One for All? State Violence and Insurgent Cohesion

Abstract

What effect does state violence have on the cohesiveness and fragmentation of insurgent organizations? This article develops a theory of how state violence against civilians affects insurgent cohesion and fragmentation in civil war. It argues that the state-led collective targeting of an armed group's alleged civilian constituency increases the probability of insurgent fragmentation, defined as the process through which insurgent organizations split into distinct entities, each with its own social composition, goals, and leadership. This effect is driven by the interaction of several mechanisms at the individual, group, and organizational levels: state-led collective targeting enlarges the supply of fresh recruits, strengthens the bonds between immediate group members (interpersonal cohesion), and disrupts intra-organizational coordination, strategic unity, and institutional arrangements that underpin the commitment of individual fighters to the organization as a whole (ideological cohesion). The implications of this argument are empirically tested in an analysis of armed groups fighting against their governments between 1946 and 2008. The results suggest that campaigns of massive state violence directed against the civilian constituency of rebel groups increase the overall risk of insurgent fragmentation, a finding that has important implications for the duration and escalation of civil wars.

State violence against civilians is a cause of immense human suffering in many countries around the globe. Between 2017 and 2018 an estimated 700,000 Rohingya were displaced from Myanmar to refugee camps in Cox's Bazar, Bangladesh, in one of the most recent waves of violence against them. A fact-finding mission sent by the UN Human Rights Council concluded that the human rights violations in Kachin, Rakhine, and Shan States were “shocking for their horrifying nature and ubiquity” and identified “the Myanmar security forces, particularly the military,” as the principal perpetrators.¹ In the armed conflict that has ravaged Syria since 2011, many thousands of civilians have died as a result of persistent state violence, often inflicted by unguided barrel bombs dropped in large numbers, and frequently hitting markets, hospitals, and schools.² The UN Special Adviser on the Prevention of Genocide expressed grave concern about noncombatants in areas under attack by the government, given that “government offensives in areas controlled by armed opposition groups in Syria have been carried out with little regard for the lives of civilians.”³ Over the course of the civil war in Guatemala from the early 1960s to the mid 1990s, most of the state's victims were civilians, and most of those were of indigenous origin.⁴ In 1999, the Commission for Historical Clarification concluded that over 80 percent of the 200,000 war victims were Mayan, and that state forces and affiliated paramilitary groups were responsible for more than 90 percent of all documented violations.⁵ Other high-profile examples of large-scale state violence against civilians include the internal armed conflicts in Chechnya, Sri Lanka, and Sudan. In short, large-scale state violence against civilians remains a horrifyingly common feature around the globe. Between 1955 and 2005 alone, ninety campaigns of state-led mass killings with 1,000 victims or more have been identified, the overwhelming majority conducted in the context of political instability, such as civil wars.⁶

Evidence from several studies suggests that the likelihood of state-led mass killings of civilians increases dramatically in times of internal conflict, when state actors deliberately target not only rebels but also members of their alleged civilian constituency, or fail to effectively distinguish between the two.⁷ Yet the consequences of this type of violence for wartime dynamics, and armed groups in particular, remain poorly understood. While previous research has suggested a critical role for state-orchestrated civilian victimization in explaining patterns of insurgent violence,⁸ insurgent recruitment,⁹ civil war occurrence,¹⁰ and insurgent territorial control,¹¹ little is known about how state-led violence against civilians affects the networks, institutions, and internal functioning of insurgent groups.

In this paper I investigate the social and institutional repercussions of state-led collective targeting, and in particular its consequences for the cohesiveness of armed groups. I explore how the targeting of an armed group's alleged civilian constituency affects the probability of insurgent fragmentation, defined as the process through which insurgent organizations split into distinct entities with their own composition, goals, and leadership. I develop a theoretical framework that specifies several mechanisms through which collective state violence influences various dimensions of insurgent cohesion and internal control, and derive the implications for an organization's vulnerability to internal splits. The theory holds that while state violence against the alleged constituency of insurgent groups strengthens bonds between immediate group members (interpersonal or primary cohesion) and multiplies individual-level motivations to fight, it also tends to weaken organizational coordination, strategic unity, and institutions that foster the commitment of individual fighters to the organization as a whole (ideological or secondary cohesion). In particular, I argue that institutions that forge and sustain secondary cohesion within armed organizations tend to be disrupted by state violence, even while the demand for these same institutions is increased by the influx of new recruits. My theory further implies that it is the distinct combination of strengthened primary cohesion and weakened secondary cohesion that, together with surges in fresh recruits, increases the probability of internal splits.

While the theoretical argument is motivated and illustrated with qualitative examples, the core empirical implication is tested in a quantitative analysis of insurgent organizations that were active between 1946 and 2008. I use a novel data set that records (for the first time in this detail, to my knowledge) whether the civilian constituency of each armed group has been affected by state-led collective targeting in the context of a mass killing episode. I find that the state-led targeting of the civilian constituency of rebel groups increases the overall probability of major insurgent splits. This finding has important implications for the duration and severity of civil wars because fragmentation may intensify conflicts and make them harder to resolve. Moreover, the study shows that the entrance of new nonstate actors into the global landscape of political violence is aided by violations committed by states in their quest to maintain a monopoly on violence.

What Do We Know?

The extent to which insurgent organizations manage to maintain internal control and cohesion varies dramatically. Some organizations successfully unite their members behind a common goal for a very long time, while others quickly disintegrate and decay, sometimes into violently competing groups.¹² The Fuerzas Armadas Revolucionarias de Colombia (FARC), for instance, was a highly unified organization for several decades, successfully managing a multitude of blocs and fronts.¹³ By contrast, the Groupe Islamique Armée, founded in Algeria in 1992, underwent several splits within the first few years of its existence, as insurgent leaders defected to form their own armed opposition, resulting in organizational fragmentation and violent confrontations between different groups.¹⁴

A thriving body of literature explores the puzzling variation in the internal cohesion of opposition movements and armed groups.¹⁵ However, insights into the impact of state violence are limited. Either studies have been primarily concerned with determinants that are exogenous to wartime dynamics,¹⁶ or they have only partially examined the effect of state violence against civilians as a causal variable.¹⁷ Moreover, this literature has wide conceptual diversity, which complicates the comparison of findings and claims. Overall, when it comes to the effect of state repression on the cohesiveness of opposition movements, the evidence is still scarce and mixed. Scholars have variously claimed that state repression tends to strengthen the collective identity and internal cohesion of targeted groups,¹⁸ that it fosters tendencies of movement fragmentation or cooperation,¹⁹ that it increases the propensity of movements to fragment,²⁰ or that it has no effect at all.²¹ Prior work on fragmentation has also often focused on either small subunits on the one hand, or overall movements or conflicts on the other,²² rather than insurgent organizations in particular.²³ Another limitation of prior work has been the reliance on generalized conceptualizations and measures of state repression, rather than distinct forms and targets of state violence.²⁴

In summary, previous work has greatly enhanced our understanding of how armed groups try to maximize cohesion and internal control.²⁵ I extend these important insights by developing a theory of how state violence against civilians affects the fragmentation of insurgent organizations. The theoretical framework incorporates several intersecting mechanisms at different levels of analysis, and two distinct notions of insurgent cohesion.²⁶ The empirical implications are evaluated based on novel data on collective state violence against the alleged civilian constituency of rebel groups. To the best of my knowledge, this study provides the first theoretical and empirical inquiry that directly addresses the effects of such targeting on insurgent cohesion and fragmentation in civil war.²⁷

A Theory of State Violence and Insurgent Fragmentation

In this section I develop a theory that integrates several mechanisms through which the collective targeting of armed groups’ purported support base promotes the fragmentation of armed groups. I argue that this targeting increases the supply of volunteers by pushing people into the opposition's ranks who would not otherwise have joined. Such upswings in the supply of volunteers committed to the fight against the state, but not necessarily to the organization they join, may threaten the cohesiveness of armed organizations, unless there are sufficient incentives and institutional capacities to screen, socialize, and indoctrinate new and prospective recruits in accordance with the organization's identity, values, and interests. These incentives and capacities are themselves endogenous to wartime dynamics, however, and I argue that they will be undermined by the consequences of state-led civilian targeting. Moreover, state violence will tend to reduce strategic unity and insurgent internal control. As a result, while bonds between immediate group members (primary or interpersonal cohesion) tend to be strengthened by collective state violence, the commitment of individual fighters to the organization overall (secondary or ideological cohesion) will be weakened. It is precisely under these conditions that aspiring leaders of defecting factions will have the greatest incentives to become first movers and launch their “own” rebellion. Collective desertion is a risky endeavor, and prospective leaders of nascent splinter groups will seize the initiative only once they are confident that their closest allies and subordinates will stay loyal when the time comes—and that, once defected, they will be able to enlarge their ranks fast. Thus, once secondary cohesion is weakened, upsurges in fighting morale, strengthened primary cohesion, and inflows of fresh recruits provide ideal conditions for concerted defections that result in insurgent splits.

Assumptions and Concepts

This article adopts the conceptual distinction between selective, collective, and indiscriminate targeting of civilians. Selective violence aims to punish behavioral noncompliance at the individual level,²⁸ while indiscriminate targeting is completely arbitrary.²⁹ In between is collective targeting,³⁰ or “group-selective violence,”³¹ which is based on collective attributes—in civil wars, often attributes that are associated with the social constituency of the opponent, such as geographic location, political affiliation, or ethnic identity.³² I focus on state-led collective targeting against the alleged civilian constituency of armed groups, and in particular where such campaigns of state violence reach a massive scale.³³ I specifically theorize the effects of collective violence against civilians who are targeted because of their suspected affiliation with or support for insurgent groups.

I use the terms armed opposition organization, rebel organization, and insurgent organization interchangeably. Like Kenny and Woldemariam, I restrict the term insurgent fragmentation to the splintering of an insurgent organization into two or more distinct ones, each with its own social composition, goals, and leadership.³⁴ Insurgent fragmentation thus refers to concerted defections within an insurgent organization, a phenomenon that has to be distinguished from individual desertions and side-switching, which may occur to a significant degree without constituting fragmentation.³⁵ By insurgent internal control I mean the extent to which the leadership exerts control over the factions and individual members of a rebel organization, in the sense of being able to oversee and discipline the behavior of subordinate group members.³⁶

Regarding the concept of insurgent cohesion, Wood introduced the distinction between primary and secondary cohesion from military sociology to the study of armed groups engaged in civil wars.³⁷ Building on this work, by primary cohesion I mean the extent to which horizontal and vertical bonds between primary group members (that is, individuals that regularly interact face to face) are positively “loaded” with a sense of collective responsibility and mutual trust, while I restrict the term secondary cohesion to the extent to which individuals identify with the armed organization as a whole.³⁸ Secondary and primary cohesion are thus related to the concept of collective identity, “an individual's cognitive, moral, and emotional connection with a broader community, category, practice, or institution.”³⁹ In what follows, I will use the terms secondary cohesion and ideological cohesion as synonymous, and the terms primary cohesion and interpersonal cohesion as interchangeable as well.⁴⁰

Building on a rich body of work that has documented adverse selection problems in recruitment⁴¹ and highlighted the relevance of indoctrination for the transformation of individual preferences and armed group cohesion,⁴² I proceed from the assumption that the most secure foundations of secondary cohesion are strong institutions for indoctrination and political education. Typically tied to an organization's particular ideology and agenda, these institutions provide coherent frameworks and instruments to align the commitments and preferences of individual fighters with the organization's goals and principles as defined by the leadership. I further assume that if secondary cohesion is strong among commanders and rank-and-file members, combatants will not desert in large numbers, be it on their own or in groups. If, on the contrary, cohesion is weak across the board, and both interpersonal and secondary cohesion low, individual defections might be common, but insurgent fragmentation unlikely. Concerted defections leading to fragmentation occur when prospective leaders of splintering factions are convinced that their allies and followers will remain by their side, and that, once formed, their new group will grow quickly in size.

Supply of Recruits, Screening, and Recruitment

At the individual level, the state-led collective targeting of noncombatants that frequently occurs in civil wars increases the supply of volunteers for armed organizations. This occurs through several mechanisms: by fostering moral outrage and thus participation-related emotional rewards,⁴³ by reinforcing grievances that stir reactive mobilization through increased individual cost tolerance and radicalization,⁴⁴ and by turning protection from state violence into a selective incentive.⁴⁵

This implies that, while state violence against civilians tends to create surges of fresh recruits, it also tends to do so via motivations related to the state, rather than the long-term goals and principles of the rebel group.⁴⁶ As a result, rebel organizations are typically faced with an increased influx of volunteers driven by motives such as protection and vengeance, rather than a commitment to the ideological principles of the organization they join. Note that this does not imply these recruits are opportunistic or lack determination.⁴⁷ They are likely to be highly committed to the fight against the state, but not necessarily to the organization under whose banner they fight.⁴⁸ An abundant supply of such recruits has the potential to impair the internal control and cohesion of armed groups because the norms and preferences between combatants and leaders may be poorly aligned.⁴⁹

Armed-group leaders often screen prospective joiners in a bid to recruit followers whose preferences are already well aligned with those of the group.⁵⁰ Indeed, armed groups do not always aspire to grow in size, especially if this comes at the expense of the “quality” of their recruits.⁵¹ However, armed groups also differ markedly in their skill and capacity to select and screen prospective recruits.⁵² The extent to which (over-)supplies of certain “types” of recruits will impair insurgent cohesion will thus depend on the recruitment and screening strategies of the group, as well as the internal institutions that forge and sustain the commitment of individuals not only to their immediate peers but also to the goals and principles of the organization. At the same time, these strategies are themselves endogenous to state violence: while state violence against civilians will enlarge the supply of new recruits for insurgents, it also has a tendency to undermine the capacity of insurgent organizations to screen these volunteers carefully. The targeting of the civilian support base of insurgent groups typically enforces insurgent dislocation, strategic adjustments, and sometimes even the breakdown into smaller operating units, thus diverting resources away from the screening of fresh recruits. Moreover, as with other threats to insurgent groups’ survival, attacks on their support base increase the pressure on them both to strike back against state forces to show resilience and strength,⁵³ and to open their ranks to aspiring recruits, making more lenient screening more likely.⁵⁴ Finally, even if armed organizations have no incentives to enlarge their ranks, they may be concerned that turning down volunteers would make them available to rival armed groups.⁵⁵

Of course, insurgent organizations may be able to adapt. The recruitment and screening practices of the Provisional Irish Republican Army (PIRA), for example, underwent several major transformations. In the late 1970s the organization was radically reorganized and the number of members reduced as a reaction to an influx of recruits motivated primarily by state violence, leading to problems with infiltration.⁵⁶ And in the 1980s the Fianna Eireann (the organization's youth wing) was disbanded in a bid to tighten recruitment practices once more, for similar reasons.⁵⁷ The Liberation Tigers of Tamil Eelam (LTTE) also successfully tightened their screening mechanisms after the dramatic increase in applicants following Black July in 1983.⁵⁸

These are just two cases that illustrate that insurgent leaders are sometimes capable of adjusting their recruitment and screening strategies to changing circumstances in effective ways. However, such adjustments are not always made, or—as with the LTTE and PIRA—occur after inflows of new volunteers have already significantly altered the composition of the rank and file. Whether and to what extent such changes in membership will impair insurgent internal control will depend on the internal institutions that forge and sustain the commitment of individuals, not only to their immediate peers but also to the goals and principles of the overall organization and its leadership. In other words, successfully adapting to these changes requires strong institutions that effectively socialize and indoctrinate fresh recruits. Such institutions will be the focus of the next section, where I argue that state-led violence against civilians will not only increase the pool and influx of prospective recruits but also facilitate socialization processes that create and sustain interpersonal cohesion, while at the same time weakening those institutions that underpin the commitment of individual fighters to the leadership and the organization as a whole.

Cohesiveness in Insurgent Groups and Organizations

The collective targeting of the civilian support base of insurgent groups likely forces them to relocate away from targeted areas, to adjust their strategies, and at times even to rearrange their structural makeup. During the El Salvadoran civil war, for example, the insurgents of the Frente Farabundo Marí para la Liberación Nacional responded to indiscriminate state violence—including large-scale aerial bombings—by sending civilians to refugee camps, and by breaking down its battalion-size forces to smaller units, sometimes of no more than five combatants.⁵⁹ While such adaptations will tend to strengthen the bonds between immediate group members, they aggravate the challenges of maintaining intra-organizational coordination, strategic unity, and the institutional coherence that fosters ideological cohesion, as I will argue.

In addition to expanding the influx of fresh recruits, exposure to campaigns of state violence may increase the commitment of already mobilized combatants to fighting. Indeed, even less extreme forms of state-induced harm, as caused, for example, by systematic discrimination, have been argued to increase cost tolerance, in-group solidarity, and commitment among those identifying with victimized groups.⁶⁰ And yet, a shared commitment to a common goal—be it secession, revolution, or the removal of an abusive government or occupying force—is not sufficient to ensure organizational unity, especially as insurgent leaders often face multiple competitors claiming to fight for the same cause. The challenge of insurgent leaders is thus not just to mobilize their followers against a common enemy but to instill and maintain a strong commitment among them to the norms, values, and goals of their particular organization (secondary cohesion)—and to sustain the conviction that theirs is the only true path to political change.

Virtually all armed organizations rely on formal and informal institutions to socialize, discipline, and indoctrinate combatants according to the principles, norms, and goals of the organization as defined by the leadership.⁶¹ Just as armed organizations strive to transform combatant preferences in the sense of instilling norms that prioritize discipline and an application of violence that is consistent with the organization's norms and strategies,⁶² it is in their interest to build institutions that create and sustain strong combatant commitment to the organization and its leadership. Such institutional arrangements typically consist of collective rituals, formalized indoctrination, and ongoing political education that regularly delineate and affirm the organization's principles, values, and long-term goals.⁶³ Socialization and indoctrination might be coupled with screening and recruitment, such as when insurgents recruit from affiliated youth wings or connect their recruitment efforts with political education. The PIRA partially relied on its youth wing, Na Fianna Eireann, to socialize, screen, and recruit young volunteers until the 1980s.⁶⁴ The institutionalized coupling of indoctrination, screening, and recruitment has also been reported for cases such as the insurgencies in Nepal and El Salvador.⁶⁵ Conversely, even coerced recruits can be turned into highly loyal fighters if the institutions for socialization and indoctrination are sufficiently strong.⁶⁶

The Peruvian Sendero Luminoso (Shining Path) is an example of an insurgent organization that devoted major attention to the political education and ideological indoctrination of its cadres across all ranks. As Gorriti writes on the formal military training of Shining Path's prospective military leaders at the eve of the war: “The purpose of the Military School was not to saturate everyone in lethal technology … but rather to relate and overlap ideology with its military manifestation at every level.”⁶⁷ Shining Path indeed managed to forge “a unity that appeared unbreakable, organic,” particularly during the early stages of the war.⁶⁸

In the words of one Shining Path combatant:

I do not think as a person any longer. One feels the party as oneself, I am the party … and everything we do and think is part of the party. Such is the political mutual understanding that we have, that we draw the same conclusions no matter how far we are from each other. It is the same thing in politics, and better still in the military realm. There may be a column that wanders off because the enemy besieges and divides it. Those two commands know what to do. The unit is so strong that we all have the same initiative. Without coordinating, we coordinate.⁶⁹

Collective state violence against civilians tends to impair institutional arrangements that mold such unity. To begin with, the influx of a large and diverse group of new recruits does not just hamper careful screening, it also places existing institutions for indoctrination under immense stress.⁷⁰ Worse, harm inflicted on the insurgents’ sources of support and information, and the increased inflow of fresh recruits, are likely to force insurgents to divert resources away from regularized indoctrination and political education—practices that require expertise, coordination, and time—while simultaneously increasing the demand for them. The consequences can be profound, because, in the words of Gutiérrez Sanín, constant indoctrination is “a sine qua non for internal cohesion.”⁷¹

In short, large-scale state violence against the civilian support base of armed groups is likely to undermine insurgent secondary cohesion by disrupting institutional arrangements within armed groups that are designed to align the preferences of individuals with organizational ones. At the same time, state violence increases the very demand for these institutions due to the increased influx of volunteers.

In addition to the impact on recruitment, screening, and indoctrination, state-orchestrated violence against an armed group's constituency tends to create, reveal, and deepen preference divergences over ideology and strategy in the leadership, for example, by pitting radical leaders against more moderate ones.⁷² Radical elements, or otherwise disloyal factions, within an organization might even significantly profit from campaigns of state violence if they can exploit them to activate support for their own agenda.⁷³ In the case of the IRA, the split into the Official IRA (OIRA) and the Provisional IRA (PIRA) was eventually triggered by internal disagreements about how to respond to the discrimination and victimization of Catholics in Northern Ireland.⁷⁴ In Syria, as the government intensified its crackdown on alleged civilian supporters of armed opposition groups,⁷⁵ disagreements over tactics and strategy fostered deep divisions within the opposition as early as 2011.⁷⁶ To quote a supporter of the Free Syrian Army: “They [the Syrian National Council] just talk and are interested in politics, while the Assad regime is slaughtering our people … We favor more aggressive military action.”⁷⁷

Importantly, these dynamics can be activated by even lesser forms of state repression, or reinforced by state violence against militants, as long as aspiring leaders of subversive factions manage to put blame on the leadership and mobilize support for their own ends. The influx of new recruits can further upset internal power balances, weakening the rank-and-file bases of some commanders relative to others, and fueling intra-organizational power struggles and leadership disputes.⁷⁸ Thus, while the literature on peace agreements has largely focused on how peace processes stimulate fragmentation,⁷⁹ dynamics of violent escalation often have similar effects by forging and revealing divisions in the leadership on how to respond to state repression, and by allowing alienated elements to advance their cause.

Finally, collective state violence against the civilian constituency of armed groups tends to temporarily disrupt intra-organizational coordination,⁸⁰ a central pillar of institutional and strategic coherence even in cases where the mobility of subgroups is among an armed group's central organizational principles.

As a former FARC commander explained,

Although they were mobile guerrillas, they had instructions to maintain contact … There was a continual exchange of ideas and opinions among the leadership of each unit … And when it was necessary to convene a meeting to examine the military situation, or any other situation, then a site was agreed upon and the commanders used every means possible to arrive at the place of the meeting.⁸¹

In many instances, civilian networks form the backbone of insurgent organizations, for example, by facilitating intra-organizational communication and clandestine logistics.⁸² The weakening of these networks poses severe challenges not only to a unified central command but also to secondary cohesion, as the constant exchange of ideas and information between the commanders of individual units is interrupted, potentially giving rise to parallel command structures and diverging perceptions and preferences about how to respond to strategic challenges.⁸³ Therefore, under conditions of collective state violence, prospective leaders of deserting factions are not only more likely to emerge but also have windows of opportunity created by weakened internal control in addition to an enlarged pool of potential volunteers eager to fight back against an abusive state.⁸⁴

After defection, the potential to grow quickly is key, as a splinter group will be in deadly rivalry with its former allies. Indeed, splintering organizations often compete fiercely for new recruits.⁸⁵ The splitting of the IRA into the OIRA and the PIRA in 1969/70, for instance, “was as much a battle among political entrepreneurs for control of the flood of potential new recruits as it was the fragmentation of a pre-existing organization.”⁸⁶ Importantly, however, the defining feature of splinter groups is precisely that the initial followers are not exclusively recruited “from scratch,” which is why organizational fragmentation is unlikely if cohesion is low across the board.⁸⁷ Yet the targeting of the civilian support base of rebel groups also has ramifications for primary cohesion, as I will argue.

In contrast to ideological cohesion, interpersonal cohesion is not dependent on political education or institutionalized training tied to higher-level goals.⁸⁸ In the context of armed conflict, primary cohesion may be cultivated through repeated, shared experiences and the close observation of one's fellow combatants in situations that reveal their trustworthiness when lives are at risk.⁸⁹ The pressures associated with campaigns of state violence against civilian constituencies are likely to strengthen the bonds between combatants who remain in direct contact, through experiences such as increased isolation from noncombatant populations and other fighting units, as well as the shared experience of mutual dependence in situations of hardship and extreme risk. The strengthening of primary cohesion is reinforced if state violence also directly affects combatants. For instance, based on a qualitative study of Burmese and Irish Republican armed groups, Kenny suggests that shared sacrifices through exposure to state repression can promote organizational socialization conducive to insurgent cohesion.⁹⁰ Bonding within factions in this context is likely to be fostered through informal socialization between peers and thus may strengthen cohesion within factions but also undermine norms and loyalties promoted at higher levels of the organization.⁹¹

Of course, such bonding processes may not always occur, and are not assumed to apply uniformly to all units. If state violence is overwhelming, it may even undermine the fighting morale and certain bonding mechanisms in some units and groups.⁹² However, state-led violence against civilians also raises the costs for combatants to individually desert their units, given the increased and often disproportionate risk of victimization that noncombatants face.⁹³ The lack of outside options alone can have a strong socializing effect on combatants and can enhance cohesion without indoctrination or the internalization of group norms.⁹⁴ Overall, once vertical and horizontal social bonds at lower levels of the organization at least partially trump secondary cohesion, they can readily be exploited by prospective leaders of nascent splinter groups.

The combination of strengthened interpersonal cohesion, weakened ideological cohesion, and surges of fresh recruits significantly enhances the prospects of success for defectors who aim to desert in concert to launch their own organization. Emerging splinter groups are typically led by coalitions of mid-to-high-level commanders; these commanders will risk launching a concerted desertion only if they trust the loyalty of their allies and their (and their allies’) subordinates,⁹⁵ and if they are confident that these loyalties eclipse secondary cohesion.⁹⁶ Weak cohesion across all levels of an organization should accordingly be conducive to the disbanding of armed groups and large-scale desertions of individuals, rather than the emergence of major splinter groups.⁹⁷ Fragmentation, in short, should be rare when cohesion is low across all levels, and more likely when cohesion at lower levels of the organization is strong.⁹⁸

Table 1 encapsulates the mechanisms just discussed. The core empirical implication to be evaluated in the remainder of this article is that collective state violence against the alleged civilian constituency of insurgent groups increases the probability of insurgent splits.

TABLE 1. The impact of state violence

Data and Empirical Approach

The empirical analysis covers armed organizations engaged in intrastate conflict between 1946 and 2008; excluded are conflicts in countries too small to be covered in the Ulfelder and Valentino state-led mass killing data set,⁹⁹ extra-systemic conflicts, interstate conflicts, and cases that did not have a clearly identifiable armed group opposing the government.¹⁰⁰ I present results from both a cross-sectional and a time-series cross-section approach. The unit of analysis in the cross-sectional approach is the armed organization—or more precisely, given that these organizations are defined by their opposition to the government, the conflict dyad. In the time-series cross-section approach, it is the armed group-year, or in other words the dyad-year. Dyads are included as soon as they reach the conventional threshold of twenty-five battle-related deaths in a particular year.¹⁰¹ I include spells of inactivity to account for the fact that both state violence and fragmentation can occur in years where conflicts do not reach the fatality threshold for inclusion in conventional data sets.

The dependent variable is a binary indicator of whether a splinter group broke away from a given armed group, and the main “treatment” variable is an indicator of whether the civilian constituency of an armed group has been affected by state violence during campaigns of state-led mass killings. The cross-sectional analysis relies on linear probability and logistic regression models with covariate adjustment and entropy balancing,¹⁰² while the time-series cross-section analysis is based on Cox duration models¹⁰³ and, alternatively, a propensity score weighting approach for time-series data.¹⁰⁴ I include a series of theoretically relevant pre-treatment covariates that are potential determinants of both state-led collective violence and insurgent fragmentation.

State-Led Collective Targeting

To identify whether a certain armed group (and hence, dyadic conflict) has been affected by campaigns of state-led collective violence against civilians of their alleged constituency, this study relies on an original coding of each case, based on the state-led mass killing data set of Ulfelder and Valentino,¹⁰⁵ associated coding notes, and additional sources. State-led mass killings are defined as episodes in which “actions of state agents result in the intentional death of at least 1,000 noncombatants from a discrete group in a period of sustained violence.”¹⁰⁶ Mass killings are included in the Ulfelder and Valentino data set based on several criteria: the victims are unarmed noncombatants and residents of the perpetrator state; the killings are directed against particular social groups (which may be defined ethnically, geographically, politically, and so on); there is evidence of state agents as perpetrators or sponsors, or of state complicity; deaths are the result of direct or indirect state violence (starvation, for example); and there is evidence or context that implies perpetrator intention. The beginning of a mass killing episode is recorded during the first year in which at least 100 civilians were killed; episodes end after three consecutive years in which fewer than 100 civilian deaths are recorded.¹⁰⁷

For the purpose of this study, the state-led mass killing data set, the underlying coding notes, and additional sources were consulted to code for each insurgent group whether the alleged civilian constituency of that group had been affected by state-led mass killings during a given period. This fine-grained approach and attention to both the temporal dimension and the armed-group level departs from prior approaches that largely assumed such a connection in case of a temporal overlap of mass killings with a given conflict or country, and that did not descend to the level of each individual armed group.¹⁰⁸ The data set records whether during any given year after conflict onset there was an episode of state-led mass killing related to that particular armed group—that is, the killings targeted its alleged members or civilian constituency.

Ideally, the data set could capture whether specific fatality thresholds were reached for each affected group in a given year. However, because several victim groups, including armed-group constituencies, can simultaneously be affected by a particular episode, capturing dyad-specific fatality thresholds resulting from particular state-led mass killing episodes is beyond the scope of this project. The coding reflects whether a constituency was subject to collective targeting by state actors during episodes of state-led mass killings, without indicating that a particular fatality threshold was reached for any given victim group.¹⁰⁹ Likewise, the start and end dates reflect the onset and termination dates of the overall mass killing episode as recorded by Valentino and Ulfelder, while ideally, this could be coded more precisely for each constituency and armed group. Despite these limitations, and to the best of my knowledge, the presented data go beyond earlier efforts to link state-led mass killings to constituencies of specific armed groups.¹¹⁰

Insurgent Fragmentation

The occurrence of insurgent fragmentation is coded based on the UCDP Actor data set,¹¹¹ which provides basic information on all actors included in UCDP's data sets on organized violence, as well as the FORGE data set on the organizational origins of rebel groups.¹¹² For every insurgent group, this variable indicates whether there was a group that split from the original rebel group to form an armed organization of its own. Instances of fragmentation are included if the splinter groups end up being involved in political violence resulting in at least twenty-five related deaths—more specifically, nonstate conflict against other armed groups, one-sided violence, and/or state-based conflict as defined by UCDP. The occurrence of insurgent fragmentation is thus operationalized through the emergence of a major splinter group, where “major” means that the splinter group itself ends up being involved in organized violence that results in twenty-five deaths or more during at least one year. As a consequence, the coding has one main limitation: it does not capture splinter groups that do not end up engaged in organized violence reaching such recorded fatality levels. The inability to detect splinter groups not subsequently involved in such violence is likely to bias against detecting a positive effect of state violence, because we observe only a subset of all splinter groups.

In the cross-sectional analysis, the outcome variable is measured as the occurrence of at least one instance of insurgent fragmentation after the onset of the mass killing episode affecting the civilian constituency of the original organization, or—in the absence of such violence—after the initial phase of the conflict (see the description of quasi-pre-treatment and quasi-post-treatment periods that follows). Once an organization splits from the original movement, it is treated as an independent organization, and splinters from this new organization do not affect the fragmentation coding for the original organization. In the time-series data, the outcome is measured year by year.

For armed groups whose alleged constituency has been affected by state-led collective targeting and that also experience a split in the same year or after, about 47 percent of splits occur within the first three years of mass killing onsets,¹¹³ about 58 percent within five years, and about 81 percent within ten years. Table 2 provides more information on the time to first split for cases that experience splintering as a possible consequence of state violence.¹¹⁴

TABLE 2. Time to first split after state violence onset

Confounders

I include several variables that potentially confound the relationship of interest. In the cross-sectional analysis, these are measured at the onset of a conflict (that is, in the first year or initial conflict episode), while in the time-series cross-section analysis, they are measured over time whenever possible.

First of all, I include a “pre-treatment” or past outcome variable (Prior fragmentation) where appropriate.¹¹⁵ Different levels of fragmentation likely trigger distinct types of state responses, as governments adapt their counterinsurgency strategies partly in response to the perceived structure and cohesiveness of armed groups. Moreover, initial levels of fragmentation affect subsequent organizational dynamics and reactions to state violence.¹¹⁶ The estimates of the effect of state violence would be biased if state targeting were more or less likely to occur against groups with an already higher tendency to fragment, and if this tendency were not accounted for.

In the time-series data, fragmentation is recorded year by year. In the cross-sectional data, Prior fragmentation indicates whether the armed organization was affected by splits prior to the onset of the mass killing episode for affected dyads. For conflict dyads without exposure to state-led collective targeting, creating an equivalent measure of initial fragmentation is not straightforward. I proceed by defining quasi-pre-treatment and quasi-post-treatment periods for control units. For groups not exposed to collective state violence, Prior fragmentation thus indicates whether there were splits during the initial phase of the conflict, that is, within five years following the year of conflict onset.¹¹⁷ As it is not a priori clear how the initial period should be defined for the unaffected units, I assess the robustness of the results to an alternative definition that caps the initial period at two years following the year of conflict onset.¹¹⁸ I present the corresponding results, and additional results with no such periods, and without the Prior fragmentation variable, in the online supplement.

Another important variable is whether a war is conducted as an Irregular war, because the type of warfare should significantly affect both state violence and insurgent institutions, and insurgent cohesion and fragmentation as a result.¹¹⁹ The indirect warfare strategies of insurgents engaged in irregular war, and their dependence on the civilian population, can make it difficult for counterinsurgent forces to engage with rebel groups directly, putting civilians at special risk.¹²⁰ Insurgent institutions that underpin secondary cohesion should also be stronger under conditions of irregular civil war since such institutions can serve as partial substitutes for direct internal control.¹²¹

Several authors have argued that the social base of insurgent groups is a powerful determinant of insurgent cohesion.¹²² Most relevant for this study is the possibility that the type of social base of an insurgent group might also determine the vulnerability to state violence. I focus on the intersection of mobilization and exclusion. Specifically, I include a variable, Insurgent recruitment from excluded ethnic groups, that indicates whether the rebel group of a conflict dyad recruits its members from an ethnic group excluded from state power.¹²³ Ethnic exclusion is a strong and robust predictor of conflict duration and outcome,¹²⁴ a relationship that has been argued to be driven to a considerable extent by grievance-induced gains in individual commitment and group solidarity.¹²⁵ This variable is also an important predictor of state-led civilian targeting, as both ethnic exclusion and exclusionary ideologies have been associated with mass killings and genocide.¹²⁶ Even short of genocidal or large-scale violence, ethnicity is often one of the main “profiling” attributes used in campaigns of wartime violence against civilians in general and state violence in particular.¹²⁷ To capture the prevalence of both exclusionary elite ideologies and the extent to which the civilian population is perceived as a threat to the regime, I also include the fraction of the population that is excluded from access to political power (Excluded population).¹²⁸

Research suggests that access to External rebel support has an important influence on patterns of insurgent cohesion,¹²⁹ although theories diverge on whether they predict a positive or negative effect.¹³⁰ External support for insurgent organizations may also influence the level of threat the insurgents are perceived as posing to the regime, potentially increasing the risk of state violence.¹³¹ Finally, the availability of external support may reduce the insurgents’ reliance on the civilian population, and hence increase the resilience of insurgent organizations to state violence against noncombatants.¹³² Relative fighting capacity is another relevant confounder.¹³³ The ability of insurgent forces to challenge the state shapes the strategic environment of the armed competition, including the incentives of state forces to employ violence against civilians on a massive scale.¹³⁴ It also determines the incentives of insurgents to enlarge their ranks and to build cohesive organizations.¹³⁵ I further include a dummy variable, Territorial control, that measures whether the rebel group controls territory in the first stage of the conflict.¹³⁶ Territorial control has been shown to be associated with civilian collaboration,¹³⁷ insurgent recruitment and defection,¹³⁸ the capacity of insurgent organizations to evade state violence,¹³⁹ state and insurgent violence against civilians,¹⁴⁰ and insurgent fragmentation.¹⁴¹

An additional variable, Previous conflict activity, indicates whether a conflict related to a particular dyad was active in the past. This is a measure of potential unobserved confounders associated with the long-term intractability of the conflict, as well as social and institutional legacies from previous but related conflicts that could affect both state violence and insurgent cohesion in the long run.¹⁴² To take into account the level of competition between armed groups, the degree of choice of prospective fighters,¹⁴³ the extent of pre-existing movement fragmentation, and the overall threat to the regime, I include the number of dyads in a particular conflict during the first year of an armed conflict.¹⁴⁴ I also include an index of “Neopatrimonial” rule that is based on various indices measuring clientelism, presidentialism, and regime corruption.¹⁴⁵ This captures the idea that state weakness and fragmentation influence state violence as well as insurgent politics.¹⁴⁶

It is important to isolate government–rebel violence from violence against civilians because battlefield dynamics can affect insurgent cohesion,¹⁴⁷ in addition to having important implications for state violence against civilians due to the perceived threat of the insurgency.¹⁴⁸ High-intensity violence might also deter opportunistic recruits.¹⁴⁹ At the same time, survival threats from battle losses will make it less likely that insurgents screen new recruits carefully, even if they are aware of the need to do so.¹⁵⁰ I hence include a measurement of Conflict intensity. In the cross-sectional data set, this measure refers to the first year of the conflict and is based on conventionally used thresholds of battle-related deaths resulting from direct confrontations between state actors and rebel forces.¹⁵¹ In the yearly analysis, I also include inactive years. Specifically, unless there is clear evidence that an armed group is no longer functional, inactive spells are included; hence, the covariate here is simply whether a conflict was active (at least 25 battle-related deaths per year) or not.¹⁵² Finally, in the time-series cross-section analysis based on propensity score weighting,¹⁵³ I include a variable indicating the occurrence of Prior mass killings because units are compared to each other over only relatively short periods of time.

Analysis and Results

Cross-Sectional Analysis

In the cross-sectional analysis, the armed group is the unit of analysis. The analysis here relies on covariate adjustment and entropy balancing.¹⁵⁴ Entropy balancing re-weights the “control” units in order to approximate the covariate moments of “treated” units and to reduce model dependence in the subsequent analysis.¹⁵⁵ All covariates are measured prior to the onset of mass killings or during the first year or first phase of the conflict.

Of the 382 armed groups in the data set, about half (51%) have been affected by state-led collective targeting of their alleged civilian constituencies. About 14 percent of all insurgent organizations in the data set underwent major splits after the initial conflict period, here defined as five years following the year of conflict onset or the onset of state-led mass killings. Of those armed groups affected by state-led collective targeting, 22 percent experienced splintering, compared to 5 percent of those unaffected.¹⁵⁶

Table 3 shows the results of linear regressions with entropy weights. The binary outcome indicates whether at least one splinter group broke away from the main organization after the onset of state violence or the initial conflict period. Models 1 and 2 include the main predictor variable of interest, namely whether the constituency of the armed group in a dyad was affected by state-led mass violence.¹⁵⁷ Models 3 and 4 add an indicator for whether the same armed group had experienced fragmentation prior to the onset of state-led mass killings, or during the initial conflict phase. Columns 5 and 6 present the results with the full set of confounders. In all specifications, the effect of state violence on the probability of insurgent fragmentation is positive and statistically significant at conventional levels. In these these linear probability models, state-led collective targeting increases the probability of insurgent fragmentation by twenty-three percentage points.

TABLE 3. State violence and insurgent fragmentation (OLS)

Notes: ⁺p < .10, *p < .05, **p < .01, ***p < .001. Robust/clustered standard errors in parentheses.

The main finding of a positive effect of state violence on fragmentation holds without entropy balancing and with the alternative prior fragmentation variable, and if replicated with binary logistic regressions; it remains positive but loses significance in the models with entropy balancing in which prior fragmentation is not taken into account (detailed results are reported in the online supplement).¹⁵⁸ Figure 1 plots the predicted probabilities and 95 percent confidence intervals based on a logistic regression model, without entropy balancing or adjusted standard errors, and with all covariates included and held at their median values.¹⁵⁹ According to these calculations, exposure to state-led mass killings raises the predicted probability of subsequent insurgent fragmentation from about 3.3 percent to 30 percent.

FIGURE 1. Predicted probability of insurgent fragmentation

Time-Series Cross-Section Analysis

The analysis presented here simply considers conflict dyads (regardless of their duration) as the unit of analysis. In this section I present an alternative approach that follows armed groups over time. Given that most groups experience either no splintering or only one major split, the analysis focuses on the duration up until the first split, or until groups either cease to exist or the data set coverage ends, which is in 2008. The duration analysis thus does not require quasi-periods, yet is able to indirectly control for pre-existing tendencies of fragmentation by focusing on the first split. The unit of observation is the dyad-year, and the dependent variable equals 1 if the armed group splits and 0 if the group does not fragment, or ceases to be observed or to be at risk. I rely on a Cox proportional hazards approach, which has the advantage of reducing misspecification bias compared to fully parametric approaches.¹⁶⁰

Table 4 shows the results. Columns 1, 3, and 5 show the results for models relying on the Breslow method, and columns 2, 4, and 6 using the Efron method for handling ties. Columns 1, 2, 3, and 4 show the results of baseline models, whereas columns 5 and 6 show results for models where the territorial control variable, which exhibits nonproportional effects according to diagnostic tests based on Schoenfeld residuals, is interacted with the time at risk. The analysis suggests that state-led collective targeting increases the risk of insurgent fragmentation. The conflict-intensity variable is omitted here due to the lack of variation within risk sets.¹⁶¹ The online supplement shows additional results including this variable. It also presents additional models in which only the first mass killing episode is considered for cases that have been exposed to several episodes; this mirrors the focus on the first state violence episode in the cross-sectional analysis. The conclusions remain the same.

TABLE 4. State violence and insurgent fragmentation (Cox proportional hazards)

Notes: ⁺p < .10, *p <  .05, **p <  .01, ***p < .001. Breslow (columns 1, 3, 5) and Efron method (columns 2, 4, 6) for ties. ^†Interacted with duration in models 5 and 6.

The online supplement also presents the matching and weighting approach for time-series data proposed by Imai, Kim, and Wang.¹⁶² This approach is more fine-grained than the analysis presented earlier and matches exactly on time period and treatment history, in addition to weighting or matching based on additional information. It does, however, use only a small subset of the data as a result of the treatment distribution and exact matching technique.

The results of this analysis are presented in the online supplement. Though statistically insignificant throughout, they tentatively suggest that the effect of state violence could vary over time: while fostering fragmentation overall, as indicated in the analyses reported earlier it might have no or even a negative effect on insurgent fragmentation for some years. However, the number of analyzed units is too small here, and the bootstrap confidence intervals too large, to allow any conclusions. Further research is needed to explore the fine-grained temporal dynamics that underlie the main results.¹⁶³

Discussion

I have argued that state-led collective violence against the civilian constituency of armed groups increases the vulnerability of insurgent organizations to internal fragmentation. Based on new data on armed groups and their relation to state-led collective targeting in the context of mass killings, the analysis suggests that such violence is indeed associated with an overall increase in the probability of insurgent splits.

These findings have important implications for the trajectory of conflicts and their aftermath. Insurgent splintering is likely to complicate negotiations between governments and opposition movements,¹⁶⁴ to foster rebel competition,¹⁶⁵ and possibly even to increase the viability of insurgent groups.¹⁶⁶ Indeed, organizational splits do not necessarily reduce the strength or efficiency of armed organizations, and they should not be mistaken for predictors of rebel demise.¹⁶⁷ Insurgent fragmentation is also likely to aggravate local polarization and militarization, two potent drivers of long-term political instability.¹⁶⁸ Insurgent fragmentation may thus be one of the key mechanisms through which large-scale state violence against civilians prolongs armed conflicts and complicates their ultimate termination,¹⁶⁹ but one that previous research on the consequences of state violence has largely overlooked.¹⁷⁰

Theoretically, my argument resonates with work that has stressed the central role of armed group institutions for insurgent cohesion,¹⁷¹ while at the same time contributing to a better understanding of the determinants of institutional weakness and strength. Future work should continue to uncover further sources of institutional continuity and change in armed groups, especially given the relevance of such dynamics in explaining downstream violence against civilians in civil war.¹⁷²

The results also highlight avenues for future empirical work. First of all, this study cannot claim to have identified a causal effect, which remains challenging in this particular context. Future research could supplement these efforts as more data become available, and explore the potential time-varying effect of state violence to a greater extent as well. Second, the argument implies that the effect of state violence on fragmentation should be mitigated if the institutions that forge and sustain secondary cohesion are strong and robust. Further investigation could explore the validity of this implication in greater depth. Third, the fragmentation measures in this study were restricted to major splinter groups, and subsequent efforts could explore the proposed effects and mechanisms for other forms of organizational fragmentation as well. Finally, I have explored a consequence of wartime state violence that likely has critical implications for the trajectory of conflicts and the aftermath of civil wars. Future research should continue to expand our understanding of these impacts.