Introduction

   The guilty perished, but now there were only the guilty to survive.

   All is uniform, though extraordinary; all is monotonous, though horrible.

I.I. FOUR PUZZLES

On the hills that rise gently off the Argolid plain, in the Greek peninsula of the Peloponnese, lie the twin villages of Manesi and Gerbesi (now Midea). Located on the eastern edge of the Plain of Argos, just a few miles from the famous archaeological sites of Tiryns, Mycenae, and Argos, these villages share a social, economic, political, and cultural outlook. In the 1940s this featured a conservative, ethnically and religiously homogeneous population of mainly Albanian descent working on small family farms of roughly equal size and practicing primarily subsistence agriculture. The inhabitants of these two villages developed common reciprocity networks and intermarried frequently; indeed, they share many family names. During the German occupation of Greece, they faced similar choices and challenges: many men from the two villages joined resistance organizations and both villages suffered German reprisals. There is, however, one crucial divergence in their histories. In August 1944 a vicious massacre of five village families, including elderly people and young children, took place in Gerbesi; armed guerillas perpetrated the actual killing, but neighbors and even relatives of the victims took part in the planning. In contrast, neighboring Manesi escaped violence of this kind. Although the same guerrillas came to Manesi looking for victims, they were successfully thwarted by the villagers. Why? How were the people of Gerbesi able to inflict such violence on their neighbors? And how were the people of Manesi, similar, it seems, in every observable aspect to those of Gerbesi, able to prevent this violence?

   Microhistorical and anthropological accounts of civil wars are replete with such village dyads, in which violence strikes in a pattern that seems to defy logic. The village of “Los Olivos” in southern Spain lost thirty-eight men to the civil war, all in 1936; they were, for the most part, socialist sympathizers who died not on the battlefield but at the hands of the right-wing Falangists, who found them after receiving information from their neighbors. Nearby “Los Marines,” however, a village with a similar political and social outlook, experienced no killings (Collier 1987:163). No other area of the Colombian department of Tolima was more devastated by the civil war known as La Violencia than the municipio of Rovira, while Dolores, a similar municipio in the same department, as mountainous and politically divided as Rovira, escaped the violence (Henderson 1985:144–5). Guatemala suffered enormous levels of violence during the early 1980s; hundreds of villages were wiped out and thousands of people killed in massacres perpetrated by the army. Yet, the anthropologist Kay Warren (1998:92) found, to her surprise, that the town of San Andrés, her field site, somehow escaped the massacres that occurred in other similar towns – the same puzzle that another anthropologist, John Watanabe (1992:182), encountered during his own field research: “Despite the army occupation, almost no one died in Chimbal, in contrast to all the towns around them.…Whether by collective disposition, acts of personal courage, or even divine intervention, Chimbal survived.” Indeed, Linda Green (1995:114) observes that “one of the notable features of the military campaign known as ‘scorched earth’ [in Guatemala] is that neighboring villages fared quite differently: one might be destroyed while another was left untouched.” Jonathan Spencer (2000) was similarly surprised to discover that the village he studied in Sri Lanka had “miraculously” escaped the worst violence even though many surrounding villages did not. The ethnic Albanian village of Bukos in Kosovo suffered Serb violence, but not its equally Albanian neighbor Novo Selo (Gall 1999); likewise, the Chechen village of Primykaniye bore the brunt of Russian violence but not the neighboring Tsentora-Yurt (Gordon 1999b). An IRA cadre in Northern Ireland (Collins 1999:98) recalled that “the IRA had already obliterated every hotel in Newry, and though Warrenpoint was a ten-minute drive from Newry, it could have been in another country, so untroubled was it by the war going on around it.” The variation in violence has been detected by several students of the conflict in Northern Ireland (Smyth and Fay 2000:133; O’Leary and McGarry 1993:9).¹

   This variation has also mystified scholars: “Why this should have been,” writes Watanabe (1992:x) about the relative absence of violence from his Guatemalan field site, “evades any simple answer.” The very existence of variation has been cited as evidence, at worst, of the sheer impossibility of making sense of violence (Kann 2000:401) and, at best, of the inability to move beyond educated guesses.² Yet, despite the obvious significance of the matter, there have been few attempts to move toward a systematic explanation of variation in violence – an oversight that has puzzled more than one scholar (e.g., Klinkhammer 1997:29; Getty and Manning 1993:17).³

   This leads to a second, more general puzzle: the oft-noted and seemingly enduring brutality of civil war. In 1589 Alberico Gentili observed that the “chief incentive to cruelty [in war] is rebellion” (in Parker 1994:44). Montaigne (Essais 2:23) argued that “a foreign war is a distemper much less harsh than a civil war,” while Adam Smith (1982:155) pointed out that “the animosity of hostile factions, whether civil or ecclesiastical, is often still more furious than that of hostile nations; and their conduct towards one another is often still more atrocious.” Why are civil wars so violent–or perceived as such?

   A third puzzle is this: almost every macrohistorical account of civil war points to the importance of preexisting popular allegiances for the war’s outcome, yet almost every microhistorical account points to a host of endogenous mechanisms, whereby allegiances and identities tend to result from the war or are radically transformed by it. Consider Lynn Horton’s findings about the dynamics of allegiance during the Nicaraguan Civil War, from her research in the municipality of Quilalí. She provides plenty of evidence about how political allegiance and geography were linked and shows how the latter tended to shape the former: the contras relied on the weakness of the Sandinista state apparatus in an outlying region to generate popular collaboration with them. First, the contras began to harass Sandinista sympathizers, forcing them to abandon their farms and seek refuge in the town of Quilalí. As a result, some peasants decided to distance themselves from the Sandinista organizations and projects. As a peasant put it, “If you behaved well, you wouldn’t have problems [with the contras].” Another peasant stated, “Here we lived very close to those people [the contras]. Maybe inside we felt something else, but we could never externalize it. The Frente Sandinista abandoned us.” Horton suggests that had the Sandinistas managed to maintain a greater military and political presence in the region, the political allegiance of most peasants would have been different. In contrast, in the town of Quilalí where Sandinista rule remained effective, “civic political dissent on the part of the anti-Sandinistas was muted during the war years.” She argues that this silence reflected practical expediency rather than political preference. In fact, when the war ended, “an anti-Sandinista backlash poured over Quilalí as many residents expressed grievances they had been unwilling or afraid to openly articulate in the 1980s.”⁴

   Spain under French Napoleonic occupation (1808–14) offers an additional illustration of this puzzle. Lawrence Tone (1994:57) found that the guerrilla war against Napoleon in Spain did not take place in the strongholds of the 1808 summer revolution (towns such as Madrid and Valencia), but in the backward and isolated region of Navarre, one of the quietest provinces of Spain during that revolution. Using detailed data on tax collection and insurgent participation in 116 towns and villages of Navarre, he reports a positive correlation between geographical location (proximity to towns and major roads) and efficacy in taxation by the French, and an inverse correlation between the efficacy of taxation (which was likely to produce considerable grievance) and popular participation in the insurgency. Contrary to what one would expect, where the French were able to tax, and therefore aggrieve, the population, they did not face an insurgency. In contrast, the more remote the region, the more successful was guerrilla taxation and the more likely were men to join the insurgency. Indeed, insurgents came predominantly from the Montaña: “They lived in small villages and towns, which the French could not steadily occupy.” Tone leaves little doubt as to the direction of causation. Hard-hit town dwellers stayed put despite the exactions, he argues, because “the constant French presence in such places made it difficult for young men to join the insurgency.” Even the towns that managed to contribute some volunteers to the insurgency never became sites of armed guerrilla struggle. In a similar fashion, the clergy in Galicia collaborated with the French in the cities but favored the resistance in the countryside.⁵ These examples suggest that, contrary to widespread perception, allegiances may be endogenous to the war, and that military control of a locale may result in popular collaboration.

   Related to the question of the origins of allegiances is a final puzzle, the oft-noted presence of a disjunction between the macrolevel causes of the war and the microlevel patterns of violence. Consider Palestine in the late 1930s, where a rebellion against the British, known as thawra (revolt), has been described as a nationalist insurrection of the Palestinians against British colonialism. In his superb study, Ted Swedenburg (1995) found that the rebel military structure often reflected rather than transcended existing divisions among Palestinians. Because guerrilla bands were based on families or clans, their mobilization triggered all sorts of divisions into new disputes, turning the rebellion against the British into a civil war among the Palestinians. Competing village groups tried to exploit rival rebel factions for their own purposes, each group occasionally denouncing a member of the opposing family group as a spy in order to incite the rebel chieftain with whom it was aligned to punish that group. In the course of these disputes, a significant number of Palestinians ended up collaborating with the British and fighting against their ethnic kin. Rather than being based on ideological or programmatic concerns, this collaboration was motivated by “strictly local, family reasons,” including revenge. In the narratives Swedenburg collected, the British were often seen as incidental to the whole story, mere “tools” for settling local feuds; those of his informants who had collaborated with the British described themselves as having manipulated and even outsmarted their supposed principals. In short, the prevailing description of this conflict based on a key cleavage (British versus Palestinian) and a central political issue (nationalism) is partially misleading as to the motivations and identities of many participants and the dynamics of the violence.

   In a different formulation, the habitually cited causes of group division (e.g., ideological, social, or ethnic polarization) often fail to account for the actual dynamics of violence: the game of record is not the game on the ground. Consider again the Argolid in southern Greece, which was a remarkably homogeneous place, lacking deep cleavages. Yet it went through a savage civil war that caused the death of close to 2 percent of the rural population. Why would a place lacking all conditions that supposedly cause civil strife experience such a tragedy? This simultaneous absence of deep divisions and presence of mass violence force us to rethink approaches that trace mass violence to such divisions and ask whether violence really is the direct result of deep divisions, even when and where such divisions exist.

I.2. GOALS

This book is simultaneously conceptual and positive, theoretical and empirical. It is difficult to understate the importance of a clear conceptualization of what remains a highly confused set of issues. Emile Durkheim (1938:14–22) noted that because thought and reflection are prior to science, physical and social phenomena are represented and understood by crudely formed “lay” concepts – notiones vulgares or praenotiones, as Francis Bacon called them. These concepts, Durkheim pointed out, are freely employed with great assurance, as if they corresponded to things well known and precisely defined, whereas they awaken in us “nothing but confused ideas, a tangle of vague impressions, prejudices, and emotions. We ridicule today the strange polemics built up by the doctors of the Middle Ages upon the basis of their concepts of cold, warm, humid, dry, etc.; and we do not realize that we continue to apply the same method to that very order of phenomena which, because of its extreme complexity, admits of it less than any other.” Indeed, when it comes to “political violence,” currently fashionable terms of practice tend to impose themselves as terms of analysis (Brubaker and Laitin 1998). The emancipation from “empirical categories, which from long continued habit have become tyrannical” (Durkheim 1938:32), requires a clear specification of key conceptual categories and the scope conditions of the argument – an eminently theoretical enterprise.

   Civil war is defined as armed combat within the boundaries of a recognized sovereign entity between parties subject to a common authority at the outset of the hostilities. Within civil war, my focus is on violence committed intentionally against noncombatants. This sort of violence is a phenomenon that has long remained off research limits because of its conceptual complexity and empirical opacity. To use Antoine De Baecque’s (2002:851) felicitous words, my goal is to bring reason to circumstances when reason is pushed to its limits. From a methodological point of view, I show the importance of systematic research at the microlevel. Typically, microlevel evidence tends to be marginalized as irrelevant or too messy. It is commonplace among historians that the “local” must be integrated with the “global” (e.g., Pred 1990:15), yet efforts to do so rarely venture beyond the boundaries of the case study. Here, I show a possible way of achieving this integration.

   I begin with a simplified and abstract characterization of violence in civil war, yet one that stands on well-specified conceptual foundations. I analytically decouple civil war violence from civil war. I show that despite its many different forms and the various goals to which it is harnessed across time and place, violence in civil war often displays some critical recurring elements. Rather than just posit this point, I coherently reconceptualize observations that surface in tens of descriptive accounts and demonstrate that seemingly random anecdotes tend to be facets of the same phenomenon. The positive component of the book consists of two parts: a theory of irregular war and a microfoundational theory of violence (with two strands: indiscriminate and selective). Unlike existing work, the theory stresses the joint character of civil war violence, entailing an interaction between actors at the central and local levels, and between combatants and noncombatants. This interaction is informed by the demands of irregular war, the logic of asymmetric information, and the local dynamics of rivalries. Hence the theory differs from existing accounts of violence that stress exclusively macrolevel motivations and dynamics, pinpoint overarching and preexisting cleavage structures, and characterize violence as “wanton,” “indiscriminate,” or “optimal” from the users’ point of view.

   From the theory, I specify a model of selective violence that is consistent with the theoretical characterization, in which the interaction between actors operating at different levels results in the production of violence in a systematic and predictable way. This exercise yields counterintuitive empirical predictions about the spatial variation of violence at the microlevel, which I subject to an empirical test using data I collected in Greece. The empirical test confirms the explanatory power of the theory in a limited setting, whereas evidence from a wide array of civil wars suggests broader plausibility. Of course, the general validity of the theory awaits further empirical testing.

   Finally, I explore two implications of the theory, looking first at mechanisms of “intimate” violence and then at how the modalities of violence identified can help inform our understanding of cleavage formation – that is, how and to what degree national-level or “master" cleavages map onto local-level divisions.

   On the whole, this book diverges from studies that approach violence in a normative way (Sorel 1921) or via interpretation or hermeneutical reflection (e.g., Sofsky 1998; Keane 1996; Héritier 1996; Arendt 1973; 1970; Friedrich 1972). It also diverges from studies that rely on inductive data analysis (e.g., Harff 2003), do not venture beyond the macrolevel (e.g., Valentino, Huth, and Balch-Lindsay 2004), or rely solely on secondary accounts (e.g., Downes 2004; Valentino 2004) and one actor only, be it the state or the rebels (e.g., J. Weinstein 2003).

   This book is a first step. More and better data can be collected to allow broader empirical tests. The theory can be further refined and expanded. Civil wars and their violence are highly complex phenomena that can be tackled only by sustained research. In this book, I restrict my focus to coercive homicidal violence in irregular civil wars. The focus on a specific type of violence acts as a baseline: the goal is to see how much can be explained given the restrictions imposed. It turns out that the theory does quite well and generates implications for noncoercive violent practices, for violence that stops short of homicide (e.g., arrest, torture, displacement), and for other types of civil wars. Still, more research is needed to graft onto the theory those aspects not yet incorporated into it.

   Reflecting about civil wars began simultaneously with the writing of history, yet only recently have we been able to use the tools of social science in our investigations. This book will have achieved its goal if it succeeds in establishing a baseline that inspires an ongoing research program.

I.3. ROAD MAP

Wars, and their violence, display enormous variation – both across and within countries and time. The form and intensity of violence used at different points in the conflict by the Reds and Whites during the Russian Civil War, by diverse Serb, Muslim, and Croat factions in Bosnia, or by competing groups in Liberia vary significantly. Ernesto “Che” Guevara (1998:75–6) summarized this variation with cogency: “The enemies of the people act in a more or less intensely criminal fashion according to the specific social, historic, and economic circumstances of each place. There are places where the flight of a man into the guerrilla zone, leaving his family and his house, does not provoke any great reaction. There are other places where this is enough to provoke the burning or seizure of his belongings, and still others where the flight will bring death to all members of his family.”

   Consider Northern Ireland. Although the British authorities have committed human rights abuses including torture, they “have not ruthlessly and brutally suppressed the population which explicitly or tacitly supports insurrection in the manner experienced by Algerian Muslims, Afghan peasants, Iraqi Kurds, Kashmiri Muslims, Palestinian Muslims and Christians, South African blacks, Sri Lankan Tamils, and Vietnamese peasants” (O’Leary & McGarry 1993:19). As an IRA man was told after his arrest by the security forces, “If this was Beirut we would just take you out into that yard and shoot you” (Collins 1999:188). At the same time, the IRA has “sought to avoid any operations which had obviously sectarian overtones: a policeman could be justified as a legitimate target, his non-combatant Protestant family could not” (Collins 1999:295). In short, there has been considerable reciprocal restraint in Northern Ireland (Toolis 1997:21), unlike in many other civil conflicts.

   The sources of this variation are highly complex. Carl von Clausewitz (1976:609–10) remarked that the conduct of war is determined by the nature of societies, as well as “by their times and prevailing conditions.” The same is true of civil wars, whose violence appears bewilderingly complex and polysemic (Apter 1997; Nordstrom 1997). The cross-national variation in levels, types, and practices of violence across wars may be affected by factors that include the specific profile of political actors and their political ideology (J. L. Anderson 2004; Heer and Naumann 2000; Degregori 1998; Bartov 1992; Furet 1981:51);⁶ their organizational structure, underlying social basis, and military culture (Gumz 2001; T. Anderson 1999; Livanios 1999; Mazower 1992); their resources ( J. Weinstein 2003); their national and local leadership and strategies (Shepherd 2002; Schulte 2000); the type of challenges they face and the assistance they receive from third parties; the prevailing international norms (Ron 2000a); the level of available military technology; and factors such as geography and climate. Furthermore, these factors may converge to produce distinct endogenous dynamics, as violence spirals and rival actors often mimic each other. Isabel Hull (2004:1–2) summarizes the sheer complexity of the issue by pointing to several determinants of violence in war: “The length of war, the sheer number of belligerent nations, the technical stalemate caused by the strength of defensive weaponry, scientific and industrial capacity (which created more, and more lethal, weapons), ideologization (making it hard to end the war and easy to vilify the enemy), bad leadership (‘donkeys leading heroes’), and the escalatory force of broad public identification with the war (which meant that the soldiers kept coming and civilians pressed for victory despite increasing sacrifice). Many of these factors have a reciprocal effect; they strengthen each other as they interact over time.”

   The same variation can be observed with respect to the cultural idioms through which violence is expressed. Political actors draw from a limitless variety of cultural repertoires and models of violence ( J.-C. Martin 1998; Richards 1996; Zulaika 1988). Imagination runs wild, and the possibilities seem endless. Thucydides notes that during the civil war in Corcyra “there was death in every shape and form” and “every form of wickedness arose” (History of the Peloponnesian War 3.81, 3.83). Pedro Altamirano, a Nicaraguan rebel chieftain in the 1920s, derived much of his notoriety from his frequent use of the “vest cut,” in which “victims were decapitated, their arms cut off at the shoulders, and their abdomens sliced open, the corpses thus resembling a waistcoat or vest” (Schroeder 2000:40). insurgents in Sierra Leone resorted to mutilation, Algerian insurgents to throat slicing, Guatemalan soldiers to defacing and mutilating corpses, Filipino militiamen to beheading and “vampirelike bloodletting,” Confederate rebels in Missouri to scalping, and so on. The emergence of a vast literature devoted to the detailed documentation of this variation is, therefore, not surprising.

   Given current theoretical and empirical limitations, specifying and testing cross-national models of violence remains challenging and, perhaps, premature. Yet, these limitations do not warrant abandoning the task of understanding the dynamics of violence in a systematic way, as sometimes suggested (e.g., David 1997:575).

   An alternative route is a deductive strategy aiming at producing testable hypotheses about empirical variation. This strategy can be traced back to Thucydides’ effort to specify a general model of civil war instead of cataloging endless variations (Price 2001:12–14). Although civil wars and their violence vary extensively, core observations from observers, practitioners, and scholars often pinpoint recurring elements, suggesting an underlying logic.⁷ The American journalist Peter Arnett, who covered civil wars in Vietnam and Afghanistan, told a Soviet colleague that “traveling around Afghanistan, I always remember the Vietnam War.…I covered Vietnam for ten years, and the analogies with Afghanistan were obvious” (Borovik 1991:67). Anthropologists have pointed to “incredible” cross-cultural similarities in practices of political violence (Sluka 2000:9; Zur 1994:13; Nordstrom 1992:262). As one of them points out, “the same basic meaning to violent acts and images is likely to be imputed by people who in other respects subscribe to very different cultural ideas” (Riches 1986:25). Nordstrom (1997:89), who researched several civil war sites, found that, despite pronounced local variation, “themes of terror and hope – however different their manifestation from locale to locale – demonstrate similarities that allow understandings across time and space, village and culture.” The presence of an underlying logic has not escaped practitioners either. Although Che Guevara (1998:75–6) pointed to a wide variation in practices of violence, he hastened to add that “the general methods of repression are always the same” – a judgment shared by several British counterinsurgents, such as General George Erskine, who brought to Kenya methods used in Palestine (D. Anderson 2005:200), and Julian Paget, who recalls: “In 1965 I found myself in Aden in a staff appointment directly concerned with the planning of measures, both civil and military, to be taken to defeat the insurgents then operating in those parts. The problems that arose were remarkably diverse and complex, but they were seldom completely new; they had almost all cropped up before in some previous Emergency, such as Palestine, Kenya, Cyprus, or Malaya, and it would have been most helpful to be able to study this past experience and learn from it” (Paget 1967:11).

   It makes sense, therefore, to take seriously Eugene Walter’s (1969:vii) claim that, although violence “emerges in unique contexts and, in each case, is expressed and understood in a local idiom, conforms to specific values, and serves the needs of a particular power system, it is a universal process formed by recurrent elements and organized in systems with regular structural features.” In the same vein, and more generally, this book subscribes to the view that there exists “a deep structure to human conflict that is masked by observable cultural variation” (Gould 2003:101). The challenge is to specify this “deep structure” in terms that are general enough to accommodate the appropriate analysis without falling into the trap of maximal extension and conceptual stretching (Sartori 1970).

   Toward this end, I rely on two strategies. First, I match abstract theoretical conjectures and highly specific illustrations from a great variety of empirical contexts to demonstrate the plausibility of these conjectures. Throughout the book, there is a constant dialogue between empirics and theory. I draw from the very best of fine-grained analysis of particular cases to suggest that, while contexts may differ, mechanisms recur. It is often forgotten that theory building should be grounded in credible intuitions, and examples from a broad comparative canvas serve to demonstrate the credibility of each and every building block in this theoretical edifice – though, obviously not the theory’s validity, which can only be derived from rigorous testing. Second, I adopt a strategy of disaggregation. I specify three levels of analysis, moving from macro to micro. The first level focuses on interactions between unitary (state and nonstate) political actors; the second level deals with the interaction between political actors and the populations they rule; and the third level concentrates on interactions within small groups and among individuals. Most research on political violence, rebellions, revolutions, and civil wars tends either to conflate these three levels or to focus on just one – usually the first.

   The first (or macro) level is the realm of elites, ideologies, and grand politics, where research in history, historical sociology, military strategy, area studies, comparative politics, and international relations is primarily located. Violence is usually merely a sideshow in these studies since it is seen as a natural outcome of war. Macrolevel studies share a key element: they all assume unitary actors. Elites and populations are fused and amalgamated. For example, references to either the Kosovo Liberation Army or “ethnic Albanians” in the context of the Kosovo conflict usually apply to an entity that indiscriminately includes the various factions of ethnic Albanian elites, ethnic Albanian fighters, and the entire ethnic Albanian population. The assumption is that elites determine automatically and unilaterally the course of group actions and that groups are monolithic and behave as such. This shortcut is perhaps necessary when narrating the story of a specific civil war or reflecting on the grande durée but is problematic when developing a theory of violence.

   Positing coherent, identifiable political groups with clear preferences fails to match the vast complexity, fluidity, and ambiguity one encounters on the ground. The insight that political actors at the top and individuals at the bottom cannot always be lumped together has been provided by applied military research (including insurgency and counterinsurgency perspectives), as well as by microsociological and anthropological studies of civil wars. In short, violence is often used to police groups internally and to achieve the desired (but seldom reached) total “overlap” between specific leaders and organizations, on the one hand, and underlying populations, on the other. The assumption of an unlimited and unwavering support of the population for the political actor who claims to represent it is at odds with the stark and widespread reality of forced recruitment in civil wars: these wars are often fought by conscript armies (including, in the most extreme cases, kidnapped children); desertion from these armies can be pervasive. This is true of grand “classic” civil wars (such as the American, Russian, Spanish, or Chinese civil wars), ethnic civil wars (such as the civil war in Sri Lanka), and minor rural insurgencies (such as the Kachin insurgency in Burma) (Argenti-Pillen 2003; Tucker 2001; Werth 1998; Ranzato 1994). Clearly, the relation between political actors and underlying populations must be problematized rather than assumed away – which defines the content of the second (or meso) level. This analysis requires a theoretical and empirical account of civil wars as processes, including their conduct qua wars – a difficult enterprise insofar as basic information on the war-fighting aspects of many civil wars remains scarce (Harkavy and Neuman 2001). Indeed, the study of military operations belongs to a policy-oriented literature that has remained outside the scope of mainstream social science and is primarily concerned with “practical” tasks (e.g., how to defeat insurgencies). As a result, the dynamics of civil wars, though generally understood, have seldom been the object of analytical examination by social scientists.⁸