Thinking outside the (temporal) box to explain protracted intrastate conflict

Abstract

Conflict duration has been one of the central and enduring questions driving civil war literature. Still not enough attention has been given to the interdependency of conflict duration dynamics. In an effort to bridge the gap this study introduces a new variable that positions conflict duration as a function of the duration of the pre-conflict phase. I argue proto-insurgents are able to protract conflict due to good choices made before large-scale conflict erupted – or during a period of time called ‘incubation’. After controlling for standard explanations, this study offers statistical evidence that proto-insurgent incubation duration is statistically significant and positively related to conflict duration. This study further explores the usefulness of thinking outside of the standard temporal space of wartime by moving beyond the widely accepted assumption that insurgents are empowered and constrained primarily by wartime decisionmaking and the wartime environment in which they find themselves.

Keywords

conflict duration intrastate conflict prewar temporality

Introduction

Conflict duration has been one of the central questions driving the civil war literature for at least the last two decades. Interest in the subject appreciated substantially in the wake of Fearon’s (2004) ‘Why do some civil wars last so much longer than others?’ and Collier, Hoeffler & Söderbom’s (2004) ‘On the duration of civil war’. Off the presses within days of each other, both probed causes and branded culprits of protracted conflict. The former found illicit economies like opium and cocaine as well as the sons of the soil dynamic tended to produce conflicts of longer duration. In addition to low per capita income and high inequality, the latter found ethnic fractionalization to be a significant predictor of protracted conflicts. Since these impressive findings, other noteworthy studies on the subject include Sobek & Thies (2015), which studied the causal relationship between lootable resources, state capacity, and conflict duration. Conrad et al. (2019) investigated natural resources and conflict duration. And Metternich (2012) and Wucherpfennig et al. (2012) both assessed the impact of ethnicity on conflict duration. While rigorous in method and implementation, these and others have commonly studied protracted intrastate conflict using the same or similar variables as those introduced by Fearon and Collier et al. all those years ago. While some new pathways have been forged on the conflict duration variable side of things, scant attention has been given to the prospect of the influential nature of pre-conflict factors on conflict duration.¹ The inclusion of such variables could improve our ability to predict protracted affairs and may capture what appears to be a heretofore ignored component of such conflicts. This study makes a modest attempt to address this gap by introducing a novel variable – a variable I call ‘incubation’ – that captures the length of time proto-insurgents spent preparing for conflict. After controlling for standard explanations, statistical evidence suggests proto-insurgent incubation duration is statistically significant and positively related to conflict duration.²

Incubation – how it is missed and why that matters

According to Fearon (2004) one way to approach the puzzle of protracted conflict is to develop hypotheses about the dynamics that potentially produce them and then gather data to test for causal relationships. Scholars interested in describing conflict duration (and outcome) have mostly focused on wartime-specific variables using measures that capture state-side attributes like GDP, regime type, and state capacity, measured from conflict onset forward in time. Exceptional studies have permitted these variables to vary, when applicable, across time. Insurgent group characteristics like military capacity, resource networks, and foreign patronage have also tended to be measured beginning at conflict onset and have been permitted to vary, when applicable, in exceptional studies. Researchers have omitted pre-conflict measures more often than we might care to admit though – declaring they are unreliable predictors of wartime processes (Greig, Mason & Hammer, 2018: 524) or noting the secretive nature of the nascent phases of proto-insurgent group formation present challenges for systematic collection of pre-conflict data (Lewis, 2017).³ Whereas studies have been inclined to treat the phases of conflict (i.e. pre-conflict, conflict, and post-conflict) as isolated and unrelated processes, Chiba et al.’s well-executed 2015 study argued the phases of conflict should be treated as interdependent processes.

The tendency to regard the phases of conflict as discrete is reflected in our commonly used datasets. To illustrate, many studies have predominantly used what Bazzi & Blattman (2014: 9) aptly coined the ‘four major datasets’: Fearon & Laitin (2003), Sambanis (2004), the Correlates of War (COW) (Sarkees & Wayman, 2010), and the Uppsala Conflict Data Program/Peace Research Institute Oslo Armed Conflict Dataset (UCDP/PRIO ACD; Gleditsch et al., 2002; Pettersson & Wallensteen, 2015).⁴ These data have been exceptionally useful for conducting large-N analyses to test for relationships among variables in the temporal context of wartime. On one hand, this is a sensible approach when operating under the assumption that antecedent factors, including regime-type, economic indicators, or what insurgents did (or did not do) prior to conflict onset likely have little explanatory power for understanding conflict duration. On the other hand, it is reasonable to assume pre-conflict information, especially data on proto-insurgent groups, is critically important and could help further clarify why select groups had the capacity to sustain long-term campaigns. A variable that captures pre-conflict information, such as the length of time proto-insurgents spent preparing for conflict (incubation hereafter), could shed light on interesting causal chains that began during the initial stages of insurgent group formation and continued after conflict onset. These data could help explain how, or if, organizational structures, recruiting strategies, and resource networks set up months or perhaps years prior to conflict onset helped facilitate wartime preparedness and conditioned insurgents to endure conflict longer than those that hurried such organization-building tasks, or those that forewent them altogether.

Presumably there are some insurgent groups that formed months, years, or even a decade or more before conflict began, others that launched their campaigns straight away, and many more somewhere in between. But measures that capture insurgent group information during the nascent pre-conflict phase are in short supply.⁵ The primary methodological shortcoming of this omission is that there is often a gap in time between the precise date when conflict began (as measured by the UCDP 25 battle-related death threshold, for example) and the date when the insurgent group initially formed. Why did some wait so long after initially forming to launch their campaigns? What, if anything, did they do during the pre-conflict phase? Does this phase influence conflict later on? Focusing our controls on a singular temporal context not only gives us an incomplete picture of insurgent groups’ entire lifespans but also turns a blind eye to Pierson’s warning in Politics in Time (2004: 2): ‘social science often pays [a very high price] when it ignores the profound temporal dimensions of real social processes’.

The incubation variable proposed here is a preliminary effort at capturing a pre-conflict insurgent-specific measure and rests on the assumption that the length of time groups underwent pre-conflict preparation is a good proxy for insurgent group preparedness, which can help further explain why some endured conflict for long periods of time while others were trounced in short order. According to Chiba, Metternich & Ward (2015: 516): ‘the theoretically implied unobserved interdependence between the durations of pre-conflict peace, conflict, and post-conflict peace’ is an important but understudied notion worth developing further, and is one of the primary motivations of this study.

The theory of incubation

Setting aside time beforehand to prepare to manage the challenges and endure the hardships associated with any number of arduous endeavors is a modest notion. Conventional militaries worldwide customarily shuttle new recruits through initial pre-combat training that lasts anywhere from six to 12 weeks. Additional training lasting upwards of 52 weeks typically follows this preliminary phase. That those insurgents who withstood an undertaking as formidable and uncertain as political violence may have also set aside time to prepare beforehand is a similarly humble concept.

Incubating insurgents are exemplary units of analysis both theoretically and practically for studying the implications of behaviors across time largely because erroneous decisions early on can have habitually dire consequences down the road. Maybe nowhere else do we have greater disincentives than for those planning to use violence to challenge a sovereign’s claim to rule. More often than not insurgencies will be unpopular, underfunded, understaffed, outnumbered, and ill-equipped. Moreover, peddling political violence is an especially challenging task. The state enjoys relative legitimacy in the eyes of the public and the international community, and it has a larger checkbook, access to advanced military technology, and the added advantage of moderately better trained and equipped military, security, and police forces at its immediate disposal.

Furthermore, even the most assertive proto-insurgent leader cannot realistically expect to defy the state singlehandedly. How proto-insurgent leaders actually persuade others to sign up for an insurgency presents an intriguing puzzle. Even more puzzling is how they manage to hold an organization together even (especially) when they are losing. Assuming some buy in to the cause, a leader cannot have foot soldiers who have no idea how a rifle or bomb works going up against professional military, police, and security services and expect them to pull off attacks in any efficient way.⁶ There is basic competence to consider and turning a farmer, factory worker, or college student into a willing and efficient foot soldier does not happen overnight. For example, Green (2016: 621) wrote: ‘To succeed (indeed, even to remain in the fight), military commanders must first create a fighting force that is capable of great, unhesitating violence’ [emphasis added]. She went on to note a lot of research attention has been justifiably focused on the processes used to transform non-combatants into killers programmed to be compliant with organizational norms. Extended pre-combat training also gives leaders and mid-level commanders time to judge who is better at performing certain tasks that ought to lead to delegation of responsibilities during wartime in ways that yield more efficiency and productivity.

Staniland (2012) found social networks forged prior to conflict onset were critical for funneling important organization-building resources to proto-insurgents and bolstered internal cohesion, and improved long-term fighting power. Byman (2008: 167) argued other pre-conflict tasks include developing a ‘politically salient identity’, harnessing a ‘compelling cause’, constructing an ‘effective sanctuary’, and contending with ‘both violent and peaceful organizational rivals’. While all of these are important pre-conflict insurgency-building errands in their own right, scholars agree these tasks take time to accomplish and tend to produce especially robust organizations as a result. Sendero Luminoso for example, the most violent and long-lasting insurgency in Peru’s history, incubated for more than a decade. The FSLN in Nicaragua incubated for roughly 16 years.⁷

Table I.

Conflict duration measures: M23 (DRC) vs. al-Shabaab (Somalia)

	M23	al-Shabaab
Polity IV	Open anocracy	Open anocracy
Ethnic fractionalization score	0.878	0.812
GDP per capita	$472	$420
Terrain	903	2416
Foreign patronage	Yes	Yes
Loot	Yes	Yes
Gems	Yes	Yes
Drugs	No	No
Cross-border sanctuary	Yes	Yes
Incubation duration	64 days	824 days
Conflict duration	519 days	5,086 days^a

^aThe al-Shabaab insurgency is presently ongoing.

Polity IV: Marshall et al. (2017); Ethnicity: Fearon (2003); GDP: World Bank (2017); Terrain: Fearon & Laitin (2003); Foreign Patronage: Wood (2010); Loot, Gems, Drugs: UCDP (2016); Sanctuary: Wood (2010).

What follows then is a preliminary effort to draw a causal theoretical linkage between two temporal spaces – prewar and wartime. Given this brief literature survey, the minimum assumption that underlies the theory proposed here is that insurgent group wartime survival is positively related to the length of time spent preparing before conflict onset. The central organization-building processes that enable an insurgency to be more durable during wartime demand considerable time and attention. A lengthy incubation period is therefore a consequence of the difficulties associated with undergoing the tasks that are indispensable in the production of a robust insurgent organization. Consequently, the duration of incubation has enduring implications for insurgent wartime survival. A brief discussion of two cases illustrates how variation in the incubation variable across two cases, given largely similar environments, is helpful for assessing variation in conflict duration.

M23 vs. al Shabaab

Observe Table I. M23 and al Shabaab are two insurgencies in sub-Saharan Africa that campaigned under similar environmental conditions but exhibit very different conflict durations. M23 was initially formed on 4 April 2012 by Bosco Ntaganda, the former military chief of staff for the Congrès National pour la Défense du Peuple (CNDP). The insurgency clashed with the government of the Democratic Republic of the Congo (DRC) from 6 June 2012 to 7 November 2013.⁸ Some evidence even suggests al Shabaab may have initially formed in 2004 or even as early as 2003.⁹ According to UCDP, al Shabaab met the 25 battle-related death threshold on 2 January 2006 and was presently active at the time this study was penned.

The first variation to highlight in Table I is conflict duration – the puzzle this study seeks to explore further. The Congolese military routed M23 rather quickly – 519 days (approximately 17 months). When using the UCDP conflict onset start-date of 2 January 2006, al Shabaab has been actively campaigning for nearly 14 years, with no immediate end in sight. Otherwise these cases look quite alike across the balance of our standard controls, with but two notable disparities: terrain and incubation duration.

Terrain favorable to insurgency has been widely praised as a powerful predictor of protracted conflict. The measure of terrain utilized here speaks to the ‘proportion of the country that is “mountainous” according to the codings of geographer A.J. Gerard’ (Fearon & Laitin, 2001: 11) and as an example, Afghanistan exhibits one of the highest measures of ruggedness cataloged among many insurgency datasets (Fearon & Laitin, 2001: 11). While the terrain measure for Somalia is more than twice that of the DRC, both scores fall on the low end of the roughness scale. It is important to note, though, that this measure ‘does not pick up other sorts of rough terrain that can be favorable to guerrillas such as swamps and jungle’ (Fearon & Laitin, 2001: 11).¹⁰ Obviously the magnitude of effect on conflict duration by employing a terrain measure of this kind tells only half (or likely only part) of the story. The analyzing of other types of geography and terrain including the influence of dense cities is left for future researchers.

In Table I we also see M23 incubated for 64 days. When using the UCDP conflict onset measure, al Shabaab incubated for 824 days (or roughly two years and three months).¹¹ For M23, a 64-day discrepancy between initial formation date and the precise date the 25 battle-related death threshold was reached is not an especially excessive empirical omission (maybe), though an 824-day incubation period for al Shabaab is a significantly unobserved measure (at least from this author’s point of view). When tallied, al Shabaab incubated for 760 days longer than M23 and endured conflict 4,567 days (and counting) longer. Let us briefly unpack the incubation periods of these two cases to illustrate how significant differences in the behaviors of these two groups influenced the length of time each spent in incubation.

Exploring incubation at the micro-level: M23 vs. al Shabaab

In the wake of the Second Congo War (1998–2002) many rebels were swept up into the Congolese national army, the Forces Armées de la République Démocratique du Congo (FARDC), as part of a comprehensive two-pronged approach intended to demilitarize or integrate militia fighters that had survived the Congo wars (1996–97, 1998–2002). M23 was initially formed when approximately 300 former CNDP militants deserted the FARDC. This group of renegades was led by ex-CNDP militia leader and indicted war criminal, Bosco Ntaganda. Despite shared experiences fighting in the Congo wars, almost immediately M23 foot soldier allegiances were divided between the erstwhile CNDP general and the proto-insurgency’s second in command, Sultani Makenga. Those faithful to Ntaganda wanted to ‘protect Rwandophone interests’ and pressure the Congolese government into following through on the provisions of the 23 March 2009 peace accord, the source of the proto-insurgency’s namesake (Alida, 2014: 77).¹² Makenga loyalists however had far more lofty regional aspirations that were decidedly more violent in nature.¹³

M23 did very little during its 64-day incubation period to address this internal cohesion shortcoming – making little to no effort in socializing members into a united group or in rallying them behind a unified goal. This unattended problem caused a volatile internal organizational culture both during its short incubation period and later on during wartime (McKnight, 2014: 4). Reaching a crescendo on 15 March 2013, the Makenga faction violently routed Ntaganda and most of his loyalists from the insurgency’s ranks (Nangini et al., 2014: 4). After this internal conflict came to a head, many among those remaining surrendered or fled. Shortly thereafter the insurgency dissolved.

Neglecting to address this internal flaw sparked hasty organization-building elsewhere. M23’s contradictory operational agenda meant the proto-insurgency was largely inept at culling any semblance of willing support from the population. Ntaganda and Makenga turned to coercion and tried drafting new recruits by force – executing those that refused in many cases.¹⁴ Civilians residing in the eastern provinces of North and South Kivu, who were already weary from nearly a decade of persistent violence, justifiably detested this tactic and rejected M23’s advances wholesale. Instead of attempting to develop a more extensive recruiting strategy or better means to establish a reliable support network – organizational elements that would have likely helped sustain the insurgency during wartime – M23 nonetheless launched its offensive.

As for pre-combat training, while most of M23’s original members were grizzled veterans of the Congo wars, those hurriedly conscripted were often simply handed weapons and taught how to shoot (sometimes). The cursory attention given to combat training during incubation, coupled with neglecting to take time to address poor internal cohesion, significantly undermined the insurgency after conflict onset. For example, M23 foot soldiers were woefully outmatched on the battlefield once the FARDC was able to sufficiently organize the efforts of the police with its own. There were also high rates of desertion and defection, and those that did not flee proved difficult to coordinate and control. M23 militants often raped and looted at will, especially in the North Kivu provincial capital of Goma, the city of more than a million people that would eventually be the setting for M23’s last stand.

Comparatively, the al Shabaab insurgency is an amalgam of individuals hailing from different clans and rallied behind a religious-nationalist ideological banner. That al Shabaab was able to unite and hold together a group of individuals from otherwise distinct clans was described by Solomon (2014) as an extraordinary but puzzling organization-building marvel. Al Shabaab’s central leadership did indeed fashion a cohesive insurgency in the midst of a largely divided society suffering from more than a decade of conflict, though it did not happen overnight. One cannot make sense of either al Shabaab’s astonishingly successful recruitment efforts or the extensive support network it managed to build without first understanding the way it became systematically embedded in Somalia’s Islamic court system during incubation.

By the end of the 20th century, Somalia’s previously uncoordinated Islamic courts reached something of a rapprochement in northern Mogadishu and the loosely affiliated Islamic Court Union (ICU) emerged. Al Shabaab’s would-be central leadership realized the ICU’s potential for peddling its particular brand of militancy and thus set out to infiltrate it over the next few years. This opportunistic relationship proved critical in helping build al Shabaab’s fighting force and for establishing its extensive support network.

While the prominence of traditional clanism that had long dominated much of Somalia should not be understated, al Shabaab was able to transcend this and unify its members by employing a marketing strategy that fused radical religion-inspired ideology with ethnic nationalism (Hansen, 2013: 34). At this time most Somalis were ‘not especially religious […] [and practiced a] moderate Sufi branch of Islam’ (Hansen, 2013: 34). However, thanks to the ICU’s promises to reduce crime and usurp the highly unpopular warlord system that presided over much of Mogadishu, al Shabaab gained near unfettered access to recruits all across the city. Between 2000 and 2004 al Shabaab’s size and power grew slowly but progressively while cloaked in the shadow of the ICU.

Between 2000 and 2004 militant training camps shuttered during the 1990s were repopulated with new al Shabaab recruits. They were divided into small groups and ‘trained and educated in a way that minimized clan allegiance’ (Marchal, 2011: 384), promoted strong in-group loyalty, and boosted morale. Foot soldiers were likewise programed to obey a rigid command structure (Marchal, 2011: 384). In addition to training in Somalia, many among al Shabaab’s central leadership traveled to Afghanistan to train, courtesy of al Qaeda who had established a presence there in 1998 at the behest of Usama bin Laden (Joseph & Maruf, 2018: 24; Hansen 2013: 22). Abu Mansoor (a deputy leader in al Shabaab), the Turk (a military leader in Somalia’s Islamic Court Union), al Afghani (another of al Shabaab’s deputy leaders), and Godane (al Shabaab’s first Emir) all traveled to Afghanistan to learn from and be inspired by the fight against the country’s western occupiers (Vidino, Pantucci & Kohlmann, 2010: 219). Between 2000 and 2004 some 100 Somali militants reportedly made the trip to Afghanistan for this kind of on-the-job training in preparation for the coming battle back home (Jones, Liepman & Chandler, 2016: 11).

As for building a support network during incubation, Somali warlords were detested for levying high taxes on businesses still desperately trying to operate in Mogadishu. The ICU and its militias understandably made quite an impression when it ran many of these warlords out of town, reinstating a measure of order long absent in the war-torn capital city. Many business owners were justifiably delighted to support the ICU as a result and as its wallet grew, al Shabaab benefited. The ICU, however, did not at first recognize it was helping set the financial and social support groundwork for the proto-insurgency hidden within its ranks which, according to Hansen (2013: 36), was every bit ‘a parasite on the Sharia Courts’ back, thriving from their growth’.

Al Shabaab took a great deal of time to develop clever talent acquisition strategies and an effective marketing campaign to sell their product (i.e. revolution) to consumers. M23 did not. Al Shabaab took years establishing training camps, developing tactics for training and fostering cohesion, and building an extensive support network that fed a steady flow of resources and recruits, even still today – essential mechanisms for maintaining an insurgency. M23 did not adequately address these tasks and rather chose to launch its offensive after only two months of ill-conceived preparation. This introductory case study comparison demonstrates that incubation duration for al Shabaab was a key element for a high-quality prewar preparation and suggests lengthier incubation is theoretically linked to wartime survival capacity.

Exploring incubation at the macro level

If the difference in incubation of these two cases even marginally reflects variations exhibited in the broader insurgency case study pool, it is worth larger-scale testing. What follows is the coding methodology for incubation and statistical tests for its effects on conflict duration. A reasonable starting point for our analysis, however, would be to take some time defining: what counts as insurgency?

Fearon & Laitin (2001: 7) defined insurgency as: ‘a technology of military conflict characterized by small, lightly armed bands practicing guerilla warfare from rural base areas [and] are weak relative to the governments they are fighting’. Lyall & Wilson (2009: 70) described it as: ‘protracted violent struggle by non-state actors to obtain their political objectives – often independence, greater autonomy, or subversion of existing authorities – against the current political authority (the incumbent)’. Salehyan (2009: 291–293) referred to insurgents as ‘nongovernmental actors’ who use ‘organized violence against the state’ for ‘political ends’. The US Government Counterinsurgency Guide defined insurgency as the ‘organized use of subversion and violence to seize, nullify or challenge political control of a region’ (US Department of State, 2009: 2). Insurgency is defined here as: violence against an incumbent perpetrated by a non-state actor in an attempt to further political, social, or economic objectives.

This study makes a distinction between insurgency and military or political elite-led coups d’état such that conflict must have occurred between agents of the state (i.e. military, police, or other state-backed armed forces) and at least one non-state actor not backed by any factions of the challenged incumbent. These data likewise do not include terrorist groups, mob or factional violence where the state does not constitute at least one of the actors. The inclusion threshold for observations that fit this definition was the UCDP’s 25 battle-related death threshold, which was operationalized here as the start of conflict. Data collection pursuant to this operationalization strategy yielded a large universe of cases such that the inclusive list of observations (since 1800) topped out at more than 300. The data for this initial observation pool were derived primarily from UCDP and cross-referenced with Fearon & Laitin (2003), Sambanis (2004), Lyall & Wilson (2009), the COW (Sarkees & Wayman, 2010), and Wood (2010).

Incubation

The independent variable of interest here is operationalized as: the duration in days from the precise date of proto-insurgent organization initial formation to the precise date the conflict reached 25 battle-related deaths. To build the incubation variable from scratch a great deal of information about each observation needed to be collected – particularly the precise time and circumstances surrounding each individual proto-insurgent’s initial formation. I borrowed but modified Lewis’s (2017) methodology and operationalized initial insurgent formation as the most precise date when the organization had a discernable leader or leadership assembly and exhibited at least 50 members.¹⁵

The remaining observations were subject to a restricted time series, given that crucial data on standard controls for pre-1945 insurgencies were significantly lacking.¹⁶ The historical record for pre-1945 insurgencies is also far less clear regarding the specific type of warfare, and incorporating such cases would likely combine ‘disparate types of conflict that may not represent a homogenous population’ (Lyall & Wilson, 2009: 71).¹⁷ The potential pitfalls of pooling disparate observations has been explored elsewhere at length by Bartels (1996). These constraints were applicable for a non-trivial share of the 300 original cases identified.

Given the time series restriction the Viet Minh rebel forces of the first Indochina War (start date 5 January 1945) was the earliest observation recorded in these data whereby the full complement of controls could be collected with precision (Lyall & Wilson, 2009).¹⁸ This information came from a variety of sources including extant literature, elite-level biographies, news media outlets, primary source documents, and documents penned by a whole host of various government and nongovernmental agencies.¹⁹ The collection effort proceeded from the Viet Minh rebels to each consecutive recorded observation in the roughly 140, post-1944 cases that remained. Observations persisted in the dataset when at least two separate sources for initial formation could be obtained. The earliest recorded initial formation date was used in the event there were disparate start dates among sources. For example, I used the Stanford Center for International Security and Cooperation (CISAC) initial formation date (circa 1 April 2003) for the al Mehdi Army in Iraq, though Enders (2009) claimed the organization formed on 1 June 2003. Given the dearth of available information for some observations, there were typical but expected difficulties associated with observing and measuring the initial formation date for approximately ten proto-insurgents, which were dropped from the dataset. The coding methodology described here resulted in a final tally of 120 insurgent organizations (1945–2017).

To measure the incubation period for each observation remaining in the dataset, I counted forward in time (in days) from the date the organization had a discernable leader or leadership assembly and exhibited at least 50 members to the precise date each observation met ‘at least 25 battle-related deaths in a calendar year’ (UCDP codebook, Themnér, 2016: 2).²⁰ The mean incubation duration is 2,725 days with a standard deviation of 3,360. The lengthiest incubation period in these data is 13,814 days, and where 34 out of the 120 insurgent organizations incubated for three weeks or less.

Dependent variable

Conflict duration was measured beginning at the 25 battle-related death threshold forward in time, using years as the metric of time (denoted as Time in the model) to count the number of years between conflict duration onset and the point at which each case exited the data, per the six-point coding rules in the UCDP Conflict Termination Dataset Codebook (v.2-20). The first year of conflict was numbered 1 and the last year was set equal to the total number of years the group survived. For example, an insurgency with a conflict duration of nine years meant that case ranged from 1 to 9, where year 1 corresponds with the conflict onset year, as measured by the 25 battle-related death threshold, and where year 9 denotes the last year the observation appeared in these data. Using months or weeks (or even days) as opposed to years would require all of the controls to be measured in the same scale of time, which was not feasible for a variety of controls. Years as the metric of time was used because it represented the most widely existing specified time-series interval for most other covariates in the model. There is considerable variation in the months corresponding to ‘intro’ and ‘exit’ for observations, which suggests there is no systematic bias in this measurement approach.

The binary variable (Exit) was set equal to 0 for all of the non-final years of the group’s existence and 1 for the year of exit (Carter & Signorino, 2010; Lindstädt, Vander Wielen & Green, 2017: appendix). That is, the year of exit signifies the point at which conflict terminated, per the six-point coding rules in the UCDP Conflict Termination Dataset Codebook (v.2-2015). Observations in progress (i.e. ongoing conflicts) were coded 0 for every numbered year to denote that they had not yet exited the dataset.

Figure 1.

Kaplan-Meier insurgency survival estimate (conflict duration)

Figure 2.

Incubation and duration

Figure 1 shows the Kaplan-Meier insurgency survival estimate for the annual rate of exit across conflict duration. The likelihood of exit for any given insurgent group during the first ten years of conflict is approximately 26%. This increases 30 percentage points to around 56% in years 11–20, which suggests insurgent conflicts are more likely to end at some point within the second decade. The likelihood of exit decreases at a mean of 3% for every ten years of survival thereafter. In sum, protracted affairs are more likely to come to an end during the conflict’s second decade, with exit probability exhibiting a 3% decrease (on average) for every decade a given conflict extends past the 20-year mark.

Figure 2 plots a simple bivariate relationship between conflict duration and incubation duration, providing preliminary evidence of a positive association. Notice the greatest density of cases is where incubation duration and conflict duration are short. It should also be noted that while there is variation in outcome, many cases fall fairly close the regression line, which denotes the relationship between incubation and duration (excluding covariates).

Controls

In addition to incubation, the following standard conflict controls were considered: state capacity, regime type, GDP, terrain, ethnic fractionalization, foreign patronage (material and financing), cross-border sanctuary (both with and without host state consent) and evidence insurgents were involved in looting or illicit smuggling, trafficking, or sale of drugs and gems. The summary statistics for the variables and controls used in the model can be found in the Online appendix.

The model

What follows is a discussion of the fine-grained logistic regression model used in this study and its expected outcomes. A model of this type provides a nuanced picture of the relationship between incubation and insurgency survival, given a variety of standard controls. While a survival model would estimate the relationship between incubation duration and conflict duration (i.e. insurgent survival), models of this type do not as capably account for variation in covariates over time (the results of two aggregated standard survival models can be found in the Online appendix). This limitation with aggregated survival models can be solved by disaggregating cases into years (event history model) and allows the use of a standard logistic regression model, which permits all controls to vary, when applicable, across individual insurgent case-years (Lindstädt, Vander Wielen & Green, 2017: 11, appendix). It has been demonstrated elsewhere that splitting observations into shorter periods (disaggregating cases into years, in this case) ‘such that the values of the covariates are constant within periods but not across periods’ (Chiba, Metternich & Ward, 2015: 522) helps to assess the contributions for each disaggregated case-year individually.

Carter & Signorino (2010) demonstrated that the inclusion of a cubic polynomial of time in these types of logistic regression models is a highly effective means of modeling myriad types of hazards, and is therefore a straightforward way to account for time dependence.²¹ Best practices of modeling time with disaggregated event history data entail the use of a cubic polynomial term (denoted as: Time, Time², Time³), employed here since conflict initiation (Carter & Signorino, 2010; Lindstädt, Vander Wielen & Green, 2017: appendix). A joint F test on the three coefficients indicates the non-standard measure of duration dependence should be retained (see the Online appendix). There is an emerging methodological precedent among conflict scholars using disaggregated event history data in this way (Raleigh & De Bruijne, 2017; Carboni & Moody, 2018). The polynomial approximation permits controls to vary across years, when applicable, gives us a reasonable estimation for temporal dependence, and is easy to interpret because it ‘avoids problems such as quasi-complete separation’ (Carter & Signorino, 2010: 271).

The resulting logit model is as follows, where incubation_i denotes the duration (in days) of the incubation period for insurgent group i, with coefficient $β$ . Let x _it denote the vector of the controls (excluding incubation), for insurgent group i and time t, with the corresponding vector of coefficients $θ$ . Note that the inclusion of the subscript t indicates controls are permitted, when applicable, to be time-varying within individual insurgent group case-years. Note the cubic polynomial of the duration, in years, between the operational starting point of the insurgent group and its exit, denoted $y e a r s_{i t}^{j}$ for $j \in \{1, 2, 3\}$ (with corresponding coefficients $δ_{j}$ ). Because of the presence of repeated measures within groups, the model was estimated with robust standard errors clustered on unique insurgent groups to account for dependency in the errors.

{Exit}_{it} = α + βincubation + θ {x'}_{i t} + \sum_{j = 1}^{3} δ_{j} {y e a r s}_{i t}^{j}

Results

Incubation is negative and statistically significant (p < .01) as expected. This finding indicates that lengthier incubation periods reduce the probability of exit at any given period of time. Moreover, the effect of incubation proves to be of a magnitude of interest, reducing the probability of exit, on average, by approximately ten percentage points over the range of incubation. The results for the controls are relatively straightforward: terrain is statistically significant at the 99% confidence level; as a country’s elevation delta increases, the likelihood of exit decreases (negative beta). This finding is not surprising given the reputation rough terrain has amassed over the last three decades (or more) (see Fearon & Laitin, 2003; Hendrix, 2011; Gordillo, 2018; Bulloch & Morris, 1993; Mitchell, 1994; Scott, 1985, 2009). State capacity is significant at the 99% confidence interval; a shift in state capacity from 0 to 1 corresponds to an 88% increase in the likelihood of insurgent group exit. This too reflects findings

Table II.

The impact of incubation duration on conflict duration, with additional controls

Variable	Full model
Incubation	–.033**
	(.009)
Terrain	–.533**
	(.217)
Statecap	.297**
	(1.47)
GDP	–.255
	(.036)
Ethfrac	–1.87
	(.127)
Loot	–1.26
	(.776)
Drugs	–.974
	(.748)
Gems	–1.06
	(.895)
Patron	–.401**
	(1.71)
Sanctuary	–1.52
	(.359)**
Regimetype	–.019
	(.066)
Con	–2.25
	(.094)

LR chi2 (14) 130.85, prob > chi2 0.0000, Psuedo r2 0.1854; n = 1,097; *p < 0.05, **p < 0.01.

elsewhere with other measures of state capacity including Gibler & Miller (2014) and Hendrix (2011). Foreign patronage is significant at the 99% confidence interval, when an insurgency enjoys financial and/or material resources courtesy of a foreign patron, their likelihood of exit decreases by 75%. This finding aligns with studies elsewhere, most notably Lyall & Wilson (2009; see also Salehyan, Gleditsch & Cunningham, 2011). In keeping with conventional wisdom, GDP and cross-border sanctuary were both significant at the 99% confidence interval (Holtermann, 2012; Berman, Shapiro & Felter, 2011; Boix, 2008; Salehyan, 2008; Das, 2018). Ethnic fractionalization, regime type, loot, drugs, and gems were all insignificant. See Table II for the impact of incubation duration on conflict duration.

Limitations

As in all conflict studies, one concern with the approach here may be unobserved selection bias given not all proto-insurgents that underwent incubation initiated conflict, particularly one that resulted in at least 25 battle-related deaths. The empirical difficulties these unobserved non-starters present is a primary challenge for researchers of conflict processes. This study excludes groups that did not initiate conflict, examining conflict duration only among those that did. One might worry that there are selection effects resulting from the fact this study omits those groups for whom the number of battle-deaths falls below 25. However, identifying these groups so as to include them in the study is challenging, if not impossible to solve at this point. Instead, we need to consider the ways in which their omission might affect the findings presented here.

Whereas this study’s empirical analysis does not capture all known cases, selection effects would potentially be greater had a higher battle-related death threshold been used to identify insurgents. I thus determined 25 would minimize selection bias when compared to the 1,000, 500, or even 100 battle-related death threshold approaches habitually used in earlier studies. Assessing the probable direction of bias if cases with fewer than 25 battle-related deaths were included is also particularly challenging, given we do not know the quantity of unobserved proto-insurgents, and data on key variables for the non-starters we do know about are limited at this point (Byman, 2008; Lewis, 2017). To illustrate, Wickham-Crowley (1992: 16) wrote: ‘guerilla movements appeared throughout Latin America in the 1960s, but most died an early death […] [leaving] […] but few traces on the historical record’. I thus operated under the assumption that it is plausible unobserved non-starters had short incubation periods and would ostensibly strengthen my findings had the data existed to include them. Though to be sure readers would likely disagree on the magnitude and direction of possible bias here.

Fortunately gaps in our data have started to be addressed thanks to the work of scholars like Lewis (2017) and future models will be more encompassing. Current data limitations, however, do not outweigh the need for continued theorizing about conflict duration and I endeavored to get a sense of the distribution of empirical estimates for all potentially relevant observable variables. In the end, readers should be cautioned when assessing this study’s findings given the coding decisions and limitations of available data.

Conclusion

This article was motivated by the notion that the duration dynamics of civil war are powerfully interdependent processes. The specific goal of this study was to bridge the gap between the pre-conflict and conflict phases by demonstrating that the length of time proto-insurgents take to prepare for war significantly impacts their ability to protract conflict. M23 did very little to prepare during the pre-conflict phase. A mere 64-day incubation period was hardly enough time to establish the foundation for robust organization and it failed in fewer than two years as a result. Al Shabaab incubated for at least two years longer than M23 and during that time developed a successful recruiting campaign, a cohesive and well-trained fighting force, and a substantial support network. Al Shabaab endures even today, having carried out an exceptionally violent attack in Somalia’s capital city that killed 81 people and injured dozens more.

Lengthy incubation is not a unique feature of the al Shabaab insurgency, however. Sendero Luminoso and the FSLN both incubated for a decade or more. The incubation variable introduced here thus took into account the duration of the pre-conflict phase for a larger case study pool. Incubation duration had significant explanatory power on conflict duration, suggesting the duration of incubation has enduring implications for insurgent wartime survival.

From a theoretical perspective, getting a clearer sense of how proto-insurgents problem-solve in the nascent phases of formation is valuable in its own right. Understanding how the duration dynamics of conflict are powerful, measurable interconnected processes is an equally valuable empirical endeavor. From a policy perspective, the existence of long, drawn-out conflicts is one of the more disconcerting truths in today’s world. Despite evidence showing that states have found it increasingly difficult to defeat insurgents, politicians, experts, and military professionals continue to proffer variations on the same approaches. Identifying and testing the ways duration dynamics are integrally interconnected could be a useful key for developing better policy prescriptions going forward.

Footnotes

Replication data

The data for the empirical analyses in this article and the Online appendix can be accessed at .

Acknowledgements

For helpful discussions, the author thanks Hillel David Soifer, Janet I Lewis, and Ryan Vander Wielen.

ORCID iD

Joel Blaxland

Notes

References

Africa Research Bulletin (2012) Democratic Republic of the Congo: M23 mutiny escalates. 49(6): 19295A–19298C.

Albertus

Michael

Kaplan

Oliver

(2013) Land reform as a counterinsurgency policy. Journal of Conflict Resolution 57(2): 198–231.

Alida

Furaha Umutoni

(2014) Do they fight for us? Mixed discourses of conflict and the M23 rebellion among Congolese Rwandophone refugees in Rwanda. African Security 7(2): 71–90.

Allison

Michael E

(2016) Why splinter? Parties that split from the FSLN, FMLN, and URNG. Journal of Latin American Studies 48(4): 707–737.

Areiff

Alexis

Coen

Thomas

(2013) Democratic Republic of the Congo: Background and US Policy. Congressional Research Service Report for Congress 29 July.

Arjona

Ana

Kasfir

Nelson

Mampilly

Zachariah

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

Baaz

Maria Eriksson

Verweijen

Judith

(2013) The volatility of a half-baked bouillabaisse: Rebel-military integration and conflict dynamics in eastern DRC. African Affairs 112(449): 563–582.

Balcells

Laia

Kalyvas

Stathis N

(2014) Does warfare matter? Severity, duration, and outcomes of civil wars. Journal of Conflict Resolution 58(8): 1390–1418.

Barnes

Cedric

Hassan

Harun

(2007) The rise and fall of Mogadishu’s Islamic courts. Journal of Eastern African Studies 1(2): 151–160.

10.

Bartels

Larry M

(1996) Pooling disparate observations. American Journal of Political Science 40(3): 905–943.

11.

Bazzi

Samuel

Blattman

Christopher

(2014) Economic shocks and conflict: The evidence from commodity prices. American Economic Journal: Macroeconomics 6(4): 1–38.

12.

Berman

Eli

Callen

Michael

Felter

Joseph H

Shapiro

Jacob N

(2011) Do working men rebel? Insurgency and unemployment in Afghanistan, Iraq, and the Philippines. Journal of Conflict Resolution 55(4): 496–528.

13.

Berman

Eli

Shapiro

Jacob N

Felter

Joseph H

(2011) Can hearts and minds be bought? The economics of counterinsurgency in Iraq. Journal of Political Economy 119(4): 766–819.

14.

Blaxland

Joel

(2020) Insurgency Prewar Preparation and Intrastate Conflict: Latin America and Beyond. London: Palgrave Macmillan

15.

Boix

Carles

(2008) Economic roots of civil wars and revolutions in the contemporary world. World Politics 60(3): 390–437.

16.

Braithwaite

Jessica Maves

Cunningham

Kathleen Gallagher

(2020) When organizations rebel: Introducing the Foundations of Rebel Group Emergence dataset. International Studies Quarterly 64(1): 183–193.

17.

Bulloch

John

Morris

Harvey

(1993) No Friends but the Mountains: The Tragic History of the Kurds. Oxford: Oxford University Press

18.

Byman

Daniel

(2008) Understanding proto-insurgencies. Journal of Strategic Studies 31(2): 165–200.

19.

Carboni

Andrea

Moody

James

(2018) Between the cracks: Actor fragmentation and local conflict systems in the Libyan Civil War. Small Wars and Insurgencies 29(3): 456–490.

20.

Carter

David B

Signorino

Curtis S

(2010) Back to the future: Modeling time dependence in binary data. Political Analysis 18(3): 271–292.

21.

Chiba

Daina

Metternich

Nils W

Ward

Michael D

(2015) Every story has a beginning, middle, and an end (but not always in that order): Predicting duration dynamics in a unified framework. Political Science Research and Methods 3(3): 515–541.

22.

CIA (1978) Nicaragua: The Sandinista Guerillas and their international links. United States Central Intelligence Agency, DSNA Collection: Nicaragua . 6 September.

23.

Cohen

Herman J

(2012) Escalating violence in the eastern DRC. Global Perspectives: DRC, Stability Operations.

24.

Collier

Paul

Hoeffler

Anke

(2004) Greed and grievance in civil war. Oxford Economic Papers 56(4): 563–595.

25.

Collier

Paul

Hoeffler

Anke

Söderbom

Mans

(2004) On the duration of civil war. Journal of Peace Research 41(3): 253–273.

26.

Conrad

Justin M

Green

Kevin T

Walsh

James Igoe

Whitaker

Beth Elise

(2019) Rebel natural resource exploitation and conflict duration. Journal of Conflict Resolution 63(3): 591–616.

27.

Das

Pushpita

(2018) Security challenges and the management of the India–Myanmar border. Strategic Analysis 42(6): 578–594.

28.

De Veenhoop

Hans R

(2007) Opportunities and constraints for the disarmament and repatriation of foreign armed groups in the Democratic Republic of Congo: The cases of the FDLR, FNL and ADF/NALU. Conflict and Transition Consultancies, International Crisis Group (ICG).

29.

Enders

David

(2009) Behind the wall: Inside Baghdad’s Sadr City. Virginia Quarterly Review 85(3): 120–135.

30.

Fearon

James D

(2003) Ethnic and cultural diversity by country. Journal of Economic Growth 8(2): 195–222.

31.

Fearon

James D

(2004) Why do some civil wars last so much longer than others? Journal of Peace Research 41(3): 275–301.

32.

Fearon

James D

Laitin

David D

(2001) Ethnicity, insurgency, and civil war. Paper presented at the 2001 Annual Meetings of the American Political Science Association, San Francisco, CA, August 30–September 2.

33.

Fearon

James D

Laitin

David D

(2003) Ethnicity, insurgency, and civil war. American Political Science Review 97(1): 75–90.

34.

Green

Amelia Hoover

(2016) The commander’s dilemma: Creating and controlling armed group violence. Journal of Peace Research 53(5): 619–632.

35.

Gibler

Douglas M

Miller

Steven V

(2014) External territorial threat, state capacity, and civil war. Journal of Peace Research 51(5): 634–646.

36.

Gleditsch

Nils Petter

Wallensteen

Peter

Eriksson

Mikael

Sollenberg

Margareta

Strand

Håvard

(2002) Armed conflict 1946–2001: A new dataset. Journal of Peace Research 39(5): 615–637.

37.

Gordillo

Gaston

(2018) Terrain as insurgent weapon: An affective geometry of warfare in the mountains of Afghanistan. Political Geography 64(May): 53–62.

38.

Greig, J Michael; T David Mason & Jesse Hammer (2018) Win, lose, or draw in the fog of civil war. Conflict Management and Peace Science 35(5): 523–543.

39.

Hansen

Stig Jarle

(2013) Al Shabaab in Somalia: The History and Ideology of a Militant Islamist Group, 2005–2012. New York: Oxford University Press.

40.

Hendrix

Cullen S

(2011) Head for the hills? Rough terrain, state capacity, and civil war onset. Civil Wars 13(4): 345–370.

41.

Holtermann

Helge

(2012) Explaining the development–civil war relationship. Conflict Management and Peace Science 29(1): 56–78.

42.

International Crisis Group (ICG) (2005) Counter-Terrorism in Somalia: Losing Hearts and Minds? Nairobi: ICG.

43.

Jones

Seth G

Liepman

Andrew M

Chandler

Nathan

(2016) Counterterrorism and counterinsurgency in Somalia: Assessing the campaign against al-Shabaab. Santa Monica, CA: Rand (https://www.rand.org/pubs/research_reports/RR1539.html).

44.

Joseph

Dan

Maruf

Harun

(2018) Inside Al-Shabaab. Bloomington, IN: Indiana University Press.

45.

Lewis

Janet I

(2017) How does ethnic rebellion start? Comparative Political Studies 50(10): 1420–1450.

46.

Lindstädt

René

Wielen

Ryan J Vander

Green

Matthew

(2017) Diffusion in Congress: Measuring the social dynamics of legislative behavior. Political Science Research and Methods 5(3): 511–527 (Appendix).

47.

Lutmar

Carmela

Terris

Lesley G

(2019) Introducing a new dataset on leadership change in rebel groups, 1946–2010. Journal of Peace Research 56(2): 306–315.

48.

Lyall

Jason

Wilson

Isaiah

(2009) Rage against the machines: Explaining outcomes in counterinsurgency wars. International Organization 63(1): 67–106.

49.

Marchal

Roland

(2011) The rise of a Jihadi movement in a country at war: Harakat al-Shabaab al-Mujahidin in Somalia. Paris: Sciences Po: 12 (https://spire.sciencespo.fr/notice/2441/5ncvgncesl8n4a0rpg5k3c1j5p).

50.

Marshall

Monty G

Gurr

Ted Robert

(2017) Polity IV Project: Political Regime Characteristics and Transitions, 1800–2013. Center for Systemic Peace, Data file (retrieved from http://www.systemicpeace.org/inscrdata.html).

51.

McKnight

Janet

(2014) Surrendering to the big picture: Historical and legal perspectives on accountability in the Democratic Republic of Congo following the defeat of the March 23 Movement (Practice note) (Report). Stability: International Journal of Security and Development 3(1): 1–11.

52.

Metternich

Nils W

(2012) Expecting elections: Interventions, ethnic support, and the duration of civil wars. Journal of Conflict Resolution 55(6): 909–937.

53.

Mitchell

WJT

(1994) Landscape and Power. Chicago, IL: University of Chicago Press

54.

Mueller

Jason C

(2018) The evolution of political violence: The case of Somalia’s Al-Shabaab. Terrorism and Political Violence 30(1): 116–141.

55.

Nangini

Cathy

Jas

Mainak

Fernandes

Hugo L

Muggah

Robert

(2014) Visualizing armed groups: The Democratic Republic of the Congo’s M23 in focus. Stability: International Journal of Security and Development 3(1): 51–58.

56.

Pettersson

Therése

Wallensteen

Peter

(2015) Armed Conflict, 1946–2005. Journal of Peace Research 52(4): 536–550.

57.

Pierson

Paul

(2004) Politics in Time: History, Institutions, and Social Analysis. Princeton, NJ: Princeton University Press.

58.

Raleigh

Clionadh

Bruijne

Kars De

(2017) Where rebels dare to tread: A study of conflict geography and co-option of local power in Sierra Leone. Journal of Conflict Resolution 61(6): 1230–1260.

59.

Salehyan

Idean

(2008) No shelter here: Rebel sanctuaries and international conflict. Journal of Politics 70(1): 54–66.

60.

Salehyan

Idean

(2009) Rebels without Borders: Transnational Insurgencies in World Politics. Ithaca, NY: Cornell University Press.

61.

Salehyan

Idean

Gleditsch

Kristian Skrede

Cunningham

David E

(2011) Explaining external support for insurgent groups. International Organization 65(4): 709–744.

62.

Sambanis

Nicholas

(2004) What is civil war? Conceptual and empirical complexities of an operational definition. Journal of Conflict Resolution 48(6): 814–858.

63.

Sarkees

Meredith Reid

Wayman

Frank

(2010) Resort to War: 1816–2007. Washington DC: CQ.

64.

Scott

James

(1985) Weapons of the Weak. London: Yale University Press.

65.

Scott

James

(2009) The Art of Not Being Governed. London: Yale University Press.

66.

Sobek

David

Thies

Cameron G

(2015) Civil wars and contemporary state building: Rebellion, conflict duration, and lootable resources. Civil Wars 17(1): 51–69.

67.

Solomon

Hussein

(2014) Somalia’s al-Shabaab: Clans vs. Islamist nationalism. South African Journal of International Affairs 21(3): 351–366.

68.

Staniland

Paul

(2012) States, insurgents, and wartime political orders. Perspectives on Politics 10(2): 243–264.

69.

Themnér

Lotta

(2016) UCDP Dyadic Dataset Codebook, Version 1-2016. Uppsala Conflict Data Program (UCDP) Department of Peace and Conflict Research, Uppsala University.

70.

US Department of State (2009) US Government Counterinsurgency Guide. US Government Interagency Counterinsurgency Initiative, Bureau of Political-Military Affairs, US Department of State.

71.

Vidino

Lorenzo

Pantucci

Raffaello

Kohlmann

Evan

(2010) Bringing global jihad to the Horn of Africa: al Shabaab, Western fighters, and the sacralization of the Somali conflict. African Security 3(4): 216–238.

72.

Wickham-Crowley

Timothy P

(1992) Guerrillas and Revolution in Latin America: A Comparative Study of Insurgents and Regimes Since 1956. Princeton, NJ: Princeton University Press.

73.

Wood

Reed M

(2010) Rebel capability and strategic violence against civilians. Journal of Peace Research 47(5): 601–614.

74.

World Bank (2017) World Development Indicators: GDP per capita, Atlas method, Data file (retrieved from https://data.worldbank.org/).

75.

Wucherpfennig

Julian

Metternich

Nils W

Cederman

Lars-Erik

Gleditsch

Kristian Skrede

(2012) Ethnicity, the state, and the duration of civil war. World Politics 64(1): 79–115.