq no-fly zone on its level of violence against civilians. Based on both qualitative and quantitative evidence, we find support for our theory: foreign territory leads to greater insurgent violence against civilians because foreign territorial control increases the latent lethality of an insurgent organization while simultaneously incentivizing rebel combatants to use violence as a strategy for resource extraction and civilian mobilization.
This article also has important policy implications. First, this research supports recommendations by others (Byman 2005) that states refrain from supporting insurgents, especially by providing access to foreign territory. Second and counterintuitively, if facing an armed insurrection, states seeking to limit violence against civilians should discourage insurgencies from gaining foreign territory, because insurgencies with domestic territory are likely to cause fewer civilian deaths. In particular, as the United States and Turkey consider implementing a no-fly zone over northern Syria, they should consider the potential implications for greater civilian loss of life.
Theoretically, our work suggests that the nature of the territory that rebels control shapes the incentives for insurgent violence against civilians and/or governance. As previous research demonstrates, territorial control is a resource in and of itself (Kalyvas 2006), but how territorial control confers benefits on an insurgency shifts depending on who is in this territory and the set of resources that insurgencies may extract from them. In doing so, this article takes seriously theories of state formation and fiscal bargaining and find that our theoretical logic is analogous to that developed in the extensive “rentier state” literature, which holds that resource rents make resource-rich governments less responsive to citizens than tax-dependent ones, leading to stalled political and economic development.25 Therefore, we echo work by others who find that rebel groups mimic state behavior (Stewart 2015), and that insurgencies perform the role of the state (Mampilly 2015).
Finally, our theory suggests that although insurgent movements relying on foreign territorial control may become more lethal, they may not necessarily be more successful in their ultimate pursuit of control of a state or region. Though foreign territory is a benefit militarily, insurgent groups do not necessarily solely consist of rebel fighters and more often than not need a robust political apparatus. Mao’s (1961) seminal text, On Guerrilla Warfare, underscores the need for a base of political support, from which an insurgency draws its strength. While foreign territorial control might increase the likelihood of an insurgent movement’s ability to carry out violent attacks, it may do so at the cost of losing this essential base of political support. Ultimately, we remain unconvinced that foreign territorial control is so beneficial that it renders the rebel organization a viable alternative to the incumbent government.Acknowledgments
We are grateful for extremely helpful and thoughtful comments and feedback from Daniel Nexon, Daniel Byman, Morgan Kaplan, Miriam Krieger, Reed Wood, Jessica Maives Braithwaite, Aysegul Aydin, Will Moore, Ashley Murph-Schwarzer, Margit Bussman, participants in the Georgetown University Graduate Student Working Group, New York University’s Alexander Hamilton Center Graduate Student Conference on Political Economy, and three anonymous reviewers.
Megan A. Stewart ([email protected]) is an assistant professor at the school of international service, American University, Washington, DC, and a postdoctoral research associate in the department of politics, University of Virginia, Charlottesville, VA. Yu-Ming Liou ([email protected]) is a PhD candidate in the department of government, Georgetown University, Washington, DC. His research focuses on the domestic politics of economic globalization and the political economy of natural resource extraction, with particular attention to the role of political elites.
Data and supporting materials necessary to reproduce the numerical results in the article are available in the JOP Dataverse (https://dataverse.harvard.edu/dataverse/jop). An appendix with supplementary material is available at http://dx.doi.org/10.1086/688699.
1. Peralta, Eyder. NPR. January 15, 2015. “Satellite Images Show ‘Catastrophic’ Destruction of Boko Haram Attack in Nigeria.” http://www.npr.org/sections/thetwo-way/2015/01/15/377408586/satellite-images-show-vast-destruction-of-boko-haram-attack-in-nigeria.
2. The Guardian, July 19, 2014. “Boko Haram Insurgents Kill 100 People as They Take Control of Nigerian Town.”https://www.theguardian.com/world/2014/jul/19/boko-haram-kill-100-people-take-control-nigerian-town.
3. Based on data on one-sided violence (Eck and Hultman 2007), the average insurgency in our sample kills approximately 126 civilians per year. For insurgents with foreign territorial control, this number more than doubles.
4. We use the term “stationary” and “roving” bandits to be consistent with existing terms in the literature (e.g., Beardsley, Gleditsch, and Lo 2015). However, a more appropriate and apt terminology might be “rulers” and “raiders.” In the text, we use “stationary bandits” to refer to rebel “rulers” and “roving bandits” to refer to rebel “raiders.”
5. Though not as precipitously as an occupying force that victimizes civilians.
6. For example, in 1978–79, the Revolutionary Front for an Independent East Timor (FRETILIN) faced increasing pressure as Indonesia occupied more and more of East Timor. FRETILIN could no longer guarantee the safety of civilians, and the leadership debated whether to encourage people living there to leave the territory and surrender to Indonesia. Some argued that civilians had to surrender for their own safety, while others argued that FRETILIN needed the civilians because they provided key resources needed to sustain the movement (CAVR 2006, 435, 442).
7. This theoretical framework also explains patterns of insurgent violence against civilians when insurgents control refugee camps—territory in a foreign country populated by members of the political community. These rebels often develop governance systems within the camps and limit civilian predations there (Stewart 2015). For example, the FMLN built extensive education and healthcare services within refugee camps and in the later half of the civil war, recruited extensively from there (Viterna 2013, 73, 79). Similarly, Burmese ethnic rebel groups who pushed across the Thai border continue to administer social services within refugee camps (see, e.g., Vogler 2007).
8. Associated Press, March 18, 2013. “Syria Opposition Pushes to Form Interim Government.” https://www.yahoo.com/news/syria-opposition-pushes-form-interim-government-095217830.html.
9. Insurgents generally gain foreign territory in two ways: state absence in the border regions may allow insurgents to control foreign territory by default or sanctuary willingly given by states. These dynamics should still hold regardless of the way insurgents control territory. For an empirical test, see appendix table B.11.
10. Alternatively, rebels may control depopulated territory domestically, as IS does, which would be consistent with controlling territory without the political community (such as a desert). Because we are unaware of data disaggregated below the country level, we cannot test this empirically, but the logical extension is clear.
11. See Cunningham et al. (2012, 6–7). Unless otherwise specified, all variables are from the non–state actor data set.
12. The One-Sided Fatalities Data set only contains observations after 1988, which reduces the number of observations in our sample and limits our time frame to 1989–2003.
13. To avoid dropping observations, observations of zero fatalities are coded as ln(1) = 0.
14. We use the “best” estimate of battle deaths when available and substitute the low estimate of battle deaths when “best” estimates for battle deaths are unavailable.
15. For greater detail about each of these robustness checks, see the appendix.
16. Border length and number of neighbors can reflect the success of secessionist rebels. However, by definition, secession can change national borders only after the cessation of civil conflict. Consequently, this does not affect the contemporaneous relationships between our instruments and the outcome variable.
17. The Hansen J-test requires an overidentified model (instruments outnumber endogenous predictors).
18. Turkey envisioned a small, temporary refuge on the border. What was actually implemented was very different: a no-fly zone extending south to the 36th parallel that established a de facto autonomous Iraqi Kurdistan, much to the later dismay of the Turks (Kirişci and Winrow 1997, 160, 163).
19. See “Kurdish Mountain Fighters Take Their War to the Streets.” The Herald (Glasgow). February 6, 1992. Accessed via LexisNexis Academic, September 30, 2014. See also The Toronto Star, May 3, 1991. “Kurdish Rebels Step up Attacks in East Turkey.”
20. See Abadie et al. (2010) for a more detailed and technical explanation of the method.
21. We do not employ the natural log of this measure to aid in interpretation.
22. Unless otherwise noted, these data are drawn from Cunningham et al. (2009).
23. Because the availability of nonmilitary support does not vary within the donor group, we are unable to include this measure in constructing our synthetic control.
24. Similar analyses conducted with Global Terrorism Database attacks data and covering the period 1986–95 produce substantively identical results and are presented in the appendixes.
25. See, e.g., Mahdavy (1970), Ross (2001, 2012), Sachs and Warner (1995), Smith (2008).
- Abadie, Alberto, Alexis Diamond, and Jens Hainmueller. 2010. “Synthetic Control Methods for Comparative Case Studies: Estimating the Effect of California’s Tobacco Control Program.” Journal of the American Statistical Association 105 (490): 493–505.First citation in articleCrossrefGoogle Scholar
- Abadie, Alberto, Alexis Diamond, and Jens Hainmueller. 2011. “Synth: An R Package for Synthetic Control Methods in Comparative Case Studies.” Journal of Statistical Software 42 (13): 1–17.First citation in articleCrossrefGoogle Scholar
- Abadie, Alberto, Alexis Diamond, and Jens Hainmueller. 2014. “Comparative Politics and the Synthetic Control Method.” American Journal of Political Science 59 (2): 495–510.First citation in articleCrossrefGoogle Scholar
- Abadie, Alberto, and Javier Gardeazabal. 2003. “The Economic Costs of Conflict: A Case Study of the Basque Country.” American Economic Review 93 (1): 113–32.First citation in articleCrossrefGoogle Scholar
- Aydin, Aysegul, and Cem Emrence. 2015. Zones of Rebellion: Kurdish Insurgents and the Turkish State. Ithaca, NY: Cornell University Press.First citation in articleGoogle Scholar
- Azam, Jean-Paul. 2006. “On Thugs and Heroes: Why Warlords Victimize Their Own Civilians.” Economics of Governance 7 (1): 53–73.First citation in articleCrossrefGoogle Scholar
- Beardsley, Kyle, Kristian Skrede Gleditsch, and Nigel Lo. 2015. “Roving Bandits? The Geographical Evolution of African Armed Conflicts.” International Studies Quarterly 59 (3): 503–16.First citation in articleCrossrefGoogle Scholar
- Berman, Eli, and David D. Laitin. 2008. “Religion, Terrorism and Public Goods: Testing the Club Model.” Journal of Public Economics 92 (10): 1942–67.First citation in articleCrossrefGoogle Scholar
- Berman, Eli, Jacob N. Shapiro, and Joseph H. Felter. 2011. “Can Hearts and Minds Be Bought? The Economics of Counterinsurgency in Iraq.” Journal of Political Economy 119 (4): 766–819.First citation in articleLinkGoogle Scholar
- Byman, Daniel. 2005. Deadly Connections: States that Sponsor Terrorism. New York: Cambridge University Press.First citation in articleCrossrefGoogle Scholar
- Byman, Daniel. 2008. “The Changing Nature of State Sponsorship of Terrorism.” Saban Center Analysis paper (Brookings), 1–43.First citation in articleGoogle Scholar
- CAVR, Chega. 2006. “Final Report of the Commission for Reception, Truth, and Reconciliation of Timor-Leste (CAVR).” Dili, Indonesia: Truth and Reconciliation Commission.First citation in articleGoogle Scholar
- Cohen, Dara Kay. 2013a. “Explaining Rape during Civil War: Cross-National Evidence (1980–2009).” American Political Science Review 107 (3): 461–77.First citation in articleCrossrefGoogle Scholar
- Cohen, Dara Kay. 2013b. “Female Combatants and the Perpetration of Violence: Wartime Rape in the Sierra Leone Civil War.” World Politics 65 (3): 383–415.First citation in articleCrossrefGoogle Scholar
- Cornell, Svante E. 2001. “The Kurdish Question and the Turkish Political System.” Orbis 45 (1): 31–46.First citation in articleCrossrefGoogle Scholar
- Cunningham, David E, Kristian Skrede Gleditsch, and Idean Salehyan. 2009. “It Takes Two: A Dyadic Analysis of Civil War Duration and Outcome.” Journal of Conflict Resolution 53 (4): 570–97.First citation in articleCrossrefGoogle Scholar
- Cunningham, David E., Kristian Skrede Gleditsch, and Idean Salehyan. 2012. “Codebook for the Non–State Actor Data.” http://privatewww.essex.ac.uk/~ksg/eacd.html.First citation in articleGoogle Scholar
- Davis, Paul K., Eric V. Larson, Zachary Haldeman, Mustafa Oguz, and Yashodhara Rana. 2012. “Understanding and Influencing Public Support for Insurgency and Terrorism.” DTIC monograph. http://www.dtic.mil/dtic/tr/fulltext/u2/a562875.pdf.First citation in articleGoogle Scholar
- De La Calle, Luis, and Ignacio Sánchez-Cuenca. 2012. “Rebels without a Territory: An Analysis of Nonterritorial Conflicts in the World, 1970–1997.” Journal of Conflict Resolution 56 (4): 580–603.First citation in articleCrossrefGoogle Scholar
- DeRouen, Karl R., and Uk Heo, eds. 2007. Civil Wars of the World: Major Conflicts since World War II, vol. 1. Santa Barbara, CA: ABC-CLIO.First citation in articleGoogle Scholar
- Eccarius-Kelly, Vera. 2011. The Militant Kurds: A Dual Strategy for Freedom. Santa Barbara, CA: ABC-CLIO.First citation in articleGoogle Scholar
- Eck, Kristine, and Lisa Hultman. 2007. “One-Sided Violence against Civilians in War Insights from New Fatality Data.” Journal of Peace Research 44 (2): 233–46.First citation in articleCrossrefGoogle Scholar
- Eynde, Oliver Vanden. 2011. “Targets of Violence: Evidence from India’s Naxalite Conflict.” Working paper 2015-29, LSE.First citation in articleGoogle Scholar
- Fazal, Tanisha. 2013. “Secessionism and Civilian Targeting.” Presented at the annual meeting of the American Political Science Association, Chicago.First citation in articleGoogle Scholar
- Fearon, James D., and David D. Laitin. 2003. “Ethnicity, Insurgency, and Civil War.” American Political Science Review 97 (1): 75–90.First citation in articleCrossrefGoogle Scholar
- Furlong, Kathryn, and Nils Petter Gleditsch. 2003. “The Boundary Dataset.” Conflict Management and Peace Science 20 (1): 93–117.First citation in articleCrossrefGoogle Scholar
- Gunes, Cengiz. 2013. The Kurdish National Movement in Turkey: From Protest to Resistance. New York: Routledge.First citation in articleGoogle Scholar
- Gunter, Michael M. 1997. The Kurds and the Future of Turkey. New York: St. Martin’s Press.First citation in articleGoogle Scholar
- Halliday, Fred. 2002. Arabia without Sultans. London: Saqi.First citation in articleGoogle Scholar
- Heston, Alan, Robert Summers, and Bettina Aten. 2012. “Penn World Tables Dataset, version 7.1.” University of Pennsylvania: Center for International Comparisons of Production, Income and Prices.First citation in articleGoogle Scholar
- Humphreys, Macartan, and Jeremy M. Weinstein. 2006. “Handling and Manhandling Civilians in Civil War.” American Political Science Review 100 (3): 429–47.First citation in articleCrossrefGoogle Scholar
- Jalata, Asafa. 1993. Oromia and Ethiopia: State Formation and Ethnonational Conflict, 1868–1992. Boulder, CO: Rienner.First citation in articleGoogle Scholar
- Kalyvas, Stathis N. 2006. The Logic of Violence in Civil War. Cambridge: Cambridge University Press.First citation in articleGoogle Scholar
- Kalyvas, Stathis N. 2008. “Promises and Pitfalls of an Emerging Research Program: The Microdynamics.” In Tarek Masoud, Stathis N. Kalyvas, and Ian Shapiro, eds., Order, Conflict, and Violence. New York: Cambridge University Press, chap. 16. http://stathis.research.yale.edu/documents/Kalyvas_OCV.pdf.First citation in articleGoogle Scholar
- Kazimi, Nibras. 2015. “Where Is the ‘Strategic Depth’ of the ‘Islamic State’?” http://talisman-gate.com/2015/11/23/where-is-the-strategic-depth-of-the-islamic-state/ (accessed January 17, 2016).First citation in articleGoogle Scholar
- Kirişci, Kemal, and Gareth M. Winrow. 1997. The Kurdish Question and Turkey: An Example of a Trans-State Ethnic Conflict. Portland, OR: Cass.First citation in articleGoogle Scholar
- Lacina, Bethany, and Nils Petter Gleditsch. 2005. “Monitoring Trends in Global Combat: A New Dataset of Battle Deaths.” European Journal of Population 21 (2–3): 145–66.First citation in articleCrossrefGoogle Scholar
- LaFree, Gary, and Laura Dugan. 2007. “Introducing the Global Terrorism Database.” Terrorism and Political Violence 19 (2): 181–204.First citation in articleCrossrefGoogle Scholar
- Levi, Margaret. 1989. Of Rule and Revenue. Berkeley: University of California Press.First citation in articleGoogle Scholar
- Lyall, Jason, Graeme Blair, and Kosuke Imai. 2013. “Explaining Support for Combatants during Wartime: A Survey Experiment in Afghanistan.” American Political Science Review 107 (4): 679–705.First citation in articleCrossrefGoogle Scholar
- Mahdavy, Hossein. 1970. “The Pattern and Problems of Economic Development in Rentier States: The Case of Iran.” In M. A. Cook, ed., Studies in the Economic History of the Middle East. Oxford: Oxford University Press.First citation in articleGoogle Scholar
- Mampilly, Zachariah. 2015. “Performing the Nation-State: Rebel Governance and Symbolic Processes.” In Nelson Kasfir, Ana Arjona, and Zachariah Mampilly, eds., Rebel Governance in Civil Wars. New York: Cambridge University Press, chap. 4.First citation in articleGoogle Scholar
- Mampilly, Zachariah Cherian. 2011. Rebel Rulers: Insurgent Governance and Civilian Life during War. Ithaca, NY: Cornell University Press.First citation in articleGoogle Scholar
- Manekin, Devorah. 2013. “Violence against Civilians in the Second Intifada: The Moderating Effect of Armed Group Structure on Opportunistic Violence.” Comparative Political Studies 46 (10): 1273–300.First citation in articleCrossrefGoogle Scholar
- Mao, Tse-Tung. 1961. On Guerrilla Warfare. Champaign: University of Illinois Press.First citation in articleGoogle Scholar
- Marcus, Aliza. 2007. Blood and Belief: The PKK and the Kurdish Fight for Independence. New York: NYU Press.First citation in articleGoogle Scholar
- McColl, Robert W. 1969. “The Insurgent State: Territorial Bases of Revolution.” Annals of the Association of American Geographers 59 (4): 613–31.First citation in articleCrossrefGoogle Scholar
- Minorities at Risk Project. 2004. “Chronology for Kurds in Turkey.” http://www.refworld.org/docid/469f38e91e.html.First citation in articleGoogle Scholar
- Olson, Mancur. 1993. “Dictatorship, Democracy, and Development.” American Political Science Review 87 (3): 567–76.First citation in articleCrossrefGoogle Scholar
- Radu, Michael. 2002. Dangerous Neighborhood: Contemporary Issues in Turkey’s Foreign Relations. Piscataway, NJ: Transaction Publishers.First citation in articleGoogle Scholar
- Rodoplu, Ulkumen, Jeffrey Arnold, and Gurkan Ersoy. 2003. “Terrorism in Turkey.” Prehospital and Disaster Medicine 18 (2): 152–60.First citation in articleCrossrefGoogle Scholar
- Ross, Michael L. 2001. “Does Oil Hinder Democracy?” World Politics 53 (3): 325–61.First citation in articleCrossrefGoogle Scholar
- Ross, Michael L. 2012. The Oil Curse: How Petroleum Wealth Shapes the Development of Nations. Princeton, NJ: Princeton University Press.First citation in articleGoogle Scholar
- Roth, Mitchel P., and Murat Sever. 2007. “The Kurdish Workers Party (PKK) as Criminal Syndicate: Funding Terrorism through Organized Crime, a Case Study.” Studies in Conflict and Terrorism 30 (10): 901–20.First citation in articleCrossrefGoogle Scholar
- Sachs, Jeffrey D., and Andrew M. Warner. 1995. “Natural Resource Abundance and Economic Growth.” NBER Working paper no. 5398. Cambridge, MA: NBER.First citation in articleGoogle Scholar
- Salehyan, Idean, David Siroky, and Reed M Wood. 2014. “External Rebel Sponsorship and Civilian Abuse: A Principal-Agent Analysis of Wartime Atrocities.” International Organization 63 (3): 633–61.First citation in articleCrossrefGoogle Scholar
- Schneider, Gerald, Lilli Banholzer, and Roos van der Haer. 2011. “Cain’s Choice: Causes of One-Sided Violence against Civilians.” In Tor Georg Jakobsen, ed., War: An Introduction to Theories and Research on Collective Violence. 2nd ed. New York: Nova.First citation in articleGoogle Scholar
- Smith, Alastair. 2008. “The Perils of Unearned Income.” Journal of Politics 70 (3): 780–93.First citation in articleLinkGoogle Scholar
- Stanton, Jessica A. 2013. “Terrorism in the Context of Civil War.” Journal of Politics 75 (4): 1009–22.First citation in articleLinkGoogle Scholar
- Stewart, Megan. 2015. “Civil War as State Building: Strategic Governance in Civil Wars.” Working paper. http://www.meganastewart.org/research.First citation in articleGoogle Scholar
- Stinnett, Douglas M., Jaroslav Tir, Paul F. Diehl, Philip Schafer, and Charles Gochman. 2002. “The Correlates of War (COW) Project Direct Contiguity Data, Version 3.0.” Conflict Management and Peace Science 19 (2): 59–67.First citation in articleGoogle Scholar
- Turkish Democracy Foundation. 1996. Fact Book on Turkey: Kurds and PKK Terrorism. Ankara: Turkish Democracy Foundation.First citation in articleGoogle Scholar
- Unal, Mustafa Cosar. 2012. “The Kurdistan Workers’ Party (PKK) and Popular Support: Counterterrorism toward an Insurgency Nature.” Small Wars and Insurgencies 23 (3): 432–55.First citation in articleCrossrefGoogle Scholar
- US Bureau of Citizenship and Immigration Services. 2001. “Ethiopia: Information on the Oromo Ethnic Group, the Oromo Liberation Front (OLF), and the Oromo People’s Democratic Organization (OPDO).” http://www.refworld.org/docid/3decdfcf4.html.First citation in articleGoogle Scholar
- Valentino, Benjamin, Paul Huth, and Dylan Balch-Lindsay. 2004. “Draining the Sea: Mass Killing and Guerrilla Warfare.” International Organization 58 (2): 375–407.First citation in articleCrossrefGoogle Scholar
- Van Bruinessen, Martin. 1988. “Between Guerrilla War and Political Murder: The Workers’ Party of Kurdistan.” Middle East Report 153:40–46.First citation in articleCrossrefGoogle Scholar
- Viterna, Jocelyn. 2013. Women in War: The Micro-Processes of Mobilization in El Salvador. Oxford: Oxford University Press.First citation in articleGoogle Scholar
- Vogler, Pia. 2007. “Into the Jungle of Bureaucracy: Negotiating Access to Camps at the Thai-Burma Border.” Refugee Survey Quarterly 26 (3): 51–60.First citation in articleCrossrefGoogle Scholar
- Weinstein, Jeremy M. 2006. Inside Rebellion: The Politics of Insurgent Violence. Cambridge: Cambridge University Press.First citation in articleCrossrefGoogle Scholar
- Wood, Reed M. 2010. “Rebel Capability and Strategic Violence against Civilians.” Journal of Peace Research 47 (5): 601–14.First citation in articleCrossrefGoogle Scholar
- World Bank. 2012. “World Development Indicators.” Washington, DC: World Bank. http://www.data.worldbank.org/data-catalog/world-development indicators (accessed January 2012).First citation in articleGoogle Scholar