1. Introduction
How do overmatched insurgents fight civil wars? An influential explanation in the literature on civil war centers on the material constraints rebels face. This account contends that rebel military strategy is determined by state capacity. Rebels confronting powerful states are constrained in their ability to use force. They can only directly challenge security forces in places where distance, terrain, or social opposition limit state capacity. Where the state is stronger, insurgents are left to resort to guerrilla tactics—hit-and-run attacks, ambushes, and espionage—and acts of terrorism against civilians (Mack, Reference Mack1975; Arreguin-Toft, Reference Arreguin-Toft2001; Kalyvas and Balcells, Reference Kalyvas and Balcells2010). By this argument, the geography and nature of rebel violence are dictated by the balance of coercive power between state and challenger.Footnote 1
We emphasize a logic of rebel military tactics that builds on this crucial insight. We agree that the balance of power between state and insurgent shapes where and how rebels fight. But we argue that this conclusion understates both the degree of spatial and temporal variation in state capacity and the consequences of that variation for insurgent strategy. Rebels’ tactical decisions often respond not to global constraints but to immediate, local conditions: the strength of nearby security detachments, their ability to marshal timely reinforcements, the resources each side can draw on over hours or days. Our theoretical account of insurgent violence centers the explanatory role of the shifting strategic environment in which rebels operate.
Though typically conceived of as relatively static, the state’s ability to project repression varies tightly over time and space (Tao et al., Reference Tao, Strandow, Findley, Thill and Walsh2016). This variation generates a fine-grained, shifting topography of state repressive capacity. In some communities at some times, states are able to muster security forces, intelligence operations, and rapid reinforcements; in others, state agents are absent, under-resourced, or isolated. Variation in the state’s ability to quickly and reliably project coercive power shapes the calculus of insurgent commanders. Crucially, the granular, constantly shifting geography of state repressive capacity implies that opportunities for insurgent strikes are plentiful. Opportunistic tactics, we propose, are vital to understanding patterns of violence in civil war.
Not all windows of opportunity are created equal, however. Importantly, the local balance of power is not determined solely by the presence and strength of security forces. The state is rarely the sole purveyor of resistance to rebel incursions. Local civilian communities can generate their own deterrent effect. Civilians can deter insurgents in a variety of ways, including informing security forces about impending attacks, arming themselves, and using non-violent tactics to resist rebel rule (Petersen, Reference Petersen2001; Kaplan, Reference Kaplan2013; Jentzsch, Reference Jentzsch2014). Alternatively, they can supply rebels with valuable information about security forces and cooperate with insurgent offensives (Berman and Matanock, Reference Berman and Matanock2015). Civilian agency thus tilts the local balance of power between state and insurgent: opportunities to strike vulnerable security forces are magnified by anticipated civilian cooperation and rendered less appealing when the prospect of tenacious civil resistance looms.
We argue that these two factors—shifting local state capacity and the potential for civilian resistance—jointly produce opportunities for insurgents to attack coercively superior incumbents.Footnote 2 By this opportunistic logic, insurgent commanders take advantage of shifts in the local balance of power to deploy violence against temporarily vulnerable security forces. Constrained in their coercive resources, local commanders seek out propitious opportunities to attack the state: windows of time in which they are more likely to win or survive a military encounter in a given locale because (1) local state coercive capacity is low—and reinforcements are not immediately available—and (2) tenacious civil resistance is unlikely.
We evaluate our argument on data from the Colombian civil war, a decades-long conflict waged between leftist insurgents, security forces, and a range of other violent non-state actors. In doing so, we develop a novel empirical strategy for evaluating the effects of state capacity on rebel violence. Studies of state capacity and political violence face a steep inferential challenge: the extent and geography of state capacity are deeply endogenous to the political and social processes that also shape the capabilities and behavior of armed challengers.
To cut through the endogenous relationships between state capacity, civilian preferences, and violence, we draw on a source of exogenous variation in the repressive capacity of the Colombian state: temporary road closures induced by natural landslides, which hinder the delivery of reinforcements to local military detachments. We pair geolocated data on 15 years of landslides with fine-grained information on the location of every road and highway in Colombia. We employ a network-analytic approach to locate the downstream effects of landslide-induced road closures on rural communities along the road network, identifying which communities were rendered less accessible to the central security state and which were cut off entirely.
Using detailed violent event data, we find that landslide-induced road closures increased the short-term likelihood of rebel attacks. Importantly, however, this relationship was conditioned by the preferences of landslide-affected communities. In communities with strongly anti-insurgent attitudes, isolation did not increase the likelihood of insurgent attack—implying that the prospect of social opposition can substitute for the deterrent effect of state coercion. We find that this result is robust to a range of statistical specifications and inconsistent with prominent rival explanations. We conclude by discussing the implications of these findings for theories of state capacity and violent politics.
2. Constraints on insurgent behavior in civil war
2.1. State capacity and (a)symmetric war
How do we understand rebel military strategy in civil wars? An influential strand of research in political science argues that the place and form of insurgent violence are dictated by state coercive capacity. In some parts of their territory, states are able to muster dominant security forces, extract taxes, provide social services, and effectively regulate economic and political life. In others, state institutions are either nominally present but ineffective or altogether absent (O’Donnell, Reference O’Donnell1993; Herbst, Reference Herbst2014). In these latter areas, this literature argues, “financially, organizationally, and politically weak central governments render insurgency more feasible and attractive due to weak local policing or inept and corrupt counterinsurgency practices” (Fearon and Laitin, Reference Fearon and Laitin2003). When states are generally weak or factors like rugged terrain and ineffective governance limit repressive capacity, insurgents are presented with opportunities to emerge and survive.
Beyond affecting the likelihood of conflict onset, the geography of state capacity also influences the intensity and persistence of civil wars. Buhaug and Gates (Reference Buhaug and Gates2002), Buhaug et al. (Reference Buhaug, Gates and Lujala2009), and Tollefsen and Buhaug (Reference Tollefsen and Buhaug2015) show that civil wars tend to break out and persist in remote borderlands far from capitals and marked with rough terrain. In a sub-national study, Do and Iyer (Reference Do and Iyer2010) find that geographic barriers to state capacity are associated with higher levels of violence, a result Schutte (Reference Schutte2014) replicates in a cross-national setting. Through a series of related channels, then, subnational variation in state capacity affects where insurgents are able to employ violence against the state.
In addition to explaining if and where rebels fight, state capacity shapes the tactics they use. This argument was first advanced by military theorists. The concept of “small war” emerged among 18th-century military strategists who recognized that the balance of power was the key determinant of military strategy: combatants facing more powerful opponents required a different set of tactics than those used in conventional interstate war.Footnote 3 This doctrine was extended by the Marxist insurgents of the 20th century, who argued that power asymmetry necessitates the adoption of guerrilla tactics. For Mao Zedong, guerrilla warfare is “a weapon that a nation inferior in arms and military equipment may employ against a more powerful aggressor” (Mao, Reference Mao2001, 42), while Che Guevara argued that guerrilla tactics should be “used by the side which [...] possesses a much smaller number of arms” (Guevara, Reference Guevara1998, 10). Counterinsurgents recognized the pivotal role of power in shaping insurgent strategy as well: French counterinsurgency theorist David Galula claimed that the “disproportion of strength between the opponents” is the defining feature of guerrilla warfare (Galula, Reference Galula1964, 3).
Political science research echoes this conclusion. Mack (Reference Mack1975) argues that guerrilla tactics allow overpowered insurgents to win wars by turning them into protracted struggles that incumbents lose the political will to continue. Arreguin-Toft (Reference Arreguin-Toft2001) concurs about the efficacy of guerrilla warfare for weaker belligerents, but notes that they can win through conventional tactics if their powerful adversaries resort to brutality against civilians. Kalyvas and Balcells (Reference Kalyvas and Balcells2010) perhaps most directly and influentially connect the balance of power to rebel tactics, arguing that guerrilla warfare “is an expression of relative asymmetry between states and rebels.” Building on this logic, recent work has gone on to examine the effects of shifts in rebel capacity on their choice of tactics (Bueno de Mesquita, Reference Bueno de Mesquita2013; Sonin and Wright, Reference Konstantin and Wright2024).
We have no quarrel with these conclusions: the balance of power between state and insurgent has a robust and well-documented relationship with whether, where, and how insurgents are able to fight. But we propose that these studies have not sufficiently addressed how variation in state repressive capacity across space and time can shape rebel violence. In addition to varying over territory, state coercive capacity changes finely over months, days, and hours. In a given place, the state’s repressive capacity depends on the current deployment of military assets, the availability of reinforcements and logistical infrastructure, and other strategic factors—factors that continuously shift throughout war. This spatio-temporal variation in the state’s capacity to deliver repression, we argue, has important, understudied consequences for rebel tactics. Negative fluctuations in capacity may open favorable windows of opportunity for insurgents to strike, while security forces are temporarily weak.
2.2. Civilian cooperation
State capacity is not the only constraint on rebel strategy, however. Civil war is fought in and around human communities. A large body of literature shows that the preferences and actions of local civilians have important consequences for conflict dynamics and outcomes (Berman and Matanock, Reference Berman and Matanock2015). Petersen (Reference Petersen2001) argues that civilian action can be conceptualized along a continuum of increasing involvement, from limited, “unarmed and unorganized” support for one side or the other, such as attending political rallies or writing graffiti, to “direct support of or participation in” rebel or state forces (8–9). Civilians can support combatants through non-violent political action, logistical support (e.g. paying taxes, feeding, supplying, or sheltering combatants), sabotage and disruption of the enemy, or joining armed groups as recruits. Civilians may also oppose insurgents through collective non-violent mobilization (Arjona, Reference Arjona2016; Kaplan, Reference Kaplan2017) or by force, forming militias for self-protection or contracting private security (Avant, Reference Avant2005; Jentzsch, Reference Jentzsch2014).
Above all else, however, this literature focuses on one core civilian power: the ability to share—or decline to share—information with insurgents or security forces. Information is critical in civil wars. It is particularly invaluable in asymmetrical conflicts, where rebels rely on the ability to hide among civilian populations in order to avoid militarily superior counterinsurgent forces (Berman and Matanock, Reference Berman and Matanock2015). This ability to hide engenders what Kalyvas (Reference Kalyvas2006) terms the “identification problem.” Counterinsurgents, who typically lack local knowledge, often struggle to differentiate rebel combatants, spies, and sympathizers from unaligned or friendly civilians. Local residents can provide this information. Civilian-provided information is equally valuable to rebels, who rely on accurate, timely information about security force operations to plan ambushes and evade repression. In manipulating information to advantage one side or another, this literature argues, civilians can shape conflict dynamics and civil war outcomes.Footnote 4
By this line of reasoning, civilian preferences have crucial implications for when and where insurgents choose to attack the state. Civilian support enables rebels to evade state repression, maintain secrecy about their operations, and plan and execute attacks against security forces. Attacks against state forces become much harder to organize and conduct, and their gains more difficult to consolidate, in the absence of civilian cooperation (Kalyvas, Reference Kalyvas2006). We would thus expect insurgents to launch attacks against the state in places where local populations are sympathetic to their cause—communities where they can rely on civilians to donate goods and reinforce rebel ranks, to provide information about security force operations, and to hide insurgents from reprisal.
Like state capacity, civilians’ ability and interest in deterring rebel operations change over time, sometimes quickly. Wealth redistribution and anti-poverty programs can reduce popular support for insurgents (Albertus and Kaplan, Reference Albertus and Kaplan2013; Dasgupta et al., Reference Dasgupta, Gawande and Kapur2017), while rebels can increase community support by implementing popular governance programs or creating inclusive institutions (Kasfir, Reference Kasfir2005; Arjona, Reference Arjona2016). Abuse or victimization by combatants diminishes popular support (Condra and Shapiro, Reference Condra and Shapiro2012; Deglow and Sundberg, Reference Deglow and Sundberg2021). Civilian communities’ capacity to resist can shift, too: their economic and coercive endowments grow or shrink over time and the social institutions that enable collective action fray or consolidate.
Ultimately, the local balance of power thus depends on the evolving capacity and agency of both security forces and civilians. We propose that this shifting, fine-grained topography of state capacity and civilian opposition defines opportunities for local insurgent commanders to launch attacks with a higher probability of success. These opportunities, we suggest, represent a neglected explanation of insurgent military tactics. To be sure, this proposition is not new to those who fought civil wars. Both Mao and Guevara emphasized the importance of attacking security forces where and when they are weak, especially where perceived civilian support creates propitious opportunities to do so. But these claims merit further theoretical attention and empirical scrutiny. We develop this intuition in the following section.
3. A logic of opportunistic violence
We propose that insurgent violence in asymmetric conflicts is often driven by a logic of opportunism. In both centralized and decentralized military organizations, rebel commanders react to local conditions in planning operations. In doing so, these commanders seize opportunities where security forces are vulnerable to attack: where rebel strikes are more likely to succeed in conquering territory, seizing resources, reducing military effectiveness, or shifting public opinion. We argue that this likelihood of success is a joint function of the state’s local repressive capacity—its ability to effectively wield coercion to defend and quickly respond to attacks in a given location—and the chance of encountering tenacious civil resistance.
As the existing literature contends, the state’s ability to repress challengers is a key driver of insurgent violence. This ability varies across countries (Fearon and Laitin, Reference Fearon and Laitin2003), across state-insurgent dyads (Kalyvas and Balcells, Reference Kalyvas and Balcells2010), and over the territory the state lays claim to (Buhaug et al., Reference Buhaug, Gates and Lujala2009; Schutte, Reference Schutte2014). We argue that this final dimension of variation in state capacity—variation over space—is more granular than often conceived in existing work. Repressive capacity is generally high in the immediate vicinity of military bases, police stations, and centers of administrative state power (Tao et al., Reference Tao, Strandow, Findley, Thill and Walsh2016) and decays quickly but unevenly over space. A host of factors affect the state’s ability to project coercive power on demand in a particular area: how close the nearest security forces are; whether they can travel on passable roads or muster precision air power; and if they can access infrastructure for fuel, electricity, and communications.
Emphasizing the mechanisms of deploying state repression makes evident that a state’s ability to deter armed challengers also fluctuates finely over time. Some drivers of a state’s ability to deliver coercion change across months or years: national leadership shapes resources and military strategy, roads and infrastructure develop and fail, and the political environment shifts as local governments turn over and mass sentiment evolves. But many others change within days or hours. Security forces patrol, respond to threats, and return to base. Weather and natural factors impede the ability of state agents to access areas and call for reinforcements. States are also constrained by developments elsewhere; rapidly emerging security threats or natural disasters may occupy state resources and distract policymakers. In combination, this spatial and temporal variation generates a fine-grained, dynamic topography of state repressive power.
Fluctuations in the state’s ability to deter violence open temporary windows of opportunity for rebel groups to attack the state at moments and places of particular vulnerability—when local detachments are weak or under-supplied, when the arrival of reinforcements or materiel is impeded, and when policymakers and commanders are preoccupied with other priorities. In these moments, rebel attacks are more likely to succeed in damaging incumbents’ military assets, demoralizing their troops, seizing key resources or territory, or casting symbolic blows against state legitimacy. Regardless of the specific intent of a given military operation, it is more likely to succeed in its aim—and less likely to be met with immediate retaliation—when the state is temporarily weakened.
This emphasis on windows of opportunity has a long heritage in models of “routine activity” in criminology, which suggests that violent crime occurs when actors capable of maintaining order are absent “at specific locations in space and time” (Cohen and Felson, Reference Cohen and Felson1979). We believe that a similar logic underpins opportunistic insurgent violence. Although macro-level factors influence insurgent violence—the group’s ideology and grand strategy, the national balance of power between group and state—in practice, rebel strikes often respond to favorable local conditions. Temporal shifts in state repressive capacity and permissive conditions regarding anticipated civilian resistance or aid alter the local balance of power, producing frequent windows of opportunity for insurgent commanders to seize.
Crucially, the appeal and viability of these windows of opportunity depend on the expected behavior of local civilian communities (Kalyvas, Reference Kalyvas2006). Building on recent work examining how civilian agency mediates the relationship between state capacity and political violence (Carter et al., Reference Carter, Shaver and Wright2019; Müller-Crepon et al., Reference Müller-Crepon, Hunziker and Cederman2021), we argue that insurgent commanders consider the potential for civilian compliance or resistance when selecting opportunities to attack. In doing so, they rely on information about civilian attitudes as signals of how they would respond to rebel attacks on security forces. Civilian attitudes toward the state and insurgents vary widely across space and time. Some civilians are hostile toward both state and challenger, while others have mixed perceptions or may outright support one side or the other. Importantly, much of this variation occurs at the community level (Matanock and García-Sánchez, Reference Matanock and García-Sánchez2017).
That popular attitudes toward combatants vary over space and time has important implications for opportunistic rebel strategy, particularly when rebels’ objective is to strike a blow against the state.Footnote 5 Communities hostile to the state may communicate to insurgents that they will not use their capacities to resist attacks on security forces—and that they may even use violent or non-violent tactics to support such an attack or resist reprisals. In contrast, communities with strongly pro-state attitudes may explicitly or implicitly threaten to resist armed groups who attack the state, using force to defend state interests, warning security forces of impending assaults, or using non-violent action to resist rebel rule. Communities with both anti-state and anti-insurgent attitudes may similarly promise to resist attacks on state agents, if only to keep the war out of their homes. We anticipate that rebel commanders will be aware of these community-level attitudes and will account for them in their tactical decision-making.
To be sure, window of state weakness could also offer opportunities to wrest strategically valuable state-aligned communities from government control. But consolidating these gains is likely to be fraught. When local state capacity returns to its baseline level, incumbents will be highly motivated to regain control over loyal communities and will be able to capitalize on civilian loyalties to do so. For overmatched rebel groups, the goal of attacks against the state is generally to weaken available military targets, consolidate political or territorial gains, and live to fight another day. Attacking the state where communities are more sympathetic to rebels offers a surer means of reaching that goal.
Under an opportunistic logic,Footnote 6 then, state capacity and civilian preferences jointly determine rebel military tactics. Viable opportunities for attack require both windows of decreased state repressive capacity and the expectation that civilian populations will support, or at least not reject, rebel incursions; we expect that neither variable will be sufficient on its own to drive opportunistic violence. We formalize this expectation as follows:
• Hypothesis: Negative shifts in state capacity increase the likelihood of insurgent attacks unless insurgents expect high levels of resistance from local communities.
Importantly, we make no claim to explain the state’s military strategy. Some shifts in state capacity are intentional—purposive movements of troops and resources as elements of broader strategic designs. Others, like the natural factors we examine in our empirical analysis, are not. While the state’s strategy and tactics are important for understanding conflict dynamics, we bracket them here: for our purposes, both intended and unintended negative shifts in local state capacity generate windows of opportunity for insurgents to strike.
Nor do we claim that opportunistic tactics are the only or even the greatest driver of rebel violence against the state. Other logics of violence obtain as well: insurgent leaders pursue grand strategies aimed at improving their bargaining position, seizing exploitable economic resources, or diverting military assets from other priorities. These logics of violence, however, are consistent with different patterns of violence observed in civil war (Gutiérrez-Sanín and Wood, Reference Gutiérrez-Sanín and Wood2017). The unanticipated and constantly shifting topography of state capacity means that opportunities for reactive strikes are plentiful. Attacks against the state under such conditions highlight the existence of an explicitly opportunistic tactical logic, one which represents an important and understudied driver of insurgent violence.
4. Empirical evidence
4.1. Case
We evaluate this argument against evidence from the Colombian civil war. Colombia has suffered through decades of multiparty civil conflict. Leftist insurgents organized under the Fuerzas Armadas Revolucionarias de Colombia (FARC) waged war against the Colombian state from 1964 until formally demobilizing in 2016. Other insurgents also participated in the internal conflict, as did a diverse set of criminal organizations and paramilitary groups. Violence peaked in the mid-2000s as the Álvaro Uribe administration led a harsh military crackdown on the FARC. Despite the rebel group’s official demobilization, FARC dissidents and a constellation of armed groups continue to fight each other and the state.
The weakness of the Colombian state—and in particular the challenge it faces in controlling and administering peripheral regions of its territory—is frequently cited as a key driver of the conflict. State weakness or absence features prominently in explanations for the onset, persistence, and patterns of violence characterizing the war (e.g. González (Reference González2003), Revelo Rebolledo and García Villegas (Reference Rebolledo, Eduardo and García Villegas2018), Dávila et al. (Reference Dávila, León, Restrepo, Ramírez and Cruz2020)). The protagonists of the war used similar rhetoric when describing the connection between state capacity and conflict. Government and FARC negotiators issued a joint press release in 2016 stating that sustaining peace requires “promoting the presence and effective performance of the state across the national territory, especially in [...] regions bowed down by abandonment [and] the lack of effective [state] functions” (CNN, Reference CNN2016).
We focus our attention on the period of the conflict between 2002 and 2016. Following a failed peace process, Álvaro Uribe won the Colombian presidency in 2002 on a hardline anti-FARC platform. With support from the United States, he launched an aggressive military campaign against the FARC. This campaign marked the bloodiest period of the war, with frequent clashes and large-scale human rights abuses committed by security forces, insurgents, and paramilitaries. By 2008, the FARC had suffered major setbacks, losing much of its leadership and suffering nearly 20,000 desertions (Nussio and Ugarriza, Reference Nussio and Ugarriza2021). The conflict continued into the administration of Juan Manuel Santos (2010–18), who took a more moderate approach to the war, reducing military operations and leading negotiations that concluded in the 2016 peace agreement.
How did civilians view the FARC? Popular attitudes toward the insurgents were heterogeneous. Some civilians were sympathetic to the insurgents’ ideology, and some elements of FARC governance—resolving disputes, punishing crime, providing public goods, and offering protection from other armed actors—generated popular support (Arjona, Reference Arjona2016). Other FARC actions were extremely unpopular, such as the group’s kidnapping and extortion campaigns and high-profile acts of terrorism. Among communities who opposed the FARC, some favored the state or another armed group; others opposed the presence of any armed actor.
Communities who opposed the FARC pursued a variety of resistance strategies. Some declared autonomy from armed groups and neutrality in the conflict, with varying degrees of success. Some civilians—in particular indigenous and Afro-Colombian communities—organized early-warning systems to detect armed group incursions, dispute resolution systems so members need not resort to combatant justice, and mass mobilization to raise awareness of aggression (Alther, Reference Alther2006; Kaplan, Reference Kaplan2017). Communities influenced or controlled by the FARC drew on their capacity for collective mobilization to resist the elements of FARC rule that they most opposed (Arjona, Reference Arjona2016). Although their strategies varied, communities who opposed the FARC often acted on that opposition to resist insurgent aggression.
4.2. Research design
To empirically extricate the effects of state capacity from other drivers of conflict, we turn to nature. A mountainous country, Colombia is afflicted by frequent landslides caused by volcanic activity, earthquakes, heavy rainfall, and human behaviors such as construction, mining, and logging (Ojeda and Donnelly, Reference Ojeda and Donnelly2006; Aristizábal and Sánchez, Reference Aristizábal and Sánchez2020). Government data reports over 8,000 such events since 2000, some of which have blocked major roads and disrupted state access to rural areas for weeks or months at a time (BBC News, Reference BBC News2016; Inzunza, Reference Inzunza2017). In 2019, a series of landslides obstructed a primary road in the department of Cundinamarca, “virtually disconnect[ing]” Colombia’s eastern plains from the rest of the country (Alsema, Reference Alsema2019). In recent years, other landslides have closed roads in the regions of Antioquia (Ospina Zapata, Reference Ospina Zapata2019), Cauca (Romoleroux, Reference Romoleroux2019), and Chocó (RCN Radio, Reference RCN Radio2020), among others.
We posit that the timing of these landslide-induced road closures produces local fluctuations in state repressive capacity that are exogenous to the dynamics of the Colombian armed conflict. For communities that rely on the affected roads for access to the central Colombian state, these landslides generate a temporary negative shock to state capacity, discontinuously increasing the cost of deploying reinforcements or supplies to respond to attacks on local security forces. For armed groups, these disruptions mark windows of opportunity to strike while local military detachments are temporarily isolated and response is costly.Footnote 7
The location of landslides is not random and may not be exogenous to social factors that shape conflict dynamics. Mountainous areas with poor infrastructure, for instance, may have both a higher risk of landslides and higher levels of poverty, or stronger legacies of collective action or ethnic exclusion (Daly, Reference Daly2012). The timing of landslides, however, is primarily driven by natural phenomena: heavy rain, seismic activity, and natural disasters (Highland and Bobrowsky, Reference Highland and Bobrowsky2008, 30–33). We therefore rely on within-municipality temporal variation to identify the causal effect of road closures.
We construct a municipality-day panel dataset encompassing 1,052 of the 1,122 municipalities in ColombiaFootnote 8 spanning from January 1, 2002, to May 30, 2016.Footnote 9 To construct our treatment variable, isolation from the coercive state due to landslide-induced road closures, we use data from the Colombian Geological Service to identify the date and geographic coordinates of every landslide since 2002. We pair these data with a comprehensive spatial model of the Colombian road network using data collected by Open Street Map, an open source geospatial database.Footnote 10 We unite all road and street segments in the country to create a continuous representation of the national road network, which we segment at 100-meter intervals to produce a final network consisting of 5,832,866 nodes and 6,127,582 edges.
With this network constructed, we subset our list of landslide coordinates to those events that occurred within 50 m of a road — landslides that likely damaged or blocked roads. We count 2,768 such events across 902 unique days in our 15-year period of study. We also collect data on the locations of hamlets (veredas), the most disaggregated rural territorial unit in the country. Colombia has roughly 30,000 hamlets, which are rural subdivisions of municipalities.Footnote 11 We use these hamlets as the unit of analysis in estimating the effect of road closures in order to provide a more fine-grained model of access to the road network for rural communities. Finally, we draw data on all military bases in Colombia from Dube and Naidu (Reference Dube and Naidu2015) and record the locations of the 128 bases that fall within 10 km of the main road network. Figure 1 provides a representation of the road network, military bases, and landslides.
Figure 1. Road network, landslides, and military bases.
To construct our treatment variable, we first calculate the baseline distance between each hamlet centroid and the nearest military installation along the road network. For each day in our sample in which a landslide occurred, we cut the road network at the point(s) of the landslide(s) that occurred on that day and drop all adjoining edges. We recalculate the shortest distance from each hamlet to each military base. Finally, we record every instance in which the distance from a hamlet to the nearest military base increased or a hamlet was wholly isolated from the road network. Some treated hamlets lie very near the landslide, but others belong to municipalities substantial distances downstream of the road closure. Toy examples of this procedure are provided in Figure 2. Our final treatment variable is dichotomous, taking the value of 1 if one or more hamlets in a given municipality day were rendered less accessible due to a landslide and 0 otherwise. Our final sample contains 3,223 treated municipality-days.Footnote 12
Figure 2. Toy examples of shocks to military accessibility. (a) Minor re-routing: hamlet in Isnos, Huila, on December 8, 2010. (b) Major re-routing: hamlet in Dibulla, La Guajira, on March 10, 2007. (c) Change in military base: hamlet in Nátagua, Huila, on February 17, 2010. (d) Disconnected: hamlet in Ituango, Antioquia, on April 20, 2013.
Two limitations of this spatial-analytic approach bear mentioning. First, we are not able to directly measure the extent or length of the disruption to the road network caused by each landslide. Some landslides leave one or more lanes passable, while others fully block the road. Likewise, the damage from some landslides is repaired in hours or days, while other roads remain unusable for long periods. Some landslides may deliver no meaningful shock at all. Second, no road closure generates full, total isolation from the state security apparatus. Some landslides do fully cut off hamlets from the road network, but even in those cases the Colombian air force and airlifted troops could access isolated communities if sufficiently motivated.Footnote 13
We argue that this treatment nonetheless reflects typical constraints on state capacity. Modern states are rarely fully incapable of bringing force to bear within their territory—in particular since the advent and widespread adoption of air power. With few exceptions, a properly motivated state can muster troops or launch airstrikes anywhere within its borders. State incapacity, then, is a function of just how difficult or costly that act of power projection is. The treatment we evaluate mirrors this notion of state capacity: landslide-induced road closures do not make it impossible to project power into remote areas but do increase the cost of doing so. This relatively weak treatment also provides a hard test of our thesis—if road closures affect insurgent behavior, stronger fluctuations in state capacity likely generate larger effects still.
To measure our outcome variable, rebel-state violence, we rely on micro-level violence data from the Colombian National Center for Historical Memory (CNMH), an independent government entity that compiles data on clashes between the state and armed actors. The CNMH codes the date and location of clashes, the actors involved, and the party that holds the “military advantage” in a clash. We also collect data on violence against civilians from the Center for Research and Popular Education (CINEP), aggregated by Osorio et al. (Reference Osorio, Mohamed, Pavon and Brewer-Osorio2019). CINEP draws on press sources, reports from human rights organizations, and original collection to code the date and location of violent activity by armed groups and security forces. Since these datasets are not available at the hamlet level, we aggregate both these measures of violence to the municipality-day level.
We also use these data to construct an index capturing the violent presence of the FARC, which we incorporate as a covariate in our models to account for the group’s local coercive capacity. In line with previous work measuring the presence of armed actors in the Colombian conflict (Acemoglu et al., Reference Acemoglu, Robinson and Santos2013; Osorio et al., Reference Osorio, Mohamed, Pavon and Brewer-Osorio2019), we sum the number of military clashes and attacks on civilians involving the FARC in the municipality in the previous 2 years. We construct an analogous index to record the violent presence of paramilitary groups in each municipality-year, since the Autodefensas Unidas de Colombia and other right-wing militias often collaborated with security forces against the insurgents.
We proxy for anti-FARC attitudes in each municipality with a measure of conservative voteshare in local elections. While Colombia’s dominant political parties had relatively little ideological valence for much of the 20th century, the failed peace process of 1999–2002 and the 2002 election of Álvaro Uribe reoriented national politics around an ideological cleavage: policy toward the FARC. The political parties that formed Uribe’s conservative coalition were defined by their strongly anti-FARC attitudes and hardline approach to the armed conflict (Gutiérrez Sanín, Reference Gutiérrez Sanín2020). We therefore use support for those parties as a community-level measure of anti-FARC attitudes. Importantly, this is not to argue that communities that supported non-conservative parties were necessarily sympathetic to the insurgents. Rather, we claim that conservative voteshare represents a clear signal of anti-FARC affect—affect that would have been apparent to locally embedded insurgent commanders.
We draw on a dataset of historical election returns collected by the Universidad de los Andes to record the voteshare of conservative parties in the most recent mayoral elections in each municipality. During our study period, mayoral elections were held in 2003, 2007, 2011, and 2015. We draw on academic and journalistic sources (López and Sevillano, Reference López and Sevillano2008; Olivella and Vélez, Reference Olivella and Vélez2017) to identify conservative parties and record the combined voteshare of mayoral candidates from these parties.Footnote 14 We dichotomize this variable for ease of interpretation, assigning a value of 1 if the combined voteshare of conservative parties in the previous mayoral elections exceeded the sample mean (50.4%) and 0 otherwise.
Finally, we gather a set of municipality-year political and economic covariates that may affect violence. We draw a series of time-varying control variables from the Universidad de los Andes’ Municipal Panel: population, per capita tax collection and local government spending to proxy for the administrative capacity of the local government, and an indicator for the presence of coca cultivation. Coca is a highly lucrative cash crop, and coca cultivation is often associated with contestation among armed groups and the state. We lag these variables a year to avoid concerns of post-treatment bias. Appendix 2 shows summary statistics for these covariates, time series plots of the treatment and outcome variables, and balance statistics between treated and control municipality-years.
We estimate linear probability panel models.Footnote 15 Our treatment variable is dichotomous, taking the value of 1 if one or more hamlets in a given municipality-day were rendered more isolated by a landslide—if the distance to the nearest base was increased or the hamlet was fully cut off. The outcome variable is also binary, coded as 1 if one or more clashes between the FARC and the state occurred in the week following the municipality-day and 0 otherwise.
We add municipality and year fixed effects to all models. Municipality fixed effects help account for spatial correlation between where landslides occur or are detected and conflict dynamics. Importantly, they also allow us to control for time-invariant social characteristics of municipalities: baseline levels of conservative orientation, the historical presence and influence of the FARC, and the underlying degree of state capacity. Year fixed effects adjust for general time trends across our sample, including macro-level changes in national politics and the strength and strategy of both the Colombian national security state and FARC central command. In all models, we cluster standard errors at the municipality level, to allow for auto-correlation within municipalities over time, and at the year level, to account for variables measured yearly and common shocks within years.
Our inferential leverage stems from the as-if random timing of landslides. Our central contention is that the day on which a landslide occurs is exogenous to the political and military dynamics of the Colombian civil war, instead responding to natural conditions such as rainfall and seismic activity. Conditional on the inclusion of municipality fixed effects—since the location of landslides may correlate with these dynamics—the precise timing of landslides should generate short-term exogenous variation in local state capacity in affected communities.
We first assess the overall effect of landslide-induced isolation on FARC-state clashes. We estimate the following equation:
\begin{align*}
\text{Clash}_{i,[t,t+6]} & = \beta_1 \text{Isolated}_{i,t} + \beta \text{Covariates}_{i,t} + \beta FE_{\textit{municipality}} + \beta FE_{\textit{year}} + \epsilon_{i,t}
\end{align*} where i and t index municipalities and days, respectively, and 
$\textit{\text{Clash}}_{i,[t,t+6]}$ indicates whether municipality i experienced a FARC-state clash in days t to t + 6. We add the covariates enumerated above alongside the municipality and year fixed effects. We then turn to evaluating our core hypothesis: that the effects of negative shocks to state capacity are conditional on civilian preferences. We therefore replicate the specification above but interact Isolated with Conservative,
\begin{align*}
\text{Clash}_{i,[t,t+6]} & = \beta_1 \text{Isolated}_{i,t} + \beta_2 \text{Conservative}_{i,t} + \beta_3 \text{Isolated} \times {\textit{Conservative}}_{i,t} \\
& \quad + \beta \text{Covariates}_{i,t} + \beta FE_{\textit{municipality}} + \beta FE_{\textit{year}} + \epsilon_{\textit{i,t}}
\end{align*}4.3. Results
Table 1 presents the main results of our analysis. We begin by evaluating the unconditional effects of landslide-induced isolation on FARC-state clashes, estimated without (Model 1) and with (Model 2) covariates. We find that isolated communities were nearly 1 percentage point more likely to experience a FARC-state clash in the week following the landslide (p < .05), an increase equaling 75% of the mean probability of clashes in the sample. In the full sample of municipalities, negative shocks to state capacity do seem to encourage insurgent attacks.
Table 1. The effect of isolation and civilian attitudes on the probability of FARC-state clashes
Two-way (municipality and year) standard errors in parentheses. Signif. Codes:
** 0.01, * 0.05.
Is this effect conditional on the level of civilian opposition that the insurgents expect in a given community, as our theory predicts? Models 3 and 4 interact Isolated with Conservative, with and without covariates. We uncover a sharp heterogeneity underpinning the results of the unconditional models, offering clear support for the notion that state capacity and civilian opposition jointly determine insurgent attacks. In non-conservative municipalities, isolation has a strikingly large and statistically significant effect on rebel violence: days in which a municipality has one or more hamlets cut off from the road network are 2.3 percentage points more likely to see FARC-state clashes in the following week—an increase that represents nearly double the mean probability of FARC-state clashes in the sample.
In conservative municipalities, in contrast, this effect is entirely erased. When these municipalities experience landslide-induced isolation, the impact is fully reversed and is no longer statistically distinguishable from zero. Nor do civilian preferences alone generate violence: in the absence of isolation, conservative ideology has no discernible effect on the likelihood of clashes. Consistent with an opportunistic logic of violence, it appears that both low local state capacity and low civilian opposition are necessary to drive insurgent attacks.
These are relatively short-term effects. Figure 3 plots the coefficient on Isolated as we expand the size of the window for observing FARC-state clashes from 1–25 days post-treatment for separate samples of conservative and non-conservative municipalities. For non-conservative municipalities, isolation appears to substantially increase the likelihood of clashes across a range of windows; this effect is statistically significant for windows of 6–18 days post-treatment. At no point does isolation increase the probability of clashes in conservative municipalities. This pattern is in line with our theoretical expectations. Landslide-induced road closures rarely last more than a week or two. After roads have been cleared or repaired, the level of local state capacity should return to its pre-shock level. FARC commanders appear to have recognized this fleeting window of opportunity and—when local social conditions were relatively favorable—seized it.
Figure 3. Effect duration.
Evidence further indicates that these windows of opportunity offered real strategic advantages for the FARC. The CNMH codes which party holds the “military advantage” in a clash. This coding is likely somewhat subjective and suffers from approximately 20% missingness. But it points to a marked difference in conflict dynamics between treated and untreated municipality-days. In municipalities unaffected by landslide-induced road closures, security forces held the upper hand in 29% of encounters with FARC units. When communities were isolated by landslides, that percentage declined to 14%. State weakness, this evidence suggests, manifested starkly on the battlefield.
4.4. Discussion
These results are robust to alternative model specifications and variable operationalizations. We conduct a series of robustness checks, detailed in full in Appendix 3. We find that our results persist if we employ continuous measures of conservative voteshare and isolation and if we adjust how we model the error structure of the data (Table A.4). Since seasonal climatic conditions could affect both landslide frequency and conflict dynamics, we add month-of-the-year fixed effects and find no change (Table A.4). In case the natural factors that increase the risk of landslides (e.g. topography, rainfall, and wind) also affect conflict dynamics, we estimate a landslide “risk score” and incorporate that in our main models (Table A.5); we find the same results. We also uncover consistent results if we increase the threshold for treatment—how much the distance to the closest military base must increase to classify a hamlet as treated—although the models lose significance as the number of treated units, already scarce, shrinks further (Figure A.3). Finally, we test whether the results vary if we measure conservative voteshare in national rather than local elections (Table A.6).
We also evaluate three potential alternative explanations for our results, presented in full in Appendix 4. A first concern may be that landslides cause general shocks to local political equilibria between the state, armed actors, and civil society, triggering wider conflagrations of violence rather than the narrowly targeted attacks on state security forces envisioned in our theory. We find no evidence for this account: treatment does not affect FARC attacks on civilians or confrontations with other armed groups, but rather solely increases violence against the state (Table A.7).
Second, we consider the role of economic grievances. In addition to limiting state repressive capacity, landslides can inflict substantial economic tolls. These economic effects, not shifts in state capacity, could explain our results: negative economic shocks may create incentives for civilians to mobilize in favor of the FARC, thus generating violence (Collier and Hoeffler, Reference Collier and Hoeffler2004). We conduct two placebo tests to assess this possibility. First, we evaluate whether landslides that did not affect roads (i.e. that may have caused economic damage but did not impact state repressive capacity) increase FARC attacks. We find that non-road landslides have no effect on FARC-state clashes (Table A.8). We also test whether landslides that damaged local economic output—which we measure with satellite-recorded nighttime luminosity (Mellander et al., Reference Mellander, Lobo, Stolarick and Matheson2015)—increased the likelihood of attacks. We find no significant interaction effect between isolation and decreased luminosity (Table A.8), further suggesting that the economic impact of landslides likely does not explain our findings.
A final set of concerns centers on our measure of conservative attitudes. While our model isolates exogenous variation in state capacity, the covariates and fixed effects in the model may not be sufficient to causally identify the effect of conservative attitudes. Perhaps civilian preferences alone explain patterns of insurgent attacks. To address this possibility, in Appendix 4 we present a difference-in-difference study around local elections to examine whether new information about civilian attitudes affected the likelihood of FARC attacks. We find no evidence in support of this thesis: civilian preferences do not seem to drive insurgent violence without concurrent shifts in state repressive capacity (Table A.9).
A related concern may be that conservative voteshare is a poor proxy for anti-FARC sentiment. We acknowledge this concern: although a hardline approach to the war was a defining element of conservative politics in our time period, conservative parties espoused a changing bundle of beliefs across issue areas. Our measure may be capturing those beliefs, rather than anti-FARC affect. Measuring popular attitudes toward combatants in civil war is notoriously difficult (Lyall et al., Reference Lyall, Blair and Imai2013; Matanock and García-Sánchez, Reference Matanock and García-Sánchez2017), and ours is an imperfect proxy.
To demonstrate that conservative voteshare tracks anti-FARC sentiment, we conduct two empirical tests. We show that self-reported affiliation with a conservative party predicts lower levels of trust in the insurgents and less willingness to negotiate with them (Table A.9). We also replicate the main analysis using a more direct proxy for attitudes toward the FARC: the municipal voteshare of the FARC political party in the 2018 Senate elections. We find that municipalities that would go on to vote for the FARC party in greater numbers were more likely to see strikes following landslide-induced isolation (Table A.10). Taken together, this evidence supports the notion that anti-FARC affect—and not a related public attitude—likely drove insurgent calculus.
Finally, we consider the possibility that our measure of anti-FARC attitudes actually captures local or central state capacity. Partisanship could correlate with state capacity if the central government is more likely to deploy coercive assets to co-partisan local authorities (Trejo and Ley, Reference Trejo and Ley2021). Leveraging variation in partisan orientation between the Uribe (2002–10) and Santos (2010–18) administrations, we find no such effect (Table A.11). Partisanship could also capture state capacity if more conservative local governments invest more in coercion. Although policing is centralized in a national agency in Colombia, mayors share authority over police operations in their municipality (Castillo, Reference Castillo2020). We find no support for this proposition. Even in communities with non-conservative mayors, higher conservative voteshare decreases the likelihood of attacks. The core results also hold if we only consider FARC clashes with the military, which mayors do not control (Table A.11). Although we cannot definitively rule out all alternative explanations, these tests suggest that our results are not explained by civilian preferences alone, other dynamics of violence, economic shocks, or unmeasured state capacity.
5. Implications
How should we understand patterns of insurgent violence against the state in civil war? We have argued that the geography and timing of insurgent attacks are often driven by the local strategic environment in which rebel commanders operate. States’ capacity to project repressive power fluctuates finely across space and time, generating temporary pockets of weakness for armed challengers to exploit. But the feasibility of these opportunities also varies according to the local social environment. When armed actors expect to confront high levels of violent or non-violent civilian resistance, windows of state weakness lose their appeal. We construct a novel dataset measuring local state capacity, civilian attitudes, and armed clashes in the Colombian civil war. Leveraging exogenous temporal variation in local state capacity caused by natural landslides, we find support for this opportunistic logic of insurgent violence.
To be sure, our results may not generalize across all settings. We focus here on conflicts in which insurgents are weaker than the state. Our findings may not hold in conventional wars, or in profoundly asymmetric ones: if states are extremely strong relative to challengers, minor fluctuations in their repressive reach may not represent meaningful opportunities. Similarly, deeply unpopular insurgents may be so limited by universal civilian opposition that windows of state weakness are never viable. Organizational factors likely matter as well: strictly hierarchical insurgent groups may be less reactive to local strategic conditions. Future research exploring these scope conditions would be fruitful; opportunistic tactics may be rampant in some conflicts and rare in others.
We believe that these findings offer two principal implications for scholars of political violence and statebuilding. First, they suggest that we reconsider the notion of state fragility. States that rely on their citizens to maintain the monopoly on violence are often viewed as weak, unable to conduct their basic Weberian function without help. This perspective neglects the fact that community agency can effectively buttress state power. Communities can defend state interests, enforce state-led social order, and resist changes to the political status quo. When civilian preferences align with those of government leaders, the threat of social mobilization can reinforce the material power and legitimacy of the state. This is consistent with a view of “state capacity as an interactive process, born of relationships between different social groups” (Berwick and Christia, Reference Berwick and Christia2018, 74).
Second, our findings offer implications for theories of subnational state capacity. We focus on the state’s capacity to repress and coerce challengers. Yet other facets of state power also vary finely over space and time. Government administrative capacity, which depends on the local presence of bureaucrats and assets, varies across towns and neighborhoods and shifts over time as resources are re-allocated and personnel turn over. The same holds for public goods provision, propaganda, and political representation. These forms of capacity likely have important implications for political violence in their own right: service provision, messaging infrastructure, and representation shape the formation of grievances that could lead to armed mobilization, while administrative capacity determines the state’s ability to monitor and co-opt challengers. Across all these dimensions, the shifting topography of state presence deserves further theoretical attention in studies of the state and political violence.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2025.8. To obtain replication material for this article, please visit https://doi.org/10.7910/DVN/LCZNCF.
Acknowledgements
We are grateful to Michael Albertus, Sarah Zukerman Daly, Benjamin Lessing, Paul Staniland, Michael Weintraub, and two anonymous reviewers, as well as to the participants in the 2020 Midwest Workshop on Empirical Political Science, the “Public goods provision during war: Evidence from the Colombian case” panel at the 2020 APSA Annual Meeting, the 2021 Latin American Peace Science Annual Meeting, the Yale-Chicago Latin American Politics Working Group, and the University of Chicago Comparative Politics Workshop and Political Violence Working Group. We declare no competing interests.
Open access
