Ethiopia’s Long War in Historical Perspective

“Cohesive Stalemates” and the Logic of Insurgent Solidarity: Ethiopia’s Long War in Historical Perspective This paper uses original quantitative and qualitative data from Ethiopia’s civil wars to develop and test a theory of rebel fragmentation. I argue that organizational performance affects the emergence of factional infighting in rebel organizations in unique and novel ways. While territorial losses increase the likelihood of fragmentation, so do territorial gains. The implication is that battlefield stalemates possess unique properties that promote organizational cohesion and provide the only sustainable basis for cooperation in war. I term such contexts “cohesive stalemates.” The paper pushes the limits of an emerging literature on the internal politics of insurgent groups by demonstrating how the character and coherence of rebel organizations is highly sensitive to the ebb and flow of war. In the first few months of 2009, Somalia’s primary anti-government insurgent group, Al Shabaab, was locked in a war of attrition with the UN backed, Transitional Federal Government of Somalia (TFG). It had been a long road for Al Shabaab, but the tide seemed to be turning in its favor. Scarcely three years before, in the wake of a debilitating Ethiopian offensive, Al Shabaab’s fortunes looked bleak; yet now, with the Ethiopians gone, and the TFG occupying no more than few city blocks of the capital Mogadishu, the insurgents had reason to be hopeful. Yet 2009 did not end on a high note for Al Shabaab. In late December, a dispute between two Mogadishu based officials developed into a full-fledged organizational split that threatened to jeopardize Al Shabaab’s hard fought gains.2 Though the organization’s Islamist ideology – with its emphasis on religious solidarity and devotion to cause – was designed to inoculate it from the sectarianism that plagued so many of Somalia’s militant groups in the past, Al Shabaab was proving to be no exception. So what happened to Al Shabaab in the winter of 2009? What accounts for the outbreak of factional infighting and fragmentation in an ostensibly cohesive rebel group? Organized rebellion is a dangerous undertaking where death can be the price of failure. With such steep penalties, rebels have strong incentives to maintain organizational solidarity and preserve a united front, for if they cannot get along together, they may go down apart. Yet despite the incentives to cooperate, they often cannot: in Sri Lanka, Afghanistan, Liberia and Sudan, in conflicts large and small, rebels have waged war against external adversaries with much success, only to unravel from within. Although rebel fragmentation is a ubiquitous feature of civil wars3 – both past and present – and is essential to understanding the duration (Cunningham 2006), intensity (Pearlman 2008), and outcomes of contemporary civil conflicts, there have been few systematic, theoretically driven analyses of this phenomenon.4 While historical and anthropological literatures on particular rebel organizations provide substantive insights, political science has relatively little to say.5 Of course, the paucity of research is not surprising, as the questions we ask are often driven by the data available. 2 See http://allafrica.com/stories/200912210004.html An unpublished data-set I have put together of over 170 African rebel organizations – presented at the Woodrow Wilson Center in Washington DC in August 2008 - reveals that 33 percent of post-colonial African rebel organizations have yielded independent, fully operational splinter groups 4 (Christia 2008) and (Mesquita 2008) are notable exceptions 5 (Kalyvas, The Ontology of "Political Violence": Action and Identity in Civil Wars 2003) makes a similar point, although framed somewhat differently. (Blattman and Miguel 2010) are more explicit about these omissions in the social science literature on civil war. (Johnson 2003) account of the Sudanese People’s Liberation Army is one historical account that delves deeply into the issue of rebel fragmentation 3 In the face of critical security concerns rebel organizations have powerful incentives to feign organizational unity, withhold information, or misrepresent the facts. Even long after conflicts have concluded data is rarely reliable, as popular narratives, and the reputational concerns of the combatants themselves, weigh heavily on the way past events are remembered and retold. This paper offers a modest corrective. It seeks to explain the occurrence of fragmentation within rebel organizations by offering a clear, falsifiable theory with explicit causal mechanisms. It then tests that theory through an analysis of original statistical data on Ethiopia’s civil wars, and a more comprehensive longitudinal case study of one of the major protagonists in Ethiopia’s civil conflicts — the Eritrean Liberation Front (1960-1982) – based on interviews with ex-combatants, colorcoded district level maps, and extensive archival research. I conclude with a careful discussion of alternative explanations. The evidence suggests that organizational performance structures intra- organizational cooperation, its maintenance, and its collapse. While significant battlefield losses – measured as a function of territory – encourage fragmentation by creating a series of pernicious commitment problems, gains drive fragmentation by exacerbating preference divergence. Somewhat counter-intuitively, stalemate or sustained periods of particularly slow marginal battlefield change, promote intra-organizational cohesion and solidarity. I dub these favorable conditions for intra-organizational rebel cohesion “cohesive stalemates.” By describing how three very different military contexts – gains, losses, and stalemate – create different incentives for cooperation in war, and identifying the unique political implications of military stalemate, the paper pushes the limits of an emerging literature on the internal politics of organized rebellion. At a broader level, the paper provides a rare window into how the normal processes of warfare affect the solidarity and commitment that is crucial to the trajectory of modern civil conflicts. Organizations, Coalitions, and Elites The framework I employ begins with the assumption that rebel organizations are amalgamations of identifiable groups that possess distinct interests. To a greater or lesser degree, they are coalitions united by the pursuit of a common goal – the violent contestation of state power – rather than an identical set of preferences.6 At the apex of each of these groups – respective coalition members if you will – are a class of elites who command the loyalty of rank and file sub-group members. Though the source of their authority differs, these elites engender allegiances that provide a capacity to mobilize those around them. Similar to what Charles Tilly’s has called “political entrepreneurs,” who “specialize in activation, connection, coordination, and representation” of particular intra-organizational constituencies, these individuals are the focal points of collective action, and thus cooperation in war hinges on their behavior (Tilly, The Politics of Collective 2003).7 6 This view of rebel organizations as internally complex and heterogeneous has increasingly come into vogue in both academic and policy circles, and draws its intellectual inspiration from behavioral theories of the firm first developed by economists in the early 1960’s. For an example of this work, see (Cyert and March 1963) 7 Of course, careful readers will note that the focus on organizational elites seems arbitrary and assumes more than it explains: surely, the motives and behavior of rank and file combatants are in many ways unique, and distinct from, the organizational elites who lead them. Can we assume away a unity of thought and action between elite and rank and file combatants? Robert Michels (Michels 1968), the well-known theorist of political organization, was the first to underline the most persuasive grounds for this assumption. To Michels, almost all political and social organizations converge on “oligarchy,” where a select few dominate decision making processes. The reason for this state of affairs, Michel contended, is because in large political organizations members must delegate decision making authority to elites in the interests of efficiency: Since mass participation in decision making would be unwieldy and chaotic, “oligarchy” is the only way to manage an organization’s affairs. In the context of rebel organizations, the implication of Michel’s argument is clear: rebels cede authority to sub-group elites who make strategic decisions on their behalf. Focusing on elite behavior within rebel organizations is warranted, the logic goes, because no organization – particularly military organization – can function without The fragmentation of a rebel organization logically represents the rupture of an existing coalition, or in game theoretic terms, the breakdown of a cooperative equilibrium where an equilibrium is defined as “any state of affairs which no actor has an individual incentive to disturb” (Reeve 2003). Whereas at t+1 the constituent units of a rebel organization are able to adhere to the cooperative status quo, at t +2 they cannot, since it is no longer in the interests of one or more of the constituent units to do so. The obvious question that follows from these representations – and the most important to this paper – is why and how the conditions that sustained the cooperative equilibrium within a rebel organization in one period, no longer obtain in the next. Without a theory of these shifting “conditions,” one cannot construct a theory of the breakdown of a cooperative equilibrium, and hence, fragmentation. There are several implications that follow from the recognition that explaining rebel fragmentation requires a careful exposition of changes in the conditions that underpin the cooperative equilibrium within a rebel coalition. First, path dependent arguments that focus on the initial conditions that rebel organizations confront in explaining behavior downstream are ill suited to investigating the collapse of the cooperative equilibrium within a rebel organization. While path dependent arguments would be useful in understanding differences in the propensity of certain kinds of rebel groups to fragment, they cannot explain variation in fragmentation across time within groups since the causal variables they identify do not change temporally. 8 Second, the reverse is also true. While the framework adopted can explain temporal variations in the fragmentation of rebel organizations – that is, why these organizations fragment when they do – it is the full delegation of decision making authority to a subset of leaders. Thus in war, the politics of coalition construction and destruction is an elite level game in which rank and file combatants are marginal players. 8 For the best example of this sort of work in the civil war literature see (Reno 2007) and (Weinstein, Inside Rebellion: The Politics of Insurgent Violence 2007) not particularly well-suited to explaining significant cross-sectional variation in the fragmentation of rebel organizations. Thus, I make no claims in this paper about why some kinds of rebel organizations, on average, fragment more than others. Third, the model proposed is partial in that framing fragmentation as a breakdown of a cooperative equilibrium places analytic primacy on the “motives” and “incentives “ to fragment, rather than the “capacity.” This is an important point. While rebel organizations can be viewed as a coalition of groups that voluntarily contract into a cooperative arrangement for the production of a joint good, they are “organizations” nonetheless; they possess hierarchies and the ability to sanction members who seek to disrupt the cooperative equilibrium. Thus, the fragmentation of rebel organizations is not simply about how the incentives to cooperate may change, but about the failure of power, threats, and coercion to enforce the organizational status quo. In other words, a complete theory of fragmentation would not only focus on the demand to fragment, but also its supply. Gains, Losses, and the Shifting Incentives to Cooperate So what exactly drives the breakdown of cooperative relations within rebel organizations? Theories of punctuated equilibrium posit that environmental shocks are the central cause of dramatic organizational and institutional change.9 Such shocks produce “revolutionary” moments where the structures and processes of organizations, hitherto static, go into rapid and dramatic flux. This paper picks up on this general theme, by focusing on how military shocks create contexts conducive to factional infighting, fragmentation, and the rapid re-ordering of rebel groups. 9 See (Gersick 1991), for a good review and application of this literature to organizations. The literature on punctuated equilibrium is vast. For a well-known application in American politics, see (Baumgartner and Jones 1993) My core claim is that organizational performance affects the cohesion of rebel organization in novel and unique ways. More specifically, I argue that ceteris paribus, rebel organizations are most susceptible to fragmentation at two critical moments: the first, when they have incurred significant losses, and the second when they have achieved significant gains. The implication is that periods of military stalemate, or particularly slow, marginal battlefield change, tend to promote organizational cohesion and underpin organizational stability. The language of “gain” and “loss” is not without import, for the argument proposes that what triggers cooperation breakdown is not so much the military position a rebel organization occupies in absolute terms – say near victory or failure – but rather the change in military position relative to some recent reference point or identifiable status quo. This is a critical if trying to predict exactly where in the evolution of a conflict rebel organizations fall apart. Gains and the Origins of Preference Divergence So what are the causal mechanisms linking situations of gain and loss, to the dependent variable of interest, fragmentation? My theory suggests the mechanisms differ depending on whether a gain, or loss, has been incurred. In order to understand the basis of the argument we need to address the critical issue of actor preferences. My starting point is the work of Realists in the field of international relations, who argue that states have a hierarchy of preferences and seek to satisfy their core preference above, and sometimes at the expense, of all others. Arthur Stein (Stein 1990) typifies the insights of this perspective when he suggests that states have lexical preferences where actors “maximize in sequence rather than make trade-offs…and thus choose an option that maximizes the main objective regardless of how it does on secondary ones.”10 For Realists and the states they theorize about, this “main objective” is survival (Waltz 1979) (Mearsheimer 2001). Given the nature of the international system where “self-help” and “anarchy” are organizing principles, survival concerns must be paramount since failure to guarantee this objective can be fatal. For our purposes here, Realism’s key intuition is this: in insecure environments, where existential threats loom large, actors will strive to secure immediate survival above all else. The intuition of IR Realists about hierarchical preferences is echoed in comparative politics. James Scott, in writing about the dynamics of peasant rebellion in South East Asia, notes that the political economy of peasant life is characterized by pervasive uncertainty about crop yields (Scott 1976). The precarious nature of food supply, in turn, leads the peasantry to a “security first” outlook in which they spurn risky innovation that could lead to large improvements in aggregate yields in favor of lower yields that guarantee survival. As with states in the Realist version of international politics, environments characterized by systemic insecurity, produce a behavioral logic wherein the management of security risk becomes an objective that supersedes all others. Somewhat fortuitously, the idea of hierarchical preferences is the implicit basis of much of Realist theory about why alliances form and why they fall apart. Since survival is at the top of the hierarchy of objectives, states “balance against threats,” and thus “the internal cohesion of an alliance depends to a large extent on the intensity and duration of the threat” (Walt 1987). With a coalition’s raison d’etre gone – guaranteeing security in an insecure environment – it becomes difficult for that coalition to cohere. Yet how is this relevant to understanding fragmentation within rebel organizations? My sense is that ideas of hierarchical preferences and uncertainty about survival are useful in thinking about the conditions that sustain cooperation in a rebel organization. Since civil wars are anarchic environments where self-help is the rule, it may be the case that the constituent units of a rebel organization – and by extension the leaders of these distinct units – are in a position analogous to that of states in the international system. What this means, if we follow the comparison to its logical conclusion, is that when rebel organizations face insecurity and uncertainty about short term, immediate survival, its constituent units have incentives to ensure they maintain a united front. Since the primary objective of members of rebel coalition is survival, the constituent units will put their divergent preferences on a whole range of issues aside to face the common challenge. Of course, wars can, and often do, turn quite rapidly. As rebel organizations incur military gains, often quite rapidly through a set of major shocks, the conditions that sustained cooperation begin to shift. Surging militarily, rebels no longer perceive grave uncertainty about immediate survival. As a result, the hierarchy of objectives within a rebel organization is re-ordered. With immediate survival guaranteed in the minds of rebels – a basic objective that everybody in the rebel coalition could unite around – other issues about which rebels have heterogeneous preferences increase in importance and dominate the issue arena. Under such circumstances, the incentives to cooperate have been significantly altered. No longer pre-occupied with the common concern for survival, constituent units of a rebel organization begin to take actions that undermine cooperation in ways that heretofore they had not— approaching potentially divisive issues with greater vigor and intensity, bargaining harder, and even resorting to coercion. Ceteris paribus, consensus gives way to division, and the cooperative equilibrium collapses. In this scenario, the fragmentation of a rebel organization is one of preference divergence. Yet, critics may ask, preference divergence over what? The model is agnostic over what sort issues drive preference divergence precisely because the issues in contention vary across groups and within groups over time. As the comparative analysis will show, by dint of history and circumstance, the lines of cleavage within rebel organizations are often different; ideology, ethnicity, religion, personality conflicts, can all be raw material that determine the anatomy of a particular rebel organization. Yet the underlying issues over which preference divergence emerge are less important than the recognition that this divergence happens at certain moments and not others. Losses and the Origins of Commitment Problems The notion that actors within rebel organizations attempt to maximize outcomes on the core objective of survival, has implications for fragmentation in situations quite different from those in which rebel organizations have achieved significant military gains. However, the mechanism here is one of credible commitment rather than preference divergence. What do we mean by credible commitment? Problems of credibility tend to emerge in situations in which one or more parties to a formal or informal agreement cannot reliably commit to performing the obligations that agreement entails in the future. According to Shepsle (1991), commitments can be credible in two possible ways. First, commitments can be imperatively credible if the participants are unable to do otherwise because either they will be coerced to do so by an external authority or lack the means and resources to ignore obligations. Second, and more important for the argument here, commitments can be motivationally credible if the participants expect to receive sufficient rewards for them to honor the commitment at the time of performance – thus the commitment is incentive compatible and self-enforcing. The constituent units of a rebel organization contract into an organizational arrangement on the understanding that in the production of violence for the pursuit of broad political objectives, they are maximizing on the core objective of survival in an insecure and uncertain environment. Thus at their core, rebel organizations are institutions that manage risk. Yet cooperation of this sort also entails possible costs. While rebel organizations and the coalitions that underpin them provide collective security, when they fail the future survival of its members can be imperiled. Simply put, rebel organizations must be credible in demonstrating that they have a reasonable chance of surviving into the future. Seen in this light, the consequences of significant military losses for rebel organizations are obvious. With expectations about the future substantially altered, the constituent units of a rebel organization calculate that continuing to cooperate under the status quo will likely lead to their individual demise in the next period. As a consequence they seek to challenge the cooperative status quo, with fragmentation the frequent outcome. Since a rebel organization can collectively no longer credibly commit to providing basic benefits it was contracted to provide – survival – the risks of continued cooperation become prohibitively high. Such an argument does not necessarily mean that those who challenge the status quo do better than if they continued to observe the cooperative equilibrium. In a world of perfect information, they would always do better. However, in the context of war in which incomplete information about capabilities is widespread – what Karl von Clauswitz famously called the “fog of war” (Clauswitz 1968) – what rebels believe to be the best course of action in preserving their survival and managing risk, may not be. “Cohesive Stalemates” and the Origins of Cooperation The implication of the arguments offered here, is that there is a static, stalemated middle ground where rebel organizations tend to cohere. Yet what exactly differentiates, at the micro-level, the stalemated military context from situations of gains and loss I have described? How does it affect the behavior of rebels? The idea here, is that stalemated contexts are uniquely positioned to provide the basis for cooperation because it allows two conditions to be true: 1) organization members perceive that there is significant insecurity and are uncertain about survival and 2) organization members perceive that there is a reasonable chance that the organization can continue to provide the benefits it was contracted to provide if the constituent units of a rebel organization abide by the cooperative status quo. Since survival is at the top of the hierarchy of objectives, cooperation in such contexts makes perfect sense: rebels can realize the obvious gains of cooperation, with the recognition that this cooperation will not imperil their survival in the future. Cooperation is thus incentive compatible and selfenforcing. Such contexts can be referred to as “cohesive stalemates.” The obvious implication of such arguments is that there is a connection between the maintenance of cooperation in war and the origins of cooperation in war. Indeed, it would seem to make sense that the conditions that sustain a cooperative equilibrium will be similar to the conditions under which it originates. As has already been suggested, rebel organizations and the coalitions that sustain them emerge because they allow potential organization members to maximize on the core objective of survival. Why? Applying the tenets of organizational economics, civil war theorists have recognized the value of organization in capturing the benefits of economies of scale and reducing transaction costs (Weinstein, Inside Rebellion: The Politics of Insurgent Violence 2007). My contention here is that rebels with heterogeneous preferences tend to coalesce under a hierarchical organizational framework when they are uncertain about future survival, but consider the new organizational framework to have a reasonable chance of surviving into the future. While the early evolution of a rebel organization does not resemble the situation of stalemate I have described, such contexts evoke similar perceptions about the uncertainty of immediate survival, but the plausibility that if the coalition stays united it can survive. In fact, many observers of civil wars note that insurgents, in the early days of war, have the strategic advantage because they are first-movers; they can pick targets at their choosing, without having to deal with a military that is sufficiently deployed and prepared (Kalyvas, The Logic of Violence in Civil War 2006). Therefore insurgents feel as though there is a legitimate chance at success. At the same time, having never truly encountered the long arm of the state in its most serious and violent manifestations, there is uncertainty about the true military potential of the organization and ambiguity about the immediate survival of its members.11 Such an environment would seem to be fertile ground for cooperation as it engenders similar incentives to periods of military stalemate. 11 This point is echoed by Che Guevara in his discussion of early “guerilla strategy”: “At the outset, the essential task of the guerilla fighter is to keep himself from being destroyed” (Guevara 1961, 15) Diagram 1: Outline of Argument and Causal Mechanisms Stalemate Cooperation Organizational Formation (cooperation) Commitment Problems = Factionalism/ Fragmentation Preference Divergence = Factionalism/ Fragmentation Gains, Losses and Absolute Values One should note that the theory I have outlined is fundamentally a theory about perceptions and expectations— in particular, rebel’s perceptions of military gains and losses, and their expectations about the future of the organization. When rebels incur losses they believe that the organization cannot survive, and when they achieve gains, they conjecture that problems of acute insecurity have been resolved. But it is important to note at a theory of perceptions and expectations that hinges on the language of gains and losses implies that what shapes rebel’s behavior is change relative to some reference point or status quo, rather than where a rebel organization sits in absolute terms. In other words, in explaining the timing of fragmentation it matters less whether a rebel organization is close to winning or losing, but rather, whether it is incurring a gain or loss at a given time. Yet does this make sense? Shouldn’t the absolute position of a rebel organization affect perceptions and expectations about the nature of external threats and the sustainability of the organizational status quo? My response is simple, but also counter-intuitive. Military stalemates – no matter where an organization may sit along the continuum of victory to defeat – convey very specific information to the constituent units of a rebel organization. On the one hand, the ability to maintain the military status quo reflects organizational viability both now and in the future, and thus prevents the emergence of commitment problems; yet on the other hand, the inability to make positive progress suggests that the organization still faces very real collective threats, thus minimizing preference divergence. Simply put, stalemated military contexts have unique properties, and provide the conditions that make cooperation durable, irrespective of an organization’s absolute or overarching military position. This intuition is consistent with prospect theory, in which a key principal – as reflected in the work of Kahneman and Tversky (Khaneman and Tversky 1979) and later echoed by political scientists like Jervis (Jervis, The Implications of Prospect Theory for Human Nature and Values 2004) and McDermott (McDermott 2004) – is that an actor’s absolute position is not as important in their calculation of utility and welfare as a recent change in position. Furthermore, it would seem to make sense that gains and losses convey important information about the future prospects of rebel organizations, because military change indicates the direction an organization is trending. Since in and of itself, absolute military position says little about trends, rebels look to the direction of military change in forecasting and making crucial decisions about cooperation. Thus, a rebel organization that is close to defeat, but stalemated, will be less likely to fragment than an organization whose overall position is better but is incurring a significant loss. Similarly, a rebel organization that is on the verge of victory, but stalemated, will be less likely to fragment than an organization that is doing moderately well but has experienced significant recent gains. 12 Having laid out my argument, the central hypotheses of the theory of gains and losses are as follows: H1: Gains and losses will be associated with fragmentation H2: Stalemate will be associated with organizational cohesion Data and Measurement The dependent variable this project seeks to explore is rebel fragmentation, defined as an “event where a segment of a rebel organization, formally and collectively exits that rebel organization and either a) establishes a new rebel organization, b) joins an existing rebel organization, or c) joins the incumbent government.” I view rebel fragmentation as a clearly identifiable manifestation of a breakdown of cooperation within a rebel organization, and thus a category of factional events that include other phenomenon such as coups, extra-judicial killings, extra-judicial purges/arrests, and insubordination. These events are characterized by their extra-legal nature, and sit outside the bounds of the normal, political processes of a rebel organization. One should note that the theory of cooperation breakdown enumerated in the preceding pages cannot explain why, given the desire to alter the cooperative status quo, actors choose to fragment instead of launch a coup. Or why, for instance, dissatisfaction within rebel ranks is articulated through fragmentation instead of through more limited forms of insubordination. As breakdowns of 12 The importance of relative change in battlefield position vs. the absolute battlefield position is reflected in Killcullen’s (2010) discussion of the “win/loss ratio” as a metric in counter-insurgency warfare. According to Killcullen, the win-loss ratio tells us little about performance in war since rebels usually lose most engagements. In Killcullen’s view, what is more important is relative change in win-loss ratio the cooperative equilibrium of a rebel organization, it is my contention that these factional events have the same sources and can be explained within the same framework. The case material illustrates this point, by showing how in cases of fragmentation, other factional events are tightly interwoven. In any case, it is important to note that a second level of explanation would be required to explain why, given the desire to alter the cooperative status quo we get the outcome of fragmentation rather than other forms of factionalism.13 Since fragmentation can sometimes, but not always, be an extended process, I code fragmentation as occurring in the year in which that process begins. By “begins,” I mean the first observable manifestation of factionalism that evolves into fragmentation. In many cases, cooperation breakdown occurs in such a way that fragmentation is automatic instead of the outgrowth of a drawn out process of factional infighting. The independent variable of interest – a tripartite measure of gains, losses, and stalemate – is measured as a function of shifts in “territorial control” or “operational reach,” with the former the relevant metric in cases where an organization fights a conventional war and the latter the relevant metric in contexts where guerilla warfare is the primary tactic.14 The variable is measured at the conclusion of each year in which a rebel organization effectively operates, and evaluates the ebb and flow of war by placing a rebel organization into one of the three aforementioned categories.15 In coding along these dimensions, I investigate whether in a given year, a rebel organization has seen a significant contraction or expansion of the territory they control or are able to effectively operate within. By “significant,” I mean a gain or loss in territory that involves an expansion or contraction 13 Indeed, other sections of the of broader project are devoted to outlining and testing this “second level of explanation” 14 “Operational reach,” for the purposes of this discussion, can be defined as territory in which a rebel organization demonstrates the ability to consistently launch attacks. 15 The reader may note that there is a noticeable similarity between the three substantive military contexts I describe and Mao’s “three stages of guerilla warfare”- there may be connections I have not completely thought through yet, but the similarity is entirely accidental in the aggregate number of districts a rebel organization controls or is able to effectively operate within.16 In cases where expansion into a district only leads to partial control, these are also coded as gains. Furthermore, since the expansion or contraction into large cities or towns is also a “significant” territorial exchange, they are also coded as gains or losses.17 Territorial exchanges that do not meet this threshold are excluded and coded as stalemate.18 Figure 1 shows the annual variation on the key independent and dependent variables over time for the case I examine in this paper— the Eritrean Liberation Front (from this point on, I refer to the ELF by its colloquial name, ‘Jebha’).19 One can see that for Jebha, fragmentation never occurs in stalemate years. 16 In the subsequent analysis of the former Ethiopian province of Eritrea, there are 54 districts or “woredas” (see appendix for list of districts). The remainder of Ethiopia is divided into 550 districts 17 In the subsequent analysis this would exclude all population centers below 1200 individuals. However, in practical terms, there is a high level of co-linearity between district level exchange and the exchange of towns and cities 18 In coding territorial shifts in the Ethiopian context, I use a range of available sources: archival material (American and British consular reports in particular), news reports, and interviews. Admittedly, evidence can be spotty in certain districts and during particular historical periods. 19 Jebha is the Arabic word for “front” Figure 1: Jebha: Gains, Losses, and the Onset of Fragmentation Over Time Year 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 Gain/Loss/ Stalemate Onset of Fragmentation Stalemate No Stalemate No Gain No Stalemate No Stalemate No Gain No Loss Yes*** Loss Yes* Gain No Gain No Stalemate No Stalemate No Stalemate No Gain No Gain Yes * Gain Yes* Gain No Loss No Loss No Stalemate No Stalemate No Loss Yes*** *asterisk is used to denote the number of groups that split To give an indication of the broader relationship between independent and dependent variables across Ethiopia’s civil wars, I collected annual data on fragmentation and territorial gains/stalemate/ losses of all 14 rebel participants in Ethiopia’s civil wars.20 For each rebel year, of which there are 223 total years, I code whether a rebel organization a) has experienced fragmentation, and b) whether it has experienced a significant territorial gain or loss, and then cross-tabulate these 20 Organizations identified using UCDP’s Non-State Actor Dataset variables to examine the extent to which fragmentation years and gain/loss years co-vary.21 Figure 2 displays this data in the form of a two by two matrix, with the rebel year as the unit of analyses. One can see that for a total twenty-seven fragmentation years, only three are also years in which there was territorial stalemate. Gains and losses would seem to be a necessary, but not sufficient condition for rebel fragmentation. Figure 2: Ethiopia’s 14 Rebel Organizations, 1960-2008: Cross Tab of Gains/Losses and Fragmentation (rebel year as unit of analysis, 223 rebel years total) Stalemate Gain/Loss Fragment No 120 76 Fragment Yes 3 24 Yet this correlation raises another issue: since fragmentation itself is key source of losses in war, how do we know that there is not an endogenous feedback effect driving our results? It may be the case that an initial case of fragmentation spurs a chain of loss that locks an organization into a cycle of fragmentation. Given the possible endogenous relationship between our independent and dependent variables, there is no reason to think that an organization’s fragmentation at T1 and T5 are independent one another. In order to account for this sort of endogeneity across rebel years, I analyze the data statistically using a random effects logistic regression model with rebel year as the unit of analysis. 21 When an organization no longer retains any operational capacity in the country, it is dropped from the analysis