The commander’s dilemma

Abstract

This article proposes a framework for understanding variation in armed groups’ abilities to control wartime violence, including violence against civilians. I argue that patterns (both levels and forms) of violence are shaped by armed group leaders’ attempts to meet two conflicting imperatives. To succeed, commanders must build a fighting force capable of swift, unhesitating violence; they must also maintain some control over the level, form(s), and targeting of violence. I refer to this situation as the Commander’s Dilemma. Drawing on literatures from psychology and sociology, I argue that effective behavioral control cannot be achieved via extrinsic incentives (i.e. pecuniary or non-pecuniary rewards and punishments) alone. Rather, effective control of combatant violence depends upon armed group institutions intended to align combatants’ preferences with those of commanders. I therefore focus analytically on political education, the armed group institution most likely to operate in this way. In particular, I hypothesize that armed groups with strong and consistent institutions for political education should display, on average, narrower repertoires of violence than those without. This argument finds preliminary support in a cross-national analysis of reported rape by rebel forces, as well as a qualitative investigation of armed groups during civil war in El Salvador. More broadly, this approach suggests that the creation of restraint is at least as important to our understandings of wartime violence as the production of violence.

Keywords

armed group institutions armed groups conflict-related sexual violence violence against civilians

Introduction

During the Peruvian civil war (1980–90), Sendero Luminoso (Shining Path) cadre committed nearly half of all killings (Ball et al., 2003). Yet, in most regions, Shining Path cadre committed relatively little rape (Weinstein, 2007). During the war in Vietnam, many US Army units avoided civilian casualties. Others engaged in ongoing, near-random campaigns of lethal violence against civilians (e.g. ‘Tiger Force’, cf. Sallah & Weiss, 2006). Still others, such as those who committed mass killing, rape, and destruction in the hamlet of My Lai 4, committed outbursts of several types of extreme violence (Bilton & Sim, 1992). During civil war in Sierra Leone (1991–2002), the Revolutionary United Front (RUF) practiced broad and brutal repertoires of violence, including mutilation and widespread rape. By contrast, Civilian Defense Forces (CDF) committed significantly lower levels of (reported) violence against civilians, and rape did not play a major role in CDF repertoires of violence (Conibere et al., 2004).

As these examples suggest, both levels and repertoires of violence against civilians vary considerably across time, space, armed groups, and conflicts.¹ These variations cannot be fully explained using existing theories of violence. Here, I introduce a model that views restraint, rather than violence, as the theoretical and empirical puzzle to be explained. I refer to this framework as the ‘Commander’s Dilemma’, because it emerges from armed group leaders’ efforts to meet two potentially contradictory needs: commanders must both create large groups of combatants who unhesitatingly employ violence and maintain some control over the violence that fighters wield. Both military training and combat experiences contribute to a durable increase in the average predisposition to violence among combatants. At the same time, military success (typically) relies on the ability to select specific forms and quantities of violence, and to deploy them against specific enemies, in specific times and locations. Thus, the Commander’s Dilemma focuses analytical attention on armed groups’ efforts to control combatants’ use of violence.

I argue that, following military training and combat, average combatant ‘preferences’ for many types of violence exceed those of commanders, and that therefore some mechanism for behavioral control is required.² Armed groups may employ any of several institutions to control combatant behavior (including combatant violence). These include recruitment processes, military training, political training, formal and informal socialization experiences, and disciplinary regimes (rules, rewards, and punishments). Drawing on findings from social and organizational psychology, I argue that extrinsic incentives will typically be insufficient to ensure restrained behavior. Thus, a critical characteristic of an armed group is the extent to which, in addition to providing extrinsic incentives, the group builds institutions to create intrinsic incentives for behaviors that match commander preferences.

The article proceeds as follows: I first consider existing theoretical approaches to violence against civilians during armed conflict. Next, I describe mechanisms by which armed group institutions and combat experiences contribute to increased combatant propensities toward violence, focusing particularly on the body of research findings demonstrating that increased propensities to violence are long-lasting and general. I then consider the array of armed group institutions that mediate between commander and combatant, particularly the extent to which each type of institution creates intrinsic and/or extrinsic incentives for (commanders’) preferred behaviors, focusing on the impact of institutions for political education. Quantitative and qualitative evidence follow. Finally, I consider some limitations of this theory, and look ahead to further research.

From theories of violence to a theory of restraint

Scholars have developed a number of theories of violence, including violence against civilians, in armed conflict. Among the most prominent are theories that employ rationalist models of violence against noncombatants (e.g. Kalyvas, 1999, 2006; Valentino, 2013). Scholars working in this tradition argue against a view of violence as ‘wanton and senseless’ (Kalyvas, 1999), instead modeling violence against non-combatants as an outcome of strategic decisionmaking. Results from this research program suggest that many armed group decisions about violence against civilians are broadly rational, given the constraints of considerable information asymmetry and/or extremely limited tactical options (e.g. Wood, 2010; Balcells, 2011; Fjelde & Hultman, 2013).

Yet the assumptions underlying such approaches are empirically unsatisfying. First, many (see review in Ottmann, 2015) treat armed groups as unitary, whereas in reality violence against civilians is frequently committed without, or even in defiance of, explicit orders from above (e.g. Manekin, 2013). Second, rationalist approaches frequently assume that unitary groups will choose optimal strategies with respect to violence against civilians, given their constraints. This is implausible in the context of armed conflict, particularly irregular armed conflict, for several reasons. Leaders frequently disagree about what level, targets, and types of violence are ‘optimal’, and seldom give explicit instructions about non-lethal violence. Combatants may or may not comply with leaders’ chosen strategies in any case. The framework developed here makes a more limited assumption: commanders avoid decisions that are known to endanger either the commander himself (herself) or group survival, but commanders’ abilities to enforce these decisions vary.

Variation in ability to enforce command decisions underlies a contrasting set of accounts, including those of Gates (2002), Weinstein (2007), and Mitchell (2004). In principal–agent (PA) models, commanders (principals) offer combatants (agents) a set of extrinsic, though potentially non-pecuniary, incentives to comply with commanders’ preferred behaviors. These contracts are complicated by considerable information asymmetries: commanders cannot observe combatants’ underlying preferences, and may not be able to observe combatant behaviors. For example, Gates (2002) theorizes that ‘distance’, whether ideological or informational, plays a crucial role in (lack of) compliance. However, PA accounts provide only an incomplete account. Most assume fixed preferences and rational decisionmaking for both principals and agents. While these assumptions are broadly appropriate for peacetime firms and bureaucracies – the contexts for which PA models were developed – they are less plausible when applied to wartime armed organizations, for at least two reasons.

First, most firms and bureaucracies do not subject agents to the levels of stress and fear that characterize the combat environment; stress and fear shorten agents’ time horizons and lower capacities for rational decisionmaking (e.g. Keinan, 1987; Kaufman, 1999; Lieberman et al., 2005). The PA framework accounts for these dynamics to some extent, by discounting principals’ offered (dis)incentives, relative to agents’ underlying preferences, in agent decision functions: more stress and fear implies decisionmaking that rests more strongly on underlying preferences. Very high levels of stress and fear imply decisions driven by agent preferences, to the exclusion of incentive systems (cf. Brehm & Gates, 1994). A second key difference between (most) firms and (most) armed groups also concerns combatant preferences: armed organizations (and other total institutions, cf. Goffman, 1961) alter combatants’ underlying preferences and predispositions. Below, I present evidence that these are true preference changes rather than (only) responses to changing incentives or information.

Because existing models typically do not account for predispositional changes, and because predispositions lie at the heart of decisionmaking under stress, we require a complementary account. The theory I frame below draws on both rationalist and PA accounts: commanders make limitedly rational decisions, but are subject to significant PA problems in enforcing those decisions. Importantly, it also draws on an important emerging literature that examines the effects of specific armed group institutional and organizational practices, including recruitment (Cohen, 2013), discipline (Wood, 2009), ideology (Thaler, 2012), and the management of factions (Cunningham, Bakke & Seymour, 2012).

Research on violence against civilians during armed conflict reveals a profusion of plausible theories, all supported by at least some empirical evidence. I do not attempt to adjudicate between these theories; rather, I argue that the search for one or a few causal mechanisms underlying violence against civilians is misguided. Some violence is caused by armed group strategy, some by material necessity, some by adverse combatant selection, some by individual opportunism, habituation, or panic, and some by a combination of factors – any or all of which may operate simultaneously, and at high levels, in the context of armed conflict. Given this, violence should not be a puzzling outcome – and what is necessary is not a better theory of violence. Rather, we require a theory that explains the occasional appearance of restraint.

Reframing the research question this way pushes us to ask questions differently: on observing a case of restraint, rather than asking which of the factors thought to cause violence is missing, we ask which additional factors are present. A few scholars have tackled the question from this angle. Mitchell (2004) argues that the restrained behavior of parliamentary soldiers during the English Civil War was due to commanders’ ‘commitment to toleration and self-restraint that extended to the control of the agents’ (2004: 148), not due to external restraints. However, Mitchell’s account presumes that commanders with commitments to restraint are able to enact their commitments. Below, I argue that this is not necessarily the case. Stanton’s (2009) argument, similarly, assumes the ability of commanders to implement strategic decisions. The present account pushes this research agenda forward by providing an empirically supported account of combatants’ preferences, and commanders’ attempts to shape them.

Understanding combatant predispositions to violence

To succeed (indeed, even to remain in the fight), military commanders must first create a fighting force that is capable of great, unhesitating violence. Doing so is both difficult and vitally important, and consequently military socialization – processes designed to strip away pre-military norms and identities, and to replace them with group norms and identities – has received considerable academic attention (e.g. Brotz & Wilson, 1946; Shils & Janowitz, 1948; Moskos, 1975; Faris, 1975; Cooper-Thomas & Anderson, 2002; Kelty, Kleykamp & Segal, 2010).

A fundamental goal of military training and socialization is to prepare the recruit to kill. Killing is foundational to military service as traditionally conceived, yet many new recruits are ill-prepared to kill other humans. Grossman (1996), for example, investigates resistance to killing among US troops in conflicts from the Civil War to Vietnam, relying particularly on the finding that only a small minority of World War II infantry troops fired their weapons in combat (cf. Marshall, 2000 [1947]). The accuracy of Marshall’s specific claim has been questioned (Chambers, 2003). Nevertheless, the US military recognized that resistance to killing could constitute a significant tactical disadvantage. Marshall’s work led to significant changes in soldier training (Grossman, 1996: 35). Of course, these changes were made in a major state military, rather than an ill-equipped, under-resourced insurgent group. Yet the vast majority of armed organizations, state and non-state, must prepare their recruits to kill.

These preparations are undertaken in many ways. Minimally, groups may teach recruits how to load and fire a service rifle or other firearm before an immediate combat deployment. More typically, recruits are trained intensively for a period of weeks or months before deployment. This formal training period typically introduces a set of facts and skills related to weapons, tactics, and the routines of military life. This is one aspect of ‘learning to kill’ (Grossman, 1996). It also includes an array of general behavioral expectations focused on immediate obedience to superiors, and exercises designed to desensitize recruits to the act of killing (Grossman, 1996: 35). The group exercises precise, yet seemingly arbitrary, control over nearly all aspects of recruits’ daily lives (cf. Goffman, 1961). Less formally, recruits are socialized into a culture that (typically) emphasizes honor, loyalty, toughness, and traditional masculinity (Arkin & Dobrofsky, 1978; Dunivin, 1994; Goldstein, 2003).

Thus, several institutions operate during the training period to increase combatant predispositions to violence: military skills training provides technical know-how; systems of tightly structured rules, rewards, and punishments promote obedience; training exercises habituate recruits to the physical acts of violence; and informal socialization processes valorize violence and masculinity (Dunivin, 1994). Clearly, not every group employs all (or only) these institutions. Some engage in significantly more vetting prior to recruitment, or more training afterward. Some employ significantly harsher, more punitive, or more humiliation-focused disciplinary systems. Some augment military training with other forms of education, including political education.

What are the effects of these processes on combatants’ broader predispositions to violence? Grossman (1996) argues that the elimination of natural hesitation and disgust at violence represents a permanent psychological loss to the combatant. I assume, similarly, that military training and combat experience durably increase combatant propensity to employ violence. This assumption does not imply that everyone who joins an armed group is incapable of moral reasoning or uncontrollably violent. It does mean that many who undergo military training are incrementally more predisposed to commit violence, a change that is considerably intensified by combat experiences, and that changes at the individual level aggregate to create groups of fighters in which the average predisposition to use violence exceeds those in the general population.

A range of empirical evidence supports this assumption. Heyman & Neidig (1999; and see Marshall, Panuzio & Taft, 2005), comparing representative samples from the US general population and peacetime active-duty military personnel, found that military service, regardless of combat experience, was associated with significantly increased rates of intimate partner violence, particularly severe intimate partner violence. These findings held after controlling for demographic and socio-economic differences between the military and non-military samples. Exploiting the plausibly exogenous variation of draft lotteries in the USA and Argentina, economists have estimated a significant effect of draft eligibility that cannot be accounted for by self-selection (Rohlfs, 2010; Lindo & Stoecker, 2014).

These predispositional changes are intensified by experiences of combat. Both experimental and observational studies have established that subjects who witness violence, or who experience trauma, stress, uncertainty, fear, or anger, exhibit diminished capacity for logical reasoning and increased aggression (e.g. Song, Singer & Anglin, 1998; Keinan, 1987; Beckham, Moore & Reynolds, 2000; Jakupcak & Tull, 2005).³ In studies of military populations, service members who had been deployed showed significantly increased risk for (self-reported) spousal abuse (Orcutt, King & King, 2003; Prigerson, Maciejewski & Rosenheck, 2002). Among US Vietnam veterans in a nationwide survey, deployment length was associated with increased inappropriate (illegal or excessive) violence during conflict and increased (self-reported) violent behavior in the post-deployment period (Hiley-Young et al., 1995), a result that also holds in the post-Vietnam era (McCarroll et al., 2000). Outside the USA, Manekin (2013) has found a strong association between length of deployment and atrocity participation among members of the Israeli Defense Forces. Combat also appears to predispose veterans to unemployment, homelessness, major depressive disorder, post-traumatic stress disorder, alcohol misuse, and increased risk-taking behaviors (Orcutt, King & King, 2003; Hoge et al., 2004; MacLean, 2010; Maguen et al., 2010; Suvak et al., 2002; Killgore et al., 2008; Börjesson, Österberg & Enander, 2015).

Group dynamics also play a role in changing predispositions to violence. In classic demonstrations of the ‘minimal group paradigm’ (e.g. Sherif & Sherif, 1953), people arbitrarily separated into groups quickly develop positive affect toward their own group and negative affect toward others, even in the absence of competition. Threat perception intensifies biases against the out-group (Hewstone, Rubin & Willis, 2002). Combatants’ social settings matter in part because violence, and the mechanisms by which combatants cope with violence, are social phenomena. For example, a significant body of research has implicated military culture, particularly ‘militarized masculinity’, in the perpetration of wartime sexual violence (e.g. Enloe, 1993; Baaz & Stern, 2009). In comparison with non-military persons of the same age, male enlisted soldiers endorse significantly more traditional, anti-feminine, toughness-oriented masculinities (Kurpius & Lucart, 2000). Thus, general increases in propensities for violence among fighters may be particularly heightened where sexual violence is concerned.

The durable, general increases in predisposition to violence described above are vitally important to military success. Violence, including violence against civilians, often confers military advantage (cf. Kalyvas, 2006; Valentino, 2013). Consequently we might expect that commanders generally prefer some level of violence against civilians. However, habituating recruits to violence does little to ensure that violence will be wielded only according to commanders’ preferences. Habituation to violence ‘spills over’ (Bradley, 2007), increasing combatant predispositions toward violence in general. Here I assume that, following training and particularly following combat deployment, average combatant predispositions to violence exceed commander preferences for violence, particularly for non-lethal violence and violence outside of active combat.

Given this state of affairs, we might expect to observe extremely high levels of violence, and extremely wide repertoires of violence, against civilians by all armed groups in all conflicts. We might particularly expect to see sexual violence strongly represented in the repertoire (Hoover Green, 2011). Yet not all armed groups (and not all units within armed groups) use high levels of violence against civilians. Not all use wide repertoires of violence. Not all use sexual violence. Why? In the following section I argue that restraint is possible when armed groups implement specific combinations of institutions for behavioral control.

Institutions for the creation and control of violence

Military effectiveness requires that commanders increase combatant propensities toward violence. It also requires that commanders maintain control over the timing, targeting, and types of violence their fighters deploy. At a bare minimum, commanders must ensure that combatants do not turn on their leaders. In this section, I consider specific institutions and practices through which commanders attempt to create and control violence.

PA models typically assume that conflicts of interest between principals and agents are resolved, to the extent possible, by a set of extrinsic incentives, which may be pecuniary or non-pecuniary. However, extrinsic incentive systems do not exhaust the set of institutions that may exert control over combatant behavior. Indeed, underlying preferences often outweigh extrinsic incentives when decisions are made under stress. Moreover, as I argue above, combatants’ preferences for violence following military training and combat experience – particularly unordered violence, such as violence outside of active combat and non-lethal forms of violence – exceed commanders’, on average.⁴

Therefore, I reason that, regardless of commanders’ intentions, extrinsic incentives alone will not produce restrained combatant behavior. In addition to rules, rewards, and punishment, restraint requires that commanders purposefully shape combatant preferences, such that restraint is intrinsically rewarding.⁵ While few studies of intrinsic incentives in combat exist, intrinsic incentives appear to produce restraint among police officers at least as effectively as extrinsic incentives (Tyler, Callahan & Frost, 2007; and see Benabou & Tirole, 2003). Table I lists five types of institutions that mediate between command preferences and combatant behavior, noting dimensions on which each institution varies, and mechanisms by which each institution influences combatant behavior.

Table I.

Key characteristics of armed group institutions

Institution	Variation across groups	Mechanism(s) of influence
Recruitment	Selectivity (random to highly selective) Voluntariness Population available for recruitment Offers of payment/reward	Shapes initial distribution of combatant preferences (cf. Weinstein, 2007) Shapes informal socialization processes (cf. Cohen, 2013, and see following row)
Socialization	Group culture and norms (with respect to gender/masculinity, violence, loyalty, etc.) Knowledge and influence of commanders over unordered practices	Information: group and small-group norms and practices. May shape combatant calculations about expected value of behavioral decisions Social pressure/role adoption Internalization of group and small-group norms (potentially)
Discipline	Formality Severity Willingness of commanders to enforce Ability of commanders to enforce	Information: group norms, practices, extrinsic incentives (rules, rewards, punishments, and likelihood of enforcement) may shape combatant calculations
Military training	Formality Length of pre-deployment training Timeline (single pre-deployment training period, regularly, continuously)	Information: norms, practices, extrinsic incentives may shape combatant calculations Skills acquisition, habituation to violence (cf. Grossman, 1996)
Political education	Formality Length of pre-deployment training Timeline Content (political ideology, focus on civilian protection/restraint, etc.)	Information about group norms, practices, extrinsic incentives may shape combatant calculations Information (or ‘information’) about origins of conflict Social pressure/role adoption Internalization of group norms (potentially)

Of the institutions shown in Table I, all might potentially contribute to both violence and restraint, and most employ a combination of extrinsic and intrinsic incentives to shape behavior. Political education (PE) stands out, however. PE diverges from other formal institutions (recruitment, military training, discipline) in both its content and its route(s) of influence. Like other institutions, PE may impart factual (or ‘factual’) information; however, it operates primarily via internalization of group norms (intrinsic incentives) rather than via learning or calculation (extrinsic incentives). If commanders cannot retain full control of combatant violence via extrinsic incentives alone, as argued above, we should expect some differences in behavior among groups that employ PE.

In what follows, I use ‘PE’ to mean formal instruction that explains specific social or political purposes of a particular conflict, and connects conflict purposes to specific behavioral norms. This definition requires some elaboration. First and most simply, PE is formal. Formal instruction is explicit, rather than implicit, and organized and disseminated from the highest levels of the organization. Second, PE is about the social or political purposes of a conflict. That is, its presentation of the conflict’s goals and purposes focuses on the organization of social or political life in the post-conflict era, rather than solely on military endpoints. Third, PE connects conflict purposes to behavioral norms. Rules about appropriate combatant behavior are presented in terms of the conflict’s political or social goals, rather than solely in terms of military success. Finally, both the social or political purposes of the conflict and the connection between behavioral rules and social or political purposes are specific. The specificity requirement distinguishes PE from broad statements of values, virtues or ideals.

Particularly given the secrecy that often surrounds armed groups’ internal practices, it may be difficult to distinguish empirically between PE and non-PE in the absence of detailed, qualitative data. Formal documents – political oaths, combatant codes of conduct, ideological statements by leaders, and even written political curricula – may or may not indicate PE as defined here. To qualify as PE, instruction must reach a broad majority of combatants. For example, Abdullah (1998) documents several formal ideological statements by leaders of the Revolutionary United Front (RUF, Sierra Leone) and the Mozambican National Resistance (RENAMO). However, he argues that leaders of these groups used these statements primarily as communication to external constituencies; there is little evidence that RUF or RENAMO cadre received political training. Even when most combatants receive information about political or social goals, it may not be sufficiently specific to meet the definition above.

PE is seldom intended to produce generalized restraint. (As noted above, military success demands that commanders produce a force capable of considerable violence.) Some PE valorizes violence but seeks to control its forms, targets, or timing. Some glorifies violence in general. PE in the Farabundo Martí National Liberation Front (FMLN, El Salvador) was of the former type: the curriculum defended the use of violence in general, but limited the appropriate forms and targets of violence (Hoover Green, 2011). Soviet armed forces during World War II exemplify the latter type of PE. Red Army recruits complained bitterly about long hours spent learning details of Stalinist orthodoxy, indicating a clear commitment to PE on the part of the state (Merridale, 2006: 55–56). On the drive toward Berlin, however, Red Army PE encouraged violent revenge upon both German soldiers and Germans more generally (Beevor, 2003: Ch. 3).

The sharp differences between these uses of PE highlight an important distinction, between PE in general and political education for restraint (PER). PER is the subset of PE in which formal instruction about conflict purposes connects those purposes to specific behavioral norms (as in the definition of PE), and the behavioral norm under discussion limits violence to specific contexts, forms, or targets. Like PE more generally, PER operates by bringing combatants’ predispositions into alignment with commanders’ preferences about violence. Unlike some PE, PER indicates commanders’ preferences for generally limited violence. Empirically, while PE groups in general should show lower levels of unordered violence, PER groups should show lower levels of overall violence. This implication is discussed further in the following section.

Neither PE nor PER is exclusive to groups that are ‘ideological’ in a traditional sense. Groups with formal political ideologies need not build PE institutions; groups that build PE institutions need not develop formal political ideologies (although many, or perhaps most, do). Groups may implement PE based in religious or even aesthetic belief systems, as well as PE that is more traditionally ideological. The Liberation Tigers of Tamil Eelam (LTTE), for example, mandated forms of self-denial such as sexual abstinence, which were presented as necessary to the political goal of national self-determination (Wood, 2009). As a practical matter, I identify as ‘PE groups’ those armed organizations in which documentary evidence suggests that political training was organized from the top down, and in which a majority of combatants reported receiving training, separate from military skills training and continuing in some form after deployment, that offered detailed lessons about history, politics, religion or other theories of social organization. Of course, even these concrete data are difficult to obtain without significant qualitative investigation.

The Commander’s Dilemma framework primarily concerns the effects of PE institutions, and consequently a complete discussion of the origins of PE institutions is outside the scope of this article. In brief, however: PE institutions may emerge for many reasons, including instrumental rationality, imitation, or ideology. However, groups may be more likely to develop PE when such institutions are ‘baked in’ to a given ideological or strategic tradition. Gutiérrez Sanín & Wood (2014), for example, argue that communist groups are more likely to use ideological training, including PE as defined above, because revolutionary communist traditions emphasize both political consciousness among combatants and the importance of civilian support. State militaries may be less likely to develop PE, as defined here, than rebel organizations, since their systems for training and indoctrination are not formed and re-formed with each new conflict. While formal instruction – including some instruction about the goals or purposes of conflict – is common among state groups, I speculate that specific instruction about the goals of an individual conflict may be relatively rare among state troops.

Evidence

If the overarching reasoning of the Commander’s Dilemma is correct, the main empirical implication of the framework is an association between institutions for preference change, particularly PE(R), and variation in violence. In this section, I present some preliminary evidence for such an association. In testing the association between PE and patterns of violence against civilians, distinguishing between level and repertoire of violence is vital. Rational commanders, including those who implement PE, may (and frequently do) order civilian killings. Thus, it may be difficult to distinguish observationally between ordered and unordered killings. By contrast, explicit orders to commit ‘frequently opportunistic’ forms of violence (e.g. rape, looting) are relatively rare. These types of violence are therefore more likely to appear as unordered ‘bottom-across’ practices (cf. Wood, 2009). If PE groups commit lower levels of unordered violence, and sexual violence (for example) is relatively unlikely to be ordered, then we should expect PE groups – including, but not limited to, PER groups – to perpetrate rape less frequently than non-PE groups (Hoover Green, 2011). I therefore hypothesize that, all else equal,

Hypothesis 1: PE groups – including, but not limited to, PER groups – commit lower levels of sexual violence than non-PE groups.

While PE should be associated with narrower repertoires of violence, and lower levels of unordered violence, PE may not be associated with lower overall levels of violence. In particular, given that ordered killing is relatively common, the Commander’s Dilemma implies that the level of civilian killings in a PE group depends upon whether the group implemented PER (as opposed to PE more generally). That is, all else equal,

Hypothesis 2: PER groups (but not PE groups more generally) commit lower levels of civilian killings than non-PE groups.

Unfortunately, while it is plausible to hypothesize that PER groups make up a majority of PE groups – and, therefore, that PE groups more generally should also commit lower levels of civilian killings than non-PE groups – the true relative frequency of PE and PER groups remains unknown (Hoover Green, 2014). Thus, consideration of H2 emerges primarily in the qualitative, rather than the quantitative, analysis below. (A limited quantitative test is discussed in the Online appendix.)

Quantitative evidence

A comprehensive quantitative test of the Commander’s Dilemma framework, or of H1, is outside the scope of this article. However, as an initial test, I examine cross-national variation in reported rape by 75 rebel forces between 1989 and 2004. Data for this analysis are derived from Cohen (2013), Kalyvas & Balcells (2010), the Uppsala Conflict Data Programme’s (UCDP’s) One-Sided Violence Dataset (Eck & Hultman, 2007) (OSV), and UCDP’s Armed Conflict Dataset (Themnér & Wallensteen, 2015). Systematic cross-national data on PE are not available, and consequently I use communist ideology as a rough proxy for PE. Further, because data on state militaries’ ideological orientations are not included in Kalyvas & Balcells (2010), I examine only sexual violence by rebel organizations. Thus, the specific empirical implication of H1 under consideration is

Hypothesis 1a: All else equal, communist rebel organizations commit lower levels of rape than non-communist rebels.

While communism is undoubtedly a rough proxy for PE, it is a plausible one. Communist rebel organizations typically develop military doctrines in accordance with one or more key thinkers (e.g. Lenin, Mao, or Guevara), whose specific ideological frameworks differ considerably but all of whose military strategies specify that political organization (or ‘conscientization’) of cadre and civilians is vital to military success. Each prominent communist military strategist identifies specific institutional blueprints that meet the definition of PE laid out above, particularly the practice of criticism and self-criticism (Gutiérrez Sanín & Wood, 2014). To check the plausibility of this proxy, I examined information on rebel institutions from ten conflicts included in the dataset – five randomly selected from conflicts with communist primary rebel groups, and five selected randomly from conflicts with non-communist rebels, as coded by Kalyvas & Balcells (2010). In this small sample, communist rebels were much more likely to implement PE than were non-communist rebels.⁶

I test the hypothesis that communist rebels (standing in for rebels with PE institutions) commit lower average levels of sexual violence against several other theoretical frameworks. Accounts suggesting that violence against civilians is strategically motivated (e.g. Kalyvas, 2006) imply that levels of reported rape by rebels (like all violence) should vary with strategic factors such as conflict intensity, technology of warfare, or opposition tactics. I test these theories by including variables that measure lethal one-sided violence by rebels, logged battle-deaths, irregular war, and government violence against civilians. Other theories of wartime rape, reviewed in Cohen (2013), suggest that rape is associated with ethnic hatred, low status of women in a society, or simple opportunism. I test these theories by including variables for ethnic war, fertility rate, and state failure, respectively. I also test against several institutional theories. In particular, I include measures of rebel forced recruitment (cf. Cohen, 2013) and reliance on contraband (as an indicator of adverse selection, cf. Weinstein, 2007). I also include several control variables. Results of two Tobit regressions are shown in Table II.⁷

Table II.

Tobit regression results (DV = average reported level of rape by rebel forces)

	(1)	(2)
Communist rebels dummy	–0.582*	–0.732**
	(0.284)	(0.264)
Irregular war dummy	–0.472^†	–0.469^†
	(0.260)	(0.260)
Log battle deaths		–0.044
		(0.051)
Fertility rate	–0.199*	–0.278**
	(0.082)	(0.088)
Ethnic war dummy	–0.233	–0.194
	(0.291)	(0.292)
State failure index	0.155*	0.123
	(0.079)	(0.078)
Rebel forced recruitment dummy	0.547^†	0.781*
	(0.299)	(0.304)
Rebel contraband dummy	0.277	0.354
	(0.218)	(0.218)
Average government rape	0.433*	0.250
	(0.177)	(0.199)
Deaths due to rebel OSV		3x10^–5†
		(2x10^–5)
Deaths due to government OSV		–1x10^–4
		(1x10^–4)
Conflict years	0.048*	0.055**
	(0.020)	(0.020)
Log GDP	–0.633**	–0.738**
	(0.169)	(0.185)
Log population	–0.042	–0.085
	(0.085)	(0.082)
Observations	75	58
Log likelihood	–62.726	–43.173
Wald test	66.105** (df = 11)	64.045** (df = 14)

^† p < 0.1, *p < 0.05, **p < 0.01.

As hypothesized, the association between communist rebel ideology and reported rape by rebels is negative and significant in both Model 1 (which includes significantly more observations at the cost of excluding variables on battle-deaths and one-sided violence) and Model 2 (the full model, with fewer complete observations). Substantively, after controlling for other factors, communist rebels commit more than half a ‘level’ less sexual violence than other rebel groups.⁸ This finding accords with H1a, and with the overarching theory that political education is associated with lower levels of unordered violence against civilians.

Support for other theoretical perspectives is mixed. Overall conflict intensity, as measured by battle deaths, is not significantly associated with reported rape, while a dummy variable indicating irregular war is negatively and somewhat significantly associated with reported rape in this sample. One-sided lethal violence against civilians by rebel and government forces is only weakly associated with rebel sexual violence. Among institutional variables, Cohen’s (2013) forced recruitment measure has the expected positive and significant association with rape, while contraband funding of rebels is not significant in any model. Further details of this analysis can be found in the Online appendix.

Qualitative evidence

Quantitative evidence shows that communist rebels commit lower (reported) levels of rape than other groups. However, a number of possible causal mechanisms might explain this association. Communist groups may have stronger strategic incentives than other rebels to avoid sexual violence (and, perhaps, violence against civilians more generally). Communist groups may have drawn from a more ideologically committed set of recruits. More generally, some other mechanism (including other institution types) may have accounted for the differences between communist and non-communist groups observed above. In order to assess these questions, I turn to qualitative evidence about the Frente Farabundo Martí para la Liberación Nacional (FMLN, El Salvador). Evidence about the FMLN also illustrates effectively the importance of PER as a specific subtype of PE.

A number of scholars have commented upon the low levels and narrow repertoires of violence committed by the FMLN, particularly in contrast to the high levels and extremely broad repertoires of violence committed by state forces (e.g. Stanley, 1996; Wood, 2003; Viterna, 2013). Less well documented is the extent to which levels and repertoires of violence, as well as ideologies and institutions, varied within the FMLN. Here, I focus on intra-FMLN variation to consider the effects of PE institutions on violence. The very brief account of institutional variation below is drawn from both key secondary sources and original data; the original data include approximately 100 semi-structured interviews, and approximately 360 brief structured interviews, conducted with Salvadoran ex-combatants during 2008–09 (see descriptions in Hoover Green, 2011: Ch. 4 and Appendix A) and 2015 (see description in Hoover Green, 2015).

I focus in particular on differences between the two largest FMLN subgroups, the Leninist/Guevarist ERP (Ejército Revolucionario del Pueblo, Revolutionary Army of the People) and the maoist-oriented (though not explicitly maoist) FPL (Fuerzas Populares de Liberación, Popular Army of Liberation). The ERP and FPL shared a number of characteristics. Both were communist organizations that controlled a predominantly rural, poor, mountainous border region and relied upon civilians for food and materiel (Montgomery, 1995). The groups faced relatively similar Salvadoran state personnel, including large but poorly trained and equipped regular army barracks in the regions’ main towns, local National Guard garrisons, and occasional incursions by US-trained immediate reaction battalions (Binford, 1998, on the ERP; Harnecker, 1993, on the FPL). As the conflict began in 1980, both the ERP and the FPL faced significant resource challenges, and both factions adopted a roughly conventional style of warfare. However, by approximately 1985 FMLN leaders implemented a more traditional guerrilla-war strategy, with small units operating in relative isolation. Given these similarities, a strategic account might predict relatively similar patterns of ERP and FPL violence, including violence against civilians.

On the other hand, ERP and FPL ideologies suggested divergent approaches to violence, particularly during the early years of the conflict. The FPL adopted a strategy of prolonged popular war, in which political organizing among civilians takes precedence over military operations (Pearce, 1986). In addition, FPL leaders implemented a broad range of controls over its members’ consumption, sexuality, and social lives (Hoover Green, 2011: Ch. 6). ERP leaders, by contrast, were more influenced by foquismo, in which political organizing served military goals, and in which military victory would ultimately pave the way for political incorporation of civilians. The ERP also was less concerned (or, less overtly concerned) with individual moral uprightness. In line with their more militarist ideology, ERP leaders repeatedly demonstrated their willingness to order significant campaigns of violence against civilian targets, including threats of murder against families allied with the government, and the murders of mayors in the ERP zone of control (Betancur, Figueredo Planchart & Buergenthal, 1993: 139–144). Given these ideologically driven differences in strategy, we might expect higher levels of violence against civilians by ERP cadre than by FPL cadre. In addition, unlike the FPL, the ERP engaged in forced recruitment (though not recruitment via abduction) in its zone of control, particularly during 1983–84. Cohen’s (2013) results suggest that this practice should be associated with elevated levels of sexual violence.

However, while the ERP’s approach to civilian populations was generally militaristic and instrumental, it implemented this approach in a more formalized way than did the FPL (Hoover Green, 2011: Ch. 6). Both groups implemented PER (recall that this is political education for restraint, to be distinguished from PE more generally), but ERP leaders implemented a standardized, written program earlier than FPL, and in a more routinized fashion. Whereas FPL ex-combatants from the first phase of the war recalled ongoing political instruction, both meetings for instruction and the content of instruction varied somewhat across time and subgroups. In semi-structured interviews with ERP ex-combatants, most recalled receiving formal political instruction and participating in nightly criticism and self-criticism exercises (Hoover Green, 2015). Most of these also recalled specific books and readings. The greater institutionalization of PER within the ERP, particularly in the earlier years of the conflict, implies narrower repertoires (cf. H1) and lower overall levels of violence (cf. H2) by ERP than FPL cadre.

Evidence about patterns of FMLN violence is spotty, at least in part because FMLN violence was quite rare relative to state violence. Quantitative data (see Hoover Green 2011: Ch. 7–8) include records of violence against civilians from the United Nations Truth Commission for El Salvador (1993), the non-governmental Salvadoran Commission for Human Rights (CDHES), and a US NGO, El Rescate, which coded records from the Legal Aid office of the Archbishopric of San Salvador. Together, these records include 60,246 total episodes of violence. Of these, 2,546 (5%) were attributed to the FMLN. While most FMLN violence was not attributed to a single subgroup, it is possible to compare reported FMLN violence in the FPL’s zone of control to reported FMLN violence in the ERP’s zone of control. In brief, these data suggest that the ERP committed lower levels of both lethal and non-lethal violence than did the FPL, in accord with the prediction of the Commander’s Dilemma framework.⁹ Moreover, in both structured and semi-structured interviews, FPL ex-combatants were more likely to report having witnessed or committed violence against civilians than were ERP combatants. Mid-level FPL commanders frequently acknowledged that they had difficulty controlling opportunistic violence during the earliest years of the conflict. ERP commanders reported greater confidence in their ability to control unordered violence, and many attributed this to conscientization of their troops. No available evidence suggests that there was an increase in ERP sexual violence after it began forced recruitment, implying that PER may mitigate the effects of forced recruitment.

Discussion

The Commander’s Dilemma explains variations in repertoires and levels of violence without making empirically indefensible assumptions (e.g. unitary armed groups, unchanging combatant preferences). In doing so it focuses analytical attention on armed group institutions, particularly those that shape combatant preferences.

However, the theory has a number of limitations. It is bounded to groups whose commanders prefer (boundedly) rational, strategic deployment of violence, leaving out groups in which leaders are genocidal or suicidal. It simplifies the relationships between ‘commanders’ and combatants, eliminating the mediating role of mid-level officers and neglecting groups that are not organized around traditional, hierarchical military structures (e.g. networked insurgencies). Perhaps most notably, the Commander’s Dilemma is not a theory of institutional development. This inattention to institutional development suggests some avenues for future research. For example, we might hypothesize that state militaries, given their longer institutional histories, are less likely than non-state groups to develop PE as it is defined here. Or, we might expect that armed groups founded in a religious tradition implement PE more frequently than non-religious groups. Further research on the Commander’s Dilemma framework might also consider its implications for combatant behaviors beyond violence against civilians. These behaviors include violence against enemy combatants and prisoners of war, and intra-armed group violence, such as ‘fragging’ or ‘blue on blue’ sexual violence.

Absent broader, more systematic evidence about armed groups’ resources, incentives, and institutions, it remains extremely difficult to perform rigorous cross-national tests of the Commander’s Dilemma framework. In particular, while the quantitative evidence contains a weak proxy for principal–agent dynamics, there is no sufficiently rigorous way to distinguish between PA models and the Commander’s Dilemma framework here. Moreover, it remains difficult or impossible to distinguish PE from PER in a cross-national setting. Thus, future efforts to examine hypotheses related to the Commander’s Dilemma must proceed at multiple levels. Qualitative, within-group accounts, though difficult to generalize, may have the best chance of distinguishing between competing causal accounts in a given case. Cross-national analyses cannot necessarily provide strong evidence for a causal mechanism, but can show that a given relationship (for example, between PE/R and restraint) holds across many cases. Hoover Green (2014) describes a variety of cross-national data collection efforts aimed at testing institutional theories, noting that – so far – most such data collection efforts focus on only one or a few institutional variables. Finally, the Commander’s Dilemma carries implications for groups, but operates fundamentally on individual predispositions. Verifying this mechanism will require research that examines not only group and subgroup-level variation, but also individual variation.

Regardless of difficulties with empirical testing, the Commander’s Dilemma makes significant analytical advances. Two are most relevant. First, it sidesteps debates about the causes of violence, viewing violence as a default setting, and instead asks what factor or factors underlie the (empirically rarer) circumstance of restraint. As noted above, this is not the same as viewing combatants as opportunists. Rather, it reflects the multiple pathways – only some of which are related to conflict goals – by which violence against civilians can occur. Second, and relatedly, in considering the origins of restraint, the Commander’s Dilemma shows how repertoires, as well as levels of violence, can be analytically significant. Thus, in addition to arguing for a particular causal framework and a given set of empirical tests, I argue for a shift in the standard political science approach to political violence, which has viewed violence against civilians as puzzling and non-lethal violence as irrelevant.

This approach also carries important policy implications. First and foremost, it suggests that some armed groups may be less able to control violence against civilians, increasing the need for human rights monitoring in their zones of operation. Second, this framework suggests specific reforms to military leaders who are interested in improving command and control, improving civilian welfare, or both: combatants need to know why they are fighting. Third and finally, if empirical evidence suggesting that violence is the ‘default setting’ continues to accrue, this provides a useful factual basis for doctrines of command responsibility in international criminal law. In this view, arguments blaming ‘bad apples’ for illegal wartime violence (and/or requiring direct evidence of illegal orders) is invalid; top-level commanders bear responsibility when they fail to take steps to restrain violence.

Footnotes

Replication data

Datasets, codebooks, R code for data merges and data analysis, and all other replication materials are available at .

Acknowledgements

For helpful discussions of this manuscript and its many revisions, I thank Elisabeth Jean Wood, Dara Kay Cohen, Michele Leiby, Patrick Ball, Sarah Parkinson, Scott Gates, Jeffrey Checkel, Alexander Nadolishny, Greta Jusyte, colleagues at the Human Rights Data Analysis Group, and three anonymous reviewers. For research assistance, I thank Erika Murcia. For research support I thank Yale’s MacMillan Center for International Affairs, the United States Institute of Peace, the Institute for Social and Policy Studies at Yale, and the Drexel University Funded Research Co-op Program.

Notes

References

Abdullah

Ibrahim

(1998) Bush path to destruction: The origin and character of the Revolutionary United Front (RUF/SL). Journal of Modern African Studies 36(2): 203–235.

Arkin

William

Dobrofsky

Lynne R

(1978) Military socialization and masculinity. Journal of Social Issues 34(1): 151–168.

Baaz

Maria E

Stern

Maria

(2009) Why do soldiers rape? Masculinity, violence, and sexuality in the armed forces in the Congo (DRC). International Studies Quarterly 53(2): 495–518.

Balcells

Laia

(2011) Continuation of politics by two means: Direct and indirect violence in civil war. Journal of Conflict Resolution 55(3): 397–422.

Ball

Patrick

Asher

Jana

Sulmont

David

Manrique

Daniel

(2003) How Many Peruvians Have Died? An Estimate of the Total Number of Victims Killed or Disappeared in the Internal Armed Conflict Between 1980 and 2000. Washington, DC: American Association for the Advancement of Science.

Beckham

Jean C

Moore

Scott D

Reynolds

Victoria

(2000) Interpersonal hostility and violence in Vietnam combat veterans with chronic posttraumatic stress disorder: A review of theoretical models and empirical evidence. Aggression and Violent Behavior 5(5): 451–466.

Beevor

Antony

(2003) The Fall of Berlin 1945. New York: Penguin.

Benabou

Roland

Tirole

Jean

(2003) Intrinsic and extrinsic motivation. Review of Economic Studies 70(3): 489–520.

Betancur

Belisario

Planchart

Reinaldo Figueredo

Buergenthal

Thomas

(1993) From madness to hope: The 12-year war in El Salvador: Report of the United Nations Commission on the Truth for El Salvador (http://www.latinamericanstudies.org/elsalvador/ElSalvador-Report.pdf).

10.

Bilton

Michael

Sim

Kevin

(1992) Four Hours in My Lai. New York: Penguin.

11.

Binford

Leigh

(1998) Hegemony in the interior of the Salvadoran revolution: The ERP in northern Morazan. Journal of Latin American Anthropology 4(1): 2–45.

12.

Blattman

Christopher

(2009) From violence to voting: War and political participation in Uganda. American Political Science Review 103(2): 231–247.

13.

Börjesson

Marcus

Österberg

Johan

Enander

Ann

(2015) Risk propensity within the military: A study of Swedish officers and soldiers. Journal of Risk Research 18(1): 55–68.

14.

Bradley

Christopher

(2007) Veteran status and marital aggression: Does military service make a difference? Journal of Family Violence 22(4): 197–209.

15.

Brehm

John

Gates

Scott

(1994) When supervision fails to induce compliance. Journal of Theoretical Politics 6(3): 323–343.

16.

Brotz

Howard

Wilson

Everett

(1946) Characteristics of military society. American Journal of Sociology 51(5): 371–375.

17.

Chambers

John W

(2003) SLA Marshall’s Men Against Fire: New evidence regarding fire ratios. Parameters 33(3): 113–121.

18.

Cohen

Dara K

(2013) Explaining rape during civil war: Cross-national evidence (1980–2009). American Political Science Review 107(3): 461–477.

19.

Conibere

Richard

Asher

Jana

Cibelli

Kristen

Dudukovich

Jana

Kaplan

Rafe

Ball

Patrick

(2004) Statistical appendix to the report of the Truth and Reconciliation Commission of Sierra Leone (http://www.sierraleonetrc.org/index.php/view-the-final-report/download-table-of-contents/appendices/item/appendix-1-submissions-to-the-sierra-leone-truth-reconciliation-commission-2).

20.

Cooper-Thomas

Helena

Anderson

Neil

(2002) Newcomer adjustment: The relationship between organizational socialization tactics, information acquisition and attitudes. Journal of Occupational and Organizational Psychology 75(4): 423–437.

21.

Crombach

Anselm

Weierstall

Roland

Hecker

Tobias

Schalinski

Inga

Elbert

Thomas

(2013) Social status and the desire to resort to violence: Using the example of Uganda’s former child soldiers. Journal of Aggression, Maltreatment & Trauma 22(5): 559–575.

22.

Cunningham

Kathleen G

Bakke

Kristin M

Seymour

Lee J

(2012) Shirts today, skins tomorrow: Dual contests and the effects of fragmentation in self-determination disputes. Journal of Conflict Resolution 56(1): 67–93.

23.

Dunivin

Karen O

(1994) Military culture: Change and continuity. Armed Forces & Society 20(4): 531–547.

24.

Eck

Kristine

Hultman

Lisa

(2007) One-sided violence against civilians in war: Insights from new fatality data. Journal of Peace Research 44(2): 233–246.

25.

Enloe

Cynthia

(1993) The Morning After: Sexual Politics at the End of the Cold War. Berkeley, CA: University of California Press.

26.

Faris

John H

(1975) The impact of basic combat training: The role of the drill sergeant in the all-volunteer army. Armed Forces & Society 2(1): 115–127.

27.

Fjelde

Hanne

Hultman

Lisa

(2013) Weakening the enemy: A disaggregated study of violence against civilians in Africa. Journal of Conflict Resolution 58(7): 1230–1257.

28.

Gates

Scott

(2002) Recruitment and allegiance: The microfoundations of rebellion. Journal of Conflict Resolution 46(1): 111–130.

29.

Goffman

Erving

(1961) Asylums. New York: Doubleday.

30.

Goldstein

Joshua S

(2003) War and Gender: How Gender Shapes the War System and Vice Versa. New York: Cambridge University Press.

31.

Grossman

David

(1996) On Killing: The Psychological Cost of Learning to Kill in War and Society. New York: Back Bay.

32.

Gutiérrez Sanín

Francisco

Wood

Elisabeth J

(2014) Ideology in civil war: Instrumental adoption and beyond. Journal of Peace Research 51(2): 213–226.

33.

Harnecker

Marta

(1993) Con la Mirada en Alto: Historia de las FPL Farabundo Martí a través de sus dirigentes [With a look to the heavens: The history of the FPL according to its leaders]. San Salvador: UCA.

34.

Hewstone

Miles

Rubin

Mark

Willis

Hazel

(2002) Intergroup bias. Annual Review of Psychology 53(1): 575–604.

35.

Heyman

Richard E

Neidig

Peter H

(1999) A comparison of spousal aggression prevalence rates in US army and civilian representative samples. Journal of Consulting and Clinical Psychology 67(2): 239–242.

36.

Hiley-Young

Bruce

Blake

Dudley D

Abueg

Francis R

Rozynko

Vitali

Fusman

Fred D

(1995) Warzone violence in Vietnam: An examination of premilitary, military and postmilitary factors in PTSD in-patients. Journal of Traumatic Stress 8(1): 125–141.

37.

Hoge

Charles W

Castro

Carl A

Messer

Stephen C

McGurk

Dennis

Cotting

Dave I

Koffman

Robert L

(2004) Combat duty in Iraq and Afghanistan, mental health problems, and barriers to care. New England Journal of Medicine 351(1): 13–22.

38.

Hoover Green, Amelia (2011) Repertoires of violence against noncombatants: The role of armed group institutions and ideologies. Unpublished PhD thesis. Yale University.

39.

Hoover Green, Amelia (2014) Armed group institutions and conflict-related sexual violence in cross-national perspective. Paper presented at the Workshop ‘Sexual Violence and Armed Conflict: New Research Frontiers’, Harvard University, Cambridge, MA, 2–3 September.

40.

Hoover Green, Amelia (2015) Armed group institutions, combatant socialization, and violence against civilians: Evidence from El Salvador (working paper).

41.

Jakupcak

Matthew

Tull

Matthew T

(2005) Effects of trauma exposure on anger, aggression, and violence in a nonclinical sample of men. Violence and Victims 20(5): 589–598.

42.

Kalyvas

Stathis N

(1999) Wanton and senseless? The logic of massacres in Algeria. Rationality and Society 11(3): 243–285.

43.

Kalyvas

Stathis N

(2006) The Logic of Violence in Civil War. New York: Cambridge University Press.

44.

Kalyvas

Stathis N

Balcells

Laia

(2010) International system and technologies of rebellion: How the end of the Cold War shaped internal conflict. American Political Science Review 104(3): 415–429.

45.

Kaufman

Bruce E

(1999) Emotional arousal as a source of bounded rationality. Journal of Economic Behavior & Organization 38(2): 135–144.

46.

Keinan

Giora

(1987) Decision making under stress: Scanning of alternatives under controllable and uncontrollable threats. Journal of Personality and Social Psychology 52(3): 639–644.

47.

Kelty

Ryan

Kleykamp

Meredith

Segal

David R

(2010) The military and the transition to adulthood. Future of Children 20(1): 181–207.

48.

Killgore

William DS

Cotting

Dave I

Thomas

Jeffrey L

Cox

Anthony L

McGurk

Dennis

Alexander H

Castro

Carl A

Hoge

Charles W

(2008) Post-combat invincibility: Violent combat experiences are associated with increased risk-taking propensity following deployment. Journal of Psychiatric Research 42(13): 1112–1121.

49.

Kurpius

Sharon

Lucart

A Leigh

(2000) Military and civilian undergraduates: Attitudes toward women, masculinity, and authoritarianism. Sex Roles 43(34): 255–265.

50.

Lieberman

Harris R

Bathalon

Gaston P

Falco

Christina M

Morgan

Charles A

Niro

Philip J

Tharion

William J

(2005) The fog of war: Decrements in cognitive performance and mood associated with combatlike stress. Aviation, Space, and Environmental Medicine 76(Supplement 1): C7–C14.

51.

Lindo

Jason M

Stoecker

Charles

(2014) Drawn into violence: Evidence on ‘what makes a criminal’ from the Vietnam draft lotteries. Economic Inquiry 52(1): 239–258.

52.

MacLean

Alair

(2010) The things they carry: Combat, disability and unemployment among US men. American Sociological Review 75(4): 563–585.

53.

Maguen

Shira

Lucenko

Barbara A

Reger

Mark A

Gahm

Gregory A

Litz

Brett T

Seal

Karen H

Knight

Sara J

Marmar

Charles R

(2010) The impact of reported direct and indirect killing on mental health symptoms in Iraq war veterans. Journal of Traumatic Stress 23(1): 86–90.

54.

Manekin

Devorah

(2013) Violence against civilians in the Second Intifada: The moderating effect of armed group structure on opportunistic violence. Comparative Political Studies 46(10): 1273–1300.

55.

Marshall

Amy D

Panuzio

Jillian

Taft

Casey T

(2005) Intimate partner violence among military veterans and active duty servicemen. Clinical Psychology Review 25(7): 862–876.

56.

Marshall

Samuel LA

(2000 [1947]) Men Against Fire: The Problem of Battle Command. Norman, OK: University of Oklahoma Press.

57.

McCarroll

James E

Ursano

Robert J

Liu

Xian

Thayer

Laurie E

Newby

John H

Norwood

Ann E

Fullerton

Carol S

(2000) Deployment and the probability of spousal aggression by US Army soldiers. Military Medicine 165(1): 41–44.

58.

Merridale

Catherine

(2006) Ivan’s War: Life and Death in the Red Army, 1939–1945. New York: Picador.

59.

Mitchell

Neil J

(2004) Agents of Atrocity: Leaders, Followers, and the Violation of Human Rights in Civil War. New York: Palgrave Macmillan.

60.

Montgomery

Tommie Sue

(1995) Revolution in El Salvador: From Civil Strife to Civil Peace. Boulder, CO: Westview.

61.

Moskos

Charles C

(1975) The American combat soldier in Vietnam. Journal of Social Issues 31(4): 25–37.

62.

Orcutt

Holly K

King

Lynda A

King

Daniel W

(2003) Male-perpetrated violence among Vietnam veteran couples: Relationships with veteran’s early life characteristics, trauma history, and PTSD symptomatology. Journal of Traumatic Stress 16(4): 381–390.

63.

Ottmann

Martin

(2015) Rebel constituencies and rebel violence against civilians in civil conflicts. Conflict Management and Peace Science DOI: 0738894215570428.

64.

Pearce

Jenny

(1986) Promised Land: Peasant Rebellion in Chalatenango, El Salvador. London: Latin America Bureau.

65.

Prigerson

Holly G

Maciejewski

Paul K

Rosenheck

Robert A

(2002) Population attributable fractions of psychiatric disorders and behavioral outcomes associated with combat exposure among US men. American Journal of Public Health 92(1): 59–63.

66.

Rohlfs

Chris

(2010) Does combat exposure make you a more violent or criminal person? Evidence from the Vietnam draft. Journal of Human Resources 45(2): 271–300.

67.

Sallah

Michael

Weiss

Mitch

(2006) Tiger Force: A True Story of Men and War. New York: Back Bay.

68.

Sherif

Muzafer

Sherif

Carolyn W

(1953) Groups in Harmony and Tension: An Integration of Studies of Intergroup Relations. London: Harper.

69.

Shils

Edward A

Janowitz

Morris

(1948) Cohesion and disintegration in the Wehrmacht in World War II. Public Opinion Quarterly 12(2): 280–315.

70.

Song

Singer

Mark I

Anglin

Trina M

(1998) Violence exposure and emotional trauma as contributors to adolescents’ violent behaviors. Archives of Pediatrics & Adolescent Medicine 152(6): 531–536.

71.

Stanley

William

(1996) The Protection Racket State: Elite Politics, Military Extortion, and Civil War in El Salvador. Philadelphia, PA: Temple University Press.

72.

Stanton

Jessica

(2009) Strategies of violence and restraint in civil war. Unpublished PhD thesis. Columbia University.

73.

Staub

Ervin

Vollhardt

Johanna

(2008) Altruism born of suffering: The roots of caring and helping after victimization and other trauma. American Journal of Orthopsychiatry 78(3): 267–280.

74.

Suvak

Michael K

Vogt

Dawne S

Savarese

Vincent W

King

Lynda A

King

Daniel W

(2002) Relationship of war-zone coping strategies to long-term general life adjustment among Vietnam veterans: Combat exposure as a moderator variable. Personality and Social Psychology Bulletin 28(7): 974–985.

75.

Thaler

Kai M

(2012) Ideology and violence in civil wars: Theory and evidence from Mozambique and Angola. Civil Wars 14(4): 546–567.

76.

Themnér

Lotta

Wallensteen

Peter

(2015) Armed Conflict, 1946–2010. Journal of Peace Research 52(4): 536–550.

77.

Tilly

Charles

(1993) Contentious repertoires in Great Britain, 1758–1834. Social Science History 17(2): 253–280.

78.

Tyler

Tom R

Callahan

Patrick E

Frost

Jeffrey

(2007) Armed, and dangerous (?): Motivating rule adherence among agents of social control. Law & Society Review 41(2): 457–492.

79.

Valentino

Benjamin A

(2013) Final Solutions: Mass Killing and Genocide in the 20th Century. Ithaca, NY: Cornell University Press.

80.

Viterna

Jocelyn

(2013) Women in War: The Micro-Processes of Mobilization in El Salvador. New York: Oxford University Press.

81.

Weierstall

Roland

Elbert

Thomas

(2011) The appetitive aggression scale: Development of an instrument for the assessment of humans’ attraction to violence. European Journal of Psychotraumatology 2: 8430.

82.

Weinstein

Jeremy

(2007) Inside Rebellion: The Politics of Insurgent Violence. New York: Cambridge University Press.

83.

Wood

Elisabeth J

(2003) Insurgent Collective Action and Civil War in El Salvador. New York: Cambridge University Press.

84.

Wood

Elisabeth J

(2009) Armed groups and sexual violence: When is wartime rape rare? Politics & Society 37(1): 131–161.

85.

Wood

Reed M

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