Contestation,Governance,and the Production of Violence Against Civilians: Coercive Political Order in Rural Colombia

Abstract

What explains civilian victimization during civil war? Existing scholarship claims that violence against civilians is driven primarily by competition between armed actors. We argue that this explanation neglects a crucial cause of civilian victimization: in communities they rule, armed groups employ systematic violence against civilians to establish and sustain social order. Drawing on original microlevel quantitative data from Colombia, we show that areas controlled by a sole armed actor experience high levels of victimization, while places where multiple actors jointly govern exhibit significantly less violence. To explain this pattern, we draw on evidence from original interviews, focus groups, and secondary sources. We show that armed groups employ violence to govern areas they control and enact social order. But this violence is checked when multiple groups rule jointly: the factors that sustain pacted rule disincentivize victimization. These results have implications for theories of political order, violence, and governance by non-state actors.

Keywords

civil war conflict processes rebel governance civilian victimization

“Remove justice, then, and what are kingdoms but large gangs of robbers? And what are gangs of robbers but small kingdoms? […] For it was a witty and true response that a certain captured pirate made to the famous Alexander the Great. When the king asked the man what he meant by infesting the sea, he defiantly replied, ‘Just what you mean when you infest the whole world! But because I do it with one tiny ship, I am called a robber; and because you do it with a great fleet, you are called an emperor” Augustine of Hippo, City of God (2013, 4.4).

Introduction

The prevailing social order in many Colombian communities controlled by an armed group in the 21st century is perhaps unexpected in two ways. First, crime rates are remarkably low, especially in comparison to Colombia’s cities. In these rural areas, robberies are almost non-existent, drug use is rare, and the few domestic violence cases that occur are dealt with swiftly by local associations. Communal, unpaid work details, which would be impossible to imagine in most places, clean public spaces, fix roads, and build aqueducts. But the second way the social dynamics of these communities are unexpected is darker. Despite the seemingly peaceful and collaborative social order that prevails, the breaking of local rules (normas de convivencia) by civilians carries extremely harsh consequences. Possible punishments include fines, forced labor on public works, permanent eviction from the community, and assassination.

Notably, this sort of violence has little in common with the violence civilians face when armed groups contest control of their communities. Military contestation triggers a bloody competition for civilian loyalty, in which combatants use violence to compel civilian support, deter defection, and punish populations viewed as sympathetic to their rivals. Violence in controlled areas looks strikingly different. It is not committed in response to the threat of military challenge. Rather, armed actors use coercion to construct and enforce local social order in communities they rule: to structure the local society and economy according to the group’s goals and to punish noncompliance. We find that this type of victimization of civilians in controlled areas, which we term “governing violence,” is widespread.

Territorial control by armed groups and violence against civilians are two of the central phenomena in the study of civil war. Understanding the link between the two has animated much recent scholarship on political and criminal violence (Humphreys and Weinstein 2006; Kalyvas 2006; Kalyvas and Kocher 2009; Sexton 2016; De la Calle 2017; Berg and Carranza 2018; Anders 2020; Magaloni, Franco-Vivanco, and Melo 2020). Though accounts of the precise causal mechanisms that link control and violence differ, the prevailing logic is as follows. Civilians are key actors in civil conflict. Their ability to share strategic information, logistical support, and labor shapes who wins civil wars. As a result, states and armed groups are deeply invested in ensuring that civilian populations support them and not their rivals, and often use violence against civilians to compel support and deter defection. By this logic of contestation, violence against civilians should be prevalent in zones of active military contestation and uncommon in places where an armed actor has hegemonic control and need not fear civilian defection.

In this paper, we evaluate this conventional wisdom in the context of the Colombian civil war. We adopt a descriptive statistical approach to studying the relationship between territorial control and violence against civilians, and then use qualitative evidence from semi-structured interviews and secondary sources to identify causal mechanisms. Our article does not seek to refute a specific theory of the relationship between territorial control and violence, but rather aims to test a general conclusion in the literature on civil war: that contested areas experience more violence against civilians, potentially orders of magnitude more, than in controlled areas.¹ We do not find support for this claim in our statistical or qualitative evidence.

Using detailed event data to measure violence against civilians, we find that communities controlled by a single armed actor suffer from high levels of civilian victimization — levels on par with contexts of active contestation between armed groups. In contrast, areas where armed groups form pacts to jointly rule have strikingly less violence against civilians. These data allow us to examine differences in levels of violence between different statuses of control, but are limited by the types of inferences one can draw from even rich violent event data. Therefore, we turn to evidence from interviews, focus groups, and secondary sources. We argue for the importance of two factors in explaining the causal relationships between territorial control and violence against civilians: the role of armed group governance in producing violence, and the dynamics of pacts between armed actors. We do not claim that other variables, such as combatant identity or organizational structure, do not matter, but rather that these two factors better explain levels of violence across types of territorial control.

We posit that armed groups commit governing violence in areas of complete territorial control, using force not to deter civilian defection and wrest control from rivals but to punish crimes and enact their vision of social and economic order. Importantly, the pervasiveness of this violence depends on the nature of the local political order. As Idler (2019) and Staniland (2012) show, political order in war is not limited to hegemony or contestation. Criminals, insurgents, and states frequently cooperate with each other to achieve mutual objectives. We argue that this type of cooperation has important implications for violence against civilians. When two or more armed groups jointly rule an area – a phenomenon we refer to as “pacted control” – governing violence should be less common. The economic incentives that lead to the creation of pacts between armed groups disincentivize violence against civilians, and the organizational dynamics that sustain pacts further reduce victimization.

Our findings generate two main implications. First, due to the difficulty in measuring territorial control, the link between territorial control and violence may not be as straightforward as often assumed. We contend that territorial control is most precisely measured through in-depth qualitative research. While other methods may generate larger samples, the accuracy provided by qualitative sourcing is unmatched. Second, the levels of violence against civilians in controlled areas should compel researchers to think more critically about the fundamental relationship between governance and violence for all types of political authority. We argue that governing violence is primarily devoted to handling infractions in a similar way that judicial systems function in state-controlled areas, operating as a substitute for formal, state justice provision. However, non-state actors generally do not have the same institutional tools available to them as states do.² Even though they may follow carefully prescribed and widely known procedures in evaluating guilt and assigning punishment, their justice provision will necessarily be informal. Whereas states use courts, police, and prisons to discipline, regulate, and extract from their populations, armed groups use a variety of violent tactics in pursuit of the same goals (Davenport 2010). Importantly, just as the state’s coercive power can be used to target social groups seen as undesirable, so can governing violence, which can create security or terror depending on how it is wielded.

Definitions

Before moving on, two key concepts – territorial control and pacts – require defining. We define territorial control as being comprised of two jointly necessary conditions. First, an armed group must exercise a monopoly on violence in a defined geographic area. This involves using either violence or the threat of violence to dominate an area. This does not necessarily mean that the area will contain no opposing combatants or preclude the possibility that opposing combatants will sometimes attack from outside the area, but rather that opposing armed groups cannot establish a stable presence.³ Second, armed groups must at least minimally regulate some aspect of social life through coercion. This regulation of social life can be very narrow, such as taxing a single crop or only punishing murderers, but a minimal level of regulation of social life is a necessary feature of territorial control: territories are comprised of people, and to exercise control armed groups must hold some coercive power over the territory’s inhabitants.⁴

When multiple armed groups are established in an area but agree not to engage in hostilities, we define this as a pact. Pacts are theoretically distinct from military parity or an alliance (Christia 2012; Idler 2019; 2020; Magaloni, Franco-Vivanco, and Melo 2020; Kalyvas 2006; Staniland 2012; 2021). We understand pacts not merely as a military stalemate, but an affirmative agreement between armed actors to not fight each other. They do not necessarily imply the existence of a broader alliance, because pacts are often localized phenomena; two groups may pact in one place and fight each other elsewhere.

Territorial Control and Violence Against Civilians

Theories of Violence Against Civilians

Stathis Kalyvas’ (2006) theory of the relationship between territorial control and violence is foundational to the literature on violence against civilians. Kalyvas argues that the level and type of violence against civilians in irregular, two-sided civil wars between the state and insurgents are dictated by the local state of territorial control.⁵ Civilians are pivotal players in Kalyvas’ account of civil war: they possess crucial knowledge about enemy combatants’ positions and intentions, as well as about which civilians may collaborate with one side or the other. Both states and insurgents thus have incentives to use violence against civilians to compel their support and deter defection. The nature of that violence depends on how much control an armed actor exercises in a given community. Higher levels of control produce selective violence by the stronger actor against defectors, whose identities are betrayed by their neighbors. Crucially, full control of a territory should engender little or no violence: “because the propensity of defection is largely endogenous to the level of control, full control makes violence redundant” (220). By this account, the only violence that happens in these areas is indiscriminate violence committed by the weaker actor (223).

This theory foregrounds territorial contestation as the main cause of violence against civilians: changes in territorial control shift civilians’ incentives to defect and armed actors' ability and willingness to punish defection with violence. Furthermore, it argues that violence against civilians occurs primarily because armed actors fear they will help the enemy. This notion is immensely influential in the literature on civil war.⁶ Subsequent scholarship expanded on the control-violence connection: Kalyvas and Kocher (2009) provide empirical support for Kalyvas’ argument using microlevel data from the Vietnam War. Steele (2017) shows that a logic of contestation also explains types of violence beyond assassinations, demonstrating that armed actors employ forced displacements as a tool of “political cleansing” of civilians likely to defect. Lichtenheld (2020) presents a consonant argument that forced displacement represents an attempt by armed groups to identify disloyal civilians. Balcells (2017) argues that the fear of defection bites even in settings of full control; the logic of contestation thus explains civilian victimization behind the front lines of conventional wars as well.

Measuring Territorial Control

In addition to work linking it to civilian victimization, territorial control features in a wide range of descriptive and causal theories of civil war. Notably, these works vary extensively in how they operationalize and measure control – variation that has important consequences for their results and conclusions. A few studies adopt an at least partially qualitative approach, using either interviews, focus groups, or others’ experience-based codings of territorial control (Arjona 2016; Berg and Carranza 2018; Breslawski 2021; Kalyvas and Kocher 2009; Magaloni, Franco-Vivanco, and Melo 2020). However, these types of studies are in the minority. Another approach is to measure control based on the presence of military forces (Gibilisco, Kenkel and Rueda 2022; Humphreys and Weinstein 2006; Sexton 2016). While these studies leverage useful data on combatants’ presence, they lack either direct measures of non-state armed group presence or direct information on the nature of control or contestation. Furthermore, geolocation of combatant presence has limited utility in determining control; armed groups present in an area may not control it, either because of another group’s dominance or because the group is only in transit.

Finally, scholars have used violent event data as raw material, which is then combined with some mix of geospatial models, text scraping, machine learning, and theoretical models of the link between different types and levels of violence and territorial control (Acemoglu, Robinson, and Santos 2013; Anders 2020; Ch et al. 2018; Gibilisco, Kenkel, and Rueda 2022; Osorio et al. 2019; Prem, Saavedra, and Vargas 2020; Tao et al. 2016; Wood et al., 2012). There are three key concerns to this approach. First, this measurement strategy shares superficial similarities with Kalyvas’ theory of civilian victimization, but actually directly contradicts it. Kalyvas’ theory of the relationship between territorial control and violence is non-linear, so simply using levels of violence to indicate more (or less) control misses the point. Second, violent event data does not have the level of granularity to distinguish clearly between selective and indiscriminate violence and thus draw out Kalyvas’ five zones of control. Third, this approach measures both the independent and dependent variables through violence: assuming a theoretical model linking violence and territorial control rules out the possibility of testing whether such a model is correct.

Three articles that estimate territorial control in Colombia warrant specific attention. Anders (2019) employs violent event data and a theoretical model of when discriminate and indiscriminate violence should occur to infer states of territorial control between the Colombian state and FARC insurgents through machine learning. Though innovative, this approach overestimates the Colombian government’s territorial control, especially after the FARC leadership signed a peace accord (Ibid, 708). Osorio et al. (2019) and Gibilisco, Kenkel and Rueda (2022) instead use natural language processing of textual descriptions of violent events to identify the armed actors present in different locations. These measurement strategies create datasets with a larger geographic spread than our dataset, but cannot determine the difference between armed group presence and territorial control, nor how different armed groups relate to each other when present in the same municipality. Furthermore, this empirical strategy runs the risk of not detecting armed group presence and control in areas where no violence against civilians occurs. Evaluating how territorial control drives violence against civilians requires high-fidelity, qualitative measurement of control.

Theory

We argue that contestation between armed actors is only one cause of civilian victimization during war. Rather, both the degree and logic of violence against civilians depend on the nature of the prevailing local political order. As the literature on contestation suggests, contested orders produce waves of violence against alleged spies and defectors. But other orders are coercive too. Hegemonic control generates violence of a different sort: the use of coercion to impose a system of social and economic organization on local communities and to punish resistance or deviance from that system. Importantly, the degree of violence under hegemonic control depends on the structure of that rule. Whereas rule by a single armed actor is often tragically violent, joint governance between multiple political actors – a common phenomenon that we refer to as “pacts” – tends to engender highly institutionalized, largely pacific orders.

Two central problems confound existing accounts of the relationship between territorial control and violence against civilians. First, territorial control provides armed groups with better information about civilians than they have in contested areas. Armed groups may use this information to root out people they see as undesirable for their social orders, such as thieves, drug users, political opponents, or those the armed group perceives as sexually deviant (Gutiérrez-Sanín 2015; Myrttinen 2021; Steele 2017). Just as a key component of statehood is exercising a monopoly on (the legitimate use of) violence, armed groups may use their dominance to coercively shape society in their ideological image (Gutiérrez-Sanín and Wood 2014; Leader Maynard 2019, 637; Mampilly 2011). Consequently, ideologically motivated violence against civilians may occur at high levels absent a competitive threat from an opponent.

Second, the literature on territorial control and violence against civilians rarely accounts for the existence of pacts between armed groups, in which multiple armed groups are present in an area but come to an arrangement that prevents violence between them. These arrangements, particularly when not delineated strictly in terms of territory, represent neither territorial control nor territorial competition. Instead, they imply affirmative agreements between groups to work together rather than compete, at least temporarily and locally. Even in some of the work that does examine pacts, it is unclear under what conditions pacts form, and how these conditions may then affect violence dynamics (Magaloni, Franco-Vivanco, and Melo 2020; Staniland 2012; 2021).⁷

In addition to these theoretical concerns, the empirical record calls into question the notion that full territorial control reduces violence against civilians. Groups frequently commit violence aimed at social control despite not facing a local competitive threat. To enforce taxation, Colombia’s FARC and ELN both kidnapped civilians to dissuade noncompliance (Gilbert 2022). A staple of Shining Path’s governance in Peru was popular committees, in which mixed civilian-combatant councils tried cases of individuals the group deemed class enemies. While civilians often favored non-lethal punishments, Shining Path’s preference for executions, based in the group’s millenarian ideology, frequently prevailed (Degregori 2012, 120; La Serna 2012, 148; Weinstein 2007, 91–2). In Northern Ireland, paramilitary groups frequently violently harassed gay individuals, while the Islamic State harshly enforced gender norms and punished homosexuality with death (Duggan 2012, 35–9; Outright International 2014, 4–5). This violence represents both an expression of the group’s ideology about who belongs in society, and tended to occur as part of broader campaigns responding to popular demands to punish LGBTQ individuals, drug users, sex workers, and the homeless (Myrttinen 2021).

We argue that two main factors explain variation in civilian victimization across states of territorial control: the prevalence of governing violence and the fact that the conditions that lead to the formation of pacts disincentivize violence against civilians. In controlled areas, armed groups use coercion, from threats to displacements to homicides, to create the society they desire. This coercion seeks to both enforce basic law and order, and as outlined above, to punish certain types of transgressions or target certain groups of people due to either the group’s ideology or popular demands. We refer to this use of coercion as “governing violence.” Governing violence aims not to punish defection or other conflict-related offenses, but to enforce the group’s vision of law and social order (Bateson 2021; Gilbert 2022; Jung and Cohen 2020; Lebas 2013; Lessing 2021; Moncada 2017; Phillips 2017). Governing violence is most frequently exercised to adjudicate everyday crimes common in court systems around the world, like land disputes, robberies, and murders. However, it is also exerted to purge groups of people considered undesirable, in ways which state law usually is not. In terms of patterns of violence, governing violence has varying manifestations, depending on the severity of the alleged infraction. It also tends to follow often well-institutionalized and publicly-known rules. Finally, it tends to be selective, commonly only targeting a few victims at a time. Importantly, we do not expect the level of governing violence to abate substantially over time. As in state-governed communities, some civilians will continue to violate rules and suffer the resulting punishments. Civilians living under armed groups will still sometimes drink heavily, steal, or shirk on community obligations despite the risk of punishment. And for some civilians, like LGBTQ individuals governed by homophobic armed groups, it may be exceedingly difficult to adapt, because their basic identity contravenes the rules.

Beyond the understudied role of governing violence, existing models of civil war often neglect the fact that hegemonic control is frequently exercised through institutionalized pacts between armed groups. These pacts exhibit low levels of violence because of the incentives that engender pacts in the first place. The key factor here is the presence of an economic interest in the area which groups can control. For example, Idler (2020) argues that in areas of drug production, the economic interests of drug-trafficking armed groups align, leading to longer-term, stable arrangements between the groups. Rather than battling for sole control, which would likely attract the state’s attention, a consolidated pact allows all groups to profit and keep the state out. Therefore, pacts present strong incentives for armed groups to avoid violence. Consequently, these incentives are also likely to have organizational effects, as the experience of jointly administering local economies institutionalizes relations between armed groups and between combatants and civilians, all of which make it harder for groups to victimize civilians without endangering their revenue flow. This combination of economic and organizational dynamics keeps violence against civilians lower than under either territorial control by a single actor or contestation.⁸

We expect our theory of governing violence to apply to armed groups in civil wars who attempt to govern civilians. Generally, revolutionary groups are more likely to govern civilians, though there are notable exceptions like the recent incarnation of the National Liberation Army (ELN) in Colombia (Amaya Alba 2021; Stewart 2021). Even in conventional wars, militaries employ irregular forces behind front lines to coerce civilians (Joo 2022; Krcmaric 2018). However, our theory speaks most directly to armed groups dynamics in fragmented, irregular civil wars, and is likely to be less explanatory in wars that are fully conventional, happen without distinct areas of territorial control, or feature a strong state confronting a much weaker opponent. Our theory of the determinants and outcomes of pacts applies to civil wars that feature more than two actors, and possibly certain two-sided conflicts that feature pacts, but will not apply to two-sided conflicts in which strong ideological divisions or the concentration of violence around a conventional frontline make pacting infeasible.

Methodology

To explore the relationship between territorial control and violence against civilians, we construct a novel dataset of territorial control in selected Colombian municipalities from 2000 to 2016. Crucially, our measures of territorial control come from sources that are distinct from our data on violence against civilians: we code territorial control using original interviews, interviews and focus groups conducted for other research projects, and secondary sources. We first looked for variation in control dynamics and the identity of armed groups present for an initial sample of municipalities for the years 2000–2016, the years for which violent event data is available. We created initial codings of territorial control by using secondary sources, data gathered from previous research projects, and through the advice of colleagues with deep expertise in particular regions. Building on this initial data, we sought out interviews to confirm the coding of each municipality and add new municipalities. When we encountered conflicting or unclear information, we tried to gather more interview and secondary source data about the area in question until we felt confident the coding was correct. When we were not able to meet this standard, we excluded these codings from our dataset.⁹ Wherever possible, we consulted multiple sources to increase confidence in our measurements. 199 of our 368 codings were based on two or more sources; 57 of these incorporated information from three sources. In our interviews, we asked about the dynamics of territorial control and the identity of the armed actor(s) present. We also asked about how the armed actor(s) governed the area and deployed violence in order to learn more about how the link between territorial control and violence works in practice.

During this data collection process, we interviewed ex-combatants, community leaders, government officials, staff of non-governmental and international organizations, and researchers.¹⁰ There were specific profiles we looked for in participants. For ex-combatants we looked for either former mid-to-high ranking individuals, or those who fought in multiple regions, which in both cases provided greater ability to comment on variation in territorial control over space and time. For community leaders, we looked for individuals who had served in prominent roles for groups like victims’ or environmental organizations, which required significant knowledge about an area’s history. For governmental officials, we specifically looked for those who had worked for government offices tasked with investigating human rights violations, especially from Ombudsmen, who report on and represent victims. Finally, for staff of international and non-governmental organizations and researchers, we looked for participants who had conducted research projects related to armed conflict in their respective municipalities of expertise and had spent more than five years doing this type of work.

This data gathering approach entailed a tradeoff between measurement fidelity and dataset size. For questions like ours, we believe that maximizing the credibility of territorial control measurement is methodologically crucial, even at the expense of sample size. Our sample of qualitative control codings consists of 1648 municipality-year observations encompassing 99 distinct municipalities from 12 of Colombia’s 32 departments. We code three categories of control: full territorial control, active contestation between two or more armed groups, and pacted control between two or more armed groups.¹¹ As referenced above, by “pacted control” we refer to instances in which two or more armed actors present in an area have agreed to cease hostilities and have established a set of rules governing behavior.

Our sample is unquestionably a convenience sample, as it does not cover the majority of municipalities where armed groups were active or rely on a more specialized sub-national design. Instead, it was designed to maximize variation across space, control statuses, and different identities of armed actors while maintaining internal validity. Colombia’s long-running civil war demonstrates tremendous variation in terms of territorial control dynamics, the identity of participating armed actors, and how these armed actors govern and deploy violence. Our sample, mapped in Figure 1, captures this variation. 50% of the municipality-years in our sample are controlled by a single actor, 28% are contested, and 22% are pacted between two or more actors. It encompasses a diverse set of armed groups, including leftist insurgents (the Revolutionary Armed Forces of Colombia [FARC] and National Liberation Army [ELN]), a wide range of paramilitary and criminal groups,¹² and security forces. Moreover, there is no obvious geographic bias in our sample, which covers coastal areas, plains, and highland areas, municipalities with a high percentage of Afro-Colombian and indigenous communities, as well as areas of historic prominence of the major armed groups, the FARC, the ELN, and the paramilitaries.

In the paper, we focus largely on non-state armed groups, but include the state as an actor in our control dataset. There are theoretical reasons sovereignty may generate different behaviors in the state than in armed groups (Mampilly 2011), but here we treat the state as a comparable armed group. We believe this is theoretically justifiable given the highly-localized and fragmented nature of the Colombian conflict, the independence with which state forces often operated from each other, and the existence of pacts between the state and armed groups.¹³

We more systematically compare our sample to the universe of conflict-afflicted municipalities in Table A1.¹⁴ Along a series of political and socioeconomic covariates, our sample largely resembles the broader population of municipalities most involved in the conflict. The average municipality in our sample is slightly less populous, less wealthy, and more violent than the population average, but the differences are not large; to maximize representativeness we attempt to adjust for these differences in the statistical analysis using sample weights. The full sample and corresponding data sources are presented in Appendix 4.

Data

We draw data on our outcome variable, violence against civilians, from the Center for Research and Popular Education (CINEP)’s Noche y Niebla dataset. CINEP is a Colombian NGO with a long history of tracking human rights abuses in the armed conflict; its data is generally regarded as the gold standard in violent event data. Noche y Niebla uses a combination of validated media reports, testimony from victims and social leaders, data from local non-governmental organizations, and original collection to construct detailed records of violence perpetrated against civilians by armed groups and state security forces. It records many kinds of lethal and nonlethal violence, including homicides and massacres, forced displacements, threats, and sexual violence. For each violent event, the dataset codes the number and names of victims, perpetrator(s), the department and municipality where the event occurred, and a free-form textual description. Noche y Niebla covers all Colombian municipalities from 2000 to present, encompassing more than 30,000 acts of violence. We restrict our analysis to the period from 2000 to 2016, when the largest insurgent group, the FARC, signed a peace agreement. Given the dominance of urban areas by the Colombian state during this period, we use a text-based classifier to subset the event data to only rural locations.¹⁵

We generate a municipality-year panel covering our sample of 99 municipalities from 2000 to 2016. For each observation, we record the overall number of instances of civilian victimization committed by armed actors or security forces as well as the number of victims associated with those acts. To be sure, the association between control and violence could be confounded by a host of local political, social, or economic dynamics that shape the presence and behavior of armed groups, civilian communities, and the state. To adjust for some of this potential confounding and evaluate alternative explanations, we incorporate a series of covariates in our models.

In addition to the claim that contestation explains civilian victimization, we address two core alternative explanations. An influential literature argues that the identity, goals, and organizational structure of armed groups affect their use of violence against civilians (Wood 2009; Hoover Green 2018; Magaloni, Franco-Vivanco and Melo 2020; Gutiérrez-Sanín and Wood 2017).¹⁶ We therefore include indicator variables for the identity of each armed group present in each municipality-year.¹⁷ We also consider the possibility that civilian victimization is driven by conflict over illicit economies (Weinstein 2007; Andreas 2019; Yashar 2018). The dominant illicit economy in Colombia centers around coca; we incorporate an indicator for the presence of coca cultivation in each municipality-year. Finally, we include a measure of the (logged) rural population of the municipality and the per capita tax collection by the municipal government, since dynamics of violence may be shaped by the size of communities and the capacity of the local state apparatus. All of these variables are drawn from the Universidad de los Andes’ Municipal Panel dataset. Summary statistics for all variables are shown in Table A2.

We model civilian victimization — counts of violent events and number of victims — as a function of the local state of territorial control (controlled, contested, or pacted). In addition to the covariates mentioned above, we include two-way fixed effects (municipality and year) to adjust for entity- and time-invariant unobservables. We fit negative binomial models to account for the overdispersed nature of the outcome variables and cluster standard errors within municipalities.

Results

The results of these regressions are depicted in the first two columns of Table 1. They present a striking pattern: consistent with our theoretical claims about governing violence, the level of civilian victimization — both the number of violent events perpetrated against civilians and the number of victims — is statistically indistinguishable between controlled areas and contested areas. Communities controlled by a single actor do not appear to be peaceful. In contrast, places under pacted control are significantly less violent than contested municipalities, suffering 41% fewer violent events (p < .01) and 55% fewer victims (p < .01). The creation of pacts between armed groups appears to suppress the production of violence.

Table 1.

Main results.

	Violent events(1)	Victims(2)	Violent events(3)	Victims(4)
Controlled	0.111 (0.201)	−0.070 (0.233)	−0.213 (0.226)	−0.574* (0.250)
Pacted	−0.531^** (0.196)	−0.807^** (0.210)	−0.705^** (0.214)	−1.05^** (0.236)
ln(Rural population)	1.31 (0.723)	1.34 (0.818)	1.99^* (0.904)	2.01^* (1.01)
Taxes pc	0.410 (1.25)	0.691 (1.19)	0.176 (1.27)	0.793 (1.45)
Coca cultivation	−0.163 (0.169)	−0.404 (0.273)	−0.258 (0.183)	−0.419 (0.253)
FARC presence	0.389 (0.262)	0.155 (0.277)	0.349 (0.332)	−0.004 (0.343)
Paramilitary presence	−0.463^* (0.202)	−0.658^** (0.229)	−0.340 (0.263)	−0.699^** (0.261)
Criminal presence	0.026 (0.280)	−0.067 (0.278)	0.231 (0.364)	−0.065 (0.347)
State presence	0.718^** (0.259)	0.779^* (0.304)	1.11^** (0.308)	1.17^** (0.335)
Municipality FE	Yes	Yes	Yes	Yes
Year FE	Yes	Yes	Yes	Yes
Observations	1471	1455	1455	1439
Squared correlation	0.48013	0.34519	0.41547	0.26728
Pseudo R²	0.18047	0.11289	0.18415	0.12118
BIC	6269.4	7405.5	5290.8	6180.8
Over-dispersion	1.7421	0.68801	1.4589	0.55763

Reference category is contested control. Negative binomial models with standard errors clustered by municipality. Models 1 and 2 use the full set of violent events, Models 3 and 4 subset to events likely committed by local armed groups. Signif. Codes: **: 0.01, *: 0.05.

Do these results reflect the logic of violence we propose? One potential concern is that we are counting acts of violence that do not correspond to the logics of contestation or governance. Armed groups controlling or contesting an area may not be the sole local purveyors of violence. We attempt to address this concern by discarding all violent events that were committed by actors or security forces that our qualitative codings do not record as present in a given municipality-year.¹⁸ This is an imperfect strategy, since perpetrators are often unknown and codings may be misattributed or based on misleading information. We nonetheless follow this approach to replicate Models 1 and 2, which incorporated the full set of violent events, using this more conservative coding of the outcome variables, presented in columns 3 and 4 of Table 1. Controlled municipalities again experience levels of violent events indistinguishable from those in contested municipalities. They appear to suffer lower numbers of victims (p < .05) — perhaps reflecting the fact that dominant groups can target violence more precisely — but the number of victims remains more than half that in contested municipalities, a level of violence unanticipated by existing theoretical accounts. These models buttress our results for pacted municipalities: by this measure of violence, pacts produce 51% fewer violent events and 65% fewer victims than contestation (p < .01).

We find little support for either of the alternative explanations we evaluated. Armed group identity does not seem to drive levels of victimization: once we adjust for control status and other covariates, only state presence consistently predicts increased violence.¹⁹ Nor is coca cultivation associated with increased violence, suggesting drug production by itself is insufficient to generate violence against civilians.

These results hold across a range of model specifications and outcome variables. We first distinguish between selective and indiscriminate violence. By Kalyvas’ logic of contestation, armed groups should seldom employ selective violence in areas they control and even more rarely rely on indiscriminate violence. We do not find this to be true in our sample: acts of both selective and indiscriminate violence are equally common in controlled municipalities as in contested places, and are less common in communities governed by pacts (Table A3). We also evaluate whether our results hold if we only consider acts of physical or lethal violence. Table A4 replicates the analysis using only physical integrity violations (columns 1–2) and only homicides (columns 3–4) as the outcome variable; the results remain consistent.

We further test whether our results are sensitive to our exclusion of urban violence from the analysis (Table A5, columns 1–2). They change little. Finally, we attempt to account for the non-random nature of our sample in order to assess whether these relationships likely hold for the universe of conflict-afflicted municipalities. We construct sample weights based on the joint distributions of five major economic and political covariates (population, area, rurality, distance to nearest market, GDP per capita, literacy, and conservative voteshare) and employ these weights in estimation.²⁰ The core results are unchanged, suggesting they are unlikely to be driven by our sampling strategy (Table A5, columns 3–4).

We assess three final possible sources of bias. First, as mentioned above, we discarded a set of municipality-years due to vague or contradictory codings. Excluding these observations could bias our results if informational clarity is correlated with control or violence. We re-fit our main models incorporating those municipalities; the results are unchanged (Table A6, columns 1–2). Second, including the state in our analysis as an armed group like any other may muddy our results if fundamentally different dynamics govern victimization by state security forces. We exclude observations where the state is present and re-estimate our main models. The results are largely consistent, though the effect of pacted rule loses significance (Table A6, columns 3–4). Finally, it is possible that – contrary to our theoretical expectations – the violence we observe under sole or pacted control reflects initial “social cleansing” by the dominant actor(s) or a lack of public knowledge about the new rules they put in place, and that consolidated control is associated with lower levels of violence. To test this alternative explanation, we assess whether the duration of the political order affects violence. We find no significant effect: the level of civilian victimization is not decreasing in the length of time a municipality has been controlled, contested, or pacted (Table A7).

If our theoretical claims are correct, the reduction in violence observed in pacted areas stems from the economic benefit of peaceful collaboration between groups. To test the plausibility of this mechanism, we generate an indicator for whether municipalities in our sample experienced one or more years of pacted control.²¹ We regress this indicator on two proxies for the potential rewards of pacted rule: average municipal economic output as measured by government records (Model 5) and average nighttime luminosity (Model 6); we take the natural log of both variables to adjust for outliers.²² As in the previous models, we incorporate variables for rural population, coca cultivation, and indicators for the identity of present armed groups, averaged across the sample period, and add department fixed effects to account for regional confounders. We employ linear probability models with robust standard errors. Consistent with our theoretical expectations, both models suggest that municipalities with greater economic value were more likely to experience pacted rule (Table 2).²³

Table 2.

Correlates of pacts.

	Pacted control
	(5)	(6)
ln(Municipal GDP)	0.161^* (0.070)
ln(Luminosity)		0.147^** (0.036)
ln(Rural population)	−0.291^** (0.087)	−0.287^** (0.070)
Coca cultivation	0.240 (0.138)	0.351^** (0.122)
FARC presence	0.601^** (0.157)	0.538^** (0.155)
Paramilitary presence	0.151 (0.317)	0.209 (0.342)
Criminal presence	0.609^** (0.214)	0.746^** (0.218)
State presence	0.704 (0.472)	0.741 (0.432)
Department FE	Yes	Yes
Observations	99	96
R²	0.52876	0.59663
Within R²	0.31387	0.40770

Dependent variable is indicator for whether municipality experienced pacted control from 2000-16. Linear probability models with robust standard errors. Signif. Codes: **: 0.01, *: 0.05.

These results present correlations between statuses of control and violence against civilians – correlations broadly inconsistent with existing scholarship. Lacking robust identification assumptions and evidence regarding mechanisms, however, they should be treated as descriptive. Nor does this quantitative approach permit us to unpack the logics underpinning these patterns of violence. Though Noche y Niebla does classify each event according to the type of violence it entails – such as “assassination,” “collective confinement,” and “torture” – these classifications rarely relay the perpetrator’s motive. It is thus impossible to know if a given assassination occurred under a governance logic, such as punishment for repeated thefts, or if it was motivated by a logic of contestation, as when victims are accused of spying for an opposing group. This problem is not specific to Noche y Niebla but to essentially all large-scale violent event datasets, which rarely include systematic explanations of motives for individual acts of violence. Given these inferential limitations, in the next section we turn to qualitative evidence to unpack the causal mechanisms underpinning the relationships between territorial control and civilian victimization.

Discussion of Qualitative Findings

Governing Violence

In controlled areas, where armed groups face low competitive pressures from other armed groups, they often govern through the imposition of clear rules of behavior for civilians. These rules generally serve dual masters: on one hand, they create the social order the armed group aims to build; on the other hand, the armed group serves as the enforcer for rules demanded by the community (Gilbert 2022; Gutiérrez-Sanín 2015; Myrttinen 2021). In South Córdoba, the Clan del Golfo’s norms both pursue popular goals, like organizing communal work details (convites), prohibiting prostitution, and punishing suspected criminals, while also seeking to prevent civilians from demanding land redistribution or other rights.²⁴ Similarly, groups across the political spectrum committed violence against people of diverse sexual orientations and gender identities and expressions. The goal was to prevent people the armed group saw as morally deviant from forming a part of the society under the group’s control, and therefore the violence was less opportunistic than strategic and ideological (Centro Nacional de Memoria Histórica 2015, 26; Myrttinen 2021). The violence itself was sometimes tailored to explicitly humiliate the victims, such as paramilitaries in Sucre forcing men perceived as “effeminate” to fight each other in public boxing matches. Pointedly, in an interview describing the event, the interviewee recalled it fondly. This was despite the fact that one of the victims was killed and many more were displaced following the event (Grupo de Memoria Histórica 2013, cited in Serrano-Amaya 2014, 160).

This strategy of using coercion in pursuit of at least somewhat popular goals is a widespread phenomenon. For example, in Cauca, the ELN listened to community demands to enforce limits on when bars could stay open, and collected fines from civilians for public drunkenness. The group also enforced limits on deforestation, which local leaders had long attempted without success prior to receiving the ELN’s coercive backing (Amaya Alba 2021, 431–4). Another example of this sort of regulation comes from Chocó, where prior to the 2016 peace accord, the FARC vigorously protected the environment in its areas of control. The group barred the pollution of rivers and mandated that 18% of profits gained from mining be redirected to community organizations.²⁵ In Catatumbo, the ELN, with at least partial community support, directly handles the entrance of unknown people and Venezuelan migrants into their areas, which can involve both threats and the infliction of physical violence, while they let local community organizations (juntas) handle interpersonal violence between local residents.²⁶

While the findings of past literature would predict that higher territorial control leads to less violent behavior toward civilians, in some cases it was the opposite. A FARC ex-combatant, who fought in both Meta, where the FARC had high levels of territorial control, and in Catatumbo, contested by multiple groups, said that less territorial control demanded a “softer” approach toward civilians to gain their trust. Conversely, where the FARC had total control, they could be much more authoritarian toward civilians, as there was nothing stopping them, and the potential negative consequences of civilian victimization were small.²⁷

In areas of sole territorial control, the scope of regulation over everyday life could be quite wide. In southeastern Chocó, the ELN controlled and taxed movement along local rivers, dictated which animals could be hunted, taxed local commerce, prohibited the use of dynamite for fishing, banned drug usage, and adjudicated domestic conflicts. In an attempt to improve the provision of dispute resolution services, the group ran a WhatsApp account that civilians could contact with problems, and the ELN would then invite the aggrieved parties to sit down with them and solve the conflict. The ELN’s justice provision here was heavily modeled on state processes, and the ELN simultaneously fulfilled the roles of the police, notaries, prosecutors, and the department of family protection provided elsewhere by the Colombian government. However, lacking the formal mechanisms of the state, the ELN’s justice provision was centered not around courts, social workers, and prisons, but mediations, fines, and enforced displacements (Aponte González and González Ramírez 2021, 540–2).

These penalties for not following the rules are enforced through coercion and sometimes the infliction of physical violence. For example, fighting or non-participation in the communal work details in South Córdoba resulted in fines, and those who did not pay fines were forced to leave the community.²⁸ In Southeastern Chocó, the ELN gave marijuana users and sellers the choice of accepting guilt and giving up the practice, or leaving the area permanently (Aponte González and González Ramírez 2021, 540–1). The most serious offenses, such as rape, spying, or stealing drugs tend to result in either steep fines or the death penalty.²⁹ Many of these norms are years or decades old due to long histories of armed group presence, and therefore are well understood and sometimes accepted by civilians.³⁰ And just as in state-governed areas, civilians adapt as best they can to the rules, but still sometimes break them. For example, the FARC in southern Tolima prohibited the entrance of migrant workers to the area for security reasons, but farmers who required help for the coffee harvest hired migrant labor anyway (Aponte 2019, 235–6).

Still, not all groups follow a set of well-institutionalized rules for governing. Some groups tend to commit governing violence with less hesitation, such as the FARC in southeastern Chocó. Whereas the ELN in the area tended to more thoroughly investigate alleged crimes and mediate disputes, the FARC was quicker to kill the alleged offender (Aponte González and González Ramírez 2021, 542). Perhaps the starkest example of this sort of rough justice comes from the EPL and certain paramilitary groups, whose activities are more oriented towards regulating illicit economies. Their justice provision proceeds in only two steps. First, a threat to leave the area is made, and if that threat is unheeded, the second step is to kill the offender.³¹

The Logic of Pacts

The kaleidoscope of armed groups across Colombia has frequently produced pacts, in which multiple groups coexist without either fighting or merging.³² We argue that these pacts lead to lower violence against civilians because the profitability of local economies creates robust agreements between armed groups and codes of behavior toward civilians. These economies are often, but not always, illicit. In the municipality of Briceño, Antioquia, the FARC formed a pact with their erstwhile enemies, the paramilitaries, to control the local coca economy. An “invisible border” divided the municipality, and each group had an exclusive zone to buy and regulate the coca grown there. The violence during this period was significantly lower than prior to the pact, during which the two had engaged in frequent open conflict.³³ Similarly, in South Córdoba, the same two groups controlled specific zones of coca cultivation, but as part of the agreement, the FARC would process the raw coca leaves and sell the resulting product to the paramilitary Clan del Golfo.³⁴ These pacts can be quite localized. Such is the case in Bajo Catatumbo, where the ELN and FARC dissidents maintain an informal truce, even though the two are currently fighting in other areas of the country, such as Arauca. Their joint regulation of coca, which is relatively abundant in the area, is the glue that holds the truce in place.³⁵ In these examples, the coordination between groups in pursuit of economic goals produces clear rules for how both armed groups and civilians are supposed to behave. The rules, as well as the negative attention that violence against civilians brings, keeps violence against civilians relatively low.

One of the best examples of how pacts reduce violence against civilians comes from Arauca, where the FARC and the ELN formed a department-wide pact from 2011-16.³⁶ The political orientations of the groups made the alliance easier, and also strengthened their varied economic interests in oil, drug trafficking, contraband, coca, and the taxing of local licit businesses. These sources of income were jointly administered by the two groups.³⁷ The pact was delineated in a 12-page document which detailed norms of behavior for how the groups should relate to each other, how combatants should behave, the duties and rights of civilians, and the processes of investigation and punishment for when these rules were violated (Bloque Comandante Jorge Briceño y Frente de Guerra Oriental Comandante en Jefe Manuel Vasquez Castaño 2013). The pact provided civilians and combatants alike with clear expectations of how each was supposed to behave, and generally warned about the dangers of abusing the civilian population. An ex-combatant who served in Arauca noted how violations of local rules by civilians were not dealt with hastily and violently, but rather prompted deliberation between the FARC, the ELN, and the accused party. Even though the FARC had a more vertical and cohesive command structure than the ELN, both sides held up their end of the agreement.³⁸

A similar instance of a pact between the ELN and the FARC occurred in South Bolívar. Facing the threat of paramilitary incursions and the promise of regulating local mining economies, the ELN and the FARC created a well-institutionalized pact. A former FARC commander described the process: “The relationship was good, even if there were normal problems, like there are between any two armies…and we were able to mediate our differences. We had spaces for coordination between commanders, and most of the time we were coordinating actions against paramilitarism; we also coordinated social works even though we each had our own political projects. In reality, until we signed the [2016] peace accord the relationship was good, one of two brotherly organizations” (Quoted in Forero Pineda and Amaya Alba 2021, 213). Prior to the 2016 peace agreement, the FARC and the ELN had jointly taxed mining at 15% each, some of which was used for community development or given directly to local organizations. Following the demobilization of the FARC, the two groups locally agreed that the FARC’s portion of the tax would no longer be collected and that the FARC’s cut would instead be directly reinvested into community organizations. However, the ELN did not hold up its end of the deal following the FARC’s withdrawal, resulting in a doubling of their income at the expense of local communities³⁹ – illustrating how single-actor rule limits incentives to improve civilian welfare. In sum, while armed groups generally enter into pacts with primarily economic motives, the social and organizational dynamics required to maintain inter-group agreements and positive cash flow keep violence against civilians low.

Conclusion and Contributions

In this paper, we use a novel, qualitatively constructed dataset of armed group territorial control combined with rich violent event data to explore the relationship between territorial control and violence against civilians in the Colombian civil war. Contrary to conventional wisdom, we find that controlled municipalities suffer from high levels of civilian victimization, and areas where multiple armed groups are present but have agreed pacts experience substantially lower levels of violence. We argue that the higher-than-expected levels of violence in controlled areas result from two factors. First, armed groups exercise “governing violence” in the areas they control. This type of violence, which includes everything from threats to forced displacements to homicides, seeks to enforce rules maintained by the group, or punish individuals the armed actor sees as undesirable in their ideal society. Second, the existence of pacts between armed groups reduces violence against civilians due to the high organizational capacity needed to maintain a pact and the economic incentives to avoid state attention by keeping violence low.

One might interpret our findings as contradicting scholarship on civil war that suggests that ordered territorial control is preferable to disordered competition between armed groups. We do not share this interpretation: we find that territorial competition does lead to high levels of violence against civilians, and that this violence is harder to predict for civilians than governing violence. Instead, we interpret our results as demonstrating that, just as states use coercion to enforce laws, repress groups viewed as undesirable, and quell protest, armed groups use coercion to govern (Davenport 2012). Though armed groups operating in civil wars may not have formal police, prisons, and justice systems, their attempts at social control are comparable to those of states. Crucially, the state’s carceral violence is mostly absent from datasets on violence against civilians, and state presence predicts an increased level of violence in our models, whereas law and order-based violence by militant groups appears alongside violence committed under logics specific to armed conflict. Rather than see this violence as the product of armed groups operating in ungoverned or informally-governed spaces that uniquely generate violence, we argue that governance is ubiquitous and generally coercive. Therefore, the enforcement of “law of order” through violence and coercion, whether done by non-state armed groups or state actors, are theoretically comparable phenomena, and should be analyzed in the same framework. Further, we believe these results highlight the need to think of armed actors not solely as power-maximizers or engaged in zero-sum contests with opposing groups, but as one example among many of armed authority structures attempting to stabilize their regimes amid a range of challenges.

We see two directions for future research. First, we hope that further research uses qualitatively constructed measures of territorial control, rather than more easily-available but measurement error-prone codings based on violent event data. Our control codings were constructed by the authors without research assistants, and larger teams with more funding can create more comprehensive datasets. Such datasets, especially for a variety of conflicts, would allow for a range of investigation about the causes and effects of territorial control in civil war. Second, further research could also seek to determine why armed groups are more or less likely to commit governing violence, what determines how similar or different non-state dispute-resolution processes are to state processes, and how theories of state capacity tie into these explanations.

Supplemental Material

Supplemental Material - Contestation, Governance, and the Production of Violence Against Civilians: Coercive Political Order in Rural Colombia

Supplemental Material for Contestation, Governance, and the Production of Violence Against Civilians: Coercive Political Order in Rural Colombia by Andrés F. Aponte González, Daniel Hirschel-Burns, and Andres D. Uribe in Journal of Conflict Resolution

Supplemental Material

Supplemental Material - Contestation, Governance, and the Production of Violence Against Civilians: Coercive Political Order in Rural Colombia

Footnotes

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors gratefully acknowledge financial support from the Pearson Institute for the Study and Resolution of Global Conflicts.

ORCID iD

Daniel Hirschel-Burns

Supplemental Material

Supplemental material for this article is available online.

Notes

References

Acemoglu

Daron

Robinson

James A.

Santos

Rafael J

. 2013. “The Monopoly of Violence: Evidence from Colombia.” Journal of the European Economic Association 11: 5–44.

Albert

Karen E

. 2022. “What is rebel governance? Introducing a new dataset on rebel institutions, 1945–2012.” Journal of Peace Research 59(4): 622-630.

Amaya

Alba

Ricardo

Daniel

. 2021. “El Frente de Guerra Suroccidental: Entre El Arraigo Social y La Reconfiguración Militar.” In ¿Por Qué Es Tan Difícil Negociar Con El ELN?: Las Consecuencias de Un Federalismo Insurgente, 1964-2020, eds. Andrés

Aponte

González

Fernán E.

. Bogotá: Cinep/Programa por la Paz, 417–76.

Anders

Therese

. 2020. “Territorial Control in Civil Wars: Theory and Measurement Using Machine Learning.” Journal of Peace Research 57(6): 701–14.

Andreas

Peter

. 2019. “Crime, Violence, and Illicit Economies in Regional and Global Perspective.” Perspectives on Politics 17(2): 485–91.

Aponte

Andrés

. 2019. Grupos Armados y Construcción de Orden Social En La Esquina Sur Del Tolima, 1948-2016. Bogotá: Cinep/Programa por la Paz.

Aponte

González

Andrés

González Ramírez

Valentina

. 2021. “Un Poder Popular Distorsionado: Madera, Oro y Coca Como Determinantes Organizacionales y Armados Del Eln Chocoano.” In ¿Por Qué Es Tan Difícil Negociar Con El ELN?: Las Consecuencias de Un Federalismo Insurgente, 1964-2020, eds. Aponte González

Andrés F.

Fernán E

González González

. Bogotá: Cinep/Programa por la Paz, 477–582.

Arjona

Ana

. 2016. Rebelocracy: Social Order in the Colombian War. New York: Cambridge University Press.

Augustine

. 2013. The City of God. ed. Babcock

William

. New York: New City Press.

10.

Balcells

Laia

. 2017. Rivalry and Revenge: The Politics of Violence During Civil War. Cambridge: Cambridge University Press.

11.

Balcells

Laia

Stanton

Jessica A.

. 2021. “Violence Against Civilians During Armed Conflict: Moving Beyond the Macro- and Micro-Level Divide.” Annual Review of Political Science 24(1): 45–69.

12.

Bateson

Regina.

2021. “The Politics of Vigilantism.” Comparative Political Studies 54(6): 923–55.

13.

Berg

Louis-Alexandre

Carranza

Marlon

. 2018. “Organized Criminal Violence and Territorial Control: Evidence from Northern Honduras.” Journal of Peace Research 55(5): 566–581.

14.

Bloque Comandante Jorge Briceño y Frente de Guerra Oriental Comandante en Jefe Manuel Vasquez Castaño . 2013. Normas Unitarias de Comportamiento y Convivencia Del Frente de Guerra Oriental Comandante En Jefe Manuel Vazquez Castaño Del ELN y Del Bloque Comandante Jorge. Briceño de Las FARC EP.

15.

Breslawski

Jori

. 2021. “The Social Terrain of Rebel Held Territory.” Journal of Conflict Resolution 65(2–3): 453–79.

16.

Centro Nacional de Memoria Histórica . 2015. Aniquilar La Diferencia: Lesbianas, Gays, Bisexuales y Transgeneristas En El Marco Del Conflicto Armado Colombiano. Bogotá: Centro Nacional de Memoria Histórica.

17.

Rafael

Steele

Abbey

Shapiro

Jacob

Vargas

Juan F.

. 2018. “Endogenous Taxation in Ongoing Internal Conflict: The Case of Colombia.” American Political Science Review 112(4): 996–1015.

18.

Christia

Fotini

. 2012. Alliance Formation in Civil Wars. Cambridge: Cambridge University Press.

19.

Davenport

Christian

. 2010. Media Bias, Perspective and State Repression: The Black Panther Party. Cambridge: Cambridge University Press.

20.

Davenport

Christian

. 2012. “When Democracies Kill: Reflections from the US, India, and Northern Ireland.” International Area Studies Review 15(1): 3–20.

21.

Degregori

Carlos Iván

. 2012. How Difficult It Is to Be God: Shining Path’s Politics of War in Peru, 1980-1999. ed. Stern

Steve J.

. Madison: University of Wisconsin Press.

22.

De la Calle

Luis

. 2017. “Compliance vs. Constraints: A Theory of Rebel Targeting in Civil War.” Journal of Peace Research 54(3): 427–41.

23.

Duggan

Marian

. 2012. Queering Conflict: Examining Lesbian and Gay Experiences of Homophobia in Northern Ireland. Burlington: Routledge.

24.

Forero

Pineda

Diego

Juan

Daniel Ricardo

Alba Amaya

. 2021. “El Eln En El Sur de Bolívar: La Pérdida de Un Bastión Militar.” In ¿Por Qué Es Tan Difícil Negociar Con El ELN?: Las Consecuencias de Un Federalismo Insurgente, 1964-2020, eds. Aponte González

Andrés F.

Fernán E

González

. Bogotá: Cinep/Programa por la Paz, 197–258.

25.

Gibilisco

Michael

Kenkel

Brenton

Rueda

Miguel R.

. 2022. “Competition and Civilian Victimization.” Journal of Conflict Resolution 66(4–5): 809–835.

26.

Gilbert

Danielle

. 2022. “The Logic of Kidnapping in Civil War: Evidence from Colombia.” American Political Science Review 116: 1–16.

27.

Grupo de Memoria Histórica . 2013. ¡basta Ya! Colombia Bogotá: Memorias de Guerra y Dignidad.

28.

Gutiérrez-Sanín

Francisco

. 2008. “Telling the Difference: Guerrillas and Paramilitaries in the Colombian War.” Politics & Society 36(1): 3–34.

29.

Gutiérrez-Sanín

Francisco

. 2015. “Organization and Governance: The Evolution of Urban Militias in Medellín, Colombia.” In Rebel Governance in Civil War, eds. Arjona

Ana

Nelson

Kasfir

Zachariah

Cherian Mampilly

. Cambridge: Cambridge University Press, 246–64.

30.

Gutiérrez-Sanín

Francisco

Wood

Elisabeth Jean

. 2014. “Ideology in Civil War.” Journal of Peace Research 51(2): 213–26.

31.

Gutiérrez-Sanín

Francisco

Wood

Elisabeth Jean

. 2017. “What Should We Mean by ‘Pattern of Political Violence’? Repertoire, Targeting, Frequency, and Technique.” Perspectives on Politics 15(1): 20–41.

32.

Hoover Green

Amelia

. 2018. The Commander’s Dilemma: Violence and Restraint in Wartime. Ithaca: Cornell University Press.

33.

Humphreys

Macartan

Weinstein

Jeremy

. 2006. “Handling and Manhandling Civilians in Civil War.” American Political Science Review 100(3): 429–47.

34.

Idler

Annette

. 2019. Borderland Battles: Violence, Crime, and Governance at the Edges of Colombia's War. New York: Oxford University Press.

35.

Idler

Annette

. 2020. “The Logic of Illicit Flows in Armed Conflict: Explaining Variation in Violent Nonstate Group Interactions in Colombia.” World Politics 72(3): 335–76.

36.

Joo

Hojung

. 2022. “Repression during Wars: Evidence from the Korean War.” Working paper .

37.

Jung

Danielle F.

Cohen

Dara Kay

2020. Lynching and Local Justice: Legitimacy and Accountability in Weak State. Cambridge: Cambridge University Press.

38.

Kalyvas

Stathis

. 2006. The Logic of Violence in Civil War. Cambridge: Cambridge University Press.

39.

Kalyvas

Stathis N.

Kocher

Matthew Adam

2009. “The Dynamics of Violence in Vietnam: An Analysis of the Hamlet Evaluation System (HES).” Journal of Peace Research 46(3): 335–55.

40.

Krcmaric

Daniel

. 2018. “Varieties of Civil War and Mass Killing: Reassessing the Relationship between Guerrilla Warfare and Civilian Victimization.” Journal of Peace Research 55(1): 18–31.

41.

La Serna

Miguel

. 2012. The Corner of Living: Ayacucho on the Eve of the Shining Path Insurgency. Chapel Hill: University of North Carolina Press.

42.

Leader Maynard

Jonathan

. 2019. “Ideology and Armed Conflict.” Journal of Peace Research 56(5): 635–49.

43.

Lebas

Adrienne

. 2013. “Violence and Urban Order in Nairobi, Kenya and Lagos, Nigeria.” Studies in Comparative International Development 48: 240–62.

44.

Lessing

Benjamin

. 2021. “Conceptualizing Criminal Governance.” Perspectives on Politics 19(3): 854–73.

45.

Lichtenheld

Adam

. 2020. “Explaining Population Displacement Strategies in Civil Wars: A Cross-National Analysis.” International Organization 74(2): 253–94.

46.

Loyle

Cyanne E.

2021. “Rebel Justice during Armed Conflict.” Journal of Conflict Resolution 65(1): 108–134.

47.

Loyle

Cyanne E.

Cunningham

Kathleen Gallagher

Huang

Reyko

Jung

Danielle F.

. 2021. “New Directions in Rebel Governance Research.” Perspectives on Politics: 1–13, 264–276.

48.

Magaloni

Beatriz

Franco-Vivanco

Edgar

Melo

Vanessa

. 2020. “Killing in the Slums: Social Order, Criminal Governance, and Police Violence in Rio de Janeiro.” American Political Science Review 114(2): 552–572.

49.

Mampilly

Zachariah Cherian

. 2011. Rebel Rulers. Ithaca: Cornell University Press.

50.

Moncada

Eduardo

. 2017. “Varieties of Vigilantism: Conceptual Discord, Meaning and Strategies.” Global Crime 18(4): 403–423.

51.

Myrttinen

Henri

. 2021. “Violence Against the ‘Undesirables.’” MOSAIC MENA: 1-11.

52.

Osorio

Javier

Mohamed

Pavon

Viveca

Brewer-Osorio

Susan

. 2019. “Mapping Violent Presence of Armed Actors in Colombia.” Paper presented at International Cartographic Association Conference, Tokyo, Japan, 2019.

53.

Otero Bahamón

Silvia

. 2022. “¿En Dónde Combaten Los Actores Armados? Desigualdad de Tierras y Capacidad Estatal En El Conflicto Armado Colombiano.” Deasfíos 34(1): 1–45.

54.

Outright International . 2014. When Coming Out Is a Death Sentence: Persecution of LGBT Iraqis: Outright International.

55.

Phillips

Brian J

. 2017. “Inequality and the Emergence of Vigilante Organizations: The Case of Mexican Autodefensas.” Comparative Political Studies 50(10): 1358–89.

56.

Prem

Mounu

Saavedra

Santiago

Vargas

Juan F.

. 2020. “End-of-Conflict Deforestation: Evidence from Colombia's Peace Agreement.” World Development 129: 104852.

57.

Serrano-Amaya

José Fernando

. 2014. “Chiaroscuro: The Uses of ‘Homophobia’ and Homophobic Violence in Armed Conflicts and Political Transitions.” The University of Sydney.

58.

Sexton

Renard

. 2016. “Aid as a Tool against Insurgency: Evidence from Contested and Controlled Territory in Afghanistan.” American Political Science Review2 110(4): 731–49.

59.

Staniland

Paul

. 2012. “States, Insurgents, and Wartime Political Orders.” Perspectives on Politics 10(2): 243–64.

60.

Staniland

Paul

. 2021. Ordering Violence: Explaining Armed Group-State Relations From Conflict to Cooperation. Ithaca: Cornell University Press.

61.

Steele

Abbey

. 2017. Democracy and Displacement in Colombia’s Civil War. Ithaca: Cornell University Press.

62.

Stewart

Megan

, 2021. Governing for Revolution: Social Transformation in Civil War. New York: Cambridge University Press.

63.

Tao

Ran

Daniel

Strandow

Michael

Findley

, et al. 2016. “A Hybrid Approach to Modeling Territorial Control in Violent Armed Conflicts.” Transactions in GIS 20(3): 413–25.

64.

United States Army . 2013. FM 3-90-1 Offense and Defense Volume 1. Washington, DC: United States Army.

65.

Weinstein

Jeremy

. 2007. Inside Rebellion: The Politics of Insurgent Violence. Cambridge: Cambridge University Press.

66.

Wood

Elisabeth Jean

. 2009. “Armed Groups and Sexual Violence: When Is Wartime Rape Rare?” Politics & Society 37(1): 131–61.

67.

Wood

Reed M

Kathman

Jacob D

Gent

Stephen E

. 2012. “Armed Intervention and Civilian Victimization in Intrastate Conflicts.” Journal of Peace Research 49(5): 647–60.

68.

Yashar

Deborah

. 2018. Homicidal Ecologies: Illicit Economies and Complicit State in Latin America. Cambridge: Cambridge University Press.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.44 MB

0.00 MB

0.88 MB