Armed group proliferation and conflict dynamics

Abstract

How are conflict dynamics affected by contexts with many different actors compared with those dominated by a few? At the aggregate level, civil conflicts with many different rebels and militias experience higher levels of violence and human rights abuse, but the causal link between these two features remains underexplored. Are violent contexts particularly prone to organizational fragmentation or does the proliferation of armed groups lead to increased violence? This paper uses novel municipal-level monthly data from Colombia during 1975–2002 to investigate local conflict trajectories after additional non-state actors enter an area. Exploring both intrusion by (a) additional rebel groups and (b) anti-rebel militias, we find that increased militant competition leads to short-term increases in violence but that long-term trends depend on whether the “new” actors can access and form links with local social community networks.

Keywords

civil war escalation pro-government militias rebel organization

Introduction

Approximately half of contemporary civil conflicts around the world involve multiple rebel groups rather than just the regime and a challenger (Davies et al., 2023). In addition to this proliferation of anti-government organizations, current conflict contexts may also consist of pro-government militias, mercenaries, criminal organizations, and other armed non-state actors (Jentzsch et al., 2015; Thompson, 2023). Existing scholarship indicates that conflict environments with a proliferation of actors have several negative consequences, including more violence (Aliyev, 2020; Pearlman and Cunningham, 2012), more human rights abuses (Dorff et al., 2023), longer duration (Estancona and Reid, 2022), and being more likely to recur (Bara and Kreutz, 2022; Rudloff and Findley, 2016; Steinert et al., 2019).

Despite this, we have limited knowledge about the temporal and spatial consequences of actor proliferation at a more disaggregated level. We do not know whether the presence of many actors causes more violence or if the correlation is spurious, and whether the local proliferation of actors occurs before or after changes in conflict dynamics (Joo and Mukherjee, 2021; Schubiger, 2023). Theories of violent outbidding and territorial contestation suggest that more actors create more violence (Duyvesteyn, 2021; Nemeth, 2014), but empirical evidence on this important aspect of contemporary warfare remains limited.

This paper responds to this general lacuna, but we also expand the topic of study both theoretically and empirically in several ways. First, while existing scholarship on civil war violence explores battles or the use of terrorism, the vulnerability of civilian governance institutions is rarely considered. We argue that if the proliferation of actors increases the contest over local power, then civilian state structures are particularly at risk as potential targets for violence owing to their importance for local political order (Larson and Lewis, 2017; Sharif and Carranza-Franco, 2026). Second, we investigate whether the effect of actor proliferation is consistent regardless of whether the “new” actors are competitors or opponents to the existing rebel movements, connecting two currently disparate literature on inter-rebel contests and so-called pro-government militias (Abbs et al., 2020; Fjelde and Nilsson, 2018). Although the presence of more armed groups in general increases local power competition, militias espousing the narrative that the government lacks sufficient resources to provide security may avoid open hostility toward state forces. Consequently, we expect that actor proliferation through the addition of pro-government militias will target civilians rather than the military or state bureaucrats. Third, and finally, we explore if the effect of actor proliferation on violence is conditional on whether the “new” actors are mobilized locally or in other parts of the country. Since armed groups with links to local communities have better information about targets, they also need to act with care to preserve this advantage and thus be more selective in their targeting strategy (Kalyvas, 2006; Staniland, 2018).

Our empirical investigation employs unique new monthly data on violence at the municipality level in Colombia 1975–2002. During this period, the civil conflict pitted government forces against left-wing guerrilla organizations (FARC, ELN, M-19, EPL, and smaller groups), with the eventual addition of anti-rebel, right-wing militias locally referred to as paramilitaries. For our purposes, this context has several advantages. Armed group proliferation was not caused by splintering of existing rebel groups but rather the geographic spread of already established actors. The temporal dimension of our sample ensures that we capture the processes of both aggregate conflict escalation and armed group proliferation. In 1975, the conflict zone consisted of only 11 municipalities (1.2% of the total) and all of these contained just a single rebel group. At the end of our sample, in 2002, the conflict had expanded to 599 (53.6%) municipalities with almost a third (181) experiencing inter-rebel competition and the majority (354) experiencing both rebels and pro-government militias.

Following investigation of both the short- and long-term trajectories of violence in the months before and after actor proliferation in municipalities, we find general support for the proposition that violence escalates after new actors emerge in a conflict system. In addition to an immediate short-term effect in the first months after new actors have emerged, our analyses show that violence continues to persist at a higher level in multi-actor settings. However, we also identify differences in the types of violence that occur depending on the identity of the new actor. The inclusion of more rebel groups leads to more fighting (“battles”) in the conflict, while the presence of pro-government militias substantively increases the number of attacks on civilians and the state bureaucracy. In addition, we find that the escalation is temporally shorter if it involves the emergence of groups that are locally recruited (rebels or locally mobilized militias) compared with externally mobilized actors.

What do we know about conflict dynamics?

As our ambition is to explain and explore how actor proliferation influences conflict dynamics, we situate our study in the limited literature about the ebbs and flows of the use of violence in civil war. Existing research on conflict dynamics has primarily focused on (i) variation in conflict intensity across cases, (ii) the effect of negotiations on violence, and (iii) the variation in civilian targeting during conflict.

The first research program primarily explored how mobilization opportunities and military capacity influence different rebel organizations’ use of violence (Pamp et al., 2025; Staniland, 2018). Scholars have shown that mobilization along ethnic divisions means that potential opponents (i.e. co-ethnics to the rebel group, and co-ethnics to the regime) are easier to identify and this increased cross-ethnic vulnerability facilitates mobilization and subsequent violence (Abbs et al., 2020; Eck, 2009; Lewis, 2023). It has also been suggested that armed groups that form during an already active conflict will have a higher proportion of violence “experts” from the start which, in combination with the logic of outbidding, will lead to escalation (Perkoski, 2022; Robinson and Malone, 2024).

The second strand of research has drawn on disaggregated data to identify short-term fluctuations in fighting in connection with negotiations between the government and the rebels. On the one hand, there are indications that negotiations are more likely to both begin and succeed when insurgents are advancing and in particular when fighting gets closer to capital cities (Butcher, 2015; Greig, 2015), but also that the onset of negotiations reduces the number of clashes in areas of low strategic importance (Hinkkainen and Kreutz, 2019). Finally, scholarship on civilian targeting in conflict has identified that this is more common in ethnic conflicts, in more contested areas, and by actors that experience losses on the battlefield (Ottmann, 2017; Polo and González, 2020; Wood, 2014).

Although these studies reveal some insights about how governments and rebels in civil conflicts make strategic decisions about when and where to fight and abuse human rights, the dominant theoretical and empirical assumption remains that civil conflict is a two-player game between government and opposition. In what follows, we discuss the consequences of this two-player setting being disrupted by the addition of more non-state armed actors and the consequences for the trajectories of violence.

Local power contention and conflict escalation

A central aspect for understanding the use of violence in civil conflicts is the need for warring sides to establish power, and preferably outright control, over some local territory (Arjona, Kasfir and Mampilly, 2015). This ensures legitimacy about their ability to achieve their stated goals and provide access to resources and recruits, and offers protection to their constituents (Gates, 2002; Kalyvas, 2006). Armed actors’ constituency includes both fighting and non-fighting participants, including supporters who help with information and resources for reaching the goals of the organization (Kalyvas, 2006; Schubiger, 2023). This is acknowledged by rebel “practitioners” such as Sun Tzu, Mao Tse Tung and Che Guevara, who in their writing emphasize the importance of safe areas from which to expand the operational capacity of the organizations. In the words of the Vietnamese strategist Giap (1962: 96): “the building of bases for a steadfast and long resistance was an important strategic question and also a very successful experience of our Party” (emphasis in original).

It is therefore logical that an increase in the number of armed non-state groups active in an area threatens the power of previously existing actors, which motivates an increased use of force. Crucially, both existing actors and the “new” additions have incentives for escalating violence. For the new actors, this is to gain some power by disrupting the existing wartime political order while already existing actors obviously defend their position, but they may also see opportunities for offensives to further increase their power relative to all opponents in the conflict.

This study concerns situations where an already existing civil conflict expands with an additional actor, when several factors commonly discussed as challenges for organizational emergence have little explanatory power. Anti-government grievances exist, the collective action problem can be overcome, and state repression does not prevent mobilization. However, “new” actors in an area have incentives both to transmit their existence to local communities and to carve out a power base in relation to the existing warring sides. To do this, they want to emphasize how they differ from other actors by signaling greater commitment through the use of violence (Robinson and Malone, 2024; Soules, 2023). In addition, new groups must compete with existing organizations for the same pool of limited resources, in the form of recruits and civilian support, to become viable challenges (Fjelde and Nilsson, 2018). These joint aims increase the likelihood that new groups that seek to disrupt the existing wartime political order will engage in violence against all existing conflict actors in an area, but also seek to outbid the others actors in violence targeting non-combatants (Bloom, 2004).

The pre-existing organizations in an area may escalate violence either deliberately or due to processes of misattributed information or because patrols of different groups inadvertently engage with each other. It is also possible that the proliferation of armed groups in a limited geographical area presents opportunities for established groups to try to expand their power in relation to all of their opponents (Schutte and Weidmann, 2011). In the immediate aftermath of armed actor proliferation, both government and rebel forces will have to recalibrate how resources and troops are distributed as they are being attacked in different locations and ways than previously, which may weaken their capabilities.

A complementary effect is that a new actor in a conflict system may lead to an increase in certain types of violence as a consequence of inter-organizational tactical learning and inspiration. When “new” actors want to establish relevance in a new territory, they will often introduce new actions to the repertoire of violence, such as kidnappings, or war innovations, like improvised explosive devices.These will initially contribute to the escalation of violence but other actors will soon imitate or incorporate these methods in their portfolio of actions, triggering a long-term peak in violent events (Townsley et al., 2008).

The above discussion provides us with the following hypotheses:

H1: Actor proliferation in a conflict system is accompanied by increased violence among belligerents (battles).

H2: Actor proliferation in a conflict system is accompanied by increased violence against civilians.

Another key consideration relates to the potential impact of the process of consolidating power by an armed actor on the configuration of local politics. Studies on rebel governance (Arjona et al., 2015) show that insurgents often establish social and political orders that, in several cases, entail a process of capturing political structures or controlling state officials (Garay et al., 2010). This is often the case in long-running armed conflicts where insurgents develop a strategy of seizing local power when they are unable to overthrow national structures. This situation is facilitated, in cases like Colombia, by the existence of local elections, as this offers an alternative pathway to infiltrate the state. Armed groups can use violence to selectively displace supporters of opposing political parties (Steele, 2019) and influence electoral results by forcing citizens to vote for “their” candidates (García-Sánchez, 2016).

Attacking politicians or state officials is a fairly attractive strategy of targeted violence for two reasons. The visibility of a state official or a politician from a party allegedly linked to the insurgencies effectively signals to the population the arrival of the new group and its military capability. Furthermore, attacking these individuals is an effective way to dismantle the structures that sustain the existing political order. Indeed, advancing pressure directly on state officials is more effective than voter intimidation as it focuses explicitly on a visible symbol and important component of local power (Sharif and Carranza-Franco, 2026). Therefore, we argue that the entrance of a new actor in a conflict system will probably be accompanied by attacks on state representatives who form part of the social and political order established by existing actors.

This provides a third hypothesis:

H3: Actor proliferation in a conflict system is accompanied by increased violence against local state representatives.

Actor-specific features and conflict dynamics

Following on our discussion on the general effect of armed actor proliferation on escalation, our argument about territorial contestation may be conditional on some features of the specific type of actors that are introduced in a conflict. The first such consideration relates to whether actor proliferation consists of the addition of yet another rebel group or a pro-government militia. Although both types of actors are non-state armed groups with, at least in the Colombian setting, the ambition to establish themselves as local powerholders, their relationship obviously differs vis-à-vis the government forces.

Both actors have incentives to announce their presence even though they seek to legitimize their existence in different ways (Balcells, 2010). Rebel groups emphasize their challenge to the government and the need for political reform while militias advance the narrative that their presence is a response to rebel atrocities and the state's inability to uphold law and order (Molina and Castaño, 2001). This argument that the state is “weak” and “unable to respond” implies that the paramilitaries are “strong” and “more decisive”, which must be accompanied by action in order to be credible. However, the use of this narrative to legitimize the presence of the group also imposes some restrictions with regards to their repertoire of targets. To act in resonance with their statements, pro-government militias should primarily seek to engage with rebel forces and—maybe primarily—civilians who are confirmed or alleged of being rebel supporters (Mason and Krane, 1989; Mitchell et al., 2014).

This motivates our fourth hypothesis:

H4: Actor proliferation in a conflict system through pro-government militias is accompanied by increased violence against civilians.

Finally, violence is not only determined by the presence of actors in a conflict zone, but also to what extent they are capable of identifying relevant targets. Armed groups primarily recruited from within the conflict zone tend to be more selective and less extreme in their use of violence compared with those that rely on external recruitment (Lyall, 2010; Moore, 2019; Petersohn, 2014). This is because local recruits have more connections with the local communities, which gives them access to more precise and reliable information about who is actually part of competing groups and how they can be found (Kalyvas, 2006). To maintain access to such information networks, they have to tread carefully and ensure that they avoid targeting innocents, as the local community have reciprocal information about them (Staniland, 2018). We expect such local connections to be a moderating influence on the level of violent escalation after actor proliferation, leading to our final hypothesis:

H5: Actor proliferation in a conflict system through new groups that are recruited outside the conflict zone is accompanied by a more substantial increase in violence than when the “new” actors are recruited locally.

Case selection and research design

To explore actor proliferation and local conflict dynamics, we draw on a unique dataset about a multitude of violent actions undertaken by all conflict actors in Colombia during 1975–2002. In contrast to most studies of the Colombian conflict that cover the period from 1990 or more recent years, our sample is suitable specifically for investigating conflict escalation. The roots of the civil war in Colombia can be traced back to the decade of unrest (La Violencia) 1948–1958 or the formation of the rebel groups Fuerzas Armadas Revolucionarias de Colombia (FARC) and Ejército de Liberación Nacional (ELN) in 1964–1965, but by the mid-1970s, the intensity of fighting had subsided substantively. Although the organizations remained, the remaining core of the Ejército Popular de Liberacion (EPL) had fled to Cuba, while only 11 municipalities had any FARC or ELN presence at the start of our analysis (Hernández, 2006; Vallejo, 2021; Villamizar, 2022).

Figure 1 shows the distribution of countrywide violence across the time period we analyze, illustrating how violence escalated slowly in the late 1970s and then more substantively in both the 1980s and the early 1990s (Vélez, 2001). During these latter periods, the violence also expanded spatially across the country while armed actor proliferation became common in the majority of municipalities (Gutiérrez Sanín and Barón, 2005; Vélez, 2001).

Figure 1.

Violence over time in Colombia, 1975–2002.

Data

Our data on local conflict activity 1975–2002 are collected from an archive of seven local and one national newspapers from different political leanings ranging from the newspaper published by the Colombian Communist Party to a mainstream national paper, making it possible to triangulate information. The underlying data consist of reported “violent actions”, both war events and actions against individuals, which are disaggregated temporally and geographically as much as possible from these sources.¹ A violent event reported in the media may include multiple military actions and various forms of violence against individuals. To prevent overestimating political violence, we aggregate information on military and civilian-targeted actions using the following rule: we assign a value of 1 if an event involves at least one military action and likewise code 1 if it includes any violence against civilians.

This dataset includes the date and municipality of the event, the armed actor responsible for or involved in the event (as defined by the reporting source), and features of the victims such as profession and political leaning, if available. Table 1 provides a summary of the different actors reported as involved in the events, with their respective frequencies. The dataset provides detailed information on 14,277 different violent events covering the entire Colombian territory for 1975–2002.

Table 1.

Summary statistics of perpetrators.

Actor	From	To	Two-actor actions*	Single actions
State forces	1975	2002	4560	147
Paramilitaries (unspecified)	1982	2002	221	762
Autodefensas Unidas de Colombia (AUC)	1995	2002		106
Guerrilla (unspecified)	1975	2002	4609	1048
Fuerzas Armadas Revolucionarias de Colombia (FARC)	1975	2002		1795
Ejército de LiberaciónNacional (ELN)	1976	2002		855
Coordinadora Nacional Guerrillera (CNG)	1986	1995		55
Ejército Popular de Liberación (EPL)	1979	2002		45
M19	1986	1994		10
Quintín Lame	1986	1990		10
Unspecified	1975	2002		4749

*Meaning: state vs. paramilitaries, state vs. guerrillas, paramilitaries vs. guerrillas.

For our analysis, we collapse event information into municipality-months, which gives us cross-sectional time-series data consisting of 345,960 observations.² Violent actions were reported in 11,223 municipality-months (3.2%). However, since we seek to explain the escalation of violence following armed actor proliferation, we restrict our sample to comparing violence levels toward an existing pre-proliferation baseline of ongoing violence. This means that our analysis only includes municipalities that already experience armed conflict, i.e. the presence of government forces and at least one rebel group. Including the onset of the first rebel group in a municipality would bias our results toward our theoretical expectations, since the pre-armed actor period (the baseline) in those circumstances would consist of no conflict violence at all.³

Potential source bias

We acknowledge that relying on violence data from newspapers may suffer from some underreporting owing to source reliance on a network of correspondents or local sources (Croicu and Kreutz, 2017; Demarest and Langer, 2022). For this case, however, there is no alternative source of systematic data prior to the late 1980s, and we have sought to mitigate bias as well as verify the quality of the data by comparison with other material when possible. To begin with, the dataset is not derived from a single newspaper but is a combination of local and national sources with somewhat overlapping empirical focus and different ideological leanings. We have thus been able to triangulate information about events in the same geographical areas and how these are described by different sources.

In addition, since it was possible, we verified the data through comparison of some of the information collected from these sources with other sources for part of the time period. Focusing on fatalities and annual data for the time period 1989–2002, we find that the information from our newspaper data corresponds to trend data collected by offline sources (Memoria Histórica, 2021) and local non-governmental human rights organizations (Restrepo et al., 2006). The detailed output of this exercise is available in the Online Appendix.

Another specific potential bias in the Colombian conflict context is the risk that paramilitary violence is under-reported, as newspapers may have been pressured to obscure such incidents owing to collusion between paramilitaries and state forces (Davenport and Ball, 2002). However, comparing our data with other sources suggests that any such effect is marginal as our data do not consist disproportionately of guerrilla violence. Notably, this is not even the case in the years that qualitative post-conflict investigations have revealed as particularly lop-sided with regards to perpetrator violence (Comision de la Verdad, 2022). In 1996, the year that marked the final stage of FARC military expansion, our database registers 607 violent events, with 20% attributed to FARC and 10% to paramilitaries. In contrast, in 2000, a year characterized by paramilitary offensives, our database reports 837 violent events, 24% perpetrated by the FARC and 31% by paramilitaries.

Independent variables

The foundational origins of the rebel organizations in Colombia precede the time period of our study. FARC and ELN were founded in 1964, EPL in 1967, and M19 and Quintín Lame in 1974 with the latter purely non-violent until 1977 (CNG was not a unique group but a vehicle for coordinating strategy and some collaborative actions between existing rebel groups.) Thus, our independent variable does not specifically capture the formation of new rebel groups, but the month and year when an organization established a local presence in the form of camps, regular patrols, and/or use of force. Information about when such movements occurred is available from internal party decisions and communication, and official statements (at the time) in the form of proclamations, the distribution of leaflets, and public meetings.

Militias were, in contrast to rebels, formed over time, but they announced their presence in similar ways. Since these justified their presence as pro-government forces, they often announced their existence publicly at meetings, with leaflets, in local media, and by graffiti. There are many instances of similar practices that we use to identify when existing militias moved into “new” municipalities, but information about these processes is also provided by interview material from demobilized former militia members after 2003.

We have relied on numerous sources to collect information that is as specific as possible about the dates when additional actors established a presence in a local area. Besides information sourced from the excellent work of the Memoria Histórica (2021), we have surveyed in-depth investigations by Colombian academics, non-governmental organizations, and government documentation of the conflict. A selection of particularly useful sources includes Tate (2001), Vélez (2001), Ferro and Uribe (2002), Romero (2003), Gutiérrez Sanín and Barón (2005), Hernández (2006), Sánchez and del Mar Palau (2006), Troyan (2008), Barón Villa (2011), Spencer (2011) González et al. (2012), González et al. (2014), Ronderos (2014), Programa Presidencial de Derechos Humanos y DIH (2014), Daly (2016), Steele and Schubiger (2018), Vallejo (2021), Comision de la Verdad (2022), Villamizar (2022), Castellanos Bautista et al. (2023), and Roux (2023).

Our indicator about the local mobilization of armed groups only captures variation in relation to the pro-government militias. That is because all instances of armed actor proliferation by a rebel group were primarily locally recruited. For militias, there is variation in recruitment patterns in different municipalities across Colombia. In general, militias were more successful in recruitment from urban rather than rural areas, even though it was in the latter regions they eventually deployed (Arjona and Kalyvas, 2012). Furthermore, several of the larger and more organized militias later organized outright invasions into new municipalities, ostensibly to counter rebel activity there, but largely also to take control over resource extraction and other business opportunities. An example of such an invading group was the sudden appearance of a paramilitary group calling themselves Bloque Catatumbo in 1998 in several municipalities of Norte de Santander on the Venezuelan border. This group contained very few locals as it was made up of members of militias in the regions of Cordoba, Antioquia, and Uraba that were transported by their leaders more than 1000 kilometers to set up the group (Daly, 2016; Reyes Navarro, 2020).

For both practical and theoretical reasons, we define a locally recruited pro-government militia as composed of at least 40% recruits and/or leaders from the municipality of its activity. This follows the praxis set by Daly (2016) and also makes it possible for us to partly use data from that study in our analysis. We collected complementary data for the militias not included in that book using secondary accounts and local sources. As an operationalization, this threshold helps us distinguish specifically the militias with limited connections to local communities, as that is the theoretical distinction of interest (e.g. Lyall (2010)).

In our data, we indicate the presence of two or more rebel organizations and/or pro-government militias active in a municipality in a given month with a dichotomous variable.

Dependent variables

Our hypotheses specify three different types of actors that are at risk of becoming victimized when additional actors join a conflict, and we investigate each of these in separate analyses. Owing to the high granularity and quality of the conflict dataset, we are able to move beyond the standard practice of focusing only on fatalities. Our focus is thus on the victims regardless of whether these have been killed, injured, kidnapped, or threatened in an event involving any conflict actors.

To be clear, our unit-of-analysis is municipality-month and the dependent variables are the aggregate number (i.e. count) of victims for the categories presented below. Municipality-months with no violent events have zero victims.

Our first category consists of combatants, which we define as members of the armed forces, the police, pro-government militias, and guerrillas. For civilians, we include everyone else except those in public roles as part of the governance apparatus (Governor, Mayor, Council member, Congress member, Deputy, Fiscal (Prosecutor), Personero, Veedor, public servant, and human rights defender). Those listed here are coded as state representatives.

Estimations and potential confounds

To identify the effect of the entrance of a new actor into a local conflict system, we employ several estimation techniques specifically tailored to remove the possible effect of context-specific confounds. Our ambition is to identify trends in violence through quasi-experimental models comparing across relevant counterfactual environments, which motivates our use of Regression Discontinuity and multi-level models. In order to test the robustness of our findings, we also estimate a more traditional cross-sectional setup in the form of an interrupted time-series approach and a battery of control variables.

We initially zoom in to the immediate short-term effect of the emergence of a new actor in a municipality as this constitutes the main focus of our argument about violence trajectories. Following a data-driven optimization of the best fit of pre- and post-functional forms (Calonico et al., 2014), we employ the most common assumption of linear trends for the 12 months prior and post every actor proliferation event. We estimate a Regression Discontinuity Design (RDD) in the form of local-level regressions controlling for time trends (Cattaneo and Titiunik, 2022). Under mild continuity assumptions, this means that our sample consists of observations that are ex ante comparable in all other ways (on average) except variation in the number of armed actors present. A crucial assumption for RDD is that all potentially relevant factors besides the treatment and the outcome variables are continuous at the point where the treatment and outcome discontinuities occur.

The RDD design can be described as:

ρ_{i j} = α + θ_{t} + θ 2_{t} + ϵ_{i j}

where ρ is the observed victims following the entrance of a new actor i in municipality j, α is the intercept, θ represents the baseline time trends, θ2 the post-proliferation time trends, and ɛ the error term.

In the next step, we expand our analytical focus to a larger temporal window while preserving a parsimonious empirical setup by the use of a two-level mixed-effects local linear regressions to systematically address long-and short- term temporal confounds while accounting for unobserved municipality-level variation. Our two-level estimation models are given by:

ρ i j t = α + θ t β a b c + θ 2 a t, b t, c t + η j + ϵ i j t

where ρ is the number of observed victims following the entrance of a new actor i at time t in municipality j, α is the common intercept, θ represents the baseline time trends, βa the establishment of new rebel actor, βb local militia, βc external militia, θ2 time trends after observed proliferation, η is the random intercept for municipality j, and ɛ is the error term. This approach captures more long-term temporal trajectories while taking into account dependence between observations within municipalities and the conflict over time (Gelman, 2006; Hair and Fávero, 2019). We specify random intercepts to account for systematic differences in baseline risk of new actors emerging in different municipalities, while the slopes, or variable effect sizes, are assumed to be constant.

In our final estimations, we open up our modeling for the possibility of contextual factors influencing when and where armed group proliferation occurs. We use an interrupted time series (Wagner et al., 2002) (functionally similar to the RDD) with temporal and municipality-specific control variables instead of the two-level setup. This approach can be described as:

ρ i j t = α + θ t + β a b c + θ 2 a t, b t, c t + Σ i j t + ϵ i j t

where ρ is the number of observed victims following the entrance of a new actor i at time t in municipality j, α is the common intercept, θ baseline time trends, βa a new rebel actor, βb a local militia, βc an external militia, and θ2 time trends after observed proliferation. In this model, Σ is a matrix of confounds and the error term ɛ_t is a normally distributed random error and an error at time t that may be correlated to errors at preceding or subsequent time points clustered at municipality level. To ensure comparability with relevant counterfactuals, these models are contrasted with the sample of municipality-months containing one rebel group and no militias, thus excluding observations of non-armed conflict.

The control variables added into our estimations consist of overall conflict contexts that can motivate shifts in contention as identified by existing research. We collected information on the presence of institutionalized alliances among different rebel groups or militias, periods of peace talks, and periods when the formation of militias was legal. In addition, we also control for municipality-level characteristics that may influence the probability of violence. This includes natural resources in the form of gold, silver, platinum, coal, mineral deposits, and coca production provided by mindat.org (2023). We also control for local municipality statistics in the form of GDP/capita, population, area, mean altitude, and distance to the capital Bogota. These are sourced from CEDE (2022) with the data temporally backcast for non-included years. Further, we include an indicator of whether the municipality was affected by the 1948–53 La Violencia taken from Daly (2012), as this may indicate an area with high recruitment potential.

We also control for features of the local political context that may incentivize certain types of violence, specified as during elections and local power holders. Using official records, we include a variable for national elections capturing the month prior to and the month of voting for Parliament, and the two months prior and the month of voting for Presidential elections and eventual runoff rounds; and another variable that indicates the month prior to and the month of voting for municipal and regional ballots. Finally, we also control for whether the local administration was headed by leftists, Conservative Party, Liberal Party, or a candidate supported by paramilitaries. This data is provided by García-Sánchez (2009) and only available for 1988–2003, so this reduces our sample somewhat. Descriptive statistics for all of our variables are available in the Online Appendix.

Results

Out main hypotheses suggest that armed actor proliferation in general increases battlefield violence, violence against civilians, and violence against state representatives. We further explore whether conditioning factors in the form of pro-government militias will primarily increase violence against civilians, and whether locally recruited militias will lead to a more moderate increase in violence than those mobilized from other municipalities. We address each of these hypotheses in turn as we move through our different results.

Figure 2 presents the output from regression discontinuity estimations with combatant victims as the dependent variable.⁴ We employ a data-driven calculation of optimal bin size for estimation, but present the output from estimations based on the temporal window of 12 months prior and post actor proliferation across all panels for comparison. The width of the confidence interval is probably influenced by the variation in observations for the different independent variables, meaning that we will focus our discussion on the average trends. For the short-term effect, we estimate predicted probabilities of combatant victims two months prior and two months past the emergence of new actors in these samples. This four-month window was selected to limit the potential influence of extreme outliers and uncertainty about dates for actor emergence. We find that proliferation of rebel actors has the greatest escalatory effect, although militias are more likely to join into already violent settings. The emergence of an additional rebel group in a municipality increases the violence from an predicted 0.17 victims (95% confidence interval: 0.10–0.24) to 0.34 (0.28–0.40) victims in our four-month window. For local militias, the respective increases range from 0.21 (0.13–0.29) to 0.37 (0.27–0.46) victims, and for external militias from 0.46 (0.24–0.68) to 0.70 (0.46–0.94) victims.

Figure 2.

Conflict expansion and battlefield intensity. Note: shading indicates sample average within bins and lines indicate polynomial fit of order 4.

The findings provide partial support the first hypothesis suggesting a short-term increase in battles following actor proliferation owing to violent outbidding and competition over territory (Bakke et al., 2012; Fjelde and Nilsson, 2018). Although we identify an average increase in visible battles across all categories, the trend is only pronounced after the addition of more rebel movements into the conflict setting. The varied effect on battles following the introduction of pro-government militias suggests that these actors may be unwilling to risk getting drawn into open conflict with the state military.

Figure 3 focuses on armed actor proliferation in the conflict zone and the trajectories of attacks on civilians.¹¹ We find support for hypothesis two that attacks on civilians increase regardless of the type of new actor, and the narrower confidence bands illustrate less variation in this effect across municipalities and time. Following predicted estimations of the number of attacks, we see similar size of effect across types of proliferation although the absolute number of victims is larger following the introduction of pro-government militias. The short-term effect of additional rebel groups is an increase from 0.06 (0.04–0.09) to 0.20 (0.17–0.23) victims, for local militias from 0.14 (0.08–0.21) to 0.38 (0.29–0.46) victims, and for external militias from 0.19 (0.05–0.33) to 0.70 (0.47–0.92) victims.

Figure 3.

Conflict expansion and attacks on civilians. Note: shading indicates sample average within bins and lines indicate polynomial fit of order 4.

Taken together, the findings on combatant and civilian victimization presented in Figures 2 and 3 confirm hypothesis four that suggested that pro-government militias will primarily target civilians. This is consistent with many qualitative accounts of the conflict that pro-government militias perpetrated the majority of the human rights abuses of the civilian population (Stanton, 2015; Carey et al., 2016). This went beyond targeting alleged rebel collaborators to also target “delinquents, street youth, prostitutes, drug users, gays, petty thieves, and anyone else who might negatively affect businesses” (Sanford, 2003: 76). This suggests that—at least in the Colombian case—despite their rhetoric of being a reactive phenomenon to conflict dynamics, the long-term behavior of militias was to take advantage of the inability or unwillingness of state forces to hold them accountable for opportunistic killings.

We also find partial support for hypothesis five, as there is a more substantial increase in violence against civilians by militias recruited from outside the conflict zone than those with local connections. There does not seem to be any substantial variation across the categories with regard to the duration of the escalation as violence against civilians persists at a higher level than prior to the actor proliferation a full year later.

Figure 4 presents the results from estimations of attacks on state representatives and institutions following armed actor proliferation in the conflict system. We see a general increase that confirms hypothesis three. Although the relative increases are substantial across all types of actor proliferation, it is the emergence of militias that creates the greatest risk for state representatives. This is somewhat surprising since these are supposedly formed to protect the state from the insurgents. The predicted short-term effect of new rebels is an increase in victims from 0.03 (0.02–0.048) to 0.07 (0.051–0.09), for local militias from 0.02 (0.00–0.03) to 0.08 (0.05–0.12), and from external militias from 0.01 (0.00–0.030) to 0.09 (0.034–0.15).

Figure 4.

Conflict expansion and attacks on local state representatives. Note: shading indicates sample average within bins and lines indicate linear fit.

The trajectories of violence also remain similar when looking at the long-term effects, where the effect of the proliferation of rebel groups is brief. Already a few months after proliferation through rebels, the pattern of violence against state representatives has reverted to the baseline. In contrast, the establishment of pro-government militias leads to an increase in violence that persists. We interpret this finding as indicative of an ambition of the paramilitary forces to assert control over the local state apparatus and thereby institutionalize—or at least garner broad-based acceptance for—themselves as a partner in local (security) governance. This meant targeting not only those suspected of links to the rebels, but also those with concerns about the involvement of the militias in criminal ventures such as drug trafficking and resource extraction.

In sum, our RDD estimations provide support for hypotheses one, two, and three about the risk of violence escalation in the immediate aftermath of actor proliferation. We also find partial support for hypothesis four that suggest that pro-government militias will avoid engaging in battles, but our theoretical argument does not account for the substantial increase in attacks on state representatives by these militias. We also find partial support for hypothesis five about the difference between armed groups that are locally and externally recruited. The former does use less violence than the latter across all categories, but this is most substantial for battlefield interaction and attacks on civilians.

Long-term effects of multiple actor presence

In most of the local-level regressions presented above, we find that the immediate escalation after armed actor proliferation is temporary and reverts toward the baseline trend over time. Does this mean that the general effect of more violence in multiple-actor conflict settings is completely determined by these short-term spikes? Since our RDD estimations do not extend beyond the first 10 months after actor proliferation, we estimate two-level interrupted time series analyses to account for longer trends.

Table 2 presents the findings on the long-term effects of multiple rebel groups and militias (local or external) in municipalities, controlling for overall temporal conflict trends. Regardless of whether proliferation occurs through additional rebel groups or militias, we observe a significant escalation in battlefield victims, violence against civilians, and attacks on state representatives. For experts on the Colombian civil conflict, this pattern is unsurprising; however, our analysis highlights trajectories at the municipal level rather than at the aggregate conflict level (Restrepo et al., 2006). Table 2 also suggests an overall trend resembling an inverted U: the Time variable has a positive coefficient, while Time squared is negative. This confirms the short-term escalatory effect identified in our earlier analyses.

Table 2.

Two-level interrupted time series regression on the effects of conflict expansion.

	(1)	(2)	(3)
Dependent variable: victims	Combatants	Civilians	State representatives
Multiple rebels	0.208^∗∗∗	0.045^∗∗∗	0.048^∗∗∗
	(0.005)	(0.004)	(0.002)
Local militia	0.117^∗∗∗	0.159^∗∗∗	0.026^∗∗∗
	(0.005)	(0.003)	(0.001)
External militia	0.265^∗∗∗	0.227^∗∗∗	0.044^∗∗∗
	(0.006)	(0.004)	(0.002)
Time	0.001^∗∗∗	0.001^∗∗∗	0.0002^∗∗∗
	(0.00003)	(0.00002)	(0.00001)
Time (squared)	−0.000003^∗∗∗	−0.000001^∗∗∗	−0.0000004^∗∗∗
	(0.00000007)	(0.00000005)	(0.00000002)
Post multiple rebels	0.001^∗∗∗	0.002^∗∗∗	−0.0003^∗∗∗
	(0.0001)	(0.00008)	(0.00004)
Post multiple rebels (squared)	−0.000001^∗∗∗	−0.00002^∗∗∗	0.0000002
	(0.0000006)	(0.0000004)	(0.0000002)
Post militia	−0.001^∗∗∗	−0.003^∗∗∗	−0.0003^∗∗∗
	(0.0001)	(0.00008)	(0.00004)
Post militia (squared)	0.000001^∗	0.00001^∗∗∗	0.0000001
	(0.0000005)	(0.0000003)	(0.0000002)
Constant	−0.019^∗∗∗	−0.006	−0.003^∗∗
	(0.006)	(0.003)	(0.001)
Municipality level
Var (constant)	−1.724^∗∗∗	−2.230^∗∗∗	−3.438^∗∗∗
Var (residual)	(0.021) −1.027^∗∗∗	(0.022) −1.381^∗∗∗	(0.022) −2.148^∗∗∗
	(0.001)	(0.001)	(0.001)
N	345,960	345,960	345,960
Municipalities	1118	1118	1118

Standard errors in parentheses.

∗

p < 0.05, ^∗∗ p < 0.01, ^∗∗∗ p < 0.001.

Considering the partial support for hypothesis four in our previous analysis, we focus specifically on the information the long-term analysis provides us with regarding the different types of armed actor proliferation. When another rebel group has been added to the conflict system, we see that the post-multiple rebels trend remains positive over time. That means that the violence level on average remains higher in municipalities with several rebel movements than in those containing only one rebel group. This pattern is not visible in the municipalities where the armed actor proliferation consists of the presence of pro-government militias, where violence decreases somewhat over time. The size of the effect of this decline is notably small, and post-estimation analyses show that it takes on average about two decades for violence levels to return to those recorded before the pro-government militia became active.

To provide a robustness test of our analysis, we conclude our empirical investigation with an approach where we investigate our hypotheses in the presence of a battery of potential confounds that may influence the temporal and spatial distribution of actor proliferation and conflict violence. Figure 5 presents the output of this analysis, with the results for our main independent variables in the first panel and—for clarity, but also transparency—the context variables in the second. Considering that these models incorporate information from a much longer time window than the previous estimations, they are more sensitive to the uneven underlying temporal distribution of guerrilla and paramilitary presence in the country, which makes this output less relevant for comparing across these categories. We therefore merely confirm that our general finding that violence escalates in areas with multiple groups is robust even in this setup, although the effect sizes are unreliable.

Figure 5.

Multiple actors, context, and intensity of violence. (a) Type of conflict expansion. (b) Context. Note: interrupted time series estimations with robust standard errors clustered on municipality. Full output of estimations are available in Online Appendix Table A4.

This main informative point of this exercise is to discuss whether certain contexts have a higher underlying baseline for violence and that our findings could be spurious. Focusing first on instances of a greater risk of violence, we find an increased risk of battlefield interaction during periods when different rebel groups coordinated strategy in an alliance, during peace talks, in larger municipalities, and in areas with coca production. That fighting increases during peace negotiations have been identified elsewhere (Hinkkainen and Kreutz, 2019), but our analysis shows that controlling for this factor has little overall effect of the proliferation of armed groups on escalation.

When it comes to violence against civilians, we see that this occurred to a large extent during the time period(s) when some forms of pro-government militias had some legal recognition (1968–1988, and 1993–1999), and during peace talks. Both of these features coincided with the spikes in violence in the late 1980s and mid-1990s. However, if we look beyond the Colombian context, these findings are consistent with comparative studies about when violence on civilians is particularly likely (Aliyev, 2020; Thomas, 2014). Finally, we find that attacks on both civilians and state representatives are more common in municipalities with higher GDP/capita. Taken together, a variety of analyses of the relationship between armed group proliferation and conflict dynamics in Colombia show strong support for our general argument that a greater number of actors breeds more violence. In addition, we also find that the main effect of the introduction of pro-government militias in a conflict system leads to more attacks on civilians as well as state representatives.

Conclusions

This paper examines the local dynamics of civil conflict violence when additional actors join a conflict system. Following an analysis of Colombia 1975–2002, we find that the introduction of additional rebel groups or pro-government militias increases all forms of violence at the municipal level. These findings offer a contribution about the temporal ordering between the number of actors and the dynamics of violence during civil conflict, but we also identify differences depending on what type of actor emerges in a municipality. Increased competition between rebel groups is characterized by a brief increase in all types of violence, but this is largely a temporary spike. In contrast, there is variation across victimization following the introduction of pro-government militias into a conflict setting. These groups contribute not to an increase in battles, but rather a substantial increase in attacks on civilians and local state representatives.

Moreover, our analysis shows that the increase in violence was particularly severe in regions where paramilitary forces were external to the area. In these municipalities, civilians and state representatives became the primary targets of violence. Our findings align with the existing literature, which suggests that locally recruited militias are generally more selective in their actions owing to greater access to local information. Violence against state representatives and civilians also persisted over time. This finding suggests that pro-government militias may have other ambitions than to defeat the insurgents in these regions, including establishing themselves as crucial power agents in the local social and political order. Our study also draws attention to the role of local political and bureaucratic institutions as contested arenas during civil conflict even for armed actors that oppose or claim to support the state.

This illustrates a potential long-term consequence of the growing acceptance or promotion of pro-government militias as cost-effective security agents during war. Just as rebel wartime political institutions and economic networks may be difficult to dismantle when peace is achieved (Berti, 2016; Steele and Schubiger, 2018), a similar challenge exists for reclaiming governance from these semi-independent supporters of the state (Sharif and Carranza-Franco, 2026).

This and many other aspects of this project and its findings deserve further attention, as well as many potential follow-up projects that may derive from this study. This includes greater attention to detail with respect to both conflict contexts and the features of existing armed actors in the region. Scholars have shown that aspects such as the experience of the rebel leader (Doctor, 2020) and the homogeneous support base (Mosinger, 2018) create organizations less at risk of splintering. These and many other aspects will benefit from investigations of conflict environments beyond the specific case of Colombia with, specifically, a broader sample of different organizations. What our study shows is that this research benefits from not restricting the sample to parent-and splinter- rebel groups but also including armed proliferation by other actors such as independently formed rebel contenders and pro-government militias.

To wrap up, the main contribution of this article is to highlight the complex interactions between multiple actors at the local level that determine how power is distributed during civil war, and the immediate negative consequences that occur following the proliferation of armed actors. Insights about this can advance not just the academic debate about the understanding of civil war, but also help in predicting shifts in conflict dynamics. With the knowledge that actor proliferation is likely to lead to increase violence, especially toward civilians, this can help inform responses in the form of preparation for humanitarian support and extending protection to official state representatives.

Supplemental Material

sj-pdf-1-cmp-10.1177_07388942261445652 - Supplemental material for Armed group proliferation and conflict dynamics

Supplemental material, sj-pdf-1-cmp-10.1177_07388942261445652 for Armed group proliferation and conflict dynamics by Miguel García-Sánchez and Joakim Kreutz in Conflict Management and Peace Science

Footnotes

ORCID iD

Joakim Kreutz

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: García-Sánches was supported by the School of Social Sciences at Univerdidad de los Andes through the FAPA fund; Kreutz was supported by Vetenskapsrådet (grant number 2020-02368).

Declaration of conflicting interests

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

Supplemental material

Supplemental material for this article is available online.

Notes

References

Abbs

Clayton

Thomson

(2020) The ties that bind: Ethnicity, pro-government militia, and the dynamics of violence in civil war. Journal of Conflict Resolution 64(5): 903–932.

Aliyev

(2020) Why are some civil wars more lethal than others? The effect of pro-regime proxies on conflict lethality. Political Studies 68(3): 749–767.

Arjona

Kasfir

Mampilly

(2015) Rebel Governance in Civil War. New York: Cambridge University Press.

Arjona

Kalyvas

(2012) Recruitment into armed groups in Colombia: A survey of demobilized fighters. In: Guichaoua

(ed) In Understanding Collective Political Violence. London: Palgrave Macmillan, 143–171.

Bakke

Cunningham

Seymour

(2012) A plague of initials: Fragmentation, cohesion, and infighting in civil wars. Perspectives on Politics 10(2): 265–283.

Balcells

(2010) Rivalry and revenge: Violence against civilians in conventional civil wars. International Studies Quarterly 54(2): 291–313.

Bara

Kreutz

(2022) To buy a war but sell the peace? Mercenaries and post-civil war stability. Security Studies 31(3): 417–445.

Barón Villa

(2011) Apogeo y caída de las autodefensas de Puerto Boyacá: del paramilitarismo a los señores de la guerra en el Magdalena Medio. Available from: https://repositorio.unal.edu.co/handle/unal/5629

Berti

(2016) Rebel politics and the state: Between conflict and post-conflict, resistance and co-existence. Civil Wars 18(2): 118–136.

10.

Bloom

(2004) Palestinian suicide bombing: Public support, market share, and outbidding. Political Science Quarterly 119(1): 61–88.

11.

Butcher

(2015) ‘Capital punishment’: Bargaining and the geography of civil war. Journal of Peace Research 52(2): 171–186.

12.

Calonico

Cattaneo

Titiunik

(2014) Robust data-driven inference in the regression-discontinuity design. The Stata Journal 14(4): 909–946.

13.

Carey

Colaresi

Mitchell

(2016) Risk mitigation, regime security, and militias: Beyond coup-proofing. International Studies Quarterly 60(1): 59–72.

14.

Castellanos Bautista

Castillo Niño

Villamizar

(2023) Génesis del movimiento sindical y lucha armada. Historia del Movimiento 19 de abril en San Alberto, Cesar (1974–1990). Historia Regional (50): 1–17. Retrieved from https://historiaregional.org/ojs/index.php/historiaregional/article/view/805.

15.

Cattaneo

Titiunik

(2022) Regression discontinuity designs. Annual Review of Economics 14: 821–851.

16.

CEDE, Centro de Estudios sobre Desarrollo Economico (2022) Panel de caracterısticas generales. Bogota, Colombia: Universidad de Los Andes. 10.57924/IX38JI (accessed 26 February 2020).

17.

Comisión de la Verdad (2022) Hay futuro si hay verdad. Informe final de la Comision para el esclarecimiento de la verdad, la convivencia y la no repeticion. Tomo 3. No mataras: relato historico del conflicto armado interno en Colombia...

18.

Croicu

Kreutz

(2017) Communication technology and reports on political violence: Cross-national evidence using African events data. Political Research Quarterly 70(1): 19–31.

19.

Daly

(2012) Organizational legacies of violence: Conditions favoring insurgency onset in Colombia, 1964–1984. Journal of Peace Research 49(3): 473–491.

20.

Daly

(2016) Organized Violence After Civil War: The Geography of Recruitment in Latin America. New York: Cambridge University Press.

21.

Davenport

Ball

(2002) Views to a kill: Exploring the implications of source selection in the case of Guatemalan state terror, 1977–1995. Journal of Conflict Resolution 46(3): 427–450.

22.

Davies

Pettersson

Öberg

(2023) Organized violence 1989–2022, and the return of conflict between states. Journal of Peace Research 60(4): 691–708.

23.

Demarest

Langer

(2022) How events enter (or not) data sets: The pitfalls and guidelines of using newspapers in the study of conflict. Sociological Methods & Research 51(2): 632–666.

24.

Doctor

(2020) A motion of no confidence: Leadership and rebel fragmentation. Journal of Global Security Studies 5(4): 598–616.

25.

Dorff

Gallop

Minhas

(2023) Network competition and civilian targeting during civil conflict. British Journal of Political Science 53(2): 441–459.

26.

Duyvesteyn

(2021) Rebels and Conflict Escalation: Explaining the Rise and Decline in Violence. New York: Cambridge University Press.

27.

Eck

(2009) From armed conflict to war: Ethnic mobilization and conflict intensification. International Studies Quarterly 53(2): 369–388.

28.

Estancona

Reid

(2022) Pro-government militias and civil war termination. Conflict Management and Peace Science 39(3): 291–310.

29.

Ferro

Uribe

(2002) El orden de la guerra: las FARC-EP, entre la organizacion y la política. Bogotá: Pontificia Universidad Javeriana.

30.

Fjelde

Nilsson

(2018) The rise of rebel contenders: Barriers to entry and fragmentation in civil wars. Journal of Peace Research 55(5): 551–565.

31.

Garay

Salcedo

De Léon

(2010) Redes de poder en Casanare y la Costa Atlántica. In: Y refundaron la patria. De como mafiosos y políıticos reconfiguraron el Estado colombiano. Bogota: Debate, 215–266.

32.

García-Sánchez

(2009) Political violence and electoral democracy in Colombia. Participation and voting behavior in violent contexts. PhD thesis. University of Pittsburgh.

33.

García-Sánchez

(2016) Control territorial y decisión de voto en Colombia. Un enfoque multinivel. Polıtica y gobierno 23(1): 57–95.

34.

Gates

(2002) Recruitment and allegiance the microfoundations of rebellion. Journal of Conflict Resolution 46(1): 111–130.

35.

Gelman

(2006) Multilevel (hierarchical) modeling: What it can and cannot do. Technometrics 48(3): 432–435.

36.

Giap

(1962) People’s War, People’s Army. New York: Praeger.

37.

González

Nieto Matiz

Aponte González

, et al (2012) Conflicto y territorio en el oriente Colombiano. Bogotá: Odecofi-Cinep.

38.

González

Quiroga

Ospina-Posse

, et al. (2014) Territorio y Conflicto en la Costa Caribe. Bogotá: Odecofi-Cinep.

39.

Greig

(2015) Rebels at the gates: Civil war battle locations, movement, and openings for diplomacy. International Studies Quarterly 59(4): 680–693.

40.

Gutiérrez Sanín

Barón

(2005) Re-stating the state: paramilitary territorial control and political order in Colombia (1978-2004). (Crisis States Research Centre working papers series 1, 66. Crisis States Research Centre, London School of Economics and Political Science.

41.

Hair

Jr Fávero

(2019) Multilevel modeling for longitu- dinal data: Concepts and applications. RAUSP Management Journal 54: 459–489.

42.

Hernández

(2006) Rojo y negro: historia del ELN. Tafalla: Txalaparta.

43.

Hinkkainen

Kreutz

(2019) Natural resource wars in the shadow of the future: Explaining spatial dynamics of violence during civil war. Journal of Peace Research 56(4): 499–513.

44.

Jentzsch

Kalyvas

Schubiger

(2015) Militias in civil wars. Journal of Conflict Resolution 59(5): 755–769.

45.

Joo

Mukherjee

(2021) Rebel command and control, time, and rebel group splits. International Interactions 47(2): 318–345.

46.

Kalyvas

(2006) The Logic of Violence in Civil War. New York: Cambridge University Press.

47.

Larson

Lewis

(2017) Ethnic networks. American Journal of Political Science 61(2): 350–364.

48.

Lewis

(2023) Rebel group formation in Africa: Evidence from a new dataset. World Development 170: 106207.

49.

Lyall

(2010) Are coethnics more effective counterinsurgents? Evidence from the second Chechen war. American Political Science Review 104(1): 1–20.

50.

Mason

Krane

(1989) The political economy of death squads: Toward a theory of the impact of state-sanctioned terror. International Studies Quarterly 33(2): 175–198.

51.

Memoria Histórica

(2021) ¡ Basta ya! Colombia: Memorias de guerra y dignidad.” Retrieved from https://centrodememoriahistorica.gov.co/micrositios/informeGeneral/.

52.

mindat.org (2023) The Hudson Mineralogical Association. https://www.mindat.org. (Accessed December 3, 2023).

53.

Mitchell

Carey

Butler

(2014) The impact of pro-government militias on human rights violations. International Interactions 40(5): 812–836.

54.

Molina

Castaño

(2001) Mi confesion: Carlos Castaño revela sus secretos. Bogotá: Editorial Oveja Negra.

55.

Moore

(2019) When do ties bind? Foreign fighters, social embeddedness, and violence against civilians. Journal of Peace Research 56(2): 279–294.

56.

Mosinger

(2018) Brothers or others in arms? Civilian constituencies and rebel fragmentation in civil war. Journal of Peace Research 55(1): 62–77.

57.

Nemeth

(2014) The effect of competition on terrorist group operations. Journal of Conflict Resolution 58(2): 336–362.

58.

Ottmann

(2017) Rebel constituencies and rebel violence against civilians in civil conflicts. Conflict Management and Peace Science 34(1): 27–51.

59.

Pamp

Thurner

Binder

, et al. (2025) Arming to fight: Rebel-government militarization and the escalation of violence in civil wars. Conflict Management and Peace Science 42(4): 357–379.

60.

Pearlman

Cunningham

(2012) Nonstate actors, fragmentation, and conflict processes. Journal of Conflict Resolution 56(1): 3–15.

61.

Perkoski

(2022) Divided, Not Conquered: How Rebels Fracture and Splinters Behave. New York: Oxford University Press.

62.

Petersohn

(2014) The impact of mercenaries and private military and security companies on civil war severity between 1946 and 2002. International Interactions 40(2): 191–215.

63.

Polo

González

(2020) The power to resist: Mobilization and the logic of terrorist attacks in civil war. Comparative Political Studies 53(13): 2029–2060.

64.

Programa Presidencial de Derechos Humanos y DIH, Observatorio (2014) Atlas del Impacto Regional del Conflicto Armado en Colombia. Bogotá: Imprenta Nacional de Colombia.

65.

Restrepo

Spagat

Vargas

(2006) Special data feature; the severity of the Colombian conflict: cross-country datasets versus new micro-data. Journal of Peace Research 43(1): 99–115.

66.

Reyes Navarro

(2020) La incursíon de las Autodefensas Unidas de Colombia (AUC) en el municipio de Tibú en el año 1999: La consolidacíon del paramilitarismo en el Catatumbo. Revista Cambios y Permanencias 11(2): 768–788.

67.

Robinson

Malone

(2024) Militant splinter groups and the use of violence. Journal of Conflict Resolution 68(2–3): 404–430.

68.

Romero

(2003) Paramilitares y autodefensas: 1982–2003. Vol. 13. Barcelona: Temas de Hoy.

69.

Ronderos

(2014) Guerras recicladas. Bogotá: Aguilar.

70.

Roux

(2023) Comunicar la revolucion: un analisis cultural de la propaganda de las FARC y su recepcion en Colombia (1964–2022). Ph D thesis. Universidad Nacional de Colombia.

71.

Rudloff

Findley

(2016) The downstream effects of combatant fragmentation on civil war recurrence. Journal of Peace Research 53(1): 19–32.

72.

Sanchez

del Mar Palau

(2006) Conflict, Decentralisation and Local Governance in Colombia, 1974–2004, Vol. 20. Bogota, Colombia: CEDE.

73.

Sanford

(2003) Learning to kill by proxy: Colombian paramilitaries and the legacy of Central American death squads, contras, and civil patrols. Social Justice 30(3(93)): 63–81.

74.

Schubiger

(2023) One for all? State violence and insurgent cohesion. International Organization 77(1): 33–64.

75.

Schutte

Weidmann

(2011) Diffusion patterns of violence in civil wars. Political Geography 30(3): 143–152.

76.

Sharif

Carranza-Franco

(2026) Delegative peacebuilding: Explaining post-conflict selective violence. Conflict Management and Peace Science 43(2): 157–182.

77.

Soules

(2023) Recruiting rebels: Introducing the rebel appeals and incentives dataset. Journal of Conflict Resolution 67(9): 1811–1837.

78.

Spencer

(2011) The evolution and implementation of FARC strategy: Insights from its internal documents. Security and Defense Studies Review 12(1-2): 73–98.

79.

Staniland

(2018) Networks of Rebellion: Explaining Insurgent Cohesion and Collapse. Ithaca: Cornell University Press.

80.

Stanton

(2015) Regulating militias: Governments, militias, and civilian targeting in civil war. Journal of Conflict Resolution 59(5): 899–923.

81.

Steele

(2019) Civilian resettlement patterns in civil war. Journal of Peace Research 56(1): 28–41.

82.

Steele

Schubiger

(2018) Democracy and civil war: The case of Colombia. Conflict Management and Peace Science 35(6): 587–600.

83.

Steinert

Carey

(2019) Spoilers of peace: Pro-government militias as risk factors for conflict recurrence. Journal of Peace Research 56(2): 249–263.

84.

Tate

(2001) Paramilitaries in Colombia. Brown J. World Aff 8: 163.

85.

Thomas

(2014) Rewarding bad behavior: How governments respond to terrorism in civil war. American Journal of Political Science 58(4): 804–818.

86.

Thompson

(2023) Armed Groups: The Twenty-first-century Threat. Lanham: Rowman & Littlefield.

87.

Townsley

Johnson

Ratcliffe

(2008) Space time dynamics of insurgent activity in Iraq. Security Journal 21: 139–146.

88.

Troyan

(2008) Ethnic citizenship in Colombia: The experience of the regional indigenous council of the Cauca in southwestern Colombia from 1970 to 1990. Latin American Research Review 43(3): 166–191.

89.

Vallejo

(2021) Ejército de Liberación Nacional (ELN): historia de las ideas polıticas (1958–2018). Bogotá: Universidad Nacional de Colombia.

90.

Vélez

(2001) FARC-ELN: Evolución y expansión territorial. Desarrollo y Sociedad 23(47): 151–225.

91.

Villamizar

(2022) Las guerrillas en Colombia. 3rd ed. Bogotá: Penguin Random House.

92.

Wagner

Soumerai

Zhang

, et al. (2002) Segmented regression analysis of interrupted time series studies in medication use research. Journal of Clinical Pharmacy and Therapeutics 27(4): 299–309.

93.

Wood

(2014) From loss to looting? Battlefield costs and rebel incentives for violence. International Organization 68(04): 979–999.

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.

0.00 MB

0.44 MB