Relevant Literature on Rebel Defection

The conflict literature has produced clear expectations about factors that make signatories more or less likely to defect from settlements. The common theoretical framework rests on a rationalist approach adopted from international relations, whereby violent conflict is understood to reflect the inability of rivals to credibly commit to peace (Fearon Reference Fearon1995; Wagner Reference Wagner2000; Powell Reference Powell2002, Reference Powell2006; de Figueiredo and Weingast Reference De Figueiredo, Weingast, Walter and Snyder1999; Mason, Weingarten, and Fett Reference Mason, Weingarten and Fett1999).Footnote ² Similar to the state of anarchy characterizing the international system, a vacuum of state authority generates a “security dilemma” (Jervis Reference Jervis1978; Posen Reference Posen1993), such that weak states are unable to monopolize violence within their borders. Thus, even if belligerent parties can agree to the terms of peace, their true intentions are obscured by information gaps, which heighten mutual suspicions and incentivize signatories to pre-emptively abandon the settlement before being caught out by a rival. Meanwhile, group preferences are inconsistent across time, with events occuring during the implementation process that threaten to shift the balance of power and cause one side to seek to renegotiate, often with violence (Werner Reference Werner1999). From this perspective, unless a third party is present to enforce the terms and bind parties to their commitments, the likelihood of defection is high (Walter Reference Walter2002; Fortna Reference Fortna2008).

Although commitment theory has been central to coordination in the field, the underlying model makes a number of nontrivial assumptions that limit its real-world utility. Modelling negotiations as a one-shot game between two cohesive actors is helpful to building parsimonious theories, and it may be valid in an IR context, where rival actors are states.Footnote ³ However, with research indicating that weak states provide fertile ground for insurgency (Fearon and Laitin Reference Fearon and Laitin2003), states that are unable to prevent the onset of violence may be susceptible to a proliferation of armed groups, especially those sufficiently weak to consent to negotiations. Peace processes are therefore likely to involve or exclude numerous, competing rebellions, and this has implications for bargaining strategies.

Recognizing this empirical reality, the two decades since Stephen Stedman’s (Reference Stedman1997) influential paper has witnessed a growing body of work on the topic of “spoiler problems” in multilateral conflicts (Reiter Reference Reiter2016; Findley Reference Findley2013; Ayres Reference Ayres2006; Zahar Reference Zahar and Darby2008; Hampson Reference Hampson1996). This effort has been facilitated by the release of cross-national data measuring conflicts at the level of government-rebel dyad, rather than country, making it possible to address variation in the constellation of armed groups active across space and time and to pursue research questions about, for example, the benefits and challenges of all-inclusive settlements. The UCDP/PRIO Armed Conflict dataset (Gleditsch et al. Reference Gleditsch, Wallensteen, Eriksson, Sollenberg and Strand2002; Themnér and Wallensteen Reference Themnér and Wallensteen2013), in particular, has become a cornerstone for quantitative work in the field. Building on the classic rationalist approach, this literature rests on a general consensus that spoiler behavior enhances the commitment problems underlying a peace process. Not only should it be more difficult to reach agreement on the terms where more “veto players” exist (Cunningham Reference Cunningham2006), but negotiations should also be prone to violent spoiling from excluded groups hoping to gain a seat at the table, or else undermine the bargain altogether (Kydd and Walter Reference Kydd and Walter2006; Tull and Mehler Reference Tull and Mehler2005). Thus, much of the prescriptive focus again rests on international actors’ ability to facilitate a mutually acceptable bargain and subsequently reign in incentives for spoiling during the transition period.

A second problem with the assumptions of the conventional rationalist approach is in failing to account for the internal factionalism that often characterizes a rebellion. Modelling and measuring post-settlement outcomes dichotomously—rebels either demobilize and conflict terminates, or else they defect and return to war—obscures variation in elite preferences over a settlement, not only across time or across rebel groups, but also across factions within a signatory party. In recent years, a new literature has emerged to challenge this notion of non-state groups as cohesive actors and examine the potential for rebel fragmentation or splintering (Pearlman and Cunningham Reference Pearlman and Cunningham2012; Bakke, Cunningham, and Seymour Reference Bakke, Cunningham and Seymour2012; Schlichte Reference Schlichte2009). Scholars in this field recognize that bargaining parties are embedded in what has been called a “dual contest” (Cunningham, Bakke, and Seymour Reference Cunningham, Bakke and Seymour2012) or “nested game” (Pearlman Reference Pearlman2009, citing Tsebelis Reference Tsebelis1991)—the external pursuit of the group’s common goals vis-à-vis rivals, as well as an internal struggle over private advantage. Empirically, this points to an alternative defection pathway, in which one faction of a signatory party may remain committed to the terms of a settlement and disarm—a partial demobilization—while another returns to violent rebellion (see figure 1). The implications of overlooking splintering as a form of defection should not be understated, as research shows that rebel fragmentation increases both the duration of conflict and the likelihood of recurrence (Cunningham Reference Cunningham2013; Findley and Rudloff Reference Findley and Rudloff2012; Rudloff and Findley Reference Rudloff and Findley2016; Ishiyama and Batta Reference Ishiyama and Batta2011).

Figure 1 Illustrative diagram of post-settlement outcomes

In seeking to identify the ways in which internal contests over group goals, preferences, and leadership may ultimately translate into splintering, scholars emphasize the capacity of latent defectors to access the resources of war—especially financing, weapons, and recruits. More specifically, existing research points to the presence of an external patron (Greenhill and Major Reference Greenhill and Major2007; Sawyer, Cunningham, and Reed Reference Sawyer, Cunningham and Reed2017; Tamm Reference Tamm2016) or access to resource rents (Stedman Reference Stedman1997; Weinstein Reference Weinstein2007) as key to the viability of a nascent, splintering rebellion. However, the extant literature has failed to consider whether the presence of other active rebellions may similarly affect the potential for splintering. According to Pearlman and Cunningham (Reference Cunningham, Bakke and Seymour2012, 7), “we should expect external impulses to interact with internal fragmentation to open and restrict constraints on actors’ choice[s].” Yet by assuming that armed groups behave strictly as rivals during a peace process, it is unclear whether groups with a shared incentive to spoil a settlement may directly provide opportunities for splintering factions to defect, rather than merely using violence to enhance uncertainty or demonstrate bargaining power. This is an important difference in the type of spoiling behaviour, since post-settlement trajectories that result in mergers or umbrella coalitions may appear in the data as new groups and may therefore be misclassified as cases of conflict termination. With this in mind, the next section draws on insights generated from each of these lines of inquiry to develop a theory of defection by strategic alliance.

Theory of Defection by Alliance

A broad literature exists on the determinants of factionalism during bargaining processes. While some focus on the ways in which government concessions lead to infighting (Fjelde and Nilsson Reference Fjelde and Nilsson2018; Plank Reference Plank2017), others argue that, knowing this, governments may use such offers tactically in order to divide and weaken the armed opposition—a strategy referred to elsewhere as “winning away pieces” or “dividing and conceding” (Reiter Reference Reiter2016; Cunningham Reference Cunningham2011; Zartman Reference Zartman and Zartman1995).Footnote ⁴ Others point to ethnic or organizational cohesion (Cunningham Reference Cunningham2013; Ishiyama and Batta Reference Ishiyama and Batta2011; Asal, Brown, and Dalton Reference Asal, Brown and Dalton2012), shifts in the balance of power between factions (Best and Bapat Reference Best and Bapat2018; Tamm Reference Tamm2016), or the nature of conflict-related grievances (Lounsbery and Cook Reference Lounsbery and Cook2011; Mosinger Reference Mosinger2018) as key to determining the potential for factionalism. While the origins of factional rivalries may also impact the interests and strategies of rebel elites, the aim of this study is to explore the processes through which disagreement over the terms of peace ultimately translates into defection. I argue that contested settlements create shared incentives for rebel elites inside and outside of a peace process to spoil, increasing the likelihood of an opportunistic alliance between splintering factions and excluded rebellions. The theoretical logic rests on several observable and interrelated propositions.

First, it is widely acknowledged that peace processes are likely to face opposition from elites within the rebel party. Negotiations may proceed without the full support of the movement, and opposing factions may even be successful in preventing a settlement if leaders are keen to maintain cohesion and avoid weaking the group’s bargaining position (Best and Bapat Reference Best and Bapat2018). Where leaders go on to sign a settlement, however, the finalized terms provide a clear indication to rebel elites about the likely winners and losers in the rewards exchanged for a commitment to demobilize. Although much of the spoiling literature portrays such dynamics as reflecting irreconcilable divisions between rebel hard-liners and moderates (Kydd and Walter Reference Kydd and Walter2006; Ayres Reference Ayres2006), these elite-level rifts more often rest on the fact that the spoils of peace are finite and must be selectively allocated (Plank Reference Plank2017). The rebel infighting that took place over the distribution of ministerial positions in successive transitional governments in Liberia has been well documented, for example (Lidow Reference Lidow2016; Nygard and Weintraub Reference Nygård and Weintraub2015; Bekoe Reference Bekoe2005), while Reiter (Reference Reiter2016) attributes the breakdown of a 1994 Djibouti settlement to “powerful losers” (31) in the FRUD rebellion, who went on to accept a new bargain just five years later.

At a minimum, key stakeholders in the rebel party should be willing to commit to a bargain that provides a satisfactory payoff for their efforts and offsets the risk of disarming for themselves and their followers. Elites perceiving themselves as insufficiently rewarded or—in many cases—existentially threatened by the outcome have a strong incentive to mobilize their followers against it. Although they may ideally seek to renegotiate more favourable terms, these sidelined factions are generally perceived as weak relative to the rebel contingent that signs the settlement, especially in the eyes of the government and international mediators, and they may therefore “find few strictly political means at their disposal” in the immediate term (Pearlman Reference Pearlman2009, 84; Werner Reference Werner1999). Thus, with a signed settlement serving as an indication to relevant stakeholders that the metaphorical train is leaving the station (Stedman Reference Stedman1997), a recourse to violence may become the only viable alternative for disgruntled factions to accepting a bad deal.

At the same time, latent defectors are constrained in their capacity to convince followers to return to the battlefield. In order to maximize the viability of a new rebellion, they must weigh up available opportunities for accessing the resources of war in the future—especially financing, weapons, and recruits—and it is in this context that a rival rebellion may begin to appear as an attractive ally (Christia Reference Christia2012). According to Bapat and Bond (Reference Bapat and Bond2012), committing to an alliance is costly for rebel elites, which helps to explain why they are empirically much more rare than would be expected based on shared strategic interests. The first-order preference of latent defectors should therefore be to launch a new rebellion outright, without bearing any unnecessary costs to reputation or bargaining power, such as by finding access to a new state sponsor or lootable resources (Greenhill and Major Reference Greenhill and Major2007; Sawyer, Cunningham, and Reed Reference Sawyer, Cunningham and Reed2017; Tamm Reference Tamm2016). Where no such opportunities exist, rebel leaders and international actors may find it easier to reign in potential defectors and prevent splintering, forcing disgruntled factions to accept their worse-case outcome of disarming under threat of political exclusion or personal insecurity. In such contexts, the option to join forces with an active group outside the peace process may provide the only viable alternative for latent defectors to avoid surrendering.

In sum, the theoretical logic developed in this section emphasizes the ways in which the terms of peace may shift the preference ordering of rebel faction leaders. Where settlements create shared incentives to spoil—for both disgruntled factions in the signatory party as well as rebel groups excluded from a settlement—faction leaders should be more willing to bear the costs of an alliance, especially where they lack opportunities to defect unilaterally. The next section provides an in-depth study of a failed peace process in Uganda as a clear illustration of the mechanisms underlying this theory.

Defection by Alliance in Uganda

Protracted negotiations ultimately resulted in separate bargains between the government of Yoweri Museveni and the military and political wings of the Uganda People’s Democratic Army/Movement (UPDA/M): the Pece Agreement in June 1988 and the Addis Accord in July 1990. The rebellion drops out of cross-national data after 1987 (Themnér and Wallensteen Reference Themnér and Wallensteen2013), and as a result, the peace process is classified as a successful case of conflict termination in quantitative work. However, closer inspection reveals that a small fraction of the movement effectively demobilized, and that in fact, the lion’s share of the UPDA/M’s remaining military capacity joined forces with the Lord’s Resistance Army (LRA). The research collected for this study reveals that this faction represented the most highly trained, well equipped, and organized element of the LRA at a time that coincides with its own emergence as a viable rebellion.Footnote ⁵ Thus, not only does the UPDA/M provide an ideal case for illustrating the theoretical logic outlined in the previous section, but it is of particular value in demonstrating the validity problems that arise when measuring armed groups discretely. The evidence presented here is based on approximately nine months of cumulative field research, drawing on news media and radio archives, as well as interviews with former rebel officers and government representatives conducted between January 2013 and August 2014.

Case background. The UPDA/M emerged in opposition to the government of Yoweri Museveni after his own insurgency took power in a coup in January 1986, bringing an end to Uganda’s Bush War. Northerners—having been over-represented in the preceding government and military of Milton Obote—were systematically excluded from Museveni’s administration, causing widespread fear that he planned to use his newfound position to seek retribution for state-sponsored atrocities committed during the war (Tripp Reference Tripp2010; Lindemann Reference Lindemann2010).Footnote ⁶ With more than ten thousand soldiers from the deposed national army and a political wing comprising high-level members of Obote’s former ruling party, the UPDA/M represents a uniquely formidable actor in Uganda’s long history of violent conflict. By early 1988, after barely a year of direct confrontation, Museveni had agreed to negotiate a truce.

Indications of a rift among the UPDA/M leadership were present from the outset. The initial stages of peace talks in Pece were dominated by the commander of the UPDA military wing, Brigadier Odong Latek, to the exclusion of the movement’s politicians exiling in London and Nairobi (Lamwaka Reference Lamwaka1998)—including its chairman and former prime minister Otema Allimadi. The initial proposal of terms presented by Latek in March 1988 largely neglected the UPDM’s political goals, prompting Allimadi to appear on BBC’s “Focus on Africa” to disown the peace talks and reject the ceasefire. Just ten days later, however, Latek travelled to the demobilization camp in Lukome to sign a deal.

While observers expected a final settlement to be imminent, the revised draft presented by the government contingent contained a number of unilateral amendments, including a stipulation that any officer-level appointments would be conditional on confirmation by Museveni, and only after full disarmament.Footnote ⁷ Initially, the military wing was unified in its rejection of these new terms, publicly referring to themselves as the “pillars of the movement” in a display of solidarity with Latek, who refused to sign. When it became clear that the government contingent would not budge on the terms, however, the schism among the rebel leadership became apparent. While Latek reaffirmed the military campaign, the UPDA/M Central Executive Committee (CEC) announced on May 8 that both he and Allimadi had been summarily released from their positions, and that the group’s military and political affairs would be assumed by the remaining military high command. Lieutenant Colonel John Angelo Okello was named as commander and chairman of the UPDA/M, and he avowed that the majority of the movement remained loyal to the peace process.Footnote ⁸ Okello signed the Pece Agreement on behalf of the UPDA/M on June 3, and soon after, he was co-opted into the national military along with the five other CEC officers who had voted to support the settlement and oust its opponents.

Meanwhile, in London, Allimadi appeared for a second time to denounce the peace process, referring to the government’s passage of an amnesty bill as a “partial amnesty” and a “trick” by Museveni. This time, the move was effective, as rumors began to circulate among UPDA ranks that the settlement was a ruse to convince northerners to hand over their weapons. The number of soldiers arriving at Lukome began to slow considerably. In total, just 2,200 rebel troops were effectively disarmed during the post-settlement period—no more than one-fifth of the UPDA’s total forces.Footnote ⁹ As Museveni dispatched government emissaries to renegotiate the surrender of the remaining troops in hiding along the Sudanese border, Allimadi appeared again to claim uncontested leadership and recommit the UPDA/M to violent rebellion.

From schism to alliance. After two years without progress or violence, in July 1990, Allimadi surprised observers by unilaterally announcing that he had agreed to a peace deal in Addis Ababa. The negotiations were shrouded in secrecy,Footnote ¹⁰ but a statement released via Uganda’s state-owned media suggests that the verbal agreement—known as the Addis Accord—was nothing more than a reiteration of the terms reached in Pece two years earlier, which Allimadi had vehemently rejected.Footnote ¹¹ For the remaining UPDA officers, this was an indication that the savvy politician had struck a backroom deal with Museveni, advantaging himself and his inner circle while neglecting the interests and security of the broader movement. They were especially alarmed by the stipulation that soldiers would “leave their operational bases under their respective commanders and report for disarmament,” issuing a press release in which they pointed to Allimadi’s removal by the CEC as evidence that he was not a “genuine representative” of the rebellion and that he lacked authority to negotiate on its behalf.Footnote ¹² In a symbolic gesture of the extent of their opposition, one rebel battalion launched a brief wave of violence which resulted in at least seventy deaths and culminated in the destruction of Allimadi’s personal residence.Footnote ¹³

Despite Allimadi’s frequent public appeals for demobilization,Footnote ¹⁴ no more than a handful of rebel troops reported for disarmament following the Addis Accord. At the same time, his co-optation into government left little hope for further renegotiation, as the remnants of the exiled UPDM political wing had resigned from their positions in protest—which, furthermore, eliminated any prospects for future financing from abroad. The culmination of the protracted peace process thus left the remaining military faction in a substantially weakened bargaining position by late 1990. Meanwhile, security fears were at an all-time high, largely a result of Allimadi’s previous attempts to undermine peace talks by stoking suspicions about Museveni’s intentions. Rather than surrender under such conditions, the remaining UPDA officer corps, along with at least 5,000 to 8,000 troops and their weapons, were incorporated into Joseph Kony’s LRA.

Assessing rival arguments. Traditional rationalist arguments about credibility gaps fail to sufficiently account for defection in this case, as a series of early events heightened mutual suspicions, yet ultimately fell short of derailing the peace talks. In March 1988, a highly respected UPDA officer and member of the negotiating contingent was accidentally shot and killed by government forces, while Uganda’s health minister threatened via state radio that “amnesty is expiring ... Run, run for your lives.” It is therefore insufficient to merely establish that gaps in information and credibility exist; rather, valid causal inference requires a precise understanding of the ways in which the preferences of rival factions shift over the course of bargaining in response to prevailing opportunities.

Second, owing to its longevity and brutality, the LRA has received ample media attention, which tends to portray the movement and its leader as ideological extremists and rent-seeking opportunists. This may suggest that the UPDA defectors were “hard-liners” opposed to striking a deal with Museveni’s government and seeking a similarly extremist ally to help undermine the peace process with violence (Kydd and Walter Reference Kydd and Walter2006; Stedman Reference Stedman1997). Although the LRA has become increasingly elusive in recent years, however, Kony initially expressed a willingness to join the 1988 Pece peace talks if they “bore fruit.” The LRA also went on to engage in peace negotations with Museveni’s government after the UPDA merger, eventually signing a comprehensive settlement in Juba in 2008.Footnote ¹⁵ The movement’s increasingly extremist and opportunistic position therefore appears as much a result of these failed attempts to negotiate peace as a convincing explanation for defection in this case.

Finally, much of the existing work on rebel alliances suggests that an ideological or identity-based affinity is necessary if elites are to overcome commitment costs (Gade et al. Reference Gade, Gabbay, Hafez and Kelly2019; Bapat and Bond Reference Bapat and Bond2012; Bencherif and Campana Reference Bencherif and Campana2017), and at first glance, this case appears to confirm such a position. Both the UPDA/M and LRA were primarily comprised of ethnic Acholis, and they had overlapping operational bases in the areas surrounding Gulu, Kitgum, Lira, and Soroti. In fact, prior to recruiting for his own rebellion, Kony served as an advisor to the UPDA’s Black Battalion, and there is evidence that the two groups shared information about government troop movements (Doom and Vlassenroot Reference Doom and Vlassenroot1999; Van Acker Reference Van Acker2004). Arguments focusing on shared goals or ethnic and geographic overlap fail to account for the timing of the merger, however. In the period leading up to negotiations with the government in 1988, the two groups engaged in an intense competition for advantageous bargaining position, including covert missions to appropriate each other’s weapons as well as overt battlefield clashes (Branch Reference Branch2005; Allen Reference Allen1991). Suspicions intensified when UPDA negotiators accepted a stipulation that ex-combatants would be transformed into task forces responsible for locating and defeating the LRA, with Kony accusing the UPDA/M of using the peace talks to avenge previous battlefield defeats (Okello Reference Okello2002). In this light, the willingness of defecting UPDA officers to be absorbed under Kony’s banner in late 1990—a move that would have been unthinkable in the years, and even months, prior—reflects the shift in preference ordering that occurred when the bargaining process ended. With the Addis Accord signaling that no further negotiations would take place, the neglected military wing was forced to either surrender under risky and uncertain conditions, or else accept an opportunistic alliance with an erstwhile rival. It chose the latter.

The evidence presented in this section supports a strategic perspective of rebel alliances as a pathway to splintering in the post-settlement period. It also reveals the problems of measurement error that may arise when coding armed groups discretely, as the incomplete demobilizaton of the UPDA/M is pivotal to explaining the LRA’s emergence as a viable rebellion in the early 1990s.Footnote ¹⁶ Kony had previously inherited the poorly trained and ill-equipped remnants of a third northern rebellion, the Holy Spirit Movement (HSM), which suffered a crushing battlefield defeat to Museveni’s military in 1987, causing his predecessor to flee into exile (Dunn Reference Dunn2004; Day Reference Day2011; Okello Reference Okello2002; Branch et al. Reference Branch, Mwenda, Titeca, Finnström, Schomerus, Blattman and Annan2010). With the absorption of the UPDA military wing, however, his ragtag troops were fortified and trained in “conventional military tactics and objectives.”Footnote ¹⁷ Ex-UPDA officers became Kony’s top commanders and closest advisors, and they are credited with formulating the tactic of terror against civilians that would become a hallmark of the LRA conflict (Van Acker Reference Van Acker2004). This case therefore corroborates work by Debos (Reference Debos2008), who concludes that although “these groups [may] seem harmless, they can play a renewed role in the factional game when they are able to forge a new alliance ... [and] thus re-emerge as critical actors” (232). The next section describes the data and method used to analyse the extent to which this defection pathway is generalizable beyond a single case.

Data and Method

The nature of cross-national data means that quantitative work on conflict resolution is largely precluded from measuring and testing the mechanisms on which competing theories rely. By focusing on average treatment effects across covariates, statistical techniques mask heterogeneity in pathways of rebel defection from settlements. Moreover, as illustrated in the previous section, the discrete coding of armed groups may cause cases of post-settlement splintering to be misclassified as resolved conflicts. In order to test my theory of strategic alliances, therefore, this study moves beyond the extant data, relying instead on a two-stage qualitative research design. Formally elaborated by Haggard and Kaufman (Reference Haggard and Kaufman2012), the analytical method involves a systematic coding of causal process observations (CPOs) across a large sample of cases and aggregation of the results across the full sample.

The relevant sample for the analysis is all rebel parties to negotiated settlements signed in Sub-Saharan Africa between 1975 and 2015. In identifying the universe of settlements, I consulted a wide range of appropriate sources, especially UN Peacemaker, the U.S. Institute of Peace, the University of Ulster’s Transitional Justice database, and the Conciliation Resources ACCORD collection.Footnote ¹⁸ As the analytical focus of this study is to illuminate the processes through which parties defect after having committed to peace, a negotiated settlement must result from direct negotiations between at least two belligerent parties involved in an intra-state conflict, and represent a mutually accepted and final solution in order to qualify (Johnson Reference Johnson2020). While the substantive provisions may vary considerably, these basic criteria require, at minimum, that the formally recognized leadership of a rebel party commits to demobilization, therefore excluding temporary ceasefires—which are especially prone to breakdown (Toft Reference Toft2010)—and cases in which opposing factions are successful in spoiling a settlement prior to its signing. The resulting sample includes a total of eighty-four settlements signed in twenty-two countries.

Next, I relied on the settlement text to identify rebel signatory parties.Footnote ¹⁹ Consistent with Reiter (Reference Reiter2016), the analysis is therefore not limited to major actors involved in the conflict, nor to groups coded as active in the conflict-year according to the UCDP Armed Conflict dataset, although extensive research was conducted on each case to ensure that the rebellion reached a minimum violence threshold of twenty-five annual battle deaths and remained active at the time of signing. This strategy thus excludes signatories that had already effectively demobilized and any political movements lacking a direct link to an armed group or military wing—for example, Burundi’s Frodebu political party, which signed on to the 2000 Arusha Power Sharing Accord alongside three active militias. A rebel party may enter the dataset more than once if it defected from a settlement, returned to violent activities, and went on to sign a new bargain at a later date. This identification strategy yields a sample of 159 government-and-rebel settlement dyads.

Unlike other studies relying on a CPO method of analysis to test mechanisms, I do not select on the dependent variable—in this case, rebel defection. Again, the limitations of existing data make it difficult to adjudicate between cases of conflict termination and those of partial demobilization, whereby groups splinter or merge and factions may resurface under a new name. Thus, after compiling a wide range of primary and secondary source material, I took a broad, qualitative approach to tracing each group’s trajectory over a five-year post-settlement period. Descriptive background from the UCDP Actor EncyclopediaFootnote ²⁰ was combined with Lexis Nexis news media searches, U.N. Panel of Experts and U.S. State Department reports, NGO reports, and academic case studies. The compiled qualitative evidence was then used to operationalize three CPOs that correspond with the central argument. Work by John Gerring (Reference Gerring2006, Reference Gerring, Box-Steffensmeier, Brady and Collier2008), David Collier (Reference Collier and Finifter1993, Reference Collier2011), and others (George and Bennett Reference George and Bennett2005; Przeworski and Teune Reference Przeworski and Teune1970) supports the use of causal process observations (CPOs)—or “system-specific indicators”—as a tool for systematizing qualitative methods and analysing causation across a large sample. According to Gerring (Reference Gerring2006),

An inductive reading of the UPDA/M case study points to three distinct CPOs that may occur in the period between a signed settlement (t₀) and either demobilization or defection (t₁), which are illustrated in an updated diagram of post-settlement pathways in figure 2. First, an internal schism may become apparent within the rebel party—a concept that I borrow from Plank (Reference Plank2017). Rebel elites often voice opposition when fearing their interests are neglected by the negotiation process (De Juan Reference De Juan2012, 11-12; Reiter Reference Reiter2016; Schlichte Reference Schlichte2009), and in many cases they are successful in amending the terms. In light of the theoretical logic proposed here, what matters is whether a faction of the signatory party remains opposed to the final settlement, or the terms ultimately accepted by the group’s recognized leadership. Schism is theoretically and analytically distinct from splintering, whereby rebellions fragment into discrete groups with at least one new group returning to conflict. As such, it is also likely to be much more empirically common. Of 159 rebel parties in the sample, seventy exhibit schism over the settlement terms.

Figure 2 Updated diagram of post-settlement pathways for rebel parties including the proposed CPOs

A second CPO identifies the presence of an excluded rebellion—an armed group outside the peace process with both incentive and capacity to spoil. A study by Nilsson (Reference Nilsson2008) suggests that the presence of such groups should be irrelevant to the prospects for dyadic peace, as signatories have likely accounted for the behaviour of excluded spoilers prior to signing a settlement. If excluded rebellions spoil by directly facilitating splintering, however, as my argument holds, then this finding may be biased due to reliance on the UCDP Armed Conflict dataset, in which such pathways tend to appear as resolved conflicts.

The third and final CPO measures whether a merger takes place with an excluded group, which enables a dissenting faction of the signatory party to return to rebellion. Rebel alliances vary in organizational form according to groups’ preferences and capacities, with existing work suggesting that most are informal and relatively short-lived (Akcinaroglu Reference Akcinaroglu2012; Christia Reference Christia2012). Although elites may prefer to avoid the costs of combining forces, these tacit or informal pacts can easily transform into bitter rivalries (Fjelde and Nilsson Reference Fjelde and Nilsson2012), and groups may “spend as much time fighting one another as contesting the government” (Cunningham, Gleditsch, and Salehyan Reference Cunningham, Gleditsch and Salehyan2013, 519). Since the strategic logic rests on rebel elites’ willingness to bear costs in order to maximize the viability of their return to rebellion, the identification creteria requires a formal unification of forces.

All 159 government-rebel dyads were coded dichotomously across the three CPOs defined here. The complete results of this coding exercise are provided in table 1, with a corresponding list of rebel group acronyms in the online appendix.Footnote ²¹ Since “evidence for one link in the chain has no bearing on the next (or previous) link” (Gerring Reference Gerring, Box-Steffensmeier, Brady and Collier2008, 174), the three observations are treated as analytically discrete rather than causally deterministic. By allowing for deviations at each step in the CPO pathway, this strategy generates a series of nested research designs, allowing for alternative pathways to defection, as well as the possibility that the three CPOs are present even in cases of demobilization—ample opportunity to disconfirm the central theory. To this end, the second stage of the analysis entails aggregating the data across the full sample in order to analyse evidence for the observable implications of the theory.

Table 1 CPO coding for all rebel parties to Sub-Saharan African settlements (1975–2015), clustered by settlement.

Discussion of Findings

The aggregated findings summarized in table 2 corroborate a general scholarly consensus that settlement is an unstable outcome of conflict, as compared to military victory (Toft Reference Toft2010; Collier, Hoeffler, and Söderbom Reference Collier, Hoeffler and Söderbom2008; Licklider Reference Licklider1995), as less than one-third of African rebel parties—just forty-six of 159 cases (28.9%)—fully demobilized in the post-settlement period. Defection appears much more common, with conflict continuing among more than half of all settlement dyads: 101 cases, or 63.5%. The remaining twelve cases (7.6%) fail to conclusively conform with any of the post-settlement pathways identified in figure 2. Consistent with Haggard and Kaufman (Reference Haggard and Kaufman2012), any CPO analysis across a large sample is likely to face difficulties categorizing every case conclusively, and due to the scarcity or imprecision of available observable evidence, this challenge is compounded when researching conflict zones. These cases are labelled as Ambiguous in table 2 and excluded from the subsequent discussion of results.

Table 2 Aggregated results of the large-N CPO analysis

Notes:

^a Ambiguity most often results from a deficiency of observable data needed to confirm a group’s trajectory during the five-year post-settlement period. Examples include the FLEC-R in Angola (2006), LDF and LPC in Liberia, and HCUA in Mali (2013). Where uncertainty precludes coding decisions, cases are not listed in table 1. In Rwanda (1993) and Chad (2002), the government party initially defected back to the battlefield.

^b Numbered defection pathways correspond with those labelled in figure 2.

Of the 101 cases of rebel defection in the sample, less than half (46.5%) presented as a clear case of Recurrence (see table 2), whereby the signatory party returned to rebellion from the top-down. The common, dichotomous approach to conflict outcomes in existing work means that arguments about defection have overwhelmingly focused on such cases of buyer’s remorse, in which rebel leaders misjudge their ability to either control the implementation process to their advantage or win post-conflict elections, leading them to mobilize a wholesale defection when conditions begin to appear unfavourable during the post-settlement period. However, fifty-four cases of defection do not fit neatly into the Recurrence category, as although one or more factions of the signatory party returned to war, a sizeable proportion of the movement may have remained committed to the settlement and demobilized. In total, approximately one-third of cases in the sample (33.9%) and more than half of all defections (53.5%) resulted in Splintering (table 2)—a proportion that is consistent with cross-national findings on the frequency of rebel fragmentation in civil wars (Rudloff and Findley Reference Rudloff and Findley2016).Footnote ²²

As a test of the theory elaborated in the third section, the analysis provides strong support for several observable implications. First and foremost, of the fifty-four cases that exhibited both schism and the presence of an excluded rebellion (see table 1), a merger was observed in thirty-six cases (66.7%), while full demobilization resulted in just two such instances (3.7%). Thus, where the outside alliance option is available to opposing factions of a signatory party, it seems highly unlikely that committed stakeholders will be able to reign in opportunities for defection. This post-settlement pathway—labelled as pathway (5) in figure 2 and table 2—accounts for more than one-third of all cases of defection in the sample (35.6%), suggesting that the overlooked link between Uganda’s UPDA/M and LRA is a much more generalizeable phenomenon than has been recognized to date. Similar processes were observed in conflicts as varied as Mali, Sudan, Liberia, and Somalia, among others.

Moreover, in a small subset of cases exhibiting both a schism and an excluded rebellion, Splintering was enabled by a temporary, informal rebel alliance during the post-settlement period—pathway (4) in figure 2—rather than a formal merger. This is further evidence that groups finding themselves excluded from negotiations may be willing to coordinate assaults or share information, weapons, and even troops in order to spoil a settlement by directly enabling a splintering faction’s return to the fray, even where rebel elites do not formally commit to join forces. At the same time, just five cases in the sample correspond with this alternative pathway (4), accounting for 5% of defections and 9.3% of the subset of rebel Splintering (table 2). In comparison, the empirical frequency with which Splintering occurs via merger lends strong support to one of my central theoretical propositions, demonstrating that the signing of a settlement makes defecting elites more willing to bear the costly trade-off of a formal alliance in order to maximize the viability of future rebellion.

Where a schism results in Splintering without the help of an excluded rebellion, the qualitative evidence confirms that defecting factions typically enjoyed access to alternative sources of rents, such as external patrons or lootable resources, as is consistent with the proposed theory. As an example, after the Ogaden National Liberation Front (ONLF) was forced to accept the terms of Ethiopia’s 1991 Transitional Charter due to the ousting of its backer, Somalia’s Siad Barre, a splintering faction calling itself the Ogaden Liberation Army (ONLA) found its own sponsor in the newly independent government of Eritrea. Although far less common than the route to Splintering via alliance—since such opportunities are not universally available—the thirteen cases following pathway (3) (see figure 2 and table 2) corroborate existing scholarship which stresses the importance of external opportunity structures in determining the potential for a viable return to rebellion.Footnote ²³

Finally, in weighing the costs and benefits of an alliance, there is ample evidence to suggest that the strategic goal of maximizing military capacity often overrides constraints related to identity or ideology. Challenging conclusions derived from formal models of contestation or studies of alliances in a single case, a large number of mergers in the sample occurred between groups viewing themselves as bitter rivals at the time, even those that had recently met on the battlefield. Factions of the Sudan Liberation Army/Movement (SLM/A) and the Justice and Equality Movement (JEM) attacked each other multiple times before joining forces in 2007 as the United Resistance Front (URF)Footnote ²⁴ in a unified rejection of the Darfur Peace Agreement signed the previous year. Simililarly, in CAR, militias fighting on opposing sides of the 2013 Séléka/anti-Balaka conflict joined forces two years after signing the Libreville Political Accord. The aggregated data therefore supports a strategic perspective in line with Christia (Reference Christia2012), who argues that alliances are primarily driven not by ethnicity or shared demands, but by the prevailing opportunity structure and considerations about future viability.

In sum, according to the results of the large-N qualitative analysis, the post-settlement trajectories of most rebel parties in Africa fail to correspond with the dichotomous operationalization of conflict termination or recurrence that is common in the field. In fact, rebel splintering is a more empirically common outcome than either demobilization or recurrence, and the vast majority of such cases of splintering in the sample occured through a formal alliance with an excluded rebellion, lending strong support to my central argument. Where contested settlements create a schism in the signatory party, latent defectors seek to mobilize defection, looking to alternative sources of revenue as their first-order preference to maximize viability. Where such opportunities are not forthcoming, however, disgruntled elites may be forced to either accept the worst-case scenario of disarming under adverse terms, or more likely, where excluded groups share the incentive to spoil, to bear the cost of an alliance with an erstwhile rival.

Concluding Remarks

Recognizing that “the norm in more recent civil conflicts is not coherent antagonists as much as shifting coalitions of groups with malleable allegiances” (Pearlman and Cunningham Reference Pearlman and Cunningham2012), much has been made of the need for more research that challenges the notion of bargaining parties as binary and discrete. I contribute to this effort in three important ways. First, while much of the existing work on splintering focuses on the processes through which schisms emerge during bargaining, this study reveals an overlooked pathway through which opposing factions can re-emerge as viable challenges to the state. In so doing, the focus shifts away from dynamics of factionalism, which are likely to vary from case to case, towards a broader puzzle about why multiparty conflicts are especially difficult to resolve. The logic behind the theory of defection-by-strategic alliance points to some precise mechanisms that may help to explain findings in previous studies, for example, that rebel fragmentation is correlated with conflicts of longer duration (Findley and Rudloff Reference Findley and Rudloff2012; Cunningham Reference Cunningham2013).

Second, by explicitly allowing for unit heterogeneity across cases of defection, the analysis reveals the frequency of post-settlement pathways that may be censored or misclassified due to the discrete coding of armed groups in cross-national data. More than one-third of rebel parties in the sample follow a trajectory which diverges from the dichotomous conflict outcomes of termination or recurrence, resulting instead in splintering. Far beyond the Ugandan case explored in detail here, the results indicate that strategic mergers are highly prevalent in the larger sample, and the fact that such outcomes result in the identification of “new” groups may explain why quantitative work has found no direct link between the behaviour of excluded spoilers and the potential for defection among signatories (Nilsson Reference Nilsson2008; Plank Reference Plank2017). In contrast, the results of the present analysis indicate that rival groups factor directly into the strategic calculations of potential defectors, by providing disgruntled elites the opportunity to avoid their worst-case outcome of surrendering without acceptable guarantees.

In this way, these findings also promise to provide policymakers with a better understanding of the unique challenges of bargaining where conflicts are fractionalized. If settlements are prone to fail where they create shared incentives for key stakeholders to spoil, then mediators should aim to eliminate the outside option for defection by promoting all-inclusive negotiations, rather than merely attempting to reign in spoilers ex post. Reports on attempts to broker peace in newly independent South Sudan by focusing efforts on the primary challenge to President Kiir’s government—to the exclusion of an estimated two dozen other militias—reveal the ways in which politically expedient mediation strategies may ultimately prove counter-productive.Footnote ²⁵

Some important questions remain. For one thing, although there are practical and methodological justifications for relying on a regional sample—especially in order to maximize cross-case comparability in the absence of statistical controls, as well as capitalizing on the researcher’s regional knowledge in operationalizing key concepts (Collier Reference Collier2011; Gerring Reference Gerring, Box-Steffensmeier, Brady and Collier2008)—the sampling specification makes it impossible to infer whether the results are generalizable beyond Sub-Saharan Africa. Rebel fluidity may be more prevalent in weakly institutionalized or more ethnically diverse contexts, for example. On the other hand, Bapat and Bond (Reference Bapat and Bond2012) find the likelihood of rebel alliance onset to be unchanged with the addition of regional dummies, suggesting that Africa is not an outlier in this regard. Out of sample defection pathways confirming the central argument have been identified in Afghanistan and the Philippines. This highlights the need for further refinement of global, cross-national data in order to better account for the fragmentation and merger of groups over time.

Given the theoretical and analytical focus on the processes through which contested settlements break down, as previously mentioned, the initial drivers of internal contestation are beyond the scope of this study. A growing literature exists to identify factors that may affect the potential for a schism to emerge, such as the nature of concessions offered by the government (Fjelde and Nilsson Reference Fjelde and Nilsson2018; Plank Reference Plank2017), the degree of organizational cohesion (Cunningham Reference Cunningham2013; Ishiyama and Batta Reference Ishiyama and Batta2011) or social linkages (Mosinger Reference Mosinger2018), and internal shifts in the balance of power (Best and Bapat Reference Best and Bapat2018; Tamm Reference Tamm2016). Since these variables may also affect the preferences and strategies of elites who ultimately defect from a settlement, theory-building efforts should attempt to extend the model to earlier nodes of decision-making, for example, by defining the different “types” of rebel parties that enter into negotiations, the likelihood that contending factions will accept or reject various payoffs offered by the government, and the incentives of faction leaders to seek out particular groups as potential allies. This is a worthy avenue for future research.